Nanomaterials: Mechanics and mechanisms [1 ed.] 0387097821, 9780387097824

Citation preview

Nanomaterials Mechanics and Mechanisms

K.T. Ramesh

Nanomaterials Mechanics and Mechanisms

123

K.T. Ramesh The Johns Hopkins University Baltimore, MD 21218 USA [email protected]

ISBN 978-0-387-09782-4 e-ISBN 978-0-387-09783-1 DOI 10.1007/978-0-387-09783-1 Springer Dordrecht Heidelberg London New York Library of Congress Control Number: 2008942797 c Springer Science+Business Media, LLC 2009  All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

To Arjun, Rohan and Priti, for your love and patience.

Preface This book grew out of my desire to understand the mechanics of nanomaterials, and to be able to rationalize in my own mind the variety of topics on which the people around me were doing research at the time. The field of nanomaterials has been growing rapidly since the early 1990s. Initially, the field was populated mostly by researchers working in the fields of synthesis and processing. These scientists were able to make new materials much faster than the rest of us could develop ways of looking at them (or understanding them). However, a confluence of interests and capabilities in the 1990s led to the explosive growth of papers in the characterization and modeling parts of the field. That confluence came from three primary directions: the rapid growth in our ability to make nanomaterials, a relatively newfound ability to characterize the nanomaterials at the appropriate length and time scales, and the rapid growth in our ability to model nanomaterials at atomistic and molecular scales. Simultaneously, the commercial potential of nanotechnology has become apparent to most high-technology industries, as well as to some industries that are traditionally not viewed as high-technology (such as textiles). Much of the rapid growth came through the inventions of physicists and chemists who were able to develop nanotechnology products (nanomaterials) through a dizzying array of routes, and who began to interface directly with biological entities at the nanometer scale. That growth continues unabated. What has also become apparent is that much of the engineering community continues to view nanomaterials as curiosities rather than as the potentially gamechanging products that they can be. This book seeks to provide an entr´e into the field for mechanical engineers, material scientists, chemical and biomedical engineers and physicists. The objective is to provide the reader with the connections needed to understand the intense activity in the area of the mechanics of nanomaterials, and to develop ways of thinking about these new materials that could be useful to both research and application. Note that the book does not cover the areas associated with soft nanomaterials (polymer-based or biologically-derived), simply because I am not knowledgeable about such systems (and the mechanics can be quite different). This is not intended to minimize the importance of soft nanomaterials or the potential of soft nanotechnology; the reader will simply have to go elsewhere to encounter those areas. This book is intended to be read by senior undergraduates and first-year graduate students who have some background in mathematics, mechanics and materials science. It should also be of interest to scientists from outside the traditional fields of mechanical engineering and materials science who wish to develop core expertise in this area. Although senior undergraduates should be able to read this book

vii

viii

Nanomaterials

from cover to cover, they may find it somewhat heavy going as a textbook. First and second year graduate students should find this book challenging but accessible. My intent has not been to provide a review of the field but rather to provide a basic understanding. If the reader puts the book down with an appreciation of the excitement of the field of nanomaterials and the potential applications in mechanical engineering, materials science and physics, this book will have achieved its objectives. My personal objective with this book is to provide a means for integrating the discussions that currently go on in the mechanics and materials communities, and at the same time to provide an accessible path to those from outside these disciplines who wish to get involved in one of the most exciting fields of our time. While this book is intended to be read cover to cover, it can also be used as a reference work. The book should serve effectively as a textbook for a course in nanomaterials or nanomechanics as it relates to materials. It is possible that mechanical engineers using this book will also want to have a materials science reference available as they move through this volume, and conversely, material scientists using this book may want to have a basic mechanics of solids book as a companion volume. An effective scenario would be for graduate students who have taken courses in the mechanics of solids or the mechanical properties of materials to then take a course that uses this book as a text. The suggested reading at the ends of the chapter will be useful for those who wish to pursue a particular topic in greater depth than can be covered in a book such as this. I am aware that the book in its current form does not have enough problems at the end of each chapter to make it ideal as a textbook. However, this is a field at the cutting edge of nanotechnology, rather than an established area with standard problems that can be given to students without considering their specific backgrounds. The problems that are provided at the ends of the chapters are intended to provoke discussion and further study, rather than to provide training in specific solution methods. We are in the midst of a veritable explosion in nanotechnology, and nanomaterials are at the heart of it. I hope you put this book down as excited about the field as I have been in the writing of it. The Johns Hopkins University Baltimore, MD

K.T. Ramesh

Acknowledgements This book developed as a result of conversations with many students and colleagues working in the area of the mechanics of nanomaterials. What began as a simple attempt to rationalize my own thinking in this area became a concept for a book after discussions with Elaine Tham, my editor at Springer. Both she and her assistant, Lauren Denahy, deserve my thanks for their patience with me as I learned how to write a large document. My thanks go also to my assistant, Libby Starnes, who has given me a great deal of help in organizing my materials for this volume, and in rearranging my schedule to make time for this book. Her resourcefulness and ability to handle details have been terrific. I would like to give special thanks to the several anonymous reviewers of this book. Their suggestions have greatly improved the manuscript, and I appreciate all the effort involved. My research group has been bearing the weight of my absentmindedness as I worked on this book for nearly a year now, and I would like to express my gratitude to all of them. They have helped me immeasurably by doing the research that I get to talk about, by being willing to critique ideas, and by helping me make up for the times that I was unavailable. During the writing of this book, the research group included Dr. Shailendra Joshi, Reuben Kraft, Bhasker Paliwal, Dr. Qiuming Wei, Dr. Fenghua Zhou, Jessica Meulbroek, Emily Huskins, Sarthak Misra, Dr. Jamie Kimberley, Dr. Krishna Jonnalagadda, Dr. Buyang Cao, Cindy Byer, Rika Wright, Guangli Hu, Cyril Williams, Brian Schuster, Dr. Bin Li, Dr. Nitin Daphalapurkar and Dr. George Zhang. My thanks to all! Several students have been involved in reviewing chapters of this book, and to them my special thanks: Guangli Hu, Cindy Byer, Justin Jones; Christian Murphy made up some of the illustrations. My colleagues at Hopkins have been a big help, explaining (patiently) to me the physical concepts in nanomaterials and nanomaterial behavior. This is an exploding field, and my knowledge of it comes almost entirely through my interactions with the faculty and students around me at Johns Hopkins. Special thanks to Evan Ma, Kevin Hemker, Bill Sharpe, Jean-Francois Molinari and Todd Hufnagel. Finally, this book would not have been possible without the love and support of my family, who had to put up with my occasional pre-occupation, my prolonged absences, and a lack of social interaction as I wrote. They kept me sane during the difficult times, and cheerfully stepped in to help when they could. Rohan reviewed some of the chapters, and everyone has been involved with the proofreading. Arjun has been a source of good cheer throughout the process, and Priti, I could not do this without you. To my family, my love and my thanks for all that you do!

ix

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv 1

Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Length Scales and Nanotechnology . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 What are Nanomaterials? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Classes of Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Making Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Making dn Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Health Risks Associated with Nanoparticles . . . . . . . . . . . . . . 1.4.3 Making Bulk Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Closing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.7 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 1 3 5 6 6 7 8 17 18 18 19

2

Fundamentals of Mechanics of Materials . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Review of Continuum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Vector and Tensor Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.2 Kinematics of Deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.3 Forces, Tractions and Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Work and Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Field Equations of Mechanics of Materials . . . . . . . . . . . . . . . . . . . . . 2.4 Constitutive Relations, or Mathematical Descriptions of Material Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Plastic Deformation of Materials . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Fracture Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21 21 21 25 29 34 35

Nanoscale Mechanics and Materials: Experimental Techniques . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 NanoMechanics Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Characterizing Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

61 61 62 64

3

35 36 43 53 57 57 59

xi

xii

Nanomaterials

3.3.1 Scanning Electron Microscopy or SEM . . . . . . . . . . . . . . . . . . 3.3.2 Transmission Electron Microscopy or TEM . . . . . . . . . . . . . . 3.3.3 X-Ray Diffraction or XRD . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.4 Scanning Probe Microscopy Techniques . . . . . . . . . . . . . . . . . 3.3.5 Atomic Force Microscopy or AFM . . . . . . . . . . . . . . . . . . . . . 3.3.6 In situ Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Nanoscale Mechanical Characterization . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Sample and Specimen Fabrication . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Nanoindentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 Microcompression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Microtensile Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.5 Fracture Toughness Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.6 Measurement of Rate-Dependent Properties . . . . . . . . . . . . . . 3.5 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64 65 66 66 68 68 71 71 72 74 82 86 86 91 91 91

4

Mechanical Properties: Density and Elasticity . . . . . . . . . . . . . . . . . . . . . 95 4.1 Density Considered as an Example Property . . . . . . . . . . . . . . . . . . . . 95 4.1.1 The Rule of Mixtures Applied to Density . . . . . . . . . . . . . . . . 96 4.1.2 The Importance of Grain Morphology . . . . . . . . . . . . . . . . . . . 101 4.1.3 Density as a Function of Grain Size . . . . . . . . . . . . . . . . . . . . . 103 4.1.4 Summary: Density as an Example Property . . . . . . . . . . . . . . 105 4.2 The Elasticity of Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 4.2.1 The Physical Basis of Elasticity . . . . . . . . . . . . . . . . . . . . . . . . 106 4.2.2 Elasticity of Discrete Nanomaterials . . . . . . . . . . . . . . . . . . . . 107 4.2.3 Elasticity of NanoDevice Materials . . . . . . . . . . . . . . . . . . . . . 110 4.3 Composites and Homogenization Theory . . . . . . . . . . . . . . . . . . . . . . . 111 4.3.1 Simple Bounds for Composites, Applied to Thin Films . . . . . 113 4.3.2 Summary of Composite Concepts . . . . . . . . . . . . . . . . . . . . . . 116 4.4 Elasticity of Bulk Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 4.5 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 4.6 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . 118 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5

Plastic Deformation of Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 5.1 Continuum Descriptions of Plastic Behavior . . . . . . . . . . . . . . . . . . . . 121 5.2 The Physical Basis of Yield Strength . . . . . . . . . . . . . . . . . . . . . . . . . . 122 5.3 Crystals and Crystal Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 5.4 Strengthening Mechanisms in Single Crystal Metals . . . . . . . . . . . . . 132 5.4.1 Baseline Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 5.4.2 Solute Strengthening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 5.4.3 Dispersoid Strengthening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 5.4.4 Precipitate Strengthening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 5.4.5 Forest Dislocation Strengthening . . . . . . . . . . . . . . . . . . . . . . . 135

Contents

xiii

5.5 From Crystal Plasticity to Polycrystal Plasticity . . . . . . . . . . . . . . . . . 136 5.5.1 Grain Size Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 5.5.2 Models for Hall-Petch Behavior . . . . . . . . . . . . . . . . . . . . . . . . 138 5.5.3 Other Effects of Grain Structure . . . . . . . . . . . . . . . . . . . . . . . . 150 5.6 Summary: The Yield Strength of Nanomaterials . . . . . . . . . . . . . . . . . 154 5.7 Plastic Strain and Dislocation Motion . . . . . . . . . . . . . . . . . . . . . . . . . . 155 5.8 The Physical Basis of Strain Hardening . . . . . . . . . . . . . . . . . . . . . . . . 156 5.8.1 Strain Hardening in Nanomaterials . . . . . . . . . . . . . . . . . . . . . 158 5.9 The Physical Basis of Rate-Dependent Plasticity . . . . . . . . . . . . . . . . 160 5.9.1 Dislocation Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 5.9.2 Thermal Activation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 5.9.3 Dislocation Substructure Evolution . . . . . . . . . . . . . . . . . . . . . 166 5.9.4 The Rate-Dependence of Nanomaterials . . . . . . . . . . . . . . . . . 167 5.10 Case Study: Behavior of Nanocrystalline Iron . . . . . . . . . . . . . . . . . . . 172 5.11 Closing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 5.12 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 5.13 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . 176 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 6

Mechanical Failure Processes in Nanomaterials . . . . . . . . . . . . . . . . . . . . 179 6.1 Defining the Failure of Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 6.2 Failure in the Tension Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 6.2.1 Effect of Strain Hardening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 6.2.2 Effect of Rate-Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 6.2.3 Multiaxial Stresses and Microscale Processes Within the Neck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 6.2.4 Summary: Failure in the Simple Tension Test . . . . . . . . . . . . . 189 6.3 The Ductility of Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 6.4 Failure Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 6.4.1 Nucleation of Failure Processes . . . . . . . . . . . . . . . . . . . . . . . . 194 6.4.2 The Growth of Failures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 6.4.3 The Coalescence of Cracks and Voids . . . . . . . . . . . . . . . . . . . 196 6.4.4 Implications of Failure Processes in Nanomaterials . . . . . . . . 196 6.5 The Fracture of Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 6.6 Shear Bands in Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 6.6.1 Types of Shear Bands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 6.6.2 Shear Bands in Nanocrystalline bcc Metals . . . . . . . . . . . . . . . 203 6.6.3 Microstructure Within Shear Bands . . . . . . . . . . . . . . . . . . . . . 207 6.6.4 Effect of Strain Rate on the Shear Band Mechanism . . . . . . . 210 6.6.5 Effect of Specimen Geometry on the Shear Band Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 6.6.6 Shear Bands in Other Nanocrystalline Metals . . . . . . . . . . . . . 211 6.7 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 6.8 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . 211 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

xiv

Nanomaterials

7

Scale-Dominant Mechanisms in Nanomaterials . . . . . . . . . . . . . . . . . . . . 215 7.1 Discrete Nanomaterials and Nanodevice Materials . . . . . . . . . . . . . . . 215 7.1.1 Nanoparticles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 7.1.2 Nanotubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222 7.1.3 Nanofibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 7.1.4 Functionalized Nanotubes, Nanofibers, and Nanowires . . . . . 226 7.1.5 Nanoporous Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 7.1.6 Thin Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 7.1.7 Surfaces and Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 7.2 Bulk Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228 7.2.1 Dislocation Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228 7.2.2 Deformation Twinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 7.2.3 Grain Boundary Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 7.2.4 Grain Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 7.2.5 Stability Maps Based on Grain Rotation . . . . . . . . . . . . . . . . . 251 7.3 Multiaxial Stresses and Constraint Effects . . . . . . . . . . . . . . . . . . . . . . 256 7.4 Closing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256 7.5 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256 7.6 Problems and Directions for Research . . . . . . . . . . . . . . . . . . . . . . . . . 257 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

8

Modeling Nanomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 8.1 Modeling and Length Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 8.2 Scaling and Physics Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . 267 8.3 Scaling Up from Sub-Atomic Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . 268 8.3.1 The Enriched Continuum Approach . . . . . . . . . . . . . . . . . . . . . 269 8.3.2 The Molecular Mechanics Approach . . . . . . . . . . . . . . . . . . . . 269 8.4 Molecular Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 8.5 Discrete Dislocation Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 8.6 Continuum Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 8.6.1 Crystal Plasticity Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 8.6.2 Polycrystalline Fracture Models . . . . . . . . . . . . . . . . . . . . . . . . 279 8.7 Theoretically Based Enriched Continuum Modeling . . . . . . . . . . . . . . 280 8.8 Strain Gradient Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 8.9 Multiscale Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 8.10 Constitutive Functions for Bulk Nanomaterials . . . . . . . . . . . . . . . . . . 292 8.10.1 Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292 8.10.2 Yield Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293 8.11 Closing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 8.12 Suggestions for Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 8.13 Problems and Directions for Future Research . . . . . . . . . . . . . . . . . . . 295 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

Acronyms There are several acronyms that are commonly used in the field of nanomaterials. Some of the primary acronyms are listed here for use throughout the book. AFM cg CNT DDD EAM FEM LEFM MD MFM MWCNT nc ns OIM QC SEM STEM STM SWCNT TEM ufg XRD

Atomic Force Microscopy coarse-grained Carbon NanoTubes Discrete Dislocation Dynamics Embedded Atom Method Finite Element Method Linear Elastic Fracture Mechanics Molecular Dynamics Magnetic Force Microscopy Multi-Walled Carbon NanoTube NanoCrystalline NanoStructured Orientation Imaging Microscopy Quasi Continuum Scanning Electron Microscopy Scanning Transmission Electron Microscopy Scanning Tunneling Microscopy Single-Walled Carbon NanoTube Transmission Electron Microscopy ultra fine grained X-Ray Diffraction

xv

List of Figures

1.1

1.2

1.3

1.4

1.5

1.6

Length scales in mechanics and materials and in nature. The topics of interest to this book cover a large part of this scale domain, but are controlled by features and phenomena at the nm scale. Note the sophistication of natural materials and systems at very small length scales. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transmission electron micrographs (see Chapter 3) of nanocrystalline nickel material produced by electrodeposition (Integran). (a) Conventional TEM micrograph showing the grain structure. The average grain size is about 20 nm. (b) High resolution electron microscopy image of the same material. Note that there is no additional or amorphous phase at the grain boundaries, although many models in the literature postulate the existence of such a phase. Micrographs by Qiuming Wei. Reprinted from Applied Physics Letters, 81(7): 1240–1242, 2002. Q. Wei, D. Jia, K.T. Ramesh and E. Ma, Evolution and microstructure of shear bands in nanostructured fe. With permission from American Institute of Physics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transmission electron micrograph of nanocrystalline iron produced by consolidation of a nanocrystalline precursor made by ball milling. There is a range of grain sizes, and many of the grains show evidence of the prior plastic work produced by the ball-milling process. The width of the photograph represents 850 nm. . . . . . . . . . . Schematic of the die used in Equal Channel Angular Pressing (ECAP). The included angle φ in this case is about 120◦ , but can be as small as 90◦ . The angle Φ subtended by the external radius is also an important parameter. Various obtuse die angles and dimensions may be used. Very large forces are required to move the workpiece through such dies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a) Microstructure of tantalum produced by Equal Channel Angular Extrusion (ECAE), four passes at room temperature through a 90◦ die. (b) Selected Area Diffraction (SAD) pattern for this sample, showing the presence of many low angle grain boundaries. The fraction of grain boundaries that are low-angle rather than high-angle can be an important feature of the microstructure. . . . . . . . Schematic of High Pressure Torsion (HPT) process. A thin specimen is compressed and then subjected to large twist within a constraining die. The typical sample size is about 1 cm in diameter and about 1 mm thick. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

5

11

13

14

15 xvii

xviii

Nanomaterials

1.7

Microstructure of tungsten produced by high-pressure torsion (HPT). (a) A bright field TEM micrograph. (b) A dark field TEM micrograph. (c) A selected area electron diffraction pattern from this field. The grains contain a high density of defects (because of the plastic work associated with the HPT process). The grains are also elongated, with widths on the order of 80 nm and lengths of about 400 nm. Grain orientation appears to be along the shearing direction. The selected area electron diffraction shows nearly continuous rings, with no obvious intensity concentration along the rings, indicating large angle type grain boundaries in the sample. Reprinted from Acta Materialia, Vol. 52, Issue 7, P. 11, Q. Wei, L. Kecskes, T. Jiao, K.T. Hartwig, K.T. Ramesh and E. Ma, Adiabatic shear banding in ultrafine-grained Fe processed by severe plastic deformation. April 2004, with permission from Elsevier. . . . . . 16

2.1

The stress tensor σ maps the normal vector n at any point on a surface to the traction vector t at that point. . . . . . . . . . . . . . . . . . . . . . . Kinematics of deformation: a body initially in the reference configuration B0 , with material particles occupying locations X (used as particle identifiers) is deformed into the current configuration Bt with each particle occupying spatial positions x. . . . Internal forces on surfaces S and S’ within a body, and the traction t at a point on a surface with normal n. The entire body must be in equilibrium, as must be all subparts of the body. (a) The entire body, in equilibrium with the external forces. (b) Free body diagram corresponding to part B1 of the whole body, showing the traction developed at a point on the surface S because of the internal forces generated on B1 by part B2 . . . . . . . . . . . . . . . . . . . . . . . Schematic of a simple tension test. The loading direction is the x1 direction, and the gauge length is L. Note the specimen shape, designed to minimize the influence of the gripping conditions on the ends. Specimen shapes may be specified by testing standards. . . . (a) Schematic of tensile stress-strain curve for ductile metal, showing yield, an ultimate tensile strength, and subsequent failure. (b) Experimentally obtained curve for nanocrystalline nickel, with an average grain size 20 nm. Note the unloading line after initial yield, with the unloading following an elastic slope. Data provided by Brian Schuster. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a) Illustration of the biaxial tension of a sheet. (b) 2D stress space corresponding to the sheet, showing the stress path and possible yield surface bounding the elastic region. . . . . . . . . . . . . . . . . . . . . . . . .

2.2

2.3

2.4

2.5

2.6

23

26

30

44

45

48

List of Figures

2.7

2.8

3.1

3.2

3.3

3.4

3.5

xix

A sharp planar crack in a very large block of linear elastic material. The coordinate axes are defined so that the crack front (the line drawn by the crack tip) is the x3 or z axis. Note that all constant z planes look identical, so that this is essentially a plane strain problem in the x1 − x2 plane. The mechanics of this problem is dominated by the behavior as one approaches the crack tip. . . . . . . . . 54 Two views of the crack in Figure 2.7, with (a) being the side view and (b) the top view looking down on the crack plane. The crack front is typically curved, but this can be neglected in the limit as one approaches the crack tip. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Schematic of a Scanning Probe Microscopy system (not showing the feedback loop, tunneling amplifiers and other electronics). A wide variety of such systems exist, some of which contain only a subset of the features shown here. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Single crystal copper specimen and tension gripper for in situ tensile testing in the SEM, from the work of the group of Dehm, Kiener et al. (this group has also performed compression experiments using a similar apparatus) (a) SEM image showing a single-crystal copper tension sample and the corresponding tungsten sample gripper before the test at a low magnification. (b) Sample and gripper aligned prior to loading. Image taken from (Kiener et al., 2008). Reprinted from Acta Materialia, Vol 56, Issue 3, P. 13, D. Kiener, W. Grosinger, G. Dehm, R. Pippan, A Further step towards an understanding of size-dependent crystal plasticity: In situ tension experiments of miniaturized single-crystal copper samples. Feb. 2008, with permission from Elsevier. . . . . . . . . . . . . . . . Schematic of the NanoIndenter, showing both actuation and force and position sensing. Such a device may fit comfortably on a large table top, although vibration isolation may be desirable. . . . . . . . . . . . Example of force-displacement curve obtained during the nanoindentation of a nanocrystalline nickel material with an average grain size of 20 nm. Data provided by Brian Schuster. . . . . . . (a) Schematic of a microcompression experiment, showing micropillar and flat-bottomed indenter tip. (b) SEM micrograph showing a micropillar of a PdNiP metallic glass produced by focused ion beam machining. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

67

71

73

74

75

xx

Nanomaterials

3.6

(a) Finite element mesh used for 2D axisymmetric model of the micropillar in a microcompression experiment. (b) Computed effective (von Mises) stress distribution in a sample after plastic compression. Note the nonuniformity at the root of the pillar, indicating the importance of the root radius (also called the fillet radius). Reprinted from Scripta Materialia, Vol 54, Issue 2, page 6, H. Zhang, B.E. Schuster, Q. Wei and K.T. Ramesh, The design of accurate microcompression experiments, January 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7 Input and simulated stress-strain curves for various assumed root radii (detailed differences observable in the inset) as computed from the finite element model presented in the last figure. The “input” curve is the input material behavior used in the finite element simulations, and the “output” stress-strain curves are obtained from the forces and displacements computed from the simulations and then processed in the same way as the force and displacement data is processed in the experiments. If the experimental design were to be perfect, the output curves would be identical to the input curve. Note that the simulated curves are all above the input data, demonstrating the effect of the end condition. The outermost curve has the largest root radius, equal to the radius of the cylinder. As the root radius is decreased, the simulated curves approach the input curve but are always above it. Reprinted from Scripta Materialia, Vol 54, Issue 2, page 6, H. Zhang, B.E. Schuster, Q. Wei and K.T. Ramesh, The design of accurate microcompression experiments, January 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Computed variation of apparent elastic modulus (normalized by true elastic modulus) with fillet radius size at fixed aspect ratio, using the finite element simulations discussed in the text. The effects of the Sneddon and modified Sneddon corrections are also shown. Reprinted from Scripta Materialia, Vol 54, Issue 2, page 6, H. Zhang, B.E. Schuster, Q. Wei and K.T. Ramesh, The design of accurate microcompression experiments, January 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 Schematic of a microtension testing apparatus, showing the actuation and sensing systems. The air bearing is an important component. The strain is computed from displacement fields measured using Digital Image Correlation (DIC) software analysis of images from the camera. Illustration due to Chris Eberl. . . . . . . . . . 3.10 Example of stress strain curve obtained on nanocrystalline nickel (20 nm grain size) in uniaxial tension using a microtension setup. The sample was a thin film. Data due to Shailendra Joshi. . . . . . . . . . .

76

77

79

84

85

List of Figures

3.11 Experimental techniques appropriate for various ranges of strain rate. The range of rates identified as “specialized machine” is very difficult to reach. Very few laboratories in the world are able to achieve strain rates higher than 104 s−1 . . . . . . . . . . . . . . . . . . . . . . . . . . 3.12 Schematic of the compression Kolsky bar (also known, incorrectly, as the split-Hopkinson pressure bar). The projectile is usually launched towards the input bar using a gas gun. Specimen surfaces must be carefully prepared for valid experiments. . . . . . . . . . . . . . . . . . 3.13 Stress strain curves obtained on 5083 aluminum using the compression Kolsky bar at high strain rates (2500 per second). The lowest curve represents quasistatic behavior. In general, the strength appears to increase with increasing strain rate, but note the anomalous softening at the highest strain rate (perhaps due to thermal softening). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.14 The Desktop Kolsky Bar – a miniaturized compression Kolsky bar arrangement developed by Jia and Ramesh (2004). This device is capable of achieving strain rates above 104 per second, and can fit on a standard desktop. Specimen sizes can be on the order of a cubic millimeter. The large circular object behind the bar is part of a lighting system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1

4.2

4.3

4.4

4.5

4.6

xxi

87

88

90

90

The linear rule of mixtures. Calculated variation of density in a nanocrystalline material with volume fraction of grain boundary, based on Equation (4.6). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Schematic of (a) a cubical grain of size d with a grain boundary domain of thickness t, and (b) the packing of such grains to form a material. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Variation of grain volume fraction, grain boundary volume fraction, triple junction volume fraction, and corner junction volume fraction with normalized grain size β = dt for the cubic grain morphology. The junction volume fractions become major contributors when d ≈ t. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Scanning electron micrograph showing the presence of pores (dark regions) on the surface of a nanocrystalline nickel material produced by an electroplating technique. . . . . . . . . . . . . . . . . . . . . . . . . 101 Another possible space-filling morphology, using hexagonal tiles of side s and height h = α s, where α is the aspect ratio of the tile. The grain boundary domain thickness remains t. . . . . . . . . . . . . . . . . . 102 Variation of grain volume fraction, grain boundary volume fraction, triple junction volume fraction, and corner junction volume fraction with normalized grain size β = dt for the hexagonal prism grain morphology of Figure 4.5 (with an aspect ratio α = 1). The junction volume fractions become major contributors when d ≈ t. . . . 103

xxii

Nanomaterials

4.7

4.8

4.9

4.10

4.11 4.12 4.13 4.14

4.15 4.16

5.1

Variation of grain volume fraction, grain boundary volume fraction, triple junction volume fraction, and corner junction volume fraction with normalized grain size β = dt for the hexagonal prism grain morphology of Figure 4.5, with two aspect ratios (10 and 0.1) corresponding to rods (top) and plates (bottom). . . . . . . . . . . . . . . . . . . 104 Predicted variation of overall density of polycrystalline material (normalized by ρsc ) with grain size for the cube and hexagonal prism morphologies (with an aspect ratio α = 10). The assumed parameters are ρgb = 0.95ρsc , ρt j = 0.9ρsc , and ρc j = 0.81ρsc . . . . . . . 105 Typical interatomic pair potential U(r), showing the equilibrium position r0 of the atoms. The curvature at the bottom of the potential well (at the equilibrium position) corresponds to the effective elastic stiffness of the bond. . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 Surface effects on a spherical nanoparticle (e.g., surface tension). Such effects can have a significant impact on the behavior of the nanoparticle, particularly with respect to its interaction with the environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Schematic of a thin film on a substrate. The film thicknesses of interest to industry are typically submicron. . . . . . . . . . . . . . . . . . . . . . . 110 Schematic of a polycrystalline thin film on a substrate. Note the typical columnar microstructure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 Process of homogenization of a composite material. (a) Original heterogeneous material. (b) Equivalent homogenized material. . . . . . . 112 (a) Schematic of a columnar microstructure for a thin film. (b) Schematic of a layered microstructure, which can be viewed as the columnar microstructure loaded in the orthogonal direction to that shown in (a). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 Variation of effective modulus (for fictitious material) with grain boundary volume fraction, based on Equations (4.30) and (4.34). . . . . 115 Variation of effective Young’s modulus with normalized grain size in a bulk nanocrystalline material, where the modulus in the grain boundary domain is defined to be ζ Eg = ζ Esc with ζ = 0.7 in this figure. The grain size is normalized by the effective thickness t of the grain boundary domain, typically assumed to be about 1 nm. If the latter thickness is assumed, the horizontal axis corresponds to grain size in nm. E1 and E2 correspond to Equations 4.30 and 4.34. . . 117 Schematic of slip in a crystalline solid under shear. (a) Original crystal subjected to shear stress. (b) Crystal after deformation (slip) along the shearing plane shown in (a). The final atomic arrangement within the deformed crystal remains that of the perfect crystal, except near the free surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

List of Figures

5.2

5.3

5.4

5.5

5.6

5.7 5.8

5.9

xxiii

Dislocations are visible in the transmission electron microscope. The picture shows dislocations (the lines marked by the black arrow) in a magnesium alloy (ZK60). Note the length scale in the picture. This micrograph was taken by Bin Li. . . . . . . . . . . . . . . . . . . . 124 Expanded version of Figure 5.1 showing the primary deformation mechanism that leads to plasticity in metals: the motion of line defects called dislocations. Edge dislocations are shown gliding along the slip plane. The inset shows an expanded view of an edge dislocation, amounting to an extra plane of atoms in the lattice (with the trace of the extra plane on the slip plane defined as the line defect). The motion of many dislocations results in the macroscopic slip step shown in (c). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Schematic of slip with an edge dislocation (top figure) and screw dislocation (bottom figure), showing the Burgers and line vectors in each case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 Examples of basic crystal structures that are common in metals: body centered cubic (one atom in the center of the cube), face centered cubic (atoms at the center of each cube face), and hexagonal close packed structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Typical planes in a cubic crystal defined using Miller indices. The (111) plane and similar {111} planes are close-packed planes in a face-centered-cubic (fcc) crystal and therefore are part of typical operating slip systems (together with < 110 > type directions). . . . . . 129 Schematic of gliding dislocation bowing around dispersoids that are periodically spaced a distance L apart along a line. . . . . . . . . . . . . . 134 Schematic of a polycrystalline microstructure with a variety of plastic strains (represented by color or shading) within each individual grain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Schematic of grain boundary ledges in a polycrystalline material. In the grain boundary ledge model, these ledges are believed to act as dislocation sources and generate a dislocation network. . . . . . . . . . 140

xxiv

Nanomaterials

5.10 A summary of the three basic models for the observed Hall-Petch behavior. (a) A dislocation pileup at a grain boundary. (b) Grain boundary ledges in a polycrystalline material. (c) A schematic of the geometrically-necessary dislocation model. On the left side of this subpart of the figure, the individual crystals that make up this polycrystalline material are assumed to have slipped along their respective slip systems, resulting in misfits between the grains. However, the polycrystalline material must remain compatible at the grain boundaries, assuming that voids do not open up at the boundaries. These misfits can be accommodated by creating a new set of dislocations (the so-called geometrically-necessary dislocations) that then result in a dislocation distribution near the individual grain boundaries. Only one such GND distribution is shown in one grain on the right. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 5.11 The variation of the flow stress (measured at 4% strain) with grain size for consolidated iron, taken from the doctoral dissertation of Dexin Jia at Johns Hopkins. The variation of hardness (defined here as H3v , where Hv is the Vickers hardness) is also shown. . . . . . . . . . . . . 144 5.12 Evidence of reduction of hardness with decreasing grain size in a Ni-P material, from Zhou, Erb et al., Scripta Materialia, 2003 (Zhou et al., 2003). The right-hand axis represents the Vickers Hardness number HV. Reprinted from Scripta Materialia, Vol. 48, Issue 6, page 6, Y. Zhou, U. Erb, K.T. Aust and G. Palumbo, The effects of triple junctions and grain boundaries on hardness and Young’s modulus in nanostructured Ni-P. March 2003, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 5.13 Compendium of data on copper in a Hall-Petch-style plot by Meyers et al. (2006), showing the scatter in data at small grain sizes but the general deviation from classical Hall-Petch behavior. Reprinted from Progress in Materials Science, Vol. 51, Issue 4, Page 130, M.A. Myers, A. Mishra, D.J. Benson, Mechanical properties of nanocrystalline materials. May 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 5.14 Hall-Petch plots for copper, iron, nickel and titanium compiled by Meyers et al. (2006), showing the deviation from classical Hall-Petch at small grain sizes. Note the strength appears to plateau but not decrease with decreasing grain size. Reprinted from Progress in Materials Science, Vol. 51, Issue 4, Page 130, M.A. Myers, A. Mishra, D.J. Benson, Mechanical properties of nanocrystalline materials. May 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

List of Figures

xxv

5.15 Measured grain size distribution in a consolidated iron sample, based on 392 grains measured from TEM images such as that shown above (Jia et al., 2003). While some of the grains are clearly in the nanocrystalline range, others are more than 100 nm in size. Reprinted from Acta Materialia, Vol. 51, Issue 12, page 15, D. Jia, K.T. Ramesh, E. Ma, Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. July 2003, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 5.16 TEM micrographs showing microstructure of a cryomilled 5083 aluminum alloy in (a) extruded and (b) transverse directions (transverse to the extrusion axis). Note that the two images were taken at slightly different magnifications. However, elongated grains are evident in the direction of extrusion (Cao and Ramesh, 2009). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 5.17 Stress strain curve for a material, showing the elastic unloading response and the increased yield strength upon reloading (this is called strain hardening, since the material has become harder because of the increased plastic strain). . . . . . . . . . . . . . . . . . . . . . . . . . 156 5.18 Variation of normalized strain hardening with the square root of grain size for a variety of nanocrystalline and ultra-fine-grain materials (from Jia’s doctoral dissertation, Jia et al., 2003). The reference numbers correspond to those in Jia’s dissertation. . . . . . . . . 159 5.19 The dependence of dislocation velocity on shear stress for a variety of materials, from Clifton (1983). Note that a limiting velocity is expected, corresponding to the shear wave speed. There is a domain in the figure where the dislocation velocity is linear with the applied shear stress, and this is called the phonon drag regime. Reprinted from Journal of Applied Mechanics, Vol. 50, p. 941–952, R.J. Clifton, Dynamic Plasticity. 1983, with permission from original publisher, ASME. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 5.20 Rate-dependence of the flow stress of A359 aluminum alloy over a wide range of strain rates. All of the flow stresses are plotted at a fixed strain of 4% (because of the nature of high strain rate experiments, it is not generally possible to measure accurately the yield strength of materials at high strain rates). This result is due to the work of Yulong Li, and includes compression, tension and torsion data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

xxvi

Nanomaterials

5.21 Results of strain rate jump tests on severely plastically deformed copper (both cold-worked and ECAPed). Note the jump in the stress associated with the step increase in the strain rate on a specimen. One of the curves has been shifted to the right for ease of discrimination. Step increases in strain rate (jump from a lower strain rate to a higher strain rate) are used because a step decrease will exacerbate the effect from the machine compliance, and the interpretation of experimental data becomes more involved. In this case the strain rate was increased by a factor of 2 between consecutive rates. Reprinted from Materials Science and Engineering, Q. Wei, S. Cheng, K.T. Ramesh, E. Ma, Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Sep. 2004, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 5.22 The rate-sensitivity of copper as a function of the mean grain size, including the nanocrystalline, ultra-fine-grain and coarse-grain domains. Data is presented for materials made through a variety of processing routes, some of which involve severe plastic deformation. The figure is taken from the work of Wei et al. (2004a). Reprinted from Materials Science and Engineering, Q. Wei, S. Cheng, K.T. Ramesh, E. Ma, Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Sep. 2004, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 5.23 The rate-sensitivity of bcc metals as a function of grain size from a variety of sources (the figure is a variant of one published by Wei et al. (2004a). The decrease of rate sensitivity with decreasing grain size is the opposite behavior to that observed in fcc metals. Reprinted from Materials Science and Engineering, Q. Wei, S. Cheng, K.T. Ramesh, E. Ma, Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Sep. 2004, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 5.24 Schematic of motion of screw dislocations by kink pair nucleation and propagation. The dislocation is visualized as a line that needs to go over the energy barrier. Rather than move the entire line over the barrier, the dislocation nucleates a kink, which jumps over the barrier. The sides of the kink pair have an edge orientation, and so fly across the crystal because of their high mobility, resulting in an effective motion of the screw dislocation. . . . . . . . . . . . . . . . . . . . . . . . 173 6.1

Stress strain curve for a material obtained from a standard tensile specimen tested in uniaxial tension, showing the final fracture of the specimen in uniaxial tension. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

List of Figures

6.2

6.3

6.4

6.5

6.6

6.7

xxvii

The development of a neck during plastic deformation of a specimen within a simple tension experiment. The position and length of the neck are determined by geometric and material imperfections in the specimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 The influence of rate-sensitivity on the total elongation to failure of materials, from the work of Woodford (1969). Note the strong effect of the rate-dependence, and the similarity in behavior of a wide variety of materials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Form of the fracture developed in a simple tension test. (a) Original specimen configuration before loading. (b) Tensile failure of a brittle material, showing that the fracture surface is essentially perpendicular to the loading axis. (c) Tensile failure of a ductile material, showing the cup and cone failure morphology (in section). . 189 The total elongation to failure of copper materials of varying grain sizes, as presented by Ma et al. (Wang et al., 2002). It is apparent that the ductility decreases dramatically as the yield strength is increased. The point labeled E in the figure corresponds to a special copper material produced by Ma and coworkers that included both nanocrystalline and microcrystalline grain sizes, with the larger grains providing an effective strain hardening in the material. Reprinted by permission from Macmillan Publishers Ltd: Nature, Vol. 419, Issue 6910, pages 912–915, High tensile ductility in a nanostructured metal, Yinmin Wang, Mingwei Chen, Fenghua Zhou, En Ma. 2002. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 The total elongation to failure of fcc materials of nanocrystalline and microcrystalline grain sizes, as presented by Dao et al. (2007). Reprinted from Acta Materialia, Vol. 55, Issue 12, page 25, M. Dao, L. Lu, R.J. Asaro, J.T.M. DeHosson, E. Ma, Toward a quantitative understanding of mechanical behavior of nanocrystalline metals. July 2007, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . 192 A standard compact tension (CT) specimen used for fracture toughness measurements. The specimens are several centimeters in size. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

xxviii

6.8

Nanomaterials

A sequence of in situ TEM micrographs obtained by Kumar et al. (2003) during the loading of nanocrystalline nickel using multiple displacement pulses. Images a–d show the microstructural evolution and progression of damage with an increase in the applied displacement pulses. The presence of grain boundary cracks and triple-junction voids (indicated by white arrows in (a)), their growth, and dislocation emission from crack tip B in (b–d) in an attempt to relax the stress at the crack tip as a consequence of the applied displacement, can all be seen. The magnified inset in (d) highlights the dislocation activity (Kumar et al., 2003). Reprinted from Acta Materialia, Vol. 51, Issue 2, page 19, K.S. Kumar, S. Suresh, M.F. Chisholm, J.A. Horton, P. Wang, Deformation of electrodeposited nanocrystalline nickel. Jan. 2002, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199 6.9 Postmortem microscale fracture morphology observed by Kumar et al. (2003) after the loading of nanocrystalline nickel in tension. There is no direct evidence of the presence of dislocations in this image, although some of the grains appear to have necked before separation. Reprinted from Acta Materialia, Vol. 51, Issue 2, page 19, K.S. Kumar, S. Suresh, M.F. Chisholm, J.A. Horton, P. Wang, Deformation of electrodeposited nanocrystalline nickel. Jan. 2002, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . 200 6.10 Postmortem fracture surface morphology of nanocrystalline (average grain size of 38 nm) gold thin film tested in tension till failure by Jonnalagadda and Chasiotis (2008). Dimples are seen on the fracture surface, with an average size of more than 100 nm. The void size just before coalescence is therefore much larger than the grain-size, and the growth of the voids must involve plastic deformation in many many grains. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 6.11 Progressive localization of a block (a), with initially uniform shearing deformations (b) developing into a shear band (c). The final band thickness depends on the material behavior. . . . . . . . . . . . . . 202 6.12 Stress-strain curve for a material showing softening after a peak stress. This is a curve corresponding to a material undergoing thermal softening, but similar behaviors can arise from other causes such as grain reorientation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

List of Figures

xxix

6.13 Gross deformation features of coarse-grain and ultra-fine-grain iron samples after compressive deformations to nearly identical strains (Jia et al., 2003). (a–b) represent quasistatic compression, while (c) represents high-strain-rate compression. (a) Homogeneous deformation of coarse-grain (20 µm grain size) Fe. (b) Shear band pattern development in ultra-fine-grain (270 nm grain size) Fe. (c) Shear band pattern development in ultra-fine-grain Fe after dynamic compression at a strain rate of ≈ 103 s−1 . Reprinted from Acta Materialia, Vol. 51, Issue 12, page 15, D. Jia, K.T. Ramesh, E. Ma, Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. July 2003, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 6.14 Shear band patterns evolve with strain in compressed ultra-finegrain Fe (Wei et al., 2002). Note the propagation of existing shear bands, the nucleation of new shear bands, and the thickening of existing shear bands. The development of families of conjugate shear bands is also observed. Reprinted from Applied Physics Letters, Vol. 81, Issue 7, pages 1240–1242, Q. Wei, D. Jia, K.T. Ramesh, E. Ma, Evolution and microstructure of shear bands in nanostructured fe. 2002, with permission from American Institute of Physics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 6.15 (a) Shearing deformation across one band and (b) the stable intersection of multiple shear bands in 270-nm Fe after quasistatic compression. The shear offset is clearly visible across the first band. Note that no failure (in terms of void growth) is evident at the intersection of the shear bands. Reprinted from Acta Materialia, Vol. 51, Issue 12, page 15, D. Jia, K.T. Ramesh, E. Ma, Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. July 2003, with permission from Elsevier. . . 206 6.16 TEM micrographs (a) within and (b) outside a shear band in quasistatically compressed Fe with an average grain size of 138 nm (Wei et al., 2002). The shearing direction is shown by the arrow. Note the preferred orientation of the grains within the shear band, while the grains outside the band are essentially equiaxed. Reprinted from Applied Physics Letters, Vol. 81, Issue 7, pages 1240–1242, Q. Wei, D. Jia, K.T. Ramesh, E. Ma, Evolution and microstructure of shear bands in nanostructured fe. 2002, with permission from American Institute of Physics. . . . . . . . . . . . . . . . . . . 208

xxx

Nanomaterials

6.17 TEM micrograph showing the microstructure near a shear band boundary in nanocrystalline iron (Wei et al., 2002). The boundary between the material within the band and that outside the band is shown by the solid line. Note that the transition occurs over a transition width that is about one grain diameter. Reprinted from Applied Physics Letters, Vol. 81, Issue 7, pages 1240–1242, Q. Wei, D. Jia, K.T. Ramesh, E. Ma, Evolution and microstructure of shear bands in nanostructured fe. 2002, with permission from American Institute of Physics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 7.1

7.2

7.3

7.4

7.5

7.6

Schematic of a cubic nanocluster consisting of 64 atoms, approximately 1 nm on a side. The majority of the atoms are on the surface of this nanocluster, and so will have a different equilibrium spacing than atoms in a bulk sample of the same material. . . . . . . . . . . 218 Volume fraction of atoms on the surface of a cuboidal nanoparticle as a function of nanoparticle size, based on Equation (7.1) and an assumed interatomic spacing of 0.3 nm. Significant surface fractions are present below about 20 nm. Note that particle sizes below 0.6 nm are poorly defined. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 Core-shell model of a nanoparticle viewed as a composite, with a surface layer that has different properties as a result of the surface energy and its effects on binding. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 A spherical nanoparticle carrying several organic molecules on its surface. The molecules are chosen to perform a specific function, such as recognizing a molecule in the environment, and so are called functionalizing molecules. The spacing of the functionalizing molecules defines the functionalization density. Since the molecules modify the surface stress state when they attach to the surface, the functionalization density modifies the surface stresses and can even modify the net conformation of the nanoparticle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Schematic of the structure of carbon nanotubes, showing the armchair, zigzag and chiral conformations. This beautiful illustration was created by Michael Str´ock on February 1, 2006 and released under the GFDL onto Wikipedia. . . . . . . . . . . . . . . . . . . . . . . . 223 Schematic of two possible modes of deformation and failure in nanotubes. (a) Buckling of a thin column. (b) Telescoping of a multi-walled nanotube. The fracture of nanotubes is a third mode, but is not shown. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224

List of Figures

7.7

7.8

7.9

7.10

7.11

7.12

7.13

7.14

xxxi

The development of a twinned region in a material. The figures show the results of a molecular dynamics calculation of the simple shear of a single crystal of pure aluminum. (a) Atomic arrangement in undeformed crystal before shearing. The straight lines are drawn to guide the eye to the atomic arrangement. (b) After shearing, a region of the crystal has been reoriented (this region is called the twin). The twin boundaries can be viewed as mirror planes, and the lines show the new arrangement of the atoms. . . . . . . . . . . . . . . . . . . . . 231 Example of deformation twins in a metal (the hcp metal hafnium). (a) Initial microstructure before deformation. (b) Twinned microstructure after compressive deformation at low strain rates and at 298K. The twinned regions are the lenticular shapes within the original grain structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 Evolution of twin number density in titanium with applied stress, over a variety of strain rates and temperatures. The twin density is not correlated with strain rate or temperature (or strain), but only with applied stress (Chichili et al., 1998). Reprinted from Acta Materialia, Vol. 46, Issue 3, page 19, D.R. Chichili, K.T. Ramesh, K.J. Hemker, The high-strain-rate response of alpha-titanium: experiments, deformation mechanisms and modeling. Jan. 1998 with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 High resolution electron micrograph of deformation twins developed in nanocrystalline aluminum subjected to large shearing deformations (Cao et al., 2008). The diffraction pattern on the right demonstrates the twinned character. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 The possible grain boundary motions: displacements δn in the direction normal to the GB (grain growth or shrinkage), and δt in a direction tangential to the plane of the boundary (grain boundary sliding). The unit normal vector is n and the unit tangent vector is t, as in Equation (7.11). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 (a) Schematic of a grain showing its soft and hard orientations (with respect to plastic deformation, not elastic stiffness). (b) The grains are initially randomly oriented in the material, but begin to orient themselves so that the soft direction is in the direction of shearing, and the process of grain rotation into the soft orientation results in the localization of the deformation into a shear band. . . . . . 238 Schematic of simple shearing of an infinite slab, showing the terms used in examining the shear localization process in simple shear (Joshi and Ramesh, 2008b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Schematic of the rotation of ensemble of nano-grains occupying region ℜ embedded in a visco-plastic sea S subjected to shear. The background image shows the undeformed configuration. . . . . . . . . . . . 240

xxxii

Nanomaterials

7.15 A hierarchical approach to homogenization of grain rotation due to interface traction. (a) Material continuum. (b) Collection of bins in sample space. (c) Grains within a RVE. The colored shading represents the average grain orientation in that bin. (d) Interaction at the grain level. Grains with individual orientations are described by different colors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 7.16 Enlarged view of ℜ (Figure 7.14) showing intergranular interaction in the region. Rotation of the central grain is accomodated by rotation of the surrounding grains over a length L. . . . . . . . . . . . . . . . . 243 7.17 Evolution of grain orientation fraction (φ ) around the band center. Note the rapid early growth, the spatial localization, and the saturation of φ . Also note the spreading of the band, i.e., the increasing band thickness - this is also observed in experiments. . . . . 248 7.18 Evolution of plastic shear strain rate (γ˙p ) around the band center. The greatest activity in this variable is at the band boundaries, where the grains are reorienting into the soft orientation for shear. . . . 248 7.19 Evolution of the plastic shear strain γ p around the band center. After localization, γ p evolves slowly indicating that the plastic flow inside the band develops at the rate of strain hardening at higher nominal strains. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248 7.20 Overall stress-strain response for a defect-free sample (curve A) and a sample with an initial defect in φ (curve B). Curve B’ indicates the development of the shear band thickness corresponding to curve B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 7.21 Evolution of shear band thickness for different grain sizes, assuming grain rotation mechanism. The material hardening parameters are held constant for all the grain sizes. The applied strain rate is 10−3 s−1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 7.22 The critical wavelength for instability as a function of grain size for three different metals (note that this is a log-log plot), with the critical wavelength computed using Equation (7.60). This figure has been obtained assuming that j = 10 below d = 100 nm, while for 100 nm < d ≤ 1 µm, we have L = 1 µm and j = Ld . The dashed straight line represents the condition that λcritical = d, which we call the inherent instability line. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 7.23 Stability map, showing the domains of inherent instability in materials as a consequence of the rotational accommodation mechanism. The map is constructed in terms of the strength index and grain size, so that every material of a given grain size represents one point on the map, and a horizontal line represents all grain sizes of a given material. Reprinted figure with permission from S.P. Joshi and K.T. Ramesh, Physical Review Letters, Stability map for nanocrystalline and amorphous materials, 101(2), 025501. Copyright 2008 by the American Physical Society. . . . . . . . . . . . . . . . . 255

List of Figures

8.1

8.2

8.3

8.4 8.5 8.6

8.7

xxxiii

Typical sequence of steps involved in modeling a mechanics of nanomaterials problem. Note the many layers of approximations involved, pointing out the danger of taking the results of simulations at face value. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262 The typical modeling approaches of interest to the mechanics of nanomaterials, and the approximate length scales over which each approach is reasonable. Note the significant overlap in length scales for several of the modeling approaches, leading to the possibility of consistency checks and true multiscale modeling. An example of the observations that can be made at each length scale is also presented, from a materials characterization perspective. . . . . . . . . . . . 263 The length scales associated with various microstructural features in metallic materials. For most crystalline materials, the behavior at larger length scales is the convolution of the collective behavior of features at smaller length scales. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Schematic of the Lennard-Jones interatomic pair potential V (d). . . . . 272 The pair potential corresponding to the EAM potential for aluminum, using the parameters provided by Mishin et al. (1999). . . . 273 Nanocrystalline material constructed using molecular dynamics. The material is nickel, and the atoms are interacting using an EAM potential due to Jacobsen. About 30 grains are shown. The color or shade of each atom represents its coordination number (number of nearest neighbors), with the atoms in the crystals having the standard face-centered-cubic coordination number. The change in coordination number at the grain boundaries is clearly visible. This collection of atoms can now be subjected to mechanical loading (deformations) and the motions of the individual atoms can be tracked to understand deformation mechanisms. . . . . . . . . . . . . . . . . . . 275 High resolution micrograph showing the boundary between two tungsten grains in a nanocrystalline tungsten material produced by high-pressure torsion. There is no evidence of any other phase at the grain boundary, contrary to pervasive assumptions about the existence of an amorphous phase at the grain boundary in nanomaterials in crystal plasticity and composite simulations at the continuum level. Note the edge dislocations inside the grain on the right. Such internal dislocations are rarely accounted for in molecular dynamics simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

xxxiv

8.8

Nanomaterials

Finite element model of a polycrystalline mass of alumina, with the crystals modeled as elastic solids (Kraft et al., 2008). The model seeks to examine the failure of the polycrystalline mass by examining cohesive failure along grain boundaries. Reprinted from Journal of the Mechanics and Physics of Solids, Vol. 56, Issue 8, page 24, R.H. Kraft, J.F. Molinari, K.T. Ramesh, D.H. Warner, Computational micromechanics of dynamic compressive loading of a brittle polycrystalline material using a distribution of grain boundary properties. Aug. 2008 with permission from Elsevier. . . . . . 280 8.9 Schematic of the structure of a carbon nanotube, viewed as a sheet of carbon atoms wrapped around a cylinder. The sheet can be arranged with various helix angles around the tube axis, which are most easily defined in terms of the number of steps in two directions along the hexagonal array required to repeat a position on the helix. Illustration by Volokh and Ramesh (2006). Reprinted from International Journal of Solids and Structures, Vol. 43, Issue 25–26, Page 19, K.Y. Volokh, K.T. Ramesh, An approach to multi-body interactions in a continuum-atomistic context: Application to analysis of tension instability in carbon nanotubes. Dec. 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . 281 8.10 Connection between continuum deformations and atomic positions as defined by Equation (8.21). We associate atomic positions in the two configurations with material and spatial vectors. . . . . . . . . . . . . . . 284 8.11 Range of experimental measurements of the elastic modulus of carbon nanotubes, as summarized by Zhang et al. (2004). Note the theoretical predictions discussed here were 705 GPa (Zhang et al., 2002) and 1385 GPa (Volokh and Ramesh, 2006). Reprinted from Journal of the Mechanics and Physics of Solids, P. Zhang, H. Jiang, Y. Huang, P.H. Geubelle, K.C. Hwang, An atomistic-based continuum theory for carbon nanotubes: analysis of fracture nucleation. May 2004, with permission from Elsevier. . . . . . . . . . . . . . 285 8.12 Range of modeling estimates of the elastic modulus of carbon nanotubes, as summarized by Zhang et al. (2004), including a wide variety of first principles and MD simulations. Note the theoretical predictions discussed here were 705 GPa (Zhang et al., 2002) and 1385 GPa (Volokh and Ramesh, 2006). Reprinted from Journal of the Mechanics and Physics of Solids, P. Zhang, H. Jiang, Y. Huang, P.H. Geubelle, K.C. Hwang, An atomistic-based continuum theory for carbon nanotubes: analysis of fracture nucleation. May 2004, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286

List of Figures

xxxv

8.13 Simulation of a copper grain boundary by Warner et al. (2006) using the quasicontinuum method. The arrows correspond to the displacement of each atom between two loading steps. Reprinted from International Journal of Plasticity, Vol. 22, Issue 4, Page 21, D.H. Warner, F. Sansoz, J.F. Molinari, Atomistic based continuum investigation of plastic deformation in nanocrystalline copper. April 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 8.14 Computational approach for simulations of nanocrystalline copper by Warner et al. (2006). The grayscale represents different orientations of the crystals. Reprinted from International Journal of Plasticity, Vol. 22, Issue 4, Page 21, D.H. Warner, F. Sansoz, J.F. Molinari, Atomistic based continuum investigation of plastic deformation in nanocrystalline copper. April 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 8.15 Comparison of the predictions of the multiscale simulations of Warner et al. (2006) (identified with the FEM symbol) with molecular dynamics (MD) simulation results and experimental data on nanocrystalline copper. Note that the experimental results are derived from hardness measurements, while the FEM simulation results correspond to a 0.2% proof strength. Both MD and FEM simulations predict much higher strengths than are observed in the experiments. Reprinted from International Journal of Plasticity, Vol. 22, Issue 4, Page 21, D.H. Warner, F. Sansoz, J.F. Molinari, Atomistic based continuum investigation of plastic deformation in nanocrystalline copper. April 2006, with permission from Elsevier. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

List of Tables

1.1

A broad classification of nanomaterials on the basis of dimensionality and morphology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

2.1

Young’s modulus and Poisson’s ratio for some conventional grain sized materials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.1

Some measured moduli of nanotubes, nanofibers, and nanowires, viewed as 1D nanomaterials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.1 5.2

Slip systems with typical numbering scheme in typical fcc metals. . . 130 Known slip systems in a variety of metals, from a table put together by Argon (Argon, 2008). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 Hall-Petch coefficients for a variety of materials. . . . . . . . . . . . . . . . . . 143 Hall-Petch coefficients for pure aluminum and select aluminum alloys. Data taken from the work of Witkin and Lavernia (2006). . . . . 145

5.3 5.4 7.1 7.2 7.3

Scale-dominant mechanisms in nanomaterials, categorized in terms of materials classification, morphology, and length scale. . . . . . . . . . . 216 Typical intrinsic length scales that arise from dislocation mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229 Basic parameters for grain rotation model in polycrystalline iron, as developed by Joshi and Ramesh. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

xxxvii

1

There’s plenty of room at the bottom. Richard Feynman

Nanomaterials 1.1 Length Scales and Nanotechnology Nanotechnology demands the ability to control features at the nanoscale (10−9 m), and a variety of techniques have been developed recently that give humanity this ability. From a fundamental science perspective, issues of physics and chemistry must be addressed at these scales. Surface and boundary effects can dominate the response. Many of the classical distinctions between mechanics, materials and physics disappear in this range of length scales, and a new kind of thinking emerges that is commonly called nanoscience (sometimes humorously interpreted as “very little science”). The recent rapid development of nanoscience is the result of a new-found ability to observe and control structure at small length and time scales, coupled with the development of computational capabilities that are most effective at small scales. It is useful to develop a physical idea of length scale, and Figure 1.1 shows the range of length scales of common interest in mechanics and biology (the latter is included because it allows one to develop a human sense of scale). Beginning at small ˚ (one angstrom, scales, features associated with atomic radii are on the order of 1 A ˚ 10−10 m) in size. The atomic lattice spacing in most crystals is of the order of 3 A. ˚ and this correlates well The diameter of a carbon nanotube is about 2 nm or 20 A, with the diameter of a double helix of DNA (which indicates, incidentally, that the nanotube is a good approach to handling DNA). A tobacco mosaic virus is about 50 nm across (this corresponds approximately in scale with the typical radii of curvature of the tips of nanomanipulators such as AFM probes). Grains in most polycrystalline metals have sizes that range from about 1 µm to about 20 µm (grain boundary thicknesses, to the extent that they can be defined, are typically 0, K > 0 and −1 < ν ≤ 0.5 (although in most engineering materials, 0 ≤ ν ≤ 0.5). From Equation (2.90) above, we see that ν = 0 only when µ = 3k 2 , but the bulk modulus is usually greater than the shear modulus, i.e., usually K > µ , so that it is difficult to find materials with zero Poisson’s ratio (the closest approximation is cork, which is why this material is used to stop up wine bottles – it does not expand as you push it into the bottle). Materials with negative Poisson’s ratio are unknown in nature, but efforts are underway to construct metamaterials that have this property. For a single crystal, the linear elastic response represents the stiffness of the interatomic bonds within the material near the equilibrium positions of the atoms in the lattice, and the elastic moduli for a single crystal are not generally isotropic. Thus one must use the full set of linear elastic constants (making use of available symmetries) to describe the mechanical behavior of a single crystal in the linear elastic range, and one must find values for the coefficients Ci jkl corresponding to Equation (2.74) for each material. For example, for single crystal α -iron, the cubic symmetry allows us to use Equation (2.76) and one only needs to find values for C1111 , C1122 , and C2323 corresponding to the Voigt constants C11 , C12 , and C44 . Experimental determination of these constants provides C11 = 230.37 GPa, C12 = 134.07 GPa, and C44 = 115.87 GPa at a temperature of 300 K (the elastic moduli are functions of temperature). A scalar measure of the elastic anisotropy of cubic crystals that is often used is 2C44 A= . (2.92) (C11 −C12 ) This measure is usually much greater than 1 for most metals (e.g., A = 2.41 for α -iron at 300K), with the exception being tungsten, where A ≈ 1. However, most engineering materials consist of large polycrystalline assemblies, with the crystals oriented randomly, and the effective behavior of this polycrystalline mass in the linear elastic range might be nearly isotropic (assuming that there is no significant texture). For polycrystalline pure iron, therefore, an isotropic linear elastic behavior is often a reasonable approximation, and the experimental values are E ≈ 210 GPa and ν ≈ 0.29. Note that the averaging performed over a large number of randomly oriented crystals has effectively reduced the number of independent constants. A table of typical values of the moduli for a number of materials (for conventional grain sizes, in the randomly oriented polycrystalline case) is presented in Table 2.1. Note the high variability in the elastic data on some polycrystalline ceramics such as B4 C, which arises because of differences in processing approaches and the resulting flaw distributions, as well as variations with measurement processes. Up to this point, the discussion has assumed that the material behavior is linear elastic. However, most materials demonstrate linear elastic behavior only in a finite

2 Fundamentals of Mechanics of Materials

43

Table 2.1 Young’s modulus and Poisson’s ratio for some conventional grain sized materials. Material Young’s modulus E, GPa Poisson’s ratio ␯ Iron 210 0.29 Copper 130 0.34 Aluminum 70 0.33 Vanadium 128 0.37 Tantalum 186 0.34 Nickel 200 0.31 Titanium 116 0.32 Alumina, Al2 O3 396 0.25 Silicon carbide, SiC 410 0.14 Boron carbide, B4 C 362–460 0.18–0.21

range of stresses (or strains), beyond which the deformations are not reversible and some of the work done by the stresses is associated with permanent changes in the material. This non-elastic behavior is discussed in the next section.

2.4.2 Plastic Deformation of Materials Most materials undergo permanent, non-reversible deformations if they are stressed beyond some critical point, so that the basic assumptions made in the section on elasticity above are no longer valid. In particular, the behavior of the material may become such that the stress in the material depends not just on the current state of deformation, but on the history of the state of deformation. For example, the material behavior can be different during a second loading cycle than it was during the first loading cycle, a behavior sometimes called hardening. This history-dependent constitutive behavior is called plasticity – a material that obeys such a constitutive law is said to be plastic. Most metals in our everyday experience show plastic behavior at some point, and such a response is used heavily in industry in the forming of materials, components and structures. Understanding plasticity is important both from a design viewpoint (if permanent deformations are to be avoided) and from a manufacturing viewpoint (if the material is to be formed into a desired shape). The classical example of plastic behavior is observed when a rod or strip of a ductile metal is subjected to uniaxial tensile loading (Figure 2.4) . This is called the simple tension test, although—as an experiment—there is little that is simple about it. The shape of the sample is typically the dogbone shown in the figure, and it is loaded along the x1 direction with a force P. The central section of the sample has a gauge length L and a cross-section area A0 , and sees a nearly uniform stress state with only one nonzero stress component, σ11 . Note that σ11 is a component of the Cauchy stress, also sometimes called the true stress, which is defined in the deformed configuration as σ11 = σtrue = PA and so requires a knowledge of the current cross-section area A (usually very difficult to measure). The corresponding engineering stress σeng is easier to measure and is defined as the ratio of the applied

44

Nanomaterials x

P

x

P

Fig. 2.4 Schematic of a simple tension test. The loading direction is the x1 direction, and the gauge length is L. Note the specimen shape, designed to minimize the influence of the gripping conditions on the ends. Specimen shapes may be specified by testing standards.

force P to the initial cross-section area A0 :

σeng =

P . A0

(2.93)

As the sample is loaded in tension, it elongates, and the change in the length ∆ L can be measured using a variety of devices. The ratio of this change in length to the original length is a strain measure called the engineering strain:

εeng =

∆L . L

(2.94)

This is a strain measure that uses the initial configuration of the specimen as a reference. The strain measure that corresponds to the current configuration of the specimen is called the true strain, and considers the change δ l in the current length l of the specimen through ε = δll . The true strain should be used whenever the true stress is used, since they both represent variables in the current configuration of the specimen. The engineering stress is plotted against the engineering strain, providing a stress-strain curve that is characteristic of the material rather than of the specimen (at least until necking of the specimen begins, when the deformation localizes to a small region that thins rapidly and results in sample failure). A schematic of such a tensile stress-strain curve is presented in Figure 2.5a. The early part of the stressstrain curve is linear, and represents the linear elastic response of the material (the slope of this line corresponds to the Young’s modulus). In this linear domain, unloading the sample by decreasing the force P will result in retracing the stress-strain curve along the straight line, as elastic behavior demands. As the engineering stress continues to increase, the stress-strain curve deviates from linearity. The point at which the deviation from linearity begins is called the proportional limit and the corresponding stress is called the proportional limit stress σ pl . With a further increase in the stress, the point is reached at which permanent deformation of the sample begins, and the corresponding stress is called the yield strength (σy ). In the regime between the proportional limit σ pl and yield σy , unloading and reloading of the sample results in retracing the stress-strain curve along this nonlinear portion, which is a nonlinear elastic response. For most metallic materials this nonlinear elastic regime is negligible, and so one often identifies the proportional limit with yield. Note that by convention, a stress level that represents material behavior is referred to as a strength, as in the yield strength.

Engineering Stress

2 Fundamentals of Mechanics of Materials

45

σUT σY σ PL E

Engineering Strain True Strain 2000

0

0.01 0.02 0.03 0.04 0.05 0.06 0.07

Stress, MPa

Nanocrystalline Nickel

1500 1000 500

Engineering Stress True Stress

0

0

0.01 0.02 0.03 0.04 0.05 0.06 0.07

Engineering Strain Fig. 2.5 (a) Schematic of tensile stress-strain curve for ductile metal, showing yield, an ultimate tensile strength, and subsequent failure. (b) Experimentally obtained curve for nanocrystalline nickel, with an average grain size 20 nm. Note the unloading line after initial yield, with the unloading following an elastic slope. Data provided by Brian Schuster.

Once the applied stress has been taken beyond σy , permanent (plastic) deformation occurs, and unloading of the sample results in an unloading stress-strain response which does not follow the original stress-strain curve, but instead typically unloads along a straight line parallel to the linear elastic domain (this is called linear elastic unloading, as shown by the unloading line in Figure 2.5). The plastic deformation process after yield is called plastic flow. In practice, measuring the true onset of permanent deformation is not possible, and one instead identifies the yield strength by determining the point at which a permanent offset strain of 0.2% would be achieved. If loading is continued beyond the yield point, the stress-strain curve is nonlinear in two senses: first, the loading behavior does not follow a straight line response, and second, the unloading path is different from the loading path. If the stress in the sample increases after yield, the material is said to show hardening, while if the stress remains constant after yield the material is said to be perfectly plastic. Since the unloading is always elastic with a slope corresponding to the original Young’s modulus E, the permanent plastic strain left over after elastic unloading evolves

46

Nanomaterials

with increasing load. Even in a hardening material, continued loading results in a maximum in the engineering stress called the ultimate strength σut , where the subscript ut signifies that the strength referred to is the ultimate strength in tension. While there is a maximum in the engineering stress, the true stress (if it could be calculated) generally shows no maximum – what is happening is that the sample begins to develop a neck, with a region of the specimen becoming rapidly thinner and consequently with a rapidly decreasing current cross-section area A. Even though the load carried by the specimen decreases once necking begins, the current crosssection area A carrying the load is typically decreasing faster, so that the true stress continues to increase. The tensile stress-strain curve for the material thus provides two new material properties that describe the plastic response: the yield strength σy and some measure of the hardening of the material with increasing plastic strain. For example, a very common description of the tensile stress-strain curve has the power law form  n ε σ = σy (2.95) εy σ

where εy = Ey is the yield strain and the parameter n is called the strain hardening index and is a new material property. For the purposes of constitutive modeling the hardening index of the material should be defined in terms of the true stress vs true strain response, rather than the engineering stress vs engineering strain response. An example of real experimental data is presented in Figure 2.5b for a nanocrystalline nickel sample with an average grain size of 20 nm. The sample was tested using a microtension apparatus at a low strain rate of 10−4 s−1 . Both the engineering stress vs engineering strain curve and the true stress vs true strain curve are shown in the figure. Several points are worthy of note. First, the onset of yield is not immediately obvious – what should be chosen as the yield strength? The departure from linear elastic response is so smooth that yield is not clearly defined (the 0.2% onset condition would suggest that σY ≈ 1000 MPa). Second, the elastic modulus can be obtained directly by allowing a small amount of elastic unloading, as shown in the figure, and measuring the slope of the unloading line. In this case, a Young’s modulus of 160 GPa is obtained, quite close to the known bulk response for coarse-grained nickel. Third, as expected, the true stress is always greater than the engineering stress (in this tension experiment). Fourth, the material is hardening for substantial strains, up to about 6%. The necking regime and the ultimate tensile strength are not shown in this dataset.

2.4.2.1 Plastic Constitutive Relations Although Equation (2.95) is commonly used by materials scientists to describe the plastic behavior of metals, it does not provide a true constitutive relation in that it does not allow one to relate the general stress in the material (and the stress is a tensor) to the strain in the material at that point (the strain is also a tensor). The classical theory of plasticity provides a mechanism for translating the concepts presented

2 Fundamentals of Mechanics of Materials

47

in Equation (2.95) into a generalized constitutive equation that can be used to describe the behavior of materials under arbitrary loading conditions. The first step is to generalize the concept of a limiting stress (beyond which the material develops permanent or plastic deformations) to arbitrary stress states rather than the uniaxial stress state discussed above. In other words, one asks the question: when will a material deform permanently if it is loaded under a stress state that involves multiple stress components? The second step is to describe how the stresses and strains are related once plasticity begins. Here one immediately sees the need to distinguish between two parts of the total strain tensor: the elastic strain εiej (t) and the plastic strain εipj (t), since the former is recoverable on unloading while the latter accumulates with the deformation. The elastic strain components are always related to the stress components through the elastic moduli (e.g., through Equation 2.91). The generic problem in plasticity therefore reduces to the following: given σi j (t), determine the plastic strain εipj (t) that is developed, i.e., given the stress history, find the strain history. The simplest way to think about the first step is to visualize the problem in stress space. Recall that there are six independent stress components in the general stress tensor, but that we can solve the eigenvalue problem to find the three principal stresses and the three principal directions. If we assume that the material is isotropic, so that the behavior of the material is the same in all directions, then all we need to consider are the three principal stresses in order to decide whether or not the material has reached a limiting condition where permanent deformations begin. Imagine, therefore, a three-dimensional space in which the three principal stresses are plotted along the three axes of the space. This three-dimensional space is called the stress space (for the anisotropic material we would have to use a sixdimensional space corresponding to the six independent components of the stress tensor). Note that the stress tensor is defined at every point in the body, so that the three principal stresses are defined at every point in the current configuration of the body, and so there is a three-dimensional stress space associated with every point in physical space currently occupied by the body. The stress state at a point in the material is represented by the local principal stresses (σ1 , σ2 , σ3 ) which represent a point in the stress space. The stress history is the locus of stress states in stress-space (followed by a material point in the physical body). Now consider the origin in the stress space, corresponding to the case where σ1 = 0, σ2 = 0, σ3 = 0. This corresponds to an unstressed material. As we load the material, the various stress components in the stress tensor begin to increase, and so do the principal stresses. For example, the uniaxial tension experiment that we described earlier corresponds to the case where the only nonzero (and increasing) principal stress is (say) σ1 . For small stresses, our expectation is that the material will remain linear elastic, even though we have a multiaxial stress state, and Equation (2.65) will hold. Thus, there must be a region around the origin in stress-space where the material behavior is always linear elastic. What is the boundary of this region? This question is identical to the question of when permanent deformations will begin under multiaxial loading, i.e., when will yield begin, and we see that the answer to this question must represent a surface in this three-dimensional stress

48

Nanomaterials

space. This surface, bounding the elastic region in stress-space, is called the yield surface, and is analogous to the definition of σy in the preceding subsection. Indeed, by definition the intersection of this yield surface with the σ1 axis will be σy . Now consider the stress state at a point. If a single principal stress is increased sufficiently in magnitude, plastic deformation will occur. If more than one principal stress is changing, the onset of plasticity will depend on the relative magnitude of the principal stresses. To see this a little more physically, consider a thin sheet of material under biaxial tension (see Figure 2.6a), so that both σ1 = 0 and σ2 = 0 in the plane of the sheet, but σ3 = 0. The corresponding two-dimensional stress space is shown in Figure 2.6b. Consider first the case where σ2 = 0, which amounts to the simple tension problem; this case corresponds to moving out along the horizontal (σ1 ) axis in Figure 2.6b, and clearly yield will occur when σ1 = σy . If instead we consider the case where σ1 = 0, corresponding to uniaxial tension along the vertical direction, the material behavior should remain similar to that in the first case since the material is isotropic, and so yield will occur when σ2 = σy along the vertical (σ2 ) axis in Figure 2.6b. This much follows just from the simple tension test. σ σ σ

σ

σ

σ σ

σ

Fig. 2.6 (a) Illustration of the biaxial tension of a sheet. (b) 2D stress space corresponding to the sheet, showing the stress path and possible yield surface bounding the elastic region.

What happens if both σ1 = 0 and σ2 = 0? In this case the stress state will follow some arbitrary path in the σ1 − σ2 space of Figure 2.6b, for example the curved path shown. When will the onset of permanent deformations occur along this path? In general, the answer to this depends on the material and can only be determined by experiment; one possible solution is the arc shown in Figure 2.6b and defined as the yield surface (this is in fact the intersection of the full yield surface with the plane corresponding to σ3 = 0). The area within this arc and containing the origin is the elastic region in this stress space; the yield surface is the boundary of this elastic region. There are many possible shapes for this yield surface, and some of the common shapes are discussed later in this book. It is already clear that plasticity is a far more complex constitutive behavior than linear elasticity. However, some of the basic properties of yield surfaces can be determined simply by considering the physics of plasticity, considered next.

2 Fundamentals of Mechanics of Materials

49

Mathematically, the yield surface is defined by those stress states that lie on it, that is, the yield surface is defined by f (σ ,Yi ) = 0

(2.96)

where the Yi represents a set of material parameters (such as the yield strength in uniaxial tension). It is expected that a finite stress magnitude is required to produce plastic deformation, so that (2.97) f (0,Yi ) = 0, that is, the yield criterion is not satisfied when the stress tensor is zero. Further, for isotropic materials, the yield surface can only depend on the principal stresses, so that f (σ ,Yi ) = 0 =⇒ f (σ1 , σ2 , σ3 ,Yi ) = 0, (2.98) and this relationship describes the yield surface in our 3D stress space. So far we have used the concepts of the existence of an elastic region, and the isotropy of the material. Further determination of the form of this function f (σ ,Yi ) requires the use of the physics of plastic deformation in metals. Plastic deformation in crystalline metals occurs through the motion of line defects known as dislocations, and dislocation slip occurs largely due to shear stresses. Indeed, it is generally observed that crystalline metals do not yield under purely hydrostatic stress states. We thus conclude that we can replace the function f (σ ,Yi ) by a function f (S, Yi ), remembering (Equation 2.55) that the Cauchy stress tensor σ can be broken down into deviatoric and hydrostatic parts: σ = S − pI. Since p cannot cause yielding, the yield surface must be represented by f (S,Yi ) = 0,

(2.99)

and the concept of material isotropy then leads to f (S,Yi ) = 0 =⇒ f (S1 , S2 , S3 ,Yi ) = 0 =⇒ f (SI , SII , SIII ,Yi ) = 0,

(2.100)

where we have used the fact that the set of eigenvalues of S can be directly obtained from the three invariants SI , SII , SIII of the deviatoric stress tensor S. The three invariants SI , SII , SIII of the deviatoric stress tensor are conventionally called J1 , −J2 , and J3 respectively (note that J2 = −SII ). Since S is deviatoric, J1 = trace S = 0, and the yield surface can only be described by the yield function f (J2 , J3 , Yi ) = 0.

(2.101)

Thus, the assumption of dislocation driven plasticity and isotropy leads to Equation (2.101) describing the yield surface (appropriate for crystalline metals). However, what is the form of this function? Further reduction of the form of the yield surface requires experimental measurements, since a specific functional form depends on the specific deformation mechanism, and even crystalline metals show a wide variety of such mechanisms. The most common additional assumption is that

50

Nanomaterials

the dependence of the function on J3 = det S is negligible, and that there is only one material strength parameter Y that is needed to describe the behavior so that f (J2 , Y ) = 0,

(2.102)

where Y would be related, for example, to the yield strength observed in the simple uniaxial tension test. The simplest possible form of this function is f (J2 , Y ) ≡ J2 −Y = 0 =⇒ yield will occur when J2 = Y.

(2.103)

An equation such as Equation (2.103) describing the condition that must be satisfied for yield to occur (that is, for the onset of plastic deformation for an arbitrary stress path) is called a yield criterion. All yield criteria have a similar form, in that on the left-hand side of the criterion will be a function of the applied stresses, while on the right-hand side of the yield condition is a material property such as the yield strength (there are more complex forms of the yield criterion, but they are not discussed in this book). Yield occurs when the function of the applied stresses is equal to the material property. In this case, for example, we know that (using Equation 2.48) J2 = −SII = −

 1 (trace S)2 − trace S2 2

1 = trace S2 , because trace S = 0 2 1 = Si j Si j 2 1 2 2 2 2 2 2 + S22 + S33 + 2(S12 + S23 + S31 )} = {S11 2 1 = (S12 + S22 + S32 ) 2

(2.104) (2.105) (2.106) (2.107) (2.108)

where we have used the summation convention. In terms of the principal stresses (the eigenvalues of the Cauchy stress tensor), we have 1 1 J2 = {σ12 + σ22 + σ32 − (σ1 + σ2 + σ3 )2 } 2 3 1 2 = {(σ1 − σ2 ) + (σ2 − σ3 )2 + (σ3 − σ1 )2 }. 6

(2.109) (2.110)

The yield condition at Equation (2.103) can thus be written as 1 Si j Si j = Y, 2

(2.111)

1 {(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 } = Y. 6

(2.112)

or in terms of the principal stresses,

2 Fundamentals of Mechanics of Materials

51

What is Y ? Consider the onset of yield in the simple tension test, where only σ1 = 0. It follows from Equation (2.112) that at yield we have Y = 13 σy2 . It follows that for general loading as in Equation (2.112) we must still have 1 Y = σy2 , 3

(2.113)

where we remember that σy represents the yield strength in tension. Thus the full yield criterion for multiaxial loading of a crystalline metal would be 1 {(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 } = σy2 , 2

(2.114)

which essentially says that yield will occur when the combination of applied stress components on the left side reaches the value defined by the material property on the right side. This represents a surface called the yield surface in our 3D space of the principal stresses. This yield criterion is sometimes called the von Mises yield criterion, and appears to work very well for a large variety of metals. The quantity on the left-hand side of Equation (2.114) is often used to define a scalar measure of the applied stress, usually called the von Mises or effective stress σ¯ :  1 σ¯ = √ {(σ1 − σ2 )2 + (σ2 − σ3 )2 + (σ3 − σ1 )2 }, (2.115) 2 and in terms of this von Mises stress, the yield criterion can be very simply written as yield occurs when σ¯ = σy . (2.116) Note that in the two-dimensional case corresponding to the biaxial tension problem of Figure 2.6a, Equation (2.114) reduces to

σ12 − σ1 σ2 + σ22 = σy2 ,

(2.117)

which is the equation for the curve in Figure 2.6b. As the stress increases along any arbitrary stress path in Figure 2.6a, yield will occur when the path reaches the yield curve in Figure 2.6b. Equation (2.116) represents a surface called the yield surface in our 3D space of the principal stresses. It predicts, for example, that yield will occur in a torsion test when the shear stress in torsion τ satisfies 1 τ 2 = σy2 , 3

(2.118)

so that the yield strength τy observed in a simple shear or torsion experiment should be related to the yield strength in simple tension σy by 1 τy = √ σy . 3

(2.119)

52

Nanomaterials

This is experimentally observed in a very large number of engineering metals. Let us consider an example of plastic behavior in nanocrystalline metals. Suppose the yield strength in uniaxial tension of a nanocrystalline nickel material with a grain size of 20 nm (made by electrodeposition) has been measured to be 1.1 GPa (Figure 2.5). Then it follows that σy = 1.1 GPa. Let us assume that von Mises yield (Equation 2.116) describes the onset of plastic deformation in nanocrystalline nickel. This yield criterion would then predict that if we were to take a tube of nanocrystalline nickel and subject it to torsion, it would yield plastically when the applied shear stress reaches √13 σy which is 635 MPa. Unfortunately the yield stress of nanocrystalline nickel has not been measured in torsion as of this writing, so we do not know if von Mises yield is a good description of the behavior (even though it is known that conventional grain sized nickel follows von Mises yield behavior). We now have a relationship (Equation 2.114) that predicts the onset of plastic deformation, but we still do not have what we set out to obtain: a relationship between the total strain history and the stress history for a plastically deforming material. Remember that the yield surface is the boundary of the elastic region. Within the elastic region, we know how the stresses and strains are related (note that in the elastic region the history of the strain is irrelevant – the current strain determines the current stress). How are these quantities related once we reach the yield surface? This is a far more complicated problem, and we merely summarize the results here. The reader should refer to textbooks on plasticity for a detailed development of the theory. Since plasticity is a history dependent behavior, it is useful to think about changes in the stress and the strain rather than the current stress and strain. Consider the situation when the stress path has just reached the yield surface and we now have an additional increment δ σi j of the stress along the path. Our goal is to find the increment of the strain δ εi j that results from this increment in the stress. Assume that the stress increment is a loading increment, that is, the stress path is trying to move out of the elastic region. By definition, this stress increment will result in a permanent deformation of the material, that is, in a permanent plastic strain increment δ εipj . On the other hand, if the additional stress increment were to be removed, the material would recover a certain amount of elastic strain, so that there must be an elastic strain increment εiej also associated with the loading stress increment. It is therefore appropriate to view the total strain increment as having an elastic part and a plastic part, so that (2.120) δ εi j = δ εiej + δ εipj , with the additional proviso that the plastic strain increment δ εipj would only be developed if the increment in the stress is the loading increment rather than an unloading increment (all unloading is assumed to be linear elastic). The elastic strain increment can be related to the stress increment using the equations of linear elasticity (e.g., Equation 2.91):

δ εiej =

 ν 1+ν δ σi j − δ σkk δi j . E E

(2.121)

2 Fundamentals of Mechanics of Materials

53

If we can now find a relationship between the plastic strain increments and the stresses, our objective of relating the stresses and strains will have been achieved, at least in an incremental sense. First, it is important to recognize that the material behavior may or may not show hardening (usually called strain hardening). If the plastic behavior of the material is perfectly plastic, without any strain hardening, then it is not possible to apply a loading stress increment, and the plastic strain increment is indeterminate (in terms of material response – any number of deformed states can be attained after yield has occurred, limited only by the boundary conditions and the available energy). If the material behavior shows hardening after initial yield, then it is possible to determine the plastic strain increment uniquely for a loading increment in the stress. A textbook on plasticity will demonstrate that the plastic strain increment is related not so much to the stress increment as it is to the current deviatoric stress:

δ εipj =

1 δ σ¯ Si j , if and only if J2 = Y and δ σi j is a loading increment (2.122) 2 h(ε p ) σ¯

where σ¯ is the effective stress measure defined in Equation (2.115) and h(ε p ) is a function that describes how the material hardens with increasing plastic strain (this function is itself a material property, relating for example to the strain hardening index n). Note that Equation (2.122) also contains δ σ¯ , which is the scalar increment in σ¯ and which depends on δ σi j , but that individual components of the plastic strain increment depend only on the deviatoric stress components and not on the Cauchy stress components. Such a model is referred to as a J2 f low model, since it is derived from a J2 -based plasticity model and is written in an incremental strain form (the flow). It is important to recognize that δ εipj = 0 if the stress state is in the elastic region, or if δ σi j is an unloading increment. Putting Equations (2.120) and (2.122) together, we have

δ εi j =

 ν 1+ν 1 δ σ¯ Si j δ σi j − δ σkk δi j + . E E 2 h(ε p ) σ¯

(2.123)

This is the relation between the strain increments and the stress increments (and the stresses themselves, because of the history dependence). Together with the yield criterion (Equation 2.114), this incremental relationship defines the plastic constitutive response. This can be used to describe material behavior under multiaxial stress states along arbitrary stress paths, but must be solved incrementally, with a process of continually updating the stresses and strains as the nonlinear deformation develops.

2.4.3 Fracture Mechanics The concept of a stress concentration is a nearly intuitive concept in elasticity – it is apparent to most people, based on daily experience, that sudden changes in cross-section, notches and holes make it easier to cause the failure of materials

54

Nanomaterials

(imagine, for example, how easy it is to tear a sheet of paper that has already had a hole made in the middle). The concepts of fracture mechanics have similar roots, and are key concepts in structural design and failure analysis. The fundamental question of fracture mechanics is this: if a structure has a crack within it, and the structure is placed under load, will the crack grow? A secondary question is: if the crack does grow, how much will it grow, and how fast? We will address these questions in this section. The simplest form of fracture mechanics involves a crack in a linear elastic material (this is called Linear Elastic Fracture Mechanics or LEFM). Consider a sharp planar crack within a large block of material (Figure 2.7), loaded in tension in the x2 -direction at some distance. Define the coordinate system as in the figure. In real plates, the crack front (the locus of crack tips) is a curve rather than a straight line, and is sometimes called a thumbnail crack (Figure 2.8). This is because the surface of the block is essentially in a different stress state (close to plane stress) while the center of the block is closer to plane strain for a sufficiently large block, and cracks appear to grow more easily in plane strain. The crack front is the curve separating the cracked plane from the rest of the material (Figure 2.8).

Fig. 2.7 A sharp planar crack in a very large block of linear elastic material. The coordinate axes are defined so that the crack front (the line drawn by the crack tip) is the x3 or z axis. Note that all constant z planes look identical, so that this is essentially a plane strain problem in the x1 − x2 plane. The mechanics of this problem is dominated by the behavior as one approaches the crack tip.

2 Fundamentals of Mechanics of Materials

55

x

x

x x

Fig. 2.8 Two views of the crack in Figure 2.7, with (a) being the side view and (b) the top view looking down on the crack plane. The crack front is typically curved, but this can be neglected in the limit as one approaches the crack tip.

From a mechanics viewpoint, we would like to know all of the stresses, strains and displacements everywhere in the problem. However, this is more information than we perhaps need to answer the primary question in fracture mechanics. Far away from the crack tip, the material and structure do not feel the influence of the crack, and the stresses and so forth are identical to what one would have from a simple uniaxial tension problem. In fact, therefore, we only need to investigate what happens as we get close to the crack tip. This is a recurring theme in linear elastic fracture mechanics – the stress, strain and displacement fields as one approaches the crack tip are those of interest, and are indeed expected to dominate the problem (because the crack tip can be viewed as a stress concentration, and so the stresses should be highest there). Most of the results in LEFM are only valid in the limit as one approaches the crack tip (such solutions are called asymptotic solutions). Thinking of the crack tip as a stress concentration, note that the radius of the crack tip in the x1 − x2 plane is much smaller (essentially zero) than the radius of the crack front in the x1 − x3 plane (Figure 2.8b). Thus the dominant behavior as one approaches the crack tip is the behavior in the x1 − x2 plane. Another way to think of this is as follows: as one approaches the crack tip, all variations in the x3 direction are small in comparison to variations in the x1 and x2 directions, that is, none of the quantities of interest depend on x3 . Note that this implies that (as one approaches the crack tip) ε33 = 0, that is, the problem looks like a plane strain problem as one approaches the crack tip. For this reason, most measurements of the toughness of a material (the resistance to fracture) are performed in plane strain. As we approach the crack tip, therefore, we have (in terms of functional dependence) σi j = σi j (x1 , x2 ,t), εi j = εi j (x1 , x2 ,t) and ui = ui (x1 , x2 ,t). In quasistatic problems the dependence on time can also be dropped. Note that u3 = 0 in general. The problems in LEFM are typically separated into problems with u1 , u2 = 0, u3 = 0 and u1 , u2 = 0, u3 = 0, using the principle that for linear problems we can add the solutions for the two problems to obtain solutions to more general problems. The

56

Nanomaterials

case where u1 = 0, u2 = 0, u3 = 0 is called the plane problem, and the case where u1 = 0, u2 = 0, u3 = 0 is called the antiplane problem (this corresponds to the tearing mode in fracture problems). Let us define the crack front line as the x3 = z-axis. Note that symmetry requires that every x3 = constant plane will look the same in terms of stress and strain fields. Consider the plane problem (Figure 2.8a), so that the problem reduces to considering the stresses in the x1 − x2 or x − y plane. Thus the only stresses of interest are σxx , σxy , or σyy , because the σ33 = σzz can be computed directly from the linear elastic constitutive law, given that ε33 = 0. The crack tip can be defined as the origin in this 2-D space. A transformation to polar coordinates (r, θ ) from the Cartesian coordinates (x, y) has the crack tip at r = 0, and the crack faces at θ = ±π , and in this coordinate system the stresses of interest are σrr , σrθ , or σθ θ . In either case, distance from the crack tip can be defined as r = x2 + y2 . Far away from the crack tip, the influence of the crack is negligible, and the stress state will just be σyy = σ∞ , assuming that the far field tension σ∞ is in the x2 = y direction. As we approach the crack tip, the influence of the crack is increasingly felt, and the stresses at the crack tip itself are infinite (for a sharp crack – like a stress concentration of infinity). How rapidly do the stresses grow as one approaches the crack tip? A mathematical analysis of this problem, found in any number of books on mechanics (e.g., Asaro and Lubarda, 2006), shows that all of the stress components increase in the same way: KI σi j = √ fi j (θ ), r

(2.124)

where the angle-dependent function fi j (θ ) is the only difference between the three stresses in the plane. Thus all of the stresses increase as √1r , going to ∞ as r → 0. The derivation of Equation (2.124) uses a specific physical concept: that although the stresses at the crack tip can be infinite, the displacements and energies at the crack tip must be finite. The fact that the stresses are infinite at the crack tip is called a singularity, and the crack tip in the elastic material is said to have a √1r singularity (other singularities are possible, for example in viscous materials). The coefficient KI in Equation (2.124) is called the stress intensity factor, and scales the magnitude of the stress: it is a function of the applied loading and the geometry of the particular problem. Since the stress intensity factor scales all of the stresses, it is a measure of the effect of the applied load in the crack problem. The stress intensity factor KI is tabulated for a very large number of problems, so often one only needs to look it up for a specific geometry and loading condition (e.g., three-pt bending of a cracked beam). As the applied load increases, KI increases. The answer to the first question that was raised at the beginning of this section is that the crack will grow when the applied stress intensity factor KI reaches a critical value that is a material property called the plane strain fracture toughness KIc , i.e., the fracture criterion is KI = KIc .

(2.125)

2 Fundamentals of Mechanics of Materials

57

As with all criteria of this type, the left-hand side defines something related to the applied load and geometry, and the right-hand side defines a material property (recall the yield criterion σ = σY ). The plane strain fracture toughness is a material property that must be measured in specific experiments called fracture toughness experiments, in which specimens containing a precrack are loaded until the initi√ ation of crack growth is perceived. The units of fracture toughness are MPa m. The fracture toughness is a fundamental material property in terms of the application of materials to problems involving mechanical loading, since fracture is one of the most common failure modes for materials and structures, and tables of fracture toughness can be found in most materials and mechanical √ design textbooks. m, while for most 15 and 60 MPa In general, for most metals KIc lies between √ ceramics it lies between 1 and 6 MPa m. Materials with low values of fracture toughness are said to be brittle; most ceramics are brittle. Many nanocrystalline metals are believed to be more brittle than the conventional grain size metal, but there is little real experimental data on the fracture toughness of nanomaterials (largely because of the difficulty of doing valid fracture experiments on very small scale samples). The second question in fracture mechanics, that of how the crack will grow once it begins to grow, must be answered on an energetic basis. The primary concept here is that it takes energy to grow a crack, for instance in terms of the addition of energy required for the free surface corresponding to the crack faces. The amount of crack growth that would be observed depends on the amount of energy available to grow the crack. The energetic approach to fracture mechanics is a very powerful approach, resulting in the ability to solve fairly complex problems and also suggesting ways in which we can increase the toughness of materials.

2.5 Suggestions for Further Reading 1. A. Bedford and K.M. Liechti, Mechanics of Materials. Good if you have little or no background in mechanics. Prentice Hall, Upper Saddle River, NJ, 2000. 2. Y.C. Fung and P. Tong, Classical and Computational Solid Mechanics. World Scientific Publishing, Singapore, 2001. 3. J.R. Barber, Elasticity (Solid Mechanics and its Applications). Springer, New York, 2003. 4. R. Hill, The Mathematical Theory of Plasticity. Clarendon Press, Oxford, 1986. 5. M.F. Kanninen and C. Popelaar, Fundamentals of Fracture Mechanics. Butterworths, London, 1973.

2.6 Problems and Directions for Research 1. Which of the following expressions are meaningful in our notation? Write out each of the meaningful ones completely for a 3D space: aii , ai j b j , aii bi , ai j bi .

58

2. 3.

4.

5.

6.

7.

Nanomaterials

⎤ 1 0 −1 Given the matrix A = ⎣−1 2 0 ⎦, find the quantities Akk , Ai j Ai j , Ai j A ji . 3 0 −1 Consider the simple tension of a circular cylindrical bar of elastic/plastic material (6061-T6 aluminum alloy). Assume that the tension test is performed very rapidly at a nearly constant strain rate ε˙0 = 10−3 s−1 . Assume further that the material flows with no strain hardening at a stress σy = 500 MPa. Solve the 1D energy equation (Equation 2.59) to obtain the temperature distribution in the bar 500 s after the deformation commences, assuming no heat conduction through the grips, and ignoring convective and radiative losses through the circumfer ential surface. Note that the plastic work W p is given by W p = σy ε˙ dt. Use handbook data as necessary. Provide the temperature distribution in the form of a plot. Consider the plastic deformations of a thin sheet of metal deformed in biaxial extension. The problem is one in which the forces in the two orthogonal directions in the plane are varied independently. Apply the von Mises yield criterion, and sketch the yield surface in the 2D stress space. If the metal were to be made of a sheet of nanocrystalline aluminum 1 cm square and 100 µm thick with a yield strength in tension of 1 GPa, at what load would it yield under equibiaxial stretching? (Simple torsion) Equilibrium in torsion of a circular bar gives τ = Tr J , where τ is the shear stress in the bar, T is the torque, r is the radius, and J is the polar moment of inertia of the cross-section. The constitutive law then gives γ = rLθ  µτ , where γ is the shear strain, θ is the angle of twist in radians, and L is the length of the bar. Now consider a specific problem. A 1 m long steel tube with a diameter of 3 cm and a wall thickness of 1 mm is subjected to torsional loading until a total angle of twist of 5◦ is achieved. Compute the torque required to accomplish this, and then calculate the shear stress in the tube. Suppose the tube is simultaneously subjected to a tensile deformation, so that an axial elongation of 1 mm is obtained. Compute the axial force required to accomplish this elongation, and compute the resulting tensile stress in the tube along the axial direction. Sketch the 2D stress state on the surface of the tube after the combined tensile and torsional deformations. Now compute the principal stresses and principal directions (the directions in which the principal stresses act) for this stress state. The yield strength of a nanocrystalline aluminum sample is 700 MPa in tension. If the material’s behavior were to obey the Mises yield surface, at what shear stress would it yield in torsion? How about in compression? The surface of a 6061-T6 aluminum component of a machine tool is subjected to principal stresses of 200 MPa and 100 MPa. The third (out-of-plane) principal stress is zero. What tensile yield strength is required to provide a safety factor of 2 with respect to initial yielding? ⎡

2 Fundamentals of Mechanics of Materials

59

References Asaro, R. and V. Lubarda (2006). Mechanics of Solids and Materials. New York: Cambridge University Press. Ericksen, J. (1998). Introduction to the Thermodynamics of Solids. New York: Springer. Lai, W., D. Rubin, and E. Krempl (1999). Introduction to Continuum Mechanics. Burlington, MA: Butterworth Heinemann. Nye (1950). Physical Properties of Crystals. Oxford: Oxford Science.

Experimental science does not receive truth from superior science: she is the mistress and the other sciences are her servants. Roger Bacon

3

Nanoscale Mechanics and Materials: Experimental Techniques 3.1 Introduction One of the fundamental difficulties associated with the measurement of the mechanical properties of nanomaterials is that nanomaterial samples generally cannot be obtained in sufficiently large sizes to do traditional mechanical testing. From another perspective, one must use experimental techniques that can operate at scales that are relevant to the nanomaterial problem. For both of these reasons, there has been a drive towards the development of small-scale experimental techniques that can provide mechanical information on materials or structures. This chapter discusses a number of these techniques. Note that this is not a comprehensive discussion because these techniques are in rapid evolution, and the reader would be well advised to review the current literature in this area. We classify these techniques into two types, based on the objectives: 1. Techniques that attempt to identify or quantify a phenomenon at the nanoscale, which we refer to as nano mechanics techniques (NMT). Examples of such phenomena include grain boundary failure, dislocation motion, and adhesion of interfaces. 2. Techniques that attempt to measure the mechanical properties of materials using small sample sizes, which we refer to as nanoscale mechanical characterization techniques (NMCT). Examples of mechanical properties of interest include elastic modulus, yield strength, and fracture toughness. Both NMT and NMCT approaches have been developing rapidly since the 1990s. In practice, some of these techniques involve specimens that are microscale (10−7 to 10−4 m) in dimension, rather than nanoscale, because the observational resolution may well be nanoscale even when the overall specimen size is in the microscale regime.

K.T. Ramesh, Nanomaterials, DOI 10.1007/978-0-387-09783-1 3, c Springer Science+Business Media, LLC 2009 

61

62

Nanomaterials

3.2 NanoMechanics Techniques Understanding the behavior of nanomaterials involves understanding the phenomena that are active at the nanoscale, and this requires two steps: (i) identifying the phenomena and (ii) being able to model them. The identification of phenomena is the core objective of nanomechanics techniques. The phenomena themselves may be mechanics (physics)-dominated or chemistry-dominated, although often it is difficult to determine which parameters are the most relevant ones to measure. Examples of such phenomena include the mechanical behavior of grain boundaries in terms of the forces required to cause separation or sliding, the forces across interfaces (for example, the interfaces between two elements of a nanoelectromechanical systems or NEMS device), and the development of instabilities at the nanoscale such as the buckling of a carbon nanotube. Many of the advances in this area are fueled by the rapid growth in techniques that can examine atomic scale phenomena, particularly using scanning probe microscopy and transmission electron microscopy. What are the typical nanoscale phenomena that are relevant to the mechanical behavior of nanomaterials and the functioning of nanomaterials or nanostructures? Since the nature of nanomaterials and nanostructures implies the presence of boundaries at the nanoscale, the dominant question is the influence of such boundaries, whether these are interfaces between entities or simply surface features. For example, the behavior of grain boundaries is critical to understanding the mechanical behavior of nanomaterials, while the behavior of the surface of a cantilever beam can dominate the sensitivity of a cantilever sensor. Thus most nanomechanics techniques focus on the observation of boundaries and interfaces. Measuring such boundary phenomena requires that one be able to observe at the appropriate scale as well as be able to quantify at that scale. The classic questions of precision, accuracy and uncertainty are important for such measurements. It is often the case that we can observe phenomena at smaller scales than the scales at which we can accurately quantify those phenomena (the nanoscience version of “one’s reach is often beyond one’s grasp”). A more fundamental question is whether one can be certain that one is measuring the phenomenon under the same conditions each time the measurement is made. For example, we cannot obtain two identical grain boundaries for examination, because it is difficult to prepare two identical grain boundaries. Because of the difficulty of preparing samples, we observe what we can get as specimens, even though there is a very large variety of boundaries that should be characterized. However, the number of phenomena at the nanoscale may be very large, and our technology is just beginning to grasp the variety of features present here, even as foretold by Feynman, when he said that “there’s plenty of room at the bottom” (Feynman, 1959). In this regard, our difficulties in identifying and quantifying nanoscale phenomena have strong similarities to the travails of astronomers who are trying to identify the behavior of stars and understand the stellar lifecycle. Since we cannot make stars, we have to build our understanding on the basis of those that we can observe.

3 Nanoscale Mechanics and Materials: Experimental Techniques

63

Just so, in the nanomaterials world, we have to build understanding on the basis of the boundaries that we can observe. Note that the richness of phenomena, and the difficulty of control, arise for exactly the same reasons in the two cases of astronomy and nanotechnology: a vast difference of scale from the human condition. We must turn the difficulty that we are presented with (the variety of structures and phenomena available) into an opportunity, even as the astronomers do, and use these riches to our advantage. An analogy might be useful. Imagine an alien spacecraft arrives at the Earth, and the spaceship commander wishes to obtain an understanding of the human lifecycle. She could indeed choose to observe a human evolve from an infant to a full-grown adult and then to old age, but this would take more than 70 years. On the other hand, if she chose to examine a very large cross-section of the population, she would observe every stage of the human lifecycle in a very short time and would be able to obtain a fairly good understanding of the process (although certainly she would see variations based on race, economic status, and environment). This is what astronomers do with stars, and what nanotechnologists need to do with phenomena in nanostructures. We cannot control all of the nanostructures, but we can observe a large variety of them and try to extract an understanding of the phenomenon of interest. The difficulty would be identifying the persistent features across a spectrum of nanostructures (comparable to identifying the common aging processes across a spectrum of races and environmental conditions). As an example of this approach, consider understanding the mechanical behavior of grain boundaries in a nanocrystalline material. Since the grain boundary is the boundary between two crystals, there is a very large variety of boundaries that are possible (depending upon the relative orientations of the two crystals). It is not physically feasible to prepare a sample of every boundary with a given relative orientation of the two crystals. However, in any nanocrystalline material, it is likely that we will find representative examples of many of these possible grain boundaries. Thus by examining the population of grain boundaries in a given nanocrystalline material, and learning what their mechanical behaviors have in common, we may obtain an understanding of the likely mechanical behavior of all grain boundaries. This approach inherently requires us to use both experimental and modeling techniques simultaneously to study these phenomena, so that we can extract the persistent behavior across a variety of examples (as in grain boundaries). With that insight, let us consider NanoMechanics Techniques. The two major experimental techniques that comprise NMT are scanning probe microscopy and in situ deformation. The latter may be performed within an AFM (atomic force microscope), a TEM (transmission electron microscope) or an SEM (scanning electron microscope). Each of these techniques is described in the following sections. Note that these are complex techniques that are comprehensively discussed in the literature, so the descriptions here are brief. Other techniques such as optical tweezers and magnetic tweezer arrangements are more appropriate for biological problems, and are not discussed here.

64

Nanomaterials

3.3 Characterizing Nanomaterials The characterization of nanomaterials is no easy task. By definition, the feature sizes that must be observed are very small, being less than half the wavelength of visible light. This makes it impossible to see these features using traditional optical microscopy. The primary characterization techniques that are used are therefore scanning electron microscopy, X-ray diffraction, and transmission electron microscopy. Of these three, the highest spatial resolution is obtained with transmission electron microscopy or TEM, and this is the final arbiter of whether a specific dimension has been achieved and is necessary for the determination of specific deformation mechanisms. However, transmission electron microscopy is a difficult thing to do, and specimen preparation in particular is a laborious task requiring substantial skill. As a consequence, many researchers in the field use X-ray diffraction to obtain a quick estimate of grain size. Unfortunately X-ray diffraction is unable to distinguish between low angle grain boundaries and high angle grain boundaries, leading to some confusion in the literature. The scanning electron microscope is sufficient for the observation of many device nanomaterials, and to examine failure mechanisms (often a question when reliability of devices is an issue). When surface examination is sufficient, various forms of scanning probe microscopy may be appropriate. The atomic force microscope is often used to provide images of surfaces with sufficient resolution. In any case, the use of nanomaterials implies the ability to characterize at the nanoscale, and a substantial investment has to be made in the infrastructure required to observe at these scales. Each of these techniques is described very briefly below; the reader should examine the suggested reading to obtain more detail.

3.3.1 Scanning Electron Microscopy or SEM A scanning electron microscope generates an image by scanning a surface with an electron beam. The electron beam is usually generated from a tungsten filament, and the energy associated with the electrons can be controlled. The beam itself can be manipulated using electron optics, analogous to traditional optics (lenses and so forth). The interaction of the electron beam with the surface results in the interrogation of a certain depth of material below the surface. Electrons are scattered off the material being investigated through both elastic and inelastic scattering, and by collecting the scattered electrons one can obtain images of the surface (in practice, the scattered electrons themselves interact with some kind of display, generating intensity variations that provide the image shown in typical scanning electron micrographs). In the most common mode of operation, the scattered electrons that are collected are those that have been inelastically scattered, and this interrogates a very small depth below the surface (several nanometers). Most SEM images thus provide surface topography information, and resolutions down to a nanometer are possible. In comparison to optical microscopy, scanning electron microscopy is capable of much greater magnification and resolution, and at the same time retains

3 Nanoscale Mechanics and Materials: Experimental Techniques

65

great depth of field (in light microscopy, high magnification usually results in low depth of field). Scanning electron microscopy is therefore a great way of looking at surface information on nanomaterials (the depth of field is particularly useful when looking at fracture surfaces). If the scattered electrons that are collected are the elastically scattered electrons (so-called back-scattered electrons), then the image that results contains a significant amount of information on the chemical composition of the surface, because different materials provide different degrees of scattering in this mode (higher atomic numbers – that is, heavier atoms – scatter more). When used in combination with some form of spectral analysis, a great deal of information on chemical composition can be obtained.

3.3.2 Transmission Electron Microscopy or TEM Unlike the SEM, the transmission electron microscope uses the transmission of an electron beam through a sample to characterize the sample. The sample must be thin enough for transmission to occur, and so a good deal of specimen preparation is typically necessary. This technique is capable of much higher resolution, and in some versions is capable of examining individual atomic layers. Since the beam goes through the sample, the interaction of the beam with the sample is much more complex, and this can be used to advantage in generating various imaging modes. Some of these modes include the ability to examine the diffraction of the beam through the crystal lattice, and to select specific diffraction spots for imaging purposes. This combined with the ability to tilt the sample in the beam makes it possible to identify individual crystal structures, to see and identify the character of dislocations in the crystals, and to examine twin structures. TEM micrographs are presented throughout this book, together with selected area diffraction (SAD) patterns that provide information on the crystal structures interacting with the beam. SAD patterns display the diffraction pattern that results from the interaction of the electron beam with some volume of the sample. In the specific case of the beam interacting with only one crystal, the SAD pattern is a pattern of dots that represents the crystal structure (actually the Fourier transform of the crystal structure). When the beam interacts with a nanocrystalline material, very many grains are averaged over, resulting in patterns that are more ring-like than spot-like. The technique of electron energy loss spectroscopy or EELS can be used to provide composition information, and the combination of scanning with transmission modes in a scanning transmission electron microscope or STEM can provide very useful information on the location of specific elements at specific regions in the material. In high resolution electron microscopy or HREM, individual atomic layers can be examined, the locations of atoms can be rigorously interrogated, and grain boundaries can be examined individually. The TEM is without question the definitive instrument for use with nanomaterials, since it can examine the interior of materials with the appropriate resolution.

66

Nanomaterials

The difficulty with the use of the TEM is twofold: first, specimen preparation is tedious and exacting and can change the structure being observed if not done properly, and second, the use of the TEM and interpretation of the TEM images requires a very high degree of expertise. As instruments go, TEMs are expensive to maintain and operate (much more so than SEMs).

3.3.3 X-Ray Diffraction or XRD X-rays interact with crystalline structures through diffraction, and X-rays are relatively easily generated, so X-ray diffraction is the easiest way to examine crystal structures. A large variety of specialized techniques have been developed based on X-ray diffraction, and careful analysis can reveal information even at the level of the locations of nearest neighbor atoms in a material. Resolution can thus be very high, particularly using high-energy beamlines such as those obtained at synchrotrons. Imaging is not usually the objective, but information can be obtained on grain size (although low-angle and high-angle grain boundaries are hard to separate), elastic strain in the crystal, and composition.

3.3.4 Scanning Probe Microscopy Techniques Scanning Probe Microscopy (SPM) is a generic description for a class of techniques where a needle tip (called the probe) is scanned across the surface of the sample, and the interactions between the probe tip and the sample surface are used to generate an image that represents some characteristic of the surface. In many SPM modes, the probe (Figure 3.1) never actually touches the surface of the sample, instead being held a fixed distance (∼1 nm) away from the sample through a feedback circuit. The interactions between the tip and the surface are extremely small in scale, including atomic level interactions. Indeed, it is possible to scan across the surface of the sample and image every atom on the surface. It is also possible to image at the micrometer scale. These techniques are thus extremely powerful, and have revolutionized the study of surface physics and surface chemistry; Gert Binnig and Heinrich Rohrer won the Nobel Prize in Physics in 1986 for inventing the technique in 1981 (Binnig et al., 1983). Such systems can be used both to image and to manipulate atoms on the surface. A very large variety of interactions are possible between the probe and the surface. These include electron tunneling interactions–leading to scanning tunneling microscopy or STM (Binnig and Rohrer, 1985); contact forces–leading to atomic force microscopy or AFM (Binnig et al., 1986); and magnetic interactions–leading to magnetic force microscopy or MFM (Martin and Wickramasinghe, 1987). A variety of technologies have developed that utilize the extreme sensitivity of the measurements to these interactions. The particular interaction that is measured using

3 Nanoscale Mechanics and Materials: Experimental Techniques

67

Laser Diode Source

Piezoelectric Scanning Actuators (x,y,z) To Position Sensitive Photodetector Cantilever

Sample Surface Fig. 3.1 Schematic of a Scanning Probe Microscopy system (not showing the feedback loop, tunneling amplifiers and other electronics). A wide variety of such systems exist, some of which contain only a subset of the features shown here.

the scanning probe system determines the characteristic of the surface that will be imaged and the name of the kind of microscopy that is being used. There is a large variety of commercial versions of scanning probe microscopy instruments on the market, and many online resources that provide detailed descriptions of these instruments. There are four critical components to scanning probe microscopy techniques: • First, the nature of the probe tip determines the resolution of interaction. • Second, the ability to scan in a controllable way over very small distances with high-resolution determines the spatial resolution of the technique (this is usually done with a piezoelectric system). • Third, a feedback system is usually used to maintain the probe a fixed distance away from the surface being scanned, and there is a very important relationship (which can be designed to some degree) between the feedback control and the specific quantity being measured. • Fourth, the ability to develop images that visualize the distribution of these interactions across the surface of the sample through appropriate software is critical to the practical use of these techniques. The interpretation of these images often requires substantial expertise, and may require a real understanding of the physics of the interactions. These are very powerful techniques, but come with a strong caveat. The ease with which an image is obtained using a commercial SPM instrument may hide very

68

Nanomaterials

complex physics that might be very important to the issue of interest, because the software that provides the image may make assumptions that are hidden from the viewer. It is critical to understand (a) what the hardware is capable of doing and (b) what the software assumes in generating the image (for example, what are the defaults in the parameters used by the software). Examples of applications of scanning probe microscopy that are of interest in this book include the examination of grain boundary character at the atomic scale, the determination of the development and motion of ledges on crystal surfaces, the examination of compositional variation close to boundaries, and the characterization and fabrication of nanomaterial surfaces.

3.3.5 Atomic Force Microscopy or AFM Atomic force microscopy (AFM) is a very high-resolution SPM technique for examining the surfaces of materials. The image is formed by scanning a probe over the surface and using the interaction of the probe with the atoms on the surface to generate an image. Many different kinds of interactions can be examined, and images can be generated of surface topography, surface magnetic character, and so on. This technique is often used with discrete nanomaterials and thin films, and less so with bulk nanocrystalline materials.

3.3.6 In situ Deformation In situ deformations are defined as deformations that are developed while the sample is within an imaging environment. The ability to image during the deformation allows us to determine the mechanisms that are active and to examine the deformation processes at the microscale. The development of an in situ deformation technique requires two components: • First, an arrangement that allows one to develop controlled deformations within the microscope • Second, an imaging capability through microscopy (in order to be able to examine the length scales of interest) From a mechanics viewpoint, it is useful to be able to measure the forces within the system at the same time as the deformations are being imaged through the microscope. This can be very hard to do at nanometer scales, because the forces that are developed can be in the nano-Newton range. The primary in situ deformation techniques that are available are accomplished within AFM, TEM, and SEM systems. The major difficulty in in situ deformation techniques is that it is difficult if not impossible to obtain samples of identical structure; one has to look at what one has

3 Nanoscale Mechanics and Materials: Experimental Techniques

69

available, rather than being able to examine model structures for a given material (such as identical grain boundaries).

3.3.6.1 In Situ Deformation Within an Atomic Force Microscope Atomic force microscopy involves the measurement of the forces involved in the scanning of a probe along the surface of a material (often including contact). The probe tip is often made from single crystal silicon. Probe tip radii of curvature can be of the order of 1 nm. Typically, the probe is carried on a cantilever beam called a microcantilever. The force of interaction between the probe tip and the surface results in the bending of the microcantilever, and the forces are measured by recording the bending of the beam using an optical arrangement. The forces that are measured might be contact forces, van der Waals forces, or electrostatic forces. In terms of spatial resolution, this technique often does not have the spatial resolution of STM but is much more robust in operation. A common approach to in situ deformation within the atomic force microscope involves the use of thin films of material that are deformed using a mechanical arrangement (Nishino et al., 2000). The large majority of such experiments do not record the forces within the sample during the deformation. However, recent approaches to mechanical loading using a MEMS type device show significant promise (Chasiotis et al., 2004). Images of the deformed sample can then be correlated to measures of the current load to estimate the relationship between the stress state and the deformation. Note that the observation of deformation processes using imaging may be worthwhile even without load measures. For example, the evolution of surface features in a nanocomposite can be imaged effectively using this technique, as can grain boundary migration.

3.3.6.2 In Situ Deformation Within a Transmission Electron Microscope The Transmission Electron Microscope (TEM), as the name implies, is an instrument that uses the transmission of electrons through a material to determine the structure of the material. An electron beam is accelerated using a high voltage and is passed through a sample; an image is constructed on the basis of the electrons that have been transmitted through the sample and that strike a detector placed on the opposite side of the sample. The most common application of the TEM is to provide microstructural characterization of crystalline materials. The interaction of the electron beam with the crystal can be studied by using the principles of diffraction, and highly sophisticated techniques have been developed to infer the structure of the material on the basis of the distributions of the transmitted electrons. Among the characteristics of the material that can be inferred are the local grain size, the crystal structure, and the defect density, in addition to the characterization of boundaries and identification of composition. At nanometer length scales, the TEM is often believed to provide “ground truth” in terms of materials characterization. The primary difficulty with the use of the TEM is that the samples must be made electron

70

Nanomaterials

transparent, and the production of good TEM specimens is essentially an art that requires a great deal of experience. Most TEM specimens are disks about 3 mm in diameter with a central dimple containing a region of material that has been thinned to about 100 nm in thickness. The process of generating a good TEM specimen may itself (if done incorrectly) result in substantial modification of the structure that is being examined. With these caveats, there is no substitute for high-quality transmission electron microscopy work in the nanocrystalline materials area for determining the material characteristics. In situ deformation experiments within a TEM are also complicated by the need to develop an electron transparent region within the sample. This essentially limits the in situ samples to thin films. There are a number of commercially available testing stages that can be mounted within TEM vacuum chambers for such experiments, but the development of an effective technique requires substantial effort to ensure that the technique is optimized to the specific microscopy facility. A common way to develop local strains in the sample is to use a thermal misfit rather than to apply controlled mechanical loads. Again, recent developments in MEMS based mechanical testing techniques show strong promise in this area (Haque and Saif, 2004). Some instruments have been developed that accomplish compression and tension within a TEM using a combination of focused ion beam machining to generate the specimens and in situ nanoindentor based loading techniques (Minor et al., 2001; Stach et al., 2001), and these are just beginning to have a significant impact on our understanding of deformation mechanisms at the nanoscale.

3.3.6.3 In Situ Deformation Within a Scanning Electron Microscope The scanning electron microscope (SEM) is an imaging instrument that also uses an electron beam, but generates an image by looking at the electrons that are scattered off a surface. The electron beam is scanned across the surface, generating images of the surface over a very wide range of magnifications. This is the premier technique for examining material surfaces in materials science and engineering. One of the major advantages of this technique is that the depth of field is considerable, so that it is possible to look at very rough surfaces such as fracture surfaces. Unlike the TEM, the SEM can be used by a relatively unskilled operator and specimen preparation is not a major problem except for some special cases. The spatial resolution of SEM systems has improved dramatically over the last two decades, and nanometer resolution is now possible. Further, it is possible to identify crystal orientations using SEM systems in Orientation Imaging Microscopy mode (this is very useful for examining polycrystalline systems). In situ deformation experiments within an SEM have a particular attraction because of the ease of use of the instrument, the wide range of magnifications that are possible, and the substantial depth of field. The coupling of these capabilities with the recently developed digital image correlation (DIC) technique for determining displacements and strains from multiple images has made this a very useful technique for in situ deformation studies. Commercial in situ testing stages for the SEM

3 Nanoscale Mechanics and Materials: Experimental Techniques

71

Fig. 3.2 Single crystal copper specimen and tension gripper for in situ tensile testing in the SEM, from the work of the group of Dehm, Kiener et al. (this group has also performed compression experiments using a similar apparatus) (a) SEM image showing a single-crystal copper tension sample and the corresponding tungsten sample gripper before the test at a low magnification. (b) Sample and gripper aligned prior to loading. Image taken from (Kiener et al., 2008). Reprinted from Acta Materialia, Vol 56, Issue 3, P. 13, D. Kiener, W. Grosinger, G. Dehm, R. Pippan, A Further step towards an understanding of size-dependent crystal plasticity: In situ tension experiments of miniaturized single-crystal copper samples. Feb. 2008, with permission from Elsevier.

are now available, and are easily fitted to specific facilities. These techniques have been used, for example, to understand the size-dependence of plastic behavior in metal single crystals (Kiener et al., 2008). An example of the specimen size and shape is presented in Figure 3.2 from the work of Kiener et al. (2008).

3.4 Nanoscale Mechanical Characterization 3.4.1 Sample and Specimen Fabrication The objective of a mechanical characterization technique is to measure the behavior of the material. However, what we in fact measure is the behavior of a sample of the material, and we then infer material behavior from sample behavior. This works very well if we (a) understand the relation between sample behavior and material behavior and (b) have not changed the material response while trying to make the samples that we will test. An example might be useful. Suppose one wished to measure the tensile strength of a relatively brittle material like pure tungsten, and so one performs a standard tensile test on dogbone-shaped samples (Figure 2.4) of the tungsten. Suppose the dogbone-shaped samples are made by machining from a larger tungsten block using a milling machine, but one set of samples is tested immediately after milling, while the other set of samples is subject to a finish grinding operation (after milling) before the tensile tests are performed. A very interesting behavior is observed. The strength of the material measured from the finish-ground samples is found to be much higher than the strength of the material measured from the as-milled samples (a similar behavior has been shown for compression; Lennon and Ramesh, 2000). This is because the as-milled samples have small surface cracks generated during the milling process, and these cause premature failure

72

Nanomaterials

of the sample. The finish-ground samples do not have these surface cracks, because the top layer of damaged material has been ground away, and these samples sustain higher strengths (and higher strains to failure). Thus sample fabrication processes can be critical. Note that this is an issue that goes above and beyond the issue of the generation of high-quality material, with few defects and reproducible microstructures. Sample preparation can be a particular problem for nanocrystalline materials, which are typically high-strength and less tolerant of defects introduced by fabrication processes. Sample fabrication processes may include various kinds of machining, extrusion, and thermal cycling, and the potential effects of all of these processes must be considered.

3.4.2 Nanoindentation The mechanics of the indentation of one solid into another have been known for many years. The first technical discussions of such matters examined the scratch testing of materials, asking the question whether one material would scratch another. This led to the famous Moh’s scale of hardness (diamond was ranked as the hardest material). The mathematical solution of the problem of the elastic indentation of one material by another was provided by Hertz, who calculated the stress state beneath an indentor. This mathematical solution is one of the fundamental solutions in contact mechanics (Johnson, 1987), and describes the relationship between the applied force on the indentor P and the depth δ of the indentation: P=

16 9

1 (

1−νi2 Ei

+

2 1−νs2 Es )

Ri Rs 3/2 δ Ri + Rs

(3.1)

where the subscripts i and s refer to the indenter and the sample respectively (e.g., Ri and Rs are the radii of curvature of the indenter and sample). Note that the applied force depends on the displacement to the 3/2 power, and so that relationship would be expected for elastic behavior. If one performs an indentation experiment using a relatively sharp indentor against a flat sample surface (so that Rs is very large), measuring both the force and the displacement of the indentor tip, and then plots the force against δ 3/2 , a straight line should be obtained, and the slope of that line can be used to compute Es using Equation (3.1) if one knows the indentor radius and indentor modulus Ei . If the indented material can develop permanent deformations (i.e., has elasticplastic behavior), then a permanent indentation will remain after the removal of the load from the indentor tip. Extending the Hertz contact analysis to the elastic plastic case, the development of these permanent indentations within materials can be described mathematically (although the description can be quite complicated if the plastic behavior of the material deviates from simple classical plasticity such as described in the chapter on mechanics). For the simple classical plasticity case, the size of the remaining indent depends on the strength of the material. This has led

3 Nanoscale Mechanics and Materials: Experimental Techniques

73

to a very common testing technique in materials science called Vickers indentation where the indentor has a specific pyramidal geometry, and Vickers hardness is a common way of estimating the strength of the material. Other indentor shapes can also be used, resulting in different hardness measures such as the Knoop hardness and Brinell hardness. Oliver and Pharr (1992) extended this approach to the problem of very smallscale indentation, calling this technique nanoindentation. The primary characteristic of nanoindentation is that both the force on the indentor and the displacement of the indentor tip are measured simultaneously with very great accuracy, and very small forces and very small displacements can be measured. A force-displacement curve is obtained for each indent. If the indentor is made out of an extremely stiff material such as diamond, the indentation force-displacement curve is dominated by the response of the material being indented. An example of such a curve is presented in Figure 3.4, which shows the result of a nanoindentation experiment on nanocrystalline nickel. Commercial nanoindentation instruments are now widely available. The coupling of this technique to software for automated indentation makes it possible to develop hardness maps that contain substantial information about the mechanical characteristics of materials and structures at the small scales of interest to nanomechanics and nanomaterials. The device that provides a controlled indentation while simultaneously measuring both indentor force and indentor displacement for very small indents (submicron) is called a nanoindenter. A schematic of such a device is shown in Figure 3.3, indicating the critical components. Current commercial nanoindenters have a displacement resolution better than 0.1 A, and a load resolution of the order of 1 nano-Newton. This high precision gives the technique popular application in

Load Frame Actuator Force Sensor

Position Sensor Indenter Tip Sample Fig. 3.3 Schematic of the NanoIndenter, showing both actuation and force and position sensing. Such a device may fit comfortably on a large table top, although vibration isolation may be desirable.

74

Nanomaterials

many areas of materials science and engineering, such as the measurement of mechanical properties of very thin films (thickness < 100 nm), and tribological measurements of coatings. In a conventional nanoindentation experiment, a Berkovich (tetrahedral) tip is usually used to probe the nano-scale mechanical behavior (such as hardness and elastic modulus) as a function of indentation depth. For example, the indentation size has a strong effect on the hardness of single crystal silver (Pharr and Oliver, 1989). The typical output is a force-displacement curve that may deviate from the form presented in Equation (3.1), as shown in Figure 3.4 for nanocrystalline nickel. The primary mechanical properties extracted from a nanoindentation test are the hardness and the elastic modulus. This is extremely useful information, particularly for nanocrystalline materials. However, the nanoindentation test cannot provide a full description of the mechanical behavior of the material in terms of the stress-strain curve, and such a description is needed in order to be able to use a material within a specific application. The microcompression and microtension tests described next can provide such a constitutive description.

3.4.3 Microcompression Microcompression is an experimental technique where a modified indenter tip geometry is utilized within a nanoindenter to probe the nano-scale mechanical behavior of micro-scale or sub-micron pillars (Uchic et al., 2004). In this modification,

300

Nanocrystalline Ni, 20 nm 250

Load, mN

200

150

100

50

0 0

500

1000

1500

2000

Displacement Into Surface, nm Fig. 3.4 Example of force-displacement curve obtained during the nanoindentation of a nanocrystalline nickel material with an average grain size of 20 nm. Data provided by Brian Schuster.

3 Nanoscale Mechanics and Materials: Experimental Techniques

75

the sharp Berkovich tip is truncated, resulting in a flat end tip which enables one to convert the indentation system into a compression system. Since both the force on the indentor tip and the displacement of the indentor tip can be measured, using this flat tip to compress the top of a cylindrical sample allows one to estimate the stress strain behavior of the material of the sample. This technique is commonly called “micro-compression” or µ -compression. This technique is of great use for probing the mechanical behavior of nanomaterials which are difficult to obtain in bulk form, and whose tensile ductility is limited (which makes micro-tension tests difficult). It is currently difficult to produce bulk forms of many metals and ceramics with nanocrystalline microstructure, but they can be readily produced in thin foils and small cubes or cylinders with dimensions less than 1 mm. Conventional mechanical testing techniques are not useful due to the small size of specimens. Micro-compression makes use of a conventional nano-indenter and a flat-end tip to measure the stress-strain curves of materials using posts with a dimension as small as 250 nm in diameter (Schuster et al., 2007). A schematic of a cylindrical post sitting on a base of the same material is shown in Figure 3.5a. Technologies such as focused ion beam (FIB) enable researchers to fabricate such microposts (Figure 3.5b) with ease. These posts usually have one end (the bottom) fixed on the matrix (or the base material), while the top end is pressed by the flat-end tip. In a conventional uniaxial compression test, both ends of the specimen are free to deform in a nearly frictionless manner. When can the data from microcompression measurements be used to represent the bulk materials behavior? How sensitive is the test to variations in material behavior? Finite element simulations can be used to determine the errors resulting from the compliant base condition, and to provide guidelines for the proper design of microcompression experiments (Zhang et al., 2006). Consider first an axisymmetric finite element model (Figure 3.6a) of the micropost. A velocity boundary condition

Fig. 3.5 (a) Schematic of a microcompression experiment, showing micropillar and flat-bottomed indenter tip. (b) SEM micrograph showing a micropillar of a PdNiP metallic glass produced by focused ion beam machining.

76

Nanomaterials

(a)

(b)

Fig. 3.6 (a) Finite element mesh used for 2D axisymmetric model of the micropillar in a microcompression experiment. (b) Computed effective (von Mises) stress distribution in a sample after plastic compression. Note the nonuniformity at the root of the pillar, indicating the importance of the root radius (also called the fillet radius). Reprinted from Scripta Materialia, Vol 54, Issue 2, page 6, H. Zhang, B.E. Schuster, Q. Wei and K.T. Ramesh, The design of accurate microcompression experiments, January 2006, with permission from Elsevier.

corresponding to the imposed strain rate is applied to the post top, and the bottom of the base is not allowed to move in the x2 -direction as shown in the figure. The true stress-strain behavior of the post can be obtained from the reaction force P generated on the post top and the displacement u of the post top using the following equations:

σ¯ 22 =

1 π r2

 r 0

ε¯22 = ln(1 +

P2 |x2 =h dx1

u2 ) h

(3.2) (3.3)

where the subscripts 2 indicate the x2 components of the corresponding vector, r is the radius of the post, h is the height of the post with respect to the base, and σ22 and ε22 are the stress and strain components that would be appropriate for a uniaxial compression experiment (note that in the microcompression experiment σ22 will not be the only non-zero stress component). These equations are essentially those used in reducing experimental data. In the finite element simulations, the material of the post is assumed to have the behavior described by the Ramberg-Osgood equation

ε¯ =

σ σ0 σ + β ( )( )N E E σ0

(3.4)

where E is the Young’s modulus, N the inverse of the traditional strain hardening exponent (see Equation 2.95), β is a material constant and σ0 is a reference stress (the Poisson’s ratio ν is also needed to complete the description of the linear elastic behavior). For the calculations shown here, the following parameter set is used,

3 Nanoscale Mechanics and Materials: Experimental Techniques

77

corresponding to an aluminum (Al) alloy: E = 70 GPa, ν = 0.33, σ0 = 65 MPa, β = 2.69, and N = 4.18. The question of interest is: under what conditions would the calculated σ22 and ε22 have the same relationship as the input material behavior given by Equation (3.4)? Those conditions would represent the design of an appropriate microcompression experiment, since the computed effective behavior would be identical to the assumed material behavior. 3.4.3.1 Post Geometry Effects h ) on the output The effects of the fillet radius rc and the post aspect ratio (α = 2r stress-strain curves are considered first. We begin by fixing α = 2 and varying rc (expressed in terms of the post radius r). Figure 3.7 displays the input stress-strain curve from Equation (3.5), along with the output stress-strain curves from the simulations with different fillet radii rc = 0.1r, 0.2r, 0.5r, and r. For clarity, a portion of the curves has been enlarged in the inset of Figure 3.7. Due to the fixed end effect, all the simulated output curves are above the input curve, indicating an overestimate

Fig. 3.7 Input and simulated stress-strain curves for various assumed root radii (detailed differences observable in the inset) as computed from the finite element model presented in the last figure. The “input” curve is the input material behavior used in the finite element simulations, and the “output” stress-strain curves are obtained from the forces and displacements computed from the simulations and then processed in the same way as the force and displacement data is processed in the experiments. If the experimental design were to be perfect, the output curves would be identical to the input curve. Note that the simulated curves are all above the input data, demonstrating the effect of the end condition. The outermost curve has the largest root radius, equal to the radius of the cylinder. As the root radius is decreased, the simulated curves approach the input curve but are always above it. Reprinted from Scripta Materialia, Vol 54, Issue 2, page 6, H. Zhang, B.E. Schuster, Q. Wei and K.T. Ramesh, The design of accurate microcompression experiments, January 2006, with permission from Elsevier.

78

Nanomaterials

of the flow stress of the material in the plastic range, and the strength of this effect depends on the fillet radius/post radius ratio. The larger the fillet size (with respect to the post radius), the greater the error. When the fillet radius ratio rrc is less than 0.5, the output curves are within 3% of the input data. This suggests that if the fillet size is well controlled the fixed-end compression measurements can still be used to represent the bulk material’s behavior in terms of flow stress. For the aspect ratio α = 2, even with the largest fillet radius/post radius ratio ( rrc = 1) the error in flow stress from the microcompression test is less than 6%. The distribution of the von Mises stresses in the post and the base (Figure 3.6) for various fillet radius/post radius ratios shows that with increased fillet size, the stress concentration at the fillet is alleviated. For rrc ≥ 0.5 there is no significant stress concentration at the fillet, the maximum von Mises stress being along the post axis. Excessive stress concentrations may result in localized specimen failure prior to general yielding. Thus, the choice of fillet radius for an experiment represents a compromise between the need to accurately recover material behavior in the plastic range (requiring small rrc ) and the desire to avoid localized failure at the post root (which requires large rrc ). The simulations suggest that a choice of rrc in the range of 0.2 ∼ 0.5 represents a good compromise. The fillet size can be controlled to some degree by the FIB or micro-machining parameters. The second important geometric factor that may affect the accuracy of the meah ). The aspect ratio of the posts has a relasurement is the post aspect ratio (α = 2r tively small effect on the output flow stress curves if the aspect ratio is larger than 2. Aspect ratios of less than 2 are not appropriate, since the output strain hardening of the post material then deviates from the input value due to the constraints from the post base. However, aspect ratios much larger than 2 can cause buckling (discussed later in this chapter and in Chapter 7).

3.4.3.2 Accuracy of Measurement of Elastic Response In a standardized mechanical experiment the machine compliance is eliminated by using extensometers or strain gages and the elastic modulus of the specimen material can then be calculated. For micro-compression, this type of compliance correction is not possible. Therefore, it is important to examine the factors that will affect the accuracy of elastic modulus measurement. Figure 3.8 displays the error of elastic modulus measurements as a function of the fillet size (the apparent elastic modulus is normalized by the known input modulus). The post aspect ratio is fixed at 2. It is apparent that increased fillet size improves the accuracy with which the elastic modulus is measured, but the apparent modulus still underestimates the input modulus. The reason is that the base will always deform along with the post, and this deformation contributes to the system compliance, which is not accounted for when collecting the displacement data. However, this error may be alleviated by accounting for the base compliance using a result from Sneddon. Sneddon computed the depth of penetration D of an elastic half-space by a flat-ended rigid cylinder of radius a under a load P as

3 Nanoscale Mechanics and Materials: Experimental Techniques

79

Normalized Modulus, E/Einput

1.4 Effect of fillet size ( = 2) Sneddon Correction Modified Sneddon Correction

1.3 1.2 1.1 1 0.9 0.8 0.7

(b) 0

0.3 0.6 0.9 1.2 Normalized Fillet Radius, rc /r

Fig. 3.8 Computed variation of apparent elastic modulus (normalized by true elastic modulus) with fillet radius size at fixed aspect ratio, using the finite element simulations discussed in the text. The effects of the Sneddon and modified Sneddon corrections are also shown. Reprinted from Scripta Materialia, Vol 54, Issue 2, page 6, H. Zhang, B.E. Schuster, Q. Wei and K.T. Ramesh, The design of accurate microcompression experiments, January 2006, with permission from Elsevier.

D=

P(1 − ν ) 4µ a

(3.5)

where µ and ν are respectively the shear modulus and Poisson’s ratio of the halfspace. We then obtain the base compliance (Zhang et al., 2006) as Cb =

1−ν 1 − ν2 = 4µ a 2Eac

(3.6)

where ac = η (r + rc ) is the effective contact radius, with η as a proportionality constant to be determined. The choice of a in Equation (3.6) is difficult because of the geometry of the post-base connection and we have set a = ac . By subtracting this base compliance from the total compliance, a corrected post stiffness (and hence modulus) is obtained. If η = 1 (the Sneddon correction), the modulus is overestimated. Setting η = 1.42 provides a significantly improved correction (the modified Sneddon correction). Figure 3.8 also includes both the Sneddon and modified Sneddon corrections. Similar results on the error in the modulus measurement are found with other aspect ratios. 3.4.3.3 Accuracy of Measurement of Plastic Response The accuracy of micro-compression measurements may be affected by the plastic flow properties (strain hardening and strain rate hardening) of the post material. Finite element simulations have shown that the microcompression technique is most accurate for materials that exhibit low strain hardening. Such materials include

80

Nanomaterials

nano-structured metals, metals and alloys with ultrafine microstructure, metallic glasses, and ceramics. Strain rate effects are also important in many applications, especially when plastic instability is of serious concern. This is because the susceptibility to plastic instability of metals and alloys is strongly dependent on the strain rate sensitivity of the material. There are various ways to define the strain rate sensitivity. An engineering definition assumes a power-law strain rate hardenσ ˙ ing such that the strain rate sensitivity is written as m = ∂∂ln ln˙ε , where ε is the imposed strain rate. Finite element simulations assuming different values of m indicate that microcompression can capture the strain rate sensitivity of the material. Note that the practically useful strain rate achievable for microcompression with a nano-indenter is probably below 1 s−1 , i.e., in the quasistatic regime. 3.4.3.4 Buckling, Friction and Taper Effects Buckling, either elastic or plastic, is a common concern for a compression test. Elastic Euler buckling of a column occurs when the applied compressive load reaches a critical value given by Pcr =

π 2 EI Le 2

(3.7)

where I is the moment of inertia of the column (here the micropost) and Le is the effective length of the column (Le = 2h for the end conditions developed in microcompression). The applied load in the microcompression experiment must be at least large enough to cause yielding of the sample, so P > σY A where A is the crosssection area of the post. The minimum aspect ratio at which Euler buckling is likely to occur in microcompression can then be computed, and is found to be greater than 10 given the typical ratio of E/σY for most materials. Thus Euler buckling is not an issue in microcompression for aspect ratios less than 10. The plastic buckling problem must be addressed with 3D finite element analysis. Effects of friction between the indenter (usually made of diamond) and the post of top must also be considered, as must the effect of “taper” of the post and the system misalignment on the simulation output. In these 3D analyses the smallest fillet radius ratio used is rrc = 0.1 for α = 2. Both 3D cylindrical and cuboidal posts are considered. The compression is exerted by a rigid surface representing the flat-end nano-indenter tip. Misalignment of the experimental setup is considered by tilting the rigid surface with respect to the axis of the post. 3D simulations show that when there is no friction, aspect ratios of approximately 2 provide good measures of behavior and avoid buckling, while aspect ratios of 5 cause buckling after small overall strains. However, micro-compression experiments are usually performed without applying lubrication to the tip-post interface. A range of values for Coulomb friction coefficients exist in the open literature for a number of diamond-metal surface pairs: static coefficients of friction in the range of 0.1–0.15, and kinetic values of 0.03 ∼ 0.06. Friction between the indenter tip and the post top delays the plastic buckling

3 Nanoscale Mechanics and Materials: Experimental Techniques

81

of cylindrical posts; however, the friction does increase the estimated flow stress. The ease of development of plastic buckling is strongly dependent on the strain hardening of the material. Thus large aspect ratio posts should also be avoided due to the possible occurrence of plastic buckling when one tests low strain hardening materials: aspect ratios of 2 ∼ 3 are recommended. Another important geometric factor that may have a significant effect on the accuracy of microcompression experiments is the “taper” of the post. Taper is typically a result of fabrication inaccuracies. Taper can cause errors in the estimates of the modulus and yield stress and result in a spurious strain hardening response (due to propagation of the plastically deforming zone down the post axis). Taper should therefore be avoided by careful manipulation of the micro-post fabrication process. 3.4.3.5 Effect of Misalignment Misalignment of the system is manifested primarily in the deviation of the elastic part of the stress-strain curve, giving rise to an underestimate of the elastic modulus of the material. A decrease of the measured elastic modulus is observed with increase of misalignment (note that a lower elastic modulus will be observed even for perfect alignment is due to the compliance of the post base). In addition, excessive misalignment will result in buckling of the post. 3.4.3.6 Summary Recommendations for Microcompression Experiments 1. The post fillet radius affects both the accuracy of the microcompression test (in terms of the flow stress) and the stress concentration at the post root. Fillet radius/post radius ratios of 0.2 ∼ 0.5 are recommended to provide sufficient test accuracy. The fillet radius also has a profound effect on the accuracy of the elastic modulus due to the compliance of the base. This effect can be alleviated using a modified Sneddon correction to account for the base compliance. 2. Strain hardening of the material significantly affects the microcompression test accuracy, particularly when the fillet radius is large (close to the post radius). The results suggest that microcompression experiments are better suited for materials with low strain hardening. 3. For aspect ratios ≥ 5, first-mode plastic buckling is suppressed by the friction between the indenter tip and the post top, but higher-mode buckling may still occur, particularly for high aspect ratios and low strain hardening materials. Post aspect ratios of 2 ∼ 3 are therefore recommended. 4. Taper of the post results in overestimated elastic modulus, spurious strain hardening and increased apparent yield and thus should be minimized during micro-post fabrication. 5. Misalignment of the microcompression system significantly affects the test accuracy by underestimating the elastic modulus, and may result in buckling of the post. This should be minimized through careful post sample preparation, tip machining, and experimental alignment.

82

Nanomaterials

3.4.4 Microtensile Testing Conventional tension testing is the standard approach to obtaining the mechanical response of materials (indeed, there are very well-developed ASTM standards for such testing). These standards involve the use of specially shaped specimens, designed to develop a uniaxial tensile stress along a specific gauge length and to obtain accurate measures of strain. The result is a stress strain curve in uniaxial tension (Figure 3.10) that is the basis of most constitutive modeling in the plastic range. However, conventional tension tests require specimens that are at least several centimeters long. As of 2008, there are very few nanomaterials which can be obtained in sizes sufficiently large to permit conventional tension tests. A number of research groups have therefore developed the ability to perform tension tests on very small samples (typically obtained from thin films). Sample sizes of interest are typically lengths of 10–1000 µm, widths of 10–100 µm, and thicknesses of 10 nm to 100 µm. There are three critical requirements for a good tension test. The first is the development of a significant region of the specimen (called the gauge length) within which a uniaxial tensile stress state exists; the second is the ability to measure that tensile stress accurately; and the third is the ability to measure the tensile strain along the direction of the stress within the gauge section. Finally, there is the question of actuation: how is the tension generated?

3.4.4.1 Stress State and Specimen Design The first requirement determines the design of the specimen. The primary complexity in tension tests arises from the need to apply the tensile force to the specimen, which requires the use of some type of grip. Many, many grip types are used in conventional tension testing, ranging from hydraulic and pneumatic grips to threaded ends and adhesively bonded ends. The stress state at the grip is very complicated, and so careful specimen design is necessary to obtain a gauge length that is primarily under uniaxial tension. An important issue is the degree of stress concentration near the grips, which is particularly important for high-strength and relatively brittle materials like nanomaterials. For the case of materials with a significant degree of plastic deformation, a very common specimen design that provides a gauge length containing nearly uniaxial tensile stresses is obtained with the “dogbone” shape of conventional tension testing. Very good computational design of tensile specimens can be performed using the finite element method, accounting for the expected material behavior, and so it is no longer necessary to rely on a standard to ensure that reasonable results are being obtained. When it comes to testing very small nanomaterials samples, the total forces that are developed are small and so unique grip designs are often used that are mated to the specimen shapes that can be easily fabricated. There is also a strong correlation between the fabrication or machining techniques used to make the nanomaterial and the tension test sample design. Many micro-tension specimens end up being essentially thin films, and so many of the techniques associated with thin-film applications are relevant to such experiments.

3 Nanoscale Mechanics and Materials: Experimental Techniques

83

3.4.4.2 Accurate Measurement of Stress The second requirement (that of determining the tensile stress within the gauge length of the specimen) is relatively easy to satisfy, because the conditions of force equilibrium demand that the force along the force chain in the testing technique be constant. Thus the simple insertion of a load cell into the force chain provides a measure of the force in the specimen, and then knowing the cross-section area of the specimen allows one to calculate the tensile stress in the sample gauge length. The only difficulty involves ensuring that the load cell is sufficiently accurate and has the resolution to measure the small forces typically developed in small samples. A rough estimate of the force that would be generated in the specimen can be obtained as σy A where σy is the yield stress of the material and A is the cross-section area of the specimen, which scales as the square of the specimen width (A ∝ b2 ). Thus for a nanocrystalline material with a yield strength of 1 GPa and a specimen width of 10 µm, the force that must be measured is of the order of 0.1N. In the elastic range, the force scales as E ε A, and so to have a strain resolution of ε = 10−4 and a Young’s modulus of 100 GPa one needs a force resolution of 1 mN.

3.4.4.3 Accurate Measurement of Strain The biggest difficulty in microscale tension experiments is accurate measurement of the strain in the specimen. In traditional tension testing, strain is measured in one of two ways. The simplest and least accurate is to measure the displacement δ l of the crosshead (or actuator) and divide this by the specimen gauge length l0 :

ε=

δl . l0

(3.8)

A much more accurate approach involves measuring the strain directly on the specimen using some kind of in situ strain measurement technique such as a strain gauge. However, it is difficult to find strain gauges that are of the appropriate size for most micro sample tests. Approaches that are currently used in microsample testing include digital image correlation or DIC (Chu et al., 1985) together with in situ imaging using an optical microscope, SEM, or TEM; the measurement of the displacement of fiducial markers placed on the specimen, and specialized techniques such as the interferometric strain/displacement gauge (ISDG). A simple measure of the difficulty involved can be obtained by recognizing that the displacements that must be measured are of the order of ε l0 , and so for a specimen length l0 of the order of 100 µm and a strain resolution of ε = 10−4 one needs to measure a displacement of δ u = 10 nm. Such displacement resolutions are difficult to achieve.

3.4.4.4 Actuation Systems for Micro-Tensile Testing Uniaxial tension experiments are typically performed either under load control or under displacement control (that is, one controls the load while the displacement is

84

Nanomaterials

measured as the output of the system, or vice versa). A simple (and idealized) example of a load controlled test is a tension test performed using dead weights to provide the load. The actuator system used to generate tension in conventional tension tests is typically either a servohydraulic system or a screw driven system. These systems are capable of carrying very high loads, and control systems for such devices are highly developed. However, such systems are not appropriate for the small loads and small displacements that must be used to study microscale specimens: the excursions in force or displacement that are developed in these conventional systems before full control is established are typically larger than the forces or displacements σ required to cause failure of the microscale specimens. Initial displacements of Ey l0 σ are sufficient to cause yield of the specimen, and since for most materials Ey ∼ 10−3 , it takes only 100 nm of displacement to cause yield in a specimen that is 100 µm long. An actuator with extreme control of displacement is therefore necessary for micro-tensile testing. Examples of actuators with such control include piezoelectric actuators and MEMS-based actuator systems (Figure 3.9). The advantage of piezoelectric actuator systems is that they can be purchased rather than having to be fabricated.

Fig. 3.9 Schematic of a microtension testing apparatus, showing the actuation and sensing systems. The air bearing is an important component. The strain is computed from displacement fields measured using Digital Image Correlation (DIC) software analysis of images from the camera. Illustration due to Chris Eberl.

In most tensile tests, the compliance of the testing machine is a significant factor in the design of the experiment. Under the load, some of the displacement that is observed in the system is a result of the deformation of the machine because of machine compliance. There are standard approaches to correcting for machine compliance so that accurate measurements of specimen strain can be obtained, at least for conventional tension testing. This can be particularly important for micro-tensile testing.

3 Nanoscale Mechanics and Materials: Experimental Techniques

85

The result of a microtensile experiment on a nanocrystalline nickel film is shown in Figure 3.10. The specimen was a thin film made by electrodeposition (an Integran material), with a film thickness of 100 nm and a mean grain size of 20 nm. Note the significant flow stress that is obtained, and the unload-reload cycle used to estimate the modulus (to minimize machine compliance issues). The apparent strain hardening that is observed early in the stress-strain curve will be discussed later in the mechanical properties chapter.

True Stress, MPa

2000

1500

1000

500

0 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

True Strain Fig. 3.10 Example of stress strain curve obtained on nanocrystalline nickel (20 nm grain size) in uniaxial tension using a microtension setup. The sample was a thin film. Data due to Shailendra Joshi.

A remarkable variety of nanomechanical and micromechanical tensile testing systems have been developed over the last decade (Sharpe et al., 1997; Espinosa et al., 2003; Legros et al., 2000; Yi et al., 2000; Haque and Saif, 2001; Srikar and Spearing, 2003; Jonnalagadda and Chasiotis, 2008). Several of them have been developed with an eye towards thin-film applications rather than testing nanomaterials, but some of them can be used as nanoscale mechanics characterization techniques with application towards nanomaterials. Since these techniques involve multiple innovations and some uncertainty about accuracy and repeatability, it is important to compare the results of these experiments with the results of other experimental techniques on similar materials. Such comparisons are difficult because the research groups that operate such techniques typically do not work on identical materials, and so it is unclear whether the differences in behavior that are observed are the result of experimental technique variations or the material variations. Note, in particular, that comparisons of nanomaterial behavior measured using these systems with the result of macroscale experiments on large-sample coarse-grain materials may not be meaningful, because of the potential differences in materials processing routes and specimen fabrication.

86

Nanomaterials

3.4.4.5 Summary Recommendations for Micro-Tensile Experiments 1. Finite element methods should be used to accomplish specimen design, with the objective being to optimize the specimen design with respect to the mechanical property that must be measured (for example, optimal design for elastic modulus measurement would be distinct from optimal design for strength measurement). 2. For minimum specimen dimensions larger than several microns, piezoelectric actuators systems provide a good compromise between accuracy and controllability. 3. In situ strain measurement techniques such as digital image correlation are recommended, and are essentially required if computational design of the specimen has not been performed. 4. Ideally, experimental results with any new technique should be compared with the results of other experimental techniques on a standard material in order to obtain confidence in the particular experimental technique that is being used. Unfortunately, at this point (2008) it is not possible to recommend a specific standard material.

3.4.5 Fracture Toughness Testing The fracture toughness of nanomaterials is extremely difficult to measure. Standard fracture toughness measurement techniques require standard specimen sizes, all of which are too large to be used with most nanomaterials (as of 2008). Indeed, it is not clear that the concept of fracture toughness as it is normally defined is relevant to most small-scale systems. Thus, most nanomaterials are most easily obtained in thin-film form, and so plain strain fracture toughness is not a sensible concept. Further, it is not clear that there is a significant K-dominant (see Chapter 2) domain within a nanomaterial containing a crack. That said, estimates of fracture toughness are needed in order for materials to be used within the design of complex components, and it is important to have at least a theoretical understanding of how fracture toughness should change with grain size. It is often stated that nanomaterials are brittle, but what is typically meant is that the tensile ductility of a specific nanomaterial specimen is low. The two existing approaches to fracture toughness testing with nanomaterials are (a) examining the failure of a thin film containing a pre-crack and (b) bending techniques developed for understanding thin film behavior. We explore these techniques further in the chapter on the failure of nanomaterials, because these techniques show material behavior that is intimately coupled to test design.

3.4.6 Measurement of Rate-Dependent Properties The primary experimental techniques associated with the measurement of the rate dependent properties of materials are described in Figure 3.11 (note that the stress

10–6

10–4

10–2

100

102

Strain Rate,

104

87

Pressure-shear plate impact

Miniaturized Kolsky bar

Conventional Kolsky bars

Specialized machines

Servohydraulic machines

3 Nanoscale Mechanics and Materials: Experimental Techniques

106

108

s–1

Fig. 3.11 Experimental techniques appropriate for various ranges of strain rate. The range of rates identified as “specialized machine” is very difficult to reach. Very few laboratories in the world are able to achieve strain rates higher than 104 s−1 .

states developed within the various techniques are not necessarily identical). An excellent and relatively recent review of these methods is presented by Field et al. (2004), and a summary of the key experimental issues is presented by Ramesh in a chapter in the Handbook of Experimental Mechanics (see suggested reading). For the purposes of this discussion, strain rates above 102 s−1 are classified as high strain rates, strain rates above 104 s−1 are called very high strain rates, and strain rates above 106 s−1 are called ultra-high-strain-rates. Conventionally, strain rates at or below 10−3 s−1 are considered to represent “quasistatic” deformations, and strain rates below 10−6 s−1 are considered to be in the “creep” domain. This is an important range for some nanomaterials. Creep experiments are typically performed at relatively high temperatures, and a variety of specialized machines exist for these kinds of loading; dead loads are often of particular interest. Quasi static experiments are typically accomplished through a variety of servohydraulic machines, and ASTM standards exist for most of these experiments. Most servohydraulic machines are unable to develop strain rates larger than 100 s−1 repeatably, but some specialized servohydraulic machines can achieve strain rates of 101 s−1 . We focus here on the higher strain rates (greater than 102 s−1 ).

3.4.6.1 Split-Hopkinson or Kolsky Bars The now-classical experimental technique in the high-strain-rate domain is the Kolsky bar or split-Hopkinson pressure bar (SHPB) experiment for determining the mechanical properties of various materials (e.g., Nicholas and Rajendran, 1990;

88

Nanomaterials

Ravichandran and Subhash, 1994; Walley and Field, 1994 in metals, ceramics, and polymers respectively) in the strain rate range of 102 s−1 to 8 × 103 s−1 . The terms “split-Hopkinson pressure bar” and Kolsky bar are often used interchangeably, although the latter is more appropriate (the technique was invented by Kolsky, and neither John nor Bertram Hopkinson used a split-bar to study material behavior). One should note also that the term “split-Hopkinson pressure bar” implies the performance of compression experiments, whereas the term “Kolsky bar” is more general and may include compression, tension, torsion or combinations of all of these. Since the fundamental concept involved in this technique was developed by Kolsky (1949), we will use the term Kolsky bar. A schematic of the typical compression Kolsky bar experimental apparatus is shown in Figure 3.12. The device consists of two long bars (called the input and output bars) that are designed to remain elastic throughout the test. These bars sandwich a small specimen (usually cylindrical), which is expected to develop inelastic deformations. The bars are typically made of high-strength steels such as maraging steel, with a very high yield strength and substantial toughness. One end of the input bar is impacted by a projectile made of a material identical to that of the bars; the resulting compressive pulse propagates down the input bar to the specimen. Several reverberations of the loading wave occur within the specimen; a transmitted pulse is sent into the output bar and a reflected pulse is sent back into the input bar. Typically, resistance strain gages are placed on the input and output bars for measuring (i) the incident pulse generated by the impacting projectile, (ii) the reflected pulse from the input bar/specimen interface, and (iii) the transmitted pulse through the specimen to the output bar. The strain gage signals are typically measured using high-speed digital oscilloscopes with at least 10-bit accuracy and preferably with differential inputs to reduce noise.

Fig. 3.12 Schematic of the compression Kolsky bar (also known, incorrectly, as the splitHopkinson pressure bar). The projectile is usually launched towards the input bar using a gas gun. Specimen surfaces must be carefully prepared for valid experiments.

Let the strain in the incident pulse be denoted by εI , that in the reflected pulse by εR , and that in the transmitted pulse by εT (these are bar strains as measured by the strain gages). By assuming stress equilibrium, uniaxial stress conditions in the specimen and 1D elastic stress wave propagation without dispersion in the bars, the nominal strain rate e˙s , nominal strain es , and nominal stress ss (all in the specimen) can be estimated using the following equations:

3 Nanoscale Mechanics and Materials: Experimental Techniques

2cb εR (t) l0

(3.9)

e˙s (τ )d τ

(3.10)

Eb Ab εT (t) As

(3.11)

e˙s (t) = − es (t) =

89

t 0

and ss (t) =

where cb = Eρb is the wavespeed in the bar, l0 is the initial length of the cylinb drical specimen, Eb is the Young’s modulus of the bar material, ρb is the density of the bar material, Ab is the bar cross-section area, and As is the specimen initial cross-section area. Note the negative sign in Equation (3.9). This arises because the strain in the reflected pulse has the opposite sign of the strain in the incident pulse; the latter is compressive and is conventionally considered positive in these experiments. Thus the specimen strain rate expressed by Equation (3.9) is a positive strain-rate, i.e., in this convention, a compressive strain-rate. The true strain and true strain rate in the specimen are then given by Ramesh and Narasimhan (1996):

εs (t) = − log (1 − es (t)) and

ε˙s (t) =

e˙s (t) . 1 − es (t)

(3.12)

(3.13)

Note that compression is defined as positive. Assuming plastic incompressibility, the true stress σs in the specimen is obtained as

σs (t) = ss (t)(1 − es (t)).

(3.14)

From Equations (3.12) and (3.14), a stress-strain curve is obtained at a strain rate defined by an average taken over the strain rate history obtained in Equation (3.13); an example of such a stress-strain curve is presented in Figure 3.13. Several fully miniaturized versions of the compression Kolsky bar have been developed (e.g., see Jia and Ramesh 2004). In the Jia and Ramesh version (see Figure 3.14), the bars are 3 mm in diameter and 300 mm long, and may be made of maraging steel or tungsten carbide. Sample sizes are cubes or cylinders on the order of 600 µm to 1 mm on a side; cube specimens are used when the failure mode must be imaged using a high-speed camera, or when the amount of material is so small that only cuboidal specimens can be cut. The technique is simple, and the entire system can be designed to fit on a desktop, and so the technique is sometimes referred to as the “desktop Kolsky bar.” The miniaturized Kolsky bar or “desktop Kolsky bar” experimental technique is particularly effective when it comes to understanding the properties of nanostructured materials, or of materials with very fine grain sizes.

90

Nanomaterials

Fig. 3.13 Stress strain curves obtained on 5083 aluminum using the compression Kolsky bar at high strain rates (2500 per second). The lowest curve represents quasistatic behavior. In general, the strength appears to increase with increasing strain rate, but note the anomalous softening at the highest strain rate (perhaps due to thermal softening).

Fig. 3.14 The Desktop Kolsky Bar – a miniaturized compression Kolsky bar arrangement developed by Jia and Ramesh (2004). This device is capable of achieving strain rates above 104 per second, and can fit on a standard desktop. Specimen sizes can be on the order of a cubic millimeter. The large circular object behind the bar is part of a lighting system.

3 Nanoscale Mechanics and Materials: Experimental Techniques

91

3.5 Suggestions for Further Reading 1. K.L. Johnson, Contact Mechanics, Cambridge University Press, Cambridge, 1987. 2. W.N. Sharpe, Jr., Handbook of Experimental Mechanics. Springer, New York, 2008. See, in particular, the chapter on High Strain Rate and Impact Testing by K.T. Ramesh. 3. B. Bhushan, Springer Handbook of Nanotechnology. Springer, New York, 2007.

3.6 Problems and Directions for Research 1. Determine the resolution limits of the transmission electron microscope and scanning electron microscopes in your institution. Will they be able to resolve a 10-nm grain of nanocrystalline aluminum? Can you see a carbon nanotube in your SEM? 2. Look up the ASTM standard for testing hard materials (determining the yield strength). Determine the size of specimen that would be required to meet the standards if the specimen were to be made of a nanocrystalline aluminum that has a yield strength of 1 GPa. 3. What is the smallest force that can be measured in one of the atomic force microscopes in your institution (or the nearest AFM, if your institution does not possess one)? Can you measure the force required to develop a strain of 1% in a single-walled carbon nanotube of Young’s modulus 1 TPa, assuming elasticity? 4. Perform a tensile stress-strain test on any aluminum alloy in a servohydraulic testing machine, and then determine how to correct for the machine compliance. Provide a corrected tensile stress-strain curve.

References Binnig, G. and H. Rohrer (1985). Scanning tunneling microscopy. Surface Science 152, 17–26. Binnig, G., C. Quate, and C. Gerber (1986). Atomic force microscope. Physical Review Letters 56(9), 930–933. Binnig, G., H. Rohrer, C. Gerber, and E. Weibel (1983). 7× 7 Reconstruction on Si (111) Resolved in Real Space. Physical Review Letters 50(2), 120–123. Chasiotis, I., S. Cho, T. Friedmann, and J. Sullivan (2004). Young’s modulus, poisson’s ratio, and nanoscale deformation fields of MEMS materials. Materials Research Society Symposium Proceedings 795, 461–466. Chu, T. C., W. F. Ranson, M. A. Sutton, and W. H. Peters (1985). Applications of digital image correlation techniques to experimental mechanics. Experimental Mechanics 25(3), 232–244. Espinosa, H. D., B. C. Prorok, and M. Fischer (2003). A methodology for determining mechanical properties of freestanding thin films and mems materials. Journal of the Mechanics and Physics of Solids 51(1), 47–67. Feynman, R. (1959, December 29th 1959). There’s plenty of room at the bottom.

92

Nanomaterials

Field, J., W. Proud, S. Walley, and H. Goldrein (2004). Review of experimental techniques for high rate deformation and shock studies. International Journal of Impact Engineering 30, 725–775. Haque, M. and M. Saif (2001). Microscale materials testing using MEMS actuators. Journal of Microelectromechanical Systems 10(1), 146–152. Haque, M. A. and M. T. A. Saif (2004). Deformation mechanisms in free-standing nanoscale thin films: A quantitative in situ transmission electron microscope study. Proceedings of the National Academy of Sciences of the United States of America 101(17), 6335–6340. Jia, D. and K. Ramesh (2004). A rigorous assessment of the benefits of miniaturization in the kolsky bar system. Experimental Mechanics 44(5), 445–454. Johnson, K. (1987). Contact Mechanics. Cambridge: Cambridge University Press. Jonnalagadda, K. and I. Chasiotis (2008). Tensile behavior of thin films of nanocrystalline fcc metals. Submitted for publication. Kiener, D., W. Grosinger, G. Dehm, and R. Pippan (2008). A further step towards an understanding of size-dependent crystal plasticity: In situ tension experiments of miniaturized single-crystal copper samples. Acta Materialia 56(3), 580–592. Kolsky, H. (1949). An investigation of the mechanical properties of materials at very high rates of loading. Proceedings of the Physical Society, London 62B, 676. Legros, M., B. Elliott, M. Rittner, J. Weertman, and K. Hemker (2000). Microsample tensile testing of nanocrystalline metals. Philosophical Magazine A 80(4), 1017–1026. Lennon, A. M. and K. T. Ramesh (2000). The thermoviscoplastic response of polycrystalline tungsten in compression. Materials Science and Engineering a-Structural Materials Properties Microstructure and Processing 276(1–2), 9–21. ˚ Martin, Y. and H. Wickramasinghe (1987). Magnetic imaging by force microscopy with 1000 A resolution. Applied Physics Letters 50, 1455. Minor, A. M., J. W. Morris, and E. A. Stach (2001). Quantitative in situ nanoindentation in an electron microscope. Applied Physics Letters 79(11), 1625–1627. Nicholas, T. and A. Rajendran (1990). Material characterization at high strain-rates. In J. Zukas (Ed.), High Velocity Impact Dynamics, pp. 127–296. New York: John Wiley & Sons, Inc. Nishino, T., A. Nozawa, M. Kotera, and K. Nakamae (2000). In situ observation of surface deformation of polymer films by atomic force microscopy. Review of Scientific Instruments 71, 2094. Oliver, W. and G. Pharr (1992). An improved technique for determining hardness and elastic modulus using load and displacement sensing indentationa experiments. Journal of Materials Research 7(6), 1564–1583. Pharr, G. and W. Oliver (1989). Nanoindentation of silver - relations between hardness and dislocation structure. Journal of Materials Research 4(1), 94–101. Ramesh, K. and S. Narasimhan (1996). Finite deformations and the dynamic measurement of radial strains in compression kolsky bar experiments. International Journal of Solids and Structures 33(25), 3723–3738. Ravichandran, G. and G. Subhash (1994). Critical appraisal of limiting strain rates for compression testing of ceramics in a split hopkinson pressure bar. Journal of the American Ceramic Society 77, 263–267. Schuster, B. E., Q. Wei, M. H. Ervin, S. O. Hruszkewycz, M. K. Miller, T. C. Hufnagel, and K. T. Ramesh (2007). Bulk and microscale compressive properties of a pd-based metallic glass. Scripta Materialia 57(6), 517–520. Sharpe, W. N., B. Yuan, and R. L. Edwards (1997). A new technique for measuring the mechanical properties of thin films. Journal of Microelectromechanical Systems 6(3), 193–199. Srikar, V. and S. Spearing (2003). A critical review of microscale mechanical testing methods used in the design of microelectromechanical systems. Experimental Mechanics 43(3), 238–247. Stach, E. A., T. Freeman, A. M. Minor, D. K. Owen, J. Cumings, M. A. Wall, T. Chraska, R. Hull, J. W. Morris, A. Zettl, and U. Dahmen (2001). Development of a nanoindenter for in situ transmission electron microscopy. Microscopy and Microanalysis 7(6), 507–517. Uchic, M. D., D. M. Dimiduk, J. N. Florando, and W. D. Nix (2004). Sample dimensions influence strength and crystal plasticity. Science 305(5686), 986–989.

3 Nanoscale Mechanics and Materials: Experimental Techniques

93

Walley, S. and J. Field (1994). Strain rate sensitivity of polymers in compression from low to high strain rates. DYMAT Journal 1, 211–228. Yi, T., L. Li, and C. Kim (2000). Microscale material testing of single crystalline silicon: process effects on surface morphology and tensile strength. Sensors & Actuators: A. Physical 83(1–3), 172–178. Zhang, H., B. E. Schuster, Q. Wei, and K. T. Ramesh (2006). The design of accurate microcompression experiments. Scripta Materialia 54(2), 181–186.

From the many, one. – A liberal translation of E Pluribus Unum

4

Mechanical Properties: Density and Elasticity Should nanomaterials have properties that are any different from conventional materials? Since we have defined nanomaterials as materials where some controllable relevant dimension is of the order of 100 nm or less, there should not be a difference in properties unless the property of interest is affected by some size scale. A big part of understanding the properties of nanomaterials will therefore be developing an understanding of the length scales associated with these properties. We will approach this discussion in two parts. First, in this chapter, we will describe the simple mechanical properties of nanomaterials: the density and the linear elastic behavior. Second, in the next chapter, we will discuss the plastic behavior of nanomaterials, where the couplings of deformation mechanisms with size scale are much more complex. The objective of the current chapter is to introduce the concepts behind the coupling of the grain size and other size scales with the mechanical properties through the simple scalar property of density and through the simple linear property of elasticity.

4.1 Density Considered as an Example Property We can learn a great deal about size effects in nanocrystalline materials by considering a simple scalar property like the density. Our intuitive sense is that the density of a metal like iron is a fixed number. Elementary physics tells us that the density must depend on both the atomic weight and the relative spacing of the atoms; the spacing of the atoms depends on interatomic interactions, and therefore – at a fundamental level – on position within the periodic table of the elements. Looking up iron in the periodic table of the elements, one would find that pure iron has a density of 7874 kg m−3 . This is the density one would obtain if one were to measure the density of a single crystal of pure iron (which is a body-centered-cubic solid at room temperature), whatever the size of that single crystal. However, pure iron is generally obtained as a polycrystalline material. If the grain size (i.e., crystal size)

K.T. Ramesh, Nanomaterials, DOI 10.1007/978-0-387-09783-1 4, c Springer Science+Business Media, LLC 2009 

95

96

Nanomaterials

were to be 10 µm, a cubic millimeter of iron would contain about a million grains. The density of this polycrystalline pure iron will be very slightly different than the density of single crystal pure iron, because some fraction of the atoms in the polycrystalline material are at grain boundaries, where their interatomic spacing is not the same as that in the single crystal. Since the spacing of atoms at grain boundaries is usually somewhat larger than the spacing of atoms within the single crystal (i.e., the atoms are more loosely packed), the polycrystal is usually slightly lower in density.

4.1.1 The Rule of Mixtures Applied to Density If the density of the material in the single crystal is ρsc , and the density of the material corresponding to atoms that are at grain boundaries is ρgb , then the overall density ρ of the iron in the polycrystalline material can be calculated by using the balance of mass: V ρ = Vsc ρsc +Vgb ρgb (4.1) where V is the total volume of the material, and Vsc and Vgb are the volumes of the atoms in the single crystal part and grain boundary part of the polycrystal respectively. Dividing Equation (4.1) throughout by V , we obtain

ρ=

Vgb Vsc ρsc + ρgb . V V

(4.2)

This can be rewritten as

ρ = fsc ρsc + fgb ρgb where fsc = fgb =

Vgb V

Vsc V

(4.3)

is the volume fraction of atoms with the single crystal packing, and

is the volume fraction of atoms with the grain boundary packing. Note that Vsc +Vgb = V,

(4.4)

fsc + fgb = 1.

(4.5)

so that It follows that Equation (4.3) can be rewritten as

ρ = (1 − fgb )ρsc + fgb ρgb .

(4.6)

An equation of the form of Equation (4.6) is called a rule of mixtures: it provides an overall property of the material (from the mechanics perspective, viewed as a composite) in terms of the corresponding properties of its constituents, with a linear dependence on the volume fraction of each constituent. This is the most common first approximation for the overall properties of any material made out of multiple

4 Mechanical Properties: Density and Elasticity

97

Overall Density

ρSC

ρgb 0

0.2 0.4 0.6 0.8 Volume fraction of grain boundary, fgb

1

Fig. 4.1 The linear rule of mixtures. Calculated variation of density in a nanocrystalline material with volume fraction of grain boundary, based on Equation (4.6).

subcomponents (in this case, crystal atoms and grain boundary atoms). The result of using such a rule of mixtures to predict the overall density of the material is shown in Figure 4.1. Note that Equation (4.6) establishes that the density of the polycrystalline material may be different from the density of the single crystal, but this does not by itself ensure that there will be a grain size effect. What is the volume fraction fgb of the grain boundary atoms? This depends, of course, on the shapes of the grains and the nature of the grain boundaries. Consider the highly idealized case of a material made up entirely of grains that are cubes (the simplest space filling shape). Assume that the grain size, given by the side of the cube, is d, and that the thickness of the grain boundary is t (Figure 4.2). Then it follows that the volume fraction of the material that is inside the grain (rather than at the grain boundary) is fsc =

d3 (d + t)3

,

(4.7)

and that the volume fraction of the material at the grain boundary is fgb = 1 − fsc = 1 −

d3 (d + t)3

.

(4.8)

The key question is whether the grain boundary thickness changes as the grain size changes. Both experimental observations and atomistic simulations suggest otherwise (Latapie and Farkas, 2003). A simple intuitive way to understand this is to ask how many layers of atoms are affected by the disordered conditions at the grain boundary: it is apparent that the number of layers of atoms should remain the same whether the grain is 1 µm or 100 µm in size. Thus, for instance, if we assume that

98

Nanomaterials

a

t

d

b

d

Fig. 4.2 Schematic of (a) a cubical grain of size d with a grain boundary domain of thickness t, and (b) the packing of such grains to form a material.

no more than three layers of atoms are affected by the grain boundary disorder, the typical grain boundary thickness will be on the order of 1 nm (in metals, the inter˚ It is apparent that the influence of the grain atomic spacing is of the order of 3 A). boundary increases dramatically at small grain sizes. This size effect arises because the grain boundary thickness provides a length scale within the problem. A rapid estimate of the effect of the grain boundary thickness can be shown by recasting Equation (4.7) in terms of the grain size ratio β , defined by β = dt . We then have β3 , (4.9) fsc = (β + 1)3 and we see that when the grain boundary thickness is about the same as the grain size, i.e., t = d so that β = 1, we have fsc = 81 and so fgb = 1− fsc = 87 ! Thus the fraction of the material that is in the grain boundary increases dramatically as the grain size decreases to the approximate size of the fixed grain boundary thickness. We refer to this limit (β = 1) as the grain boundary limit, and the corresponding grain gbl boundary volume fraction as fgb . Note that the material would be about half grain boundary ( fgb = 0.5 or 50%) and half single crystal when d50 = 3.85t, and grain boundary mechanical behaviors can be expected to begin to dominate the overall behavior at least at this point (which in this case amounts to a grain size of about 4 nm). Thus the characteristic length scale t combined with the assumed morphology (space filling cubes) gives us a characteristic grain size scale at which we should

4 Mechanical Properties: Density and Elasticity

99

expect a significant change (or transition) in behavior. Changing the assumed crystal morphology would certainly change this d50 size scale. However, at small grain sizes, other complexities arise. Examining the space filling cubic morphology, three different domains can be defined along grain boundaries. The domains that we have been discussing at this point in terms of the lower density at the grain boundaries represent planar regions where two crystal facets meet. The regions where these planes themselves meet are either nearly-linear regions of finite size that we will call triple junctions, or cuboidal regions at cube corners that are corner junctions. (Note that triple junctions get their name because in most microstructures they are formed by the intersection of three planes.) It is likely that the densities ρt j and ρc j of the material at triple junctions and corner junctions respectively are even lower than that of the grain boundaries, because of the increased lack of compatibility between the atoms in the crystal structure. The volume of each triple junction is t 2 d, and so the volume fraction ft j of triple junctions in the polycrystalline material of cubic morphology is ft j =

6t 2 d 3

4(d + t)

=

3β (β + 1)3

.

(4.10)

Similarly, the volume of each corner junction is t 3 , and the volume fraction fc j of corner junctions in the polycrystalline material of cubic morphology is fc j =

6t 3 8(d + t)3

=

3 4(β + 1)3

.

(4.11)

The variation of each of these volume fractions (for the grains, the grain boundaries, the triple junctions, and the corner junctions) with the normalized grain size (normalized by the assumed grain boundary thickness t, so that this axis represents the grain size ratio) is shown in Figure 4.3 for the cubic morphology. The single crystal grain volume fraction decreases continuously as the grain size is decreased. The grain boundary volume fraction increases rapidly as the grain size approaches t, but has a maximum at a value of d just greater than t. This is because for smaller grain sizes the contributions of first the triple junctions and then the corner junctions increase dramatically, while the sum of all the volume fractions must be equal to unity. Thus any mechanisms that are triggered at such junctions are likely to become very important at small grain sizes. Note that the triple junction volume fraction is significant in the important domain of 1–10 nm (assuming that t ≈ 1 nm). When one now considers the grain boundary, triple junction, and corner junction parts of the polycrystalline material together, the effective density of the polycrystalline material is given by

ρ = (1 − fgb − ft j − fc j )ρsc + fgb ρgb + ft j ρt j + fc j ρc j ,

(4.12)

because we must have fsc + fgb + ft j + fc j = 1.

(4.13)

100

Nanomaterials 1

Cube Morphology

fsc

0.8

fgb ftj

Volume fraction

Grains 0.6

fcj

Corner junctions

0.4

Grain boundaries 0.2

Triple junctions 0 0.1

1

10

100

1000

Normalized grain size d/t

Fig. 4.3 Variation of grain volume fraction, grain boundary volume fraction, triple junction volume fraction, and corner junction volume fraction with normalized grain size β = dt for the cubic grain morphology. The junction volume fractions become major contributors when d ≈ t.

The lower limit of the density would now be ρc j , where the atoms are typically the most disordered. How different is this from ρsc ? This is not known a priori. The density ρgb is usually assumed not to be dramatically smaller than ρsc , which would imply that the absolute density change with grain size should not normally be an issue until small grain sizes (say of the order of 50 nm) are achieved. This may of course not be true for other properties, where the property of the grain boundary may be substantially different from the property of the crystal. In fact, however, the density of many nanocrystalline materials is measurably less than the density of the corresponding conventional grain size material even for relatively large grain sizes of order 100 nm. Why is this so? The primary reason is because the processes that are used to make nanocrystalline materials typically result in the generation of significant quantities of defects in the material, particularly porosity. An example of the porosity that might be observed in a nanocrystalline material is presented in Figure 4.4, which shows the surface of a nanocrystalline nickel material made by electroplating. A significant area fraction of pores is observed on the surface. The effect of the pores on the density can be very easily understood by considering the pores to represent a third phase of the material of density ρdefect , where ρdefect = 0. The porosity φ of the material is essentially the volume fraction of the V pores, φ = fdefect = Vpores . With this fourth constituent (the defects) included, the total overall density of the polycrystalline material is now

ρ = (1− fgb − ft j − fc j − fdefect )ρsc + fgb ρgb + ft j ρt j + fc j ρc j + fdefect ρdefect , (4.14) because we must have fsc + fgb + ft j + fc j + fdefect = 1.

(4.15)

4 Mechanical Properties: Density and Elasticity

101

Fig. 4.4 Scanning electron micrograph showing the presence of pores (dark regions) on the surface of a nanocrystalline nickel material produced by an electroplating technique.

Since ρdefect = 0 when the defects are pores, it follows that

ρ = (1 − fgb − ft j − fc j − φ )ρsc + fgb ρgb + ft j ρt j + fc j ρc j .

(4.16)

The effect of the porosity can be quite dramatic, because the porosity φ itself can be substantial, depending on the processing route used to make the nanocrystalline material. Much of the literature considers material with a porosity of 3–5% to be “fully dense” even though this level of porosity has a substantial impact on mechanical properties.

4.1.2 The Importance of Grain Morphology The effects of grain morphology on the behavior of a material can be very substantial at small grain sizes. For example, consider the alternative space-filling hexagonal prism morphology shown in Figure 4.5, with side s and height h. Suppose we define the aspect ratio parameter α = hs . The effective grain size in the plane of the hexagon is d = 2s, but note that if the aspect ratio is either much less than 1 or much greater than 1, a single grain size is no longer a reasonable description of the polycrystal structure.

102

Nanomaterials

t s

h

Fig. 4.5 Another possible space-filling morphology, using hexagonal tiles of side s and height h = α s, where α is the aspect ratio of the tile. The grain boundary domain thickness remains t. √

The volume of the hexagonal prism is area × h, which is ( 3 2 3 s2 h). For this hexagonal morphology, the volume fraction of the material in the crystal phase is fsc |hex =

1 , 2α 1 1+(√ )g + ( α3 + √23 )g2 + 31 g3 + α 3

(4.17)

where the boundary-crystal size ratio is g = 2st . For the special case where α = 1 so that h = s, we find that fgb = 0.5 at a transition grain size d50 = 2.7 nm, somewhat less than in the cubic morphology case. However, if α = 0.1, corresponding to a plate morphology (e.g., from a rolling process), d50 = 12 nm; if α = 10, corresponding to a rod morphology (e.g., from an extrusion process), we obtain d50 = 1.6 nm. An even stronger effect of the morphology of the polycrystal is seen on the volume fraction of the material that is in the grain boundary when the grain boundary limit g = 1 is reached. For the space filling hexagonal prism morphology with α = 1, we gbl gbl have fgb = 0.64; and when α = 0.1, we have = 0.8; when α = 10, we have fgb gbl fgb = 0.96! We should thus expect the rod morphology and the plate morphology to produce very different rates of change of behavior as the grain size is decreased. The sensitivity to morphology is of course related to another specific issue, which is that the grain size is not a well-defined quantity for highly asymmetric structures. What single grain size number should be used for a long rod shaped crystal? Should it be the diameter of the rod or the length of the rod? One possibility, sometimes used in the literature, is to define the grain size as the size of the cube of identical volume to the grain in the actual polycrystal. In this way of thinking, the effective grain size d to be used in considering the space filling hexagonal prism morphology shown in √

1 3

Figure 4.5 would be given by d = ( 3 2 3 α ) s. This approach is useful if one needs to use a single number to define the grain size in the material. For many applications, however, it is more useful to simply remember that perhaps two effective grain sizes need to be used. For the space filling hexagonal morphology, the volume of each triple junction is t 2 h, and so the volume fraction ft j of triple junctions in the polycrystalline material is

4 Mechanical Properties: Density and Elasticity

103

1

Hexagonal Prism Morphology

fsc fgb

Volume fraction

0.8

ftj

Grain boundaries

Grains

fcj

0.6

0.4

0.2

Triple junctions

Corner junctions 0 0.1

1

10

100

1000

Normalized grain size d/t

Fig. 4.6 Variation of grain volume fraction, grain boundary volume fraction, triple junction volume fraction, and corner junction volume fraction with normalized grain size β = dt for the hexagonal prism grain morphology of Figure 4.5 (with an aspect ratio α = 1). The junction volume fractions become major contributors when d ≈ t.

ft j |hex =

2t 2 h √ . 2 3 3 √t 2 (s + 3 ) (h + t)

(4.18)

The volume fraction of the corner junctions is given by fc j |hex =

1 36(s +

√t

2

3

) (h + t)

.

(4.19)

The variation of these volume fractions with grain size for the hexagonal morphology is presented in Figure 4.6. The difference between this variation and that for the cube morphology in Figure 4.3 is considerable, particularly in terms of the importance of the triple and corner junctions (these become important at smaller grain sizes in this hexagonal case). The hexagonal prism morphology affords us the opportunity to also study the effects of grain aspect ratio (α ). The variation of the several volume fractions with grain size are presented in Figure 4.7 for aspect ratios of 10 (rods) and 0.1 (plates). The aspect ratio clearly has a strong influence, particularly in determining the grain boundary limit and the relative importance of the triple junctions.

4.1.3 Density as a Function of Grain Size What is the net effect of these variations in volume fraction of grains, grain boundaries and junctions on the variation of the density of a material as a function of grain

104

Nanomaterials 1

Hexagonal, α = 10

fsc fgb ftj fcj

Volume Fraction

0.8

0.6

0.4

0.2

0 0.1

1

10

100

1000

Normalized Grain Size, d/t 1

Hexagonal, α = 0.1 fsc fgb ftj fcj

Volume Fraction

0.8

0.6

0.4

0.2

0 0.1

1

10

100

1000

Normalized Grain Size, d/t Fig. 4.7 Variation of grain volume fraction, grain boundary volume fraction, triple junction volume fraction, and corner junction volume fraction with normalized grain size β = dt for the hexagonal prism grain morphology of Figure 4.5, with two aspect ratios (10 and 0.1) corresponding to rods (top) and plates (bottom).

size? The density of the polycrystalline material is given by Equation (4.16). The density of the material near the grain boundaries and at the triple junctions is generally not known, because there is a very wide range of atomic arrangements possible at the boundaries (some of these are discussed later in this chapter). To visualize the possibilities, let us assume that the density near the grain boundaries is 95% of the density ρsc within the single crystal, that at the triple junctions is 90% ρsc , and that at the corner junctions is 81% ρsc . The resulting variation of the density with grain size is presented in Figure 4.8 for the cube and hexagonal prism morphologies (with an aspect ratio α = 10). The overall density of the material is then predicted (using Equation (4.16)) to decrease substantially as the grain size of the polycrys-

4 Mechanical Properties: Density and Elasticity

105

1

Overall Density

0.95

0.9

0.85

Cube Morphology Hex Prism Morphology

0.8 0.1

1

10

100

1000

Normalized Grain Size, d/t

Fig. 4.8 Predicted variation of overall density of polycrystalline material (normalized by ρsc ) with grain size for the cube and hexagonal prism morphologies (with an aspect ratio α = 10). The assumed parameters are ρgb = 0.95ρsc , ρt j = 0.9ρsc , and ρc j = 0.81ρsc .

tals decreases below 50 nm or so for these assumed parameters. There is negligible change once the polycrystal size increases beyond perhaps 100t. Thus the apparent density of nanocrystalline materials should be expected to be somewhat lower than that of conventional large-grain polycrystalline versions, and this may account for some of the situations where lower densities are observed but no pores are visible even in the TEM (e.g., Jia et al., 2003).

4.1.4 Summary: Density as an Example Property In summary, this discussion of the scalar property which is the density of the material establishes several issues that are important with respect to a change in the properties of nanocrystalline materials: • There must be an internal length scale in the material (in this case the grain boundary thickness). • The properties of one part of the material (here the grain boundaries) should be substantially different from the properties of another part. • A first order estimate of the change in a property can be obtained using the rule of mixtures. • The morphology of the polycrystalline material, and in general the microstructure, can have a strong influence on the effective property. • The presence of defects can dominate the behavior, and so characterization of defects is important. Processing issues often enter into defining the effective properties through the defects that the process generates.

106

Nanomaterials

4.2 The Elasticity of Nanomaterials One of the simplest mechanical behaviors of materials is elasticity (see Chapter 2). The simplest case of elasticity is that for a homogeneous isotropic material; that is, for a material in which the properties do not change either with position (homogeneous) or with direction (isotropic). As demonstrated in the chapter on mechanics, this results in the definition of two material properties, which could be chosen to be the Lame´ moduli, or the bulk modulus and the shear modulus, or the Young’s modulus and the Poisson’s ratio.

4.2.1 The Physical Basis of Elasticity It is necessary here to understand the fundamental physics associated with elasticity. The elastic response of materials essentially arises from the interatomic forces within the material. Consider the interatomic potential U(r) that defines the interaction of two atoms separated by a distance r, usually called a pair potential. The shape of this interatomic potential for most solids is similar to that presented in Figure 4.9. The energy has a minimum at a position r0 , which corresponds to the equilibrium spacing of the atoms (e.g., in a crystal). By definition, the force between the two atoms is given by ∂U F(r) = , (4.20) ∂r

Energy, U(r)

and so of course is zero at the minimum energy. The stiffness of the interatomic bond represents the elasticity of the system, and is given by

r

Distance, r

Fig. 4.9 Typical interatomic pair potential U(r), showing the equilibrium position r0 of the atoms. The curvature at the bottom of the potential well (at the equilibrium position) corresponds to the effective elastic stiffness of the bond.

4 Mechanical Properties: Density and Elasticity

107

∂F ∂ 2U ; (4.21) = ∂r ∂ r2 i.e., the elastic stiffness is the curvature of the interatomic potential (Ashby and Jones, 1996). Different kinds of interatomic interactions lead to different elastic stiffnesses. For example, the interatomic potential that represents a covalent bond is significantly stiffer than the interatomic potential that represents a metallic bond (this is why ceramics are stiffer than metals). Generating a strain in the material amounts to moving the atoms either closer together or further apart, that is, moving the atoms away from the equilibrium position (which can be maintained with no applied force). Thus, when no force is applied, all of the atoms in the material are sitting in their equilibrium positions (although thermal vibrations about the equilibrium position will occur). In a crystalline material, these equilibrium positions define the lattice spacing. The lattice spacing may be different in different directions, corresponding to the various symmetries of crystals. In elastic deformations, when a force is applied, all of the atoms move away from their equilibrium positions into non-equilibrium positions corresponding to the magnitude of the applied force (resulting in a strain). These nonequilibrium strained positions correspond to a higher energy state (see Figure 4.9), and we say that the material has stored a certain amount of elastic strain energy. When the applied force is released, the stored energy is recovered as the atoms move back to their equilibrium locations. The idea that the elastic modulus is a constant (i.e., does not depend on strain – this defines linear elasticity) amounts to assuming constant curvature of the interatomic potential, which is typically only reasonable near the minimum energy (in effect, this amounts to assuming that the potential can be approximated by a quadratic function at the equilibrium spacing of the atoms). Nonlinear elastic approximations are also possible, although rarely used for metals (except in molecular dynamics). E(r) =

4.2.2 Elasticity of Discrete Nanomaterials Are these elastic moduli different for nanomaterials than for conventional materials? Remembering our three classes of nanomaterials (discrete nanomaterials, nano device materials, and bulk nanomaterials), we see that this question is not well posed in the case of discrete nano materials, because there is no “conventional” version of discrete nanomaterials. In this case, a different question must be asked: do the elastic properties change with the size of the particle (the 0D case) or fiber (the 1D case)? 4.2.2.1 Nanoparticles Most nanoparticles are essentially very small single crystals. On the basis of the discussion above, there is no reason for the moduli of the crystal to depend on the size of the crystal, provided the interatomic potential does not change and the

108

Nanomaterials

equilibrium positions of the atoms do not change with crystal size. However, there is reason to expect that the equilibrium positions of the atoms would be different in free crystal particles of different sizes. The argument goes as follows. The atoms at the surface of a nanoparticle (indeed at any free surface) perceive an environment that is distinctly different from that perceived by atoms within the bulk of the crystal (since there are no atoms to interact with on the outside of the free surface). The result is an effective surface tension on the surface of these particles, which implies that the atoms within the particles are under an effective force (Figure 4.10). As a consequence, the atoms within the nanoparticles should have a different equilibrium spacing than the atoms within bulk single crystals. We discuss this further in Section 7.1.1 in the chapter on scale-dominant mechanisms.

Fig. 4.10 Surface effects on a spherical nanoparticle (e.g., surface tension). Such effects can have a significant impact on the behavior of the nanoparticle, particularly with respect to its interaction with the environment.

A continuum mechanics calculation (see Section 7.1.1) based on spherical nanoparticles indicates that there should be a significant change in the lattice spacing for nanoparticles that are 5 nm or less in size (Huang et al., 2007). This change in the lattice spacing is indeed observed experimentally in the case of palladium (Lamber et al., 1995). The measured reduction in lattice spacing is of the order of 3%. This transition size of course depends on the magnitude of the surface tension for a given material. Particles larger than the transition size are not significantly affected by the surface tension. We note here that these continuum calculations essentially assumed a constant modulus and computed a change in the equilibrium lattice spacing with particle size, rather than computing a change in the modulus with particle size. It is extremely difficult to measure the modulus of a given nanoparticle while it is relatively easy to measure the lattice spacing. The effect of the decreased size in terms of the increased importance of the surface tension essentially amounts to computing the solution of an equilibrium elasticity problem on each particle, resulting in a change in the

4 Mechanical Properties: Density and Elasticity

109

particle configuration and a change in the material properties (discussed in a later chapter). There is a very different calculation involved with the effect of the insertion of nanoparticles into a polymer matrix to develop a nanocomposite of increased stiffness, and we will address this calculation in a later chapter. Another important topic is that of the properties of functionalized nanoparticles, and we will discuss this as well in Chapter 7.

4.2.2.2 Nanotubes, Nanofibers, and Nanowires The elastic response of nanotubes, nanofibers, and nanowires is generally dominated by the stiffness of a specific structure rather than by a specific interatomic interaction. For example, the apparent elastic modulus of a nanotube depends not only on the specific arrangement of atoms within the nanotube, which depends on the interatomic potential, but also on the geometry of the structure (e.g., the diameter of the nanotube, whether it is a singlewall or multiwall nanotube, and the distribution of defects within the nanotube wall). Unlike the case of nanoparticles, the elastic moduli associated with onedimensional discrete nanomaterials is of significant technical interest. Such tubes, fibers, and wires are often used as reinforcing elements within composite structures, and their high elastic stiffness is critical for successful reinforcement. The measured effective Young’s moduli of a variety of one-dimensional discrete nanomaterials is presented in Table 4.1. Size dependence is often observed, which is sometimes attributed to defects and sometimes to surface effects (Ma and Xua, 2006). Note that these one-dimensional materials have essentially zero stiffness in compression, since they generally buckle under compressive loading. This is another sense in which these discrete nanomaterials should be viewed as structures rather than as materials alone. The stiffness of carbon nanotubes is a topic of great interest, and is discussed in detail in Chapter 8. Table 4.1 Some measured moduli of nanotubes, nanofibers, and nanowires, viewed as 1D nanomaterials. Material Single-wall carbon nanotube Multi-wall carbon nanotube ZnO nanowires Silica nanowires

Modulus, GPa 1002 11–63 140–200, depending on size 20–100

Reference (Yu et al., 2000a) (Yu et al., 2000a) (Chen et al., 2006) (Silva et al., 2006)

4.2.2.3 Nanobelts Nanobelts are thin strips of material, usually a few nanometers thick and a few tens of nanometers in width, typically made of oxides or other compounds. The belts have both a tensile Young’s modulus, and a flexural modulus, but as with nanowires

110

Nanomaterials

and nanotubes, they have essentially zero stiffness in compression. The elastic response of nanobelts is also a structural response rather than a material response alone. There is very little data at present on these materials. The oxide nanobelts typically involve covalent atomic bonds, and so the intrinsic stiffness of the material is very high. A good review of the properties of nanobelts is provided by Wang (2004).

4.2.3 Elasticity of NanoDevice Materials The largest fraction of nanodevice materials are thin films; a few are nanowires or nanotubes. A schematic of a thin film structure is shown in Figure 4.11. The elastic moduli associated with thin films can be substantially different from that of the bulk material, for a variety of reasons discussed in Chapter 8. Thin films can be either single crystal or polycrystalline, depending on how they are grown. The polycrystalline thin films will be affected by the behavior of grain boundaries, but this will not be an issue for single crystal films, so we shall consider single crystal films first.

Substrate

Fig. 4.11 Schematic of a thin film on a substrate. The film thicknesses of interest to industry are typically submicron.

Consider a single crystal thin film of thickness h f . If h f is sufficiently large, the elastic behavior of the thin film material will simply be that of the bulk material. As the thickness decreases, the fraction of atoms that feel the free surface and interface conditions increases, and the overall behavior of the thin film will become perceptibly different. When the behavior of interest is elasticity, at what thickness do the surface and interface effects become significant? A good model for the elasticity of thin films should predict the transition thickness. The concepts of surface stress and surface energy have both been used to analyze the elasticity of thin-film structures. The thin film essentially has one free surface (two free surfaces if it is a freestanding film), and the atoms on the free surface have different equilibrium positions than in the bulk. The atoms at the interface with the substrate also have positions that are different from the equilibrium positions in the bulk, because they

4 Mechanical Properties: Density and Elasticity

111

Fig. 4.12 Schematic of a polycrystalline thin film on a substrate. Note the typical columnar microstructure.

must interact with the different atoms of the substrate. An excellent description of the effects of surfaces is presented by Dingreville et al. (Dingreville et al., 2005). The stiffness of a thin film is often difficult to measure, because freestanding thin films are difficult to generate, so the quality of the dataset on thin films is sometimes contested. Nanoindentation can be used to measure elastic response for thin films on a substrate, but even this technique fails when h becomes of the order of 2–3 times the indentation depth. Microtension tests on freestanding thin films have recently been performed for a small number of films, almost all of them polycrystalline rather than single crystal, e.g., Jonnalagadda et al. (2008). Polycrystalline thin films provide us with an excellent opportunity to understand the effects of size scale and microstructure on elastic behavior. In terms of elastic response, single crystals are anisotropic, with up to 21 elastic moduli Cijkl (or CIJ in the Voigt notation) as described in Chapter 2. For the purposes of modeling, we assume that the elastic behavior of the grain boundary is isotropic, and therefore described by two elastic moduli, e.g., the Young’s modulus Egb and the Poisson’s ratio νgb . A polycrystalline thin film is thus a composite (Figure 4.12) made up of an anisotropic elastic phase (the single crystals) and an isotropic elastic phase (the grain boundaries). In general, therefore, the polycrystalline thin film will be elastically anisotropic, with specific effective moduli that depend on the arrangement of the grains as well as the volume fractions of the two phases. Calculating the effective anisotropic moduli of the thin film can involve a fairly complex computation.

4.3 Composites and Homogenization Theory How does one define the effective moduli of a material like that shown in Figure 4.12? The elastic moduli reflect a linear relationship between the stress and the strain in the material, as in Equation (2.65), and the constitutive law (material behavior) of linear elasticity is defined at every point. This is a problem, however, in composite materials, because any given point in the material might be in one phase

112

Nanomaterials

(a)

(b)

Fig. 4.13 Process of homogenization of a composite material. (a) Original heterogeneous material. (b) Equivalent homogenized material.

or the other (in this case either in the crystal or within a grain boundary region). Thus, even to define a constitutive law like linear elasticity for a composite material like this polycrystal involves replacing the real heterogeneous material (with grains and grain boundaries) with an equivalent homogenized material which has material properties that in some specified way are related to the properties of the original heterogeneous material. This process of homogenization is shown schematically in Figure 4.13. The behavior of the homogenized material is now defined at every point as a relationship between the overall stress and overall strain in the homogenized material. The overall stress σ i j in the homogenized material is defined as a volume average stress over a representative volume element or RVE (in the heterogeneous material), which is a volume element that is sufficiently large that it captures the effective behavior of the composite. Mathematically, the definition of the overall stress is  1 (4.22) σij = σi j dv, VR where VR is the volume of the RVE. Similarly, the overall strain is defined by the volume average of the strain over a representative volume element: 1 εi j = VR



εi j dv.

(4.23)

Note that the stress and strain components in the two phases within the RVE will vary from point to point, but the overall stress and strain computed using Equations (4.22) and (4.23) are defined at a point corresponding to the center of the RVE. If the RVE is defined by an effective dimension (say a radius) rR , we cannot in such a homogenized approach usefully examine behavior at length scales smaller than rR . Thus the process of homogenization introduces a modeling length scale into the problem, and this should ideally be associated with a microstructural length scale in the heterogeneous material. Now that the stress and strain in the homogenized material are known by computation from those in the composite, the effective elastic properties Cijkl of the homogenized composite can be defined through the relation

σ i j = Cijkl ε kl .

(4.24)

4 Mechanical Properties: Density and Elasticity

113

Using Equation (4.22) and the definition of the elastic modulus σi j = Cijmn εmn at each point in the heterogeneous material, we have Cijkl ε kl =

1 VR



Cijmn εmn dv.

(4.25)

Because the elastic moduli Cijmn of each phase vary from point to point, it is in general not possible to pull these moduli out of the integral. Equation (4.25) therefore provides a nice formal way to define the effective elastic properties but it is not easy to compute Cijkl . In some special cases, however, this can be done explicitly, and we demonstrate this next.

4.3.1 Simple Bounds for Composites, Applied to Thin Films Consider the simpler idealized case of a polycrystalline thin film consisting of isotropic elastic grains (with elastic modulus Eg and Poisson’s Ratio νg ) with isotropic elastic grain boundaries (with elastic modulus Egb and Poisson’s Ratio νgb ). Even though both of the constituent phases are isotropic, it is possible for the composite material to be anisotropic. For example, consider one possible microstructure (Figure 4.14a), where the grains are all grown perpendicular to the plane of the film (this is called a columnar microstructure, and is fairly common in polycrystalline thin films). The effective elastic properties of this thin film are different in the directions perpendicular to the film plane and parallel to the film plane, so that this film is effectively anisotropic. A simple calculation of these effective properties follows. Let us idealize the thin film as the two-dimensional microstructure shown in Figure 4.14a, corresponding to grains that are infinitely thick perpendicular to the plane of the paper. Imagine that the thin film is being pulled along the film plane by a tensile force P. The stress in the film plane is then uniform, and is simply given by σ = PA , where A is the cross-section area of the film. That is

Fig. 4.14 (a) Schematic of a columnar microstructure for a thin film. (b) Schematic of a layered microstructure, which can be viewed as the columnar microstructure loaded in the orthogonal direction to that shown in (a).

114

Nanomaterials

σtotal = σg = σgb =

P , A

(4.26)

while the total strain in the film is the sum of the strains in the grains and the grain boundaries in the material. Given the volume fractions in the material, we have

εtotal =

1 V



ε dv = fg εg + fgb εgb

(4.27)

using an argument identical to that in Equation (4.3). Using the definitions of the Young’s modulus of the grains and the grain boundaries, we have

εtotal = fg But by definition we have εtotal =

σgb σg + fgb . Eg Egb

σtotal , E inplane

(4.28)

so that we have

σgb σg σtotal = fg + fgb , Eg Egb E inplane

(4.29)

from which it follows that (using Equation 4.26) 1 1 1 . = fg + fgb Eg Egb E inplane

(4.30)

This last equation represents the rule of mixtures for elastic compliance. The compliance is the inverse (and in this isotropic case, the reciprocal) of the elastic modulus. Solution of this equation provides the effective modulus of the thin film in the in-plane direction. If, on the other hand, we consider the elastic properties of the thin film in the direction perpendicular to the thickness of the film (putting aside for the moment the physical difficulty of carrying out such an experiment), the mechanical loading situation is essentially that corresponding to Figure 4.14b. Now the displacements of the grain and grain boundary phases are the same, and since they have the same initial lengths, the strains in the grain and grain boundary phases are the same. It follows that now we have

εtotal = εg = εgb =

δh , h

(4.31)

while the total stress in the film is the sum of the stresses in the grains and the grain boundaries in the material. Given the volume fractions in the material, we have

σtotal =

1 V



σ dv = fg σg + fgb σgb .

(4.32)

Again using the definitions of the elastic moduli, we have E outofplane εtotal = fg Eg εg + fgb Egb εgb ,

(4.33)

4 Mechanical Properties: Density and Elasticity

115

from which it follows that (using Equation 4.31) E outofplane = fg Eg + fgb Egb .

(4.34)

This last equation represents the rule of mixtures for the elastic modulus directly, and provides the effective modulus of the thin film in the out-of-plane direction. This out-of-plane effective Young’s modulus is clearly different from the in-plane effective Young’s modulus, and thus the thin film behaves in an anisotropic manner even though it is composed of two isotropic components (the grains and the grain boundaries). Real thin film microstructures may not be well-represented by the idealized case in Figure 4.14. However, it is possible to show that any other microstructure for the material will result in effective moduli that lie between the two bounds represented by Equations (4.30) and (4.34) for linear elastic materials. Thus, all polycrystalline linear elastic solids (that do not contain pores) will have moduli between these two bounds, which are represented in Figure 4.15 in terms of variation with volume fraction of grain boundary. However, there are two big caveats. First, in a composite model such as that shown in Figure 4.14, one makes an assumption about the behavior of the interface between the two materials (typically, the assumption is that the interface is perfect). This assumption seems trivial here, because we do not have a conception of an interface between a grain and a grain boundary. In general composite mechanics terms, however, the interface behavior may be a dominant issue. For example, in the simple case that the interface between the two materials sustains no stress, the material in Figure 4.14a will carry no load, while the material in Figure 4.14b will carry load, so that the behaviors are dramatically different from the behaviors shown in Figure 4.15. Second, one typically assumes that the properties of the individual phases are independent of size – only the volume fraction shows up in Equation (4.34), without a length scale.

75

75

Overall Modulus, GPa

ESC 70

70

65

65

E1 = fgbEgb + (1–fgb)Esc 60

60

1/E2 = fgb /Egb + (1–fgb)/Esc

55

55 50 Egb

50 45

0

0.2

0.4

0.6

0.8

1

45

Volume fraction of grain boundary, fgb Fig. 4.15 Variation of effective modulus (for fictitious material) with grain boundary volume fraction, based on Equations (4.30) and (4.34).

116

Nanomaterials

4.3.1.1 Size Effects in Thin Films The primary cause of the size effects that we have been discussing in this polycrystalline material is the lengthscale associated with the thickness t of the grain boundary domain. This lengthscale will exist in ANY polycrystalline material and is not specific to thin films. The size scale that is specific to thin films is the film thickness h f . As discussed earlier, even single crystal thin films have elastic properties that are a function of film thickness. In the case of polycrystalline thin films, both film thickness effects and grain size effects will exist. As the grain size decreases from ∞ (corresponding to the single crystal), the fraction fgb of the thin film material that is in the grain boundary domain will increase, while the fraction fsurf of the material that is in the surface-affected domain will remain constant. At some specific grain size, these two volume fractions will become comparable.

4.3.2 Summary of Composite Concepts There are several important concepts that should be extracted from the previous discussion on polycrystalline materials: • First, even when we are talking about a scalar parameter (the Young’s modulus), the issue of interest is that of the mechanical behavior in tension, which has a principal direction (the tensile axis). The film with the columnar microstructure of Figure 4.14 has two very different behaviors in the two tensile directions (inplane and out-of-plane). This is because the material microstructure itself has a directionality (e.g., the direction perpendicular to the grain boundary planes in Figure 4.14a, b), and the anisotropy (directionality) of the mechanical response follows from the relationship between the tensile axis and the normal to the grain boundary planes. The directionality of the arrangement of two individual phases (the grain boundary and the grains) results in this anisotropy, even though each of the phases is itself isotropic in mechanical response. This is one of the fundamental concepts of composite mechanics. • Second, most nanomaterials will not have the nice regular microstructure shown in Figure 4.11. In general, the arrangement of the phases is not periodic in any particular direction. For the general microstructure, we cannot easily extract effective moduli. However, composite mechanics theory places bounds on the moduli, as exemplified by Equations (4.30) and (4.34). Any given sample of material will have moduli that lie within those bounds. This concept of bounds on the properties is also a fundamental concept of composite mechanics. • Third, specific assumptions (related to the microstructures) had to be made to obtain the effective response of the nanomaterial. In one case it was assumed that the strains were uniform (i.e., the same in both grain and grain boundary), and in the other that the stresses were uniform. These assumptions are idealizations that lead to specific models, corresponding to Equations (4.30) and (4.34). The

4 Mechanical Properties: Density and Elasticity

117

development of models is always based on the assumption of specific idealizations, and later, specific mechanisms. • Fourth, we are NOT assuming that there is a different phase of the material at the grain boundary (as does exist, for example, in some sintered ceramics). Rather, the assumption is that the behavior of material in a finite thickness domain at the grain boundary is different from that of the bulk material. This is a point that causes some confusion in the literature. The fact that a different crystallographic or amorphous phase cannot be identified at the grain boundary does NOT imply that the models described above cannot be used – all that is needed is a different response at the grain boundary (in this case a lower apparent stiffness), which may arise merely from the increased disorder or looser packing at a boundary.

4.4 Elasticity of Bulk Nanomaterials Bulk nanomaterials are either nanocrystalline or nanostructured materials, as described in Chapter 1. Much of our discussion in the previous section applies directly to bulk nanocrystalline materials, and in particular, Equations (4.30) and (4.34) describe the bounds on the elastic properties that will be obtained (see also Figure 4.15). The predicted variation of the effective modulus of the material with grain size is presented in Figure 4.16.

Normalized Modulus E /Esc

1

0.8

0.6

0.4

0.2

0 0.1

E1

Cube Morphology 1

E2

10

100

1000

Normalized grain size d/t Fig. 4.16 Variation of effective Young’s modulus with normalized grain size in a bulk nanocrystalline material, where the modulus in the grain boundary domain is defined to be ζ Eg = ζ Esc with ζ = 0.7 in this figure. The grain size is normalized by the effective thickness t of the grain boundary domain, typically assumed to be about 1 nm. If the latter thickness is assumed, the horizontal axis corresponds to grain size in nm. E1 and E2 correspond to Equations 4.30 and 4.34.

118

Nanomaterials

Note that in either case (for either bound), a 5% drop in the effective modulus is predicted to occur only when the grain size falls below about 20t, and for t ≈ 1 nm this implies grain sizes less than about 20 nm. The behavior is essentially that of the grain boundary once d ≈ t. There is very little reduction in modulus expected for grain sizes in the range d > 50 nm (although, for the reasons pointed out in the discussion on density effects, this depends on the assumed grain morphology). However, experimental measurements of the moduli of nanocrystalline materials often show values substantially smaller than those of their conventional grain size counterparts (Shen et al., 1995). This suggests that some component of these materials has not been included in the modeling. Examination of the literature shows that the measured moduli are relatively insensitive to grain size in the nanocrystalline range, and change considerably with the route of synthesis of the nanomaterial. It is likely, therefore, that the observed reduction in modulus is due to defects such as porosity generated during the synthesis of the material (Shen et al., 1995). We will return to this issue of the effects of defects and porosity on the mechanical properties of materials in detail in a later section.

4.5 Suggestions for Further Reading 1. R.M. Christensen, Mechanics of Composite Materials, John Wiley, New York, 1979. 2. J.-M. Berthelot, Composite Materials: Mechanical Behavior and Structural Analysis, Springer, New York, 1999. 3. R. Dingreville, J.M. Qu, and M. Cherkaoui, Surface free energy and its effect on the elastic behavior of nano-sized particles, wires and films, Journal of the Mechanics and Physics of Solids, 53, 8, 1827–1854, 2005. 4. R. Cammarata, Surface and interface stress effects in thin films. Progress in Surface Science, 46, 1, 1994.

4.6 Problems and Directions for Research 1. Consider a nanocrystalline nickel material. Assuming that the grain boundary region has 90% of the Young’s modulus of the interior of the single crystal, and assuming a 1 nm thickness for the affected layer, compute the apparent Young’s modulus of nanocrystalline gold with a grain size of 20 nm and cube-morphology grains. How will the modulus change if the material has 2% porosity? 2. Solve the identical problem, but this time assume a hex-prism morphology with an aspect ratio of 5. 3. Put together a table of the elastic moduli of bulk nanocrystalline materials, using only experimental data from the literature. How do these numbers compare with the bulk response?

4 Mechanical Properties: Density and Elasticity

119

References Ashby, M. F. and D. Jones (1996). Engineering Materials I (Second ed.), Volume 1. ButterworthHeinemann. Chen, C. Q., Y. Shi, Y. S. Zhang, J. Zhu, and Y. J. Yan (2006). Size dependence of young’s modulus in zno nanowires. Physical Review Letters 96. Dingreville, R., J. Qu, and M. Cherkaoui (2005). Surface free energy and its effect on the elastic behavior of nano-sized particles, wires and films. Journal of the Mechanics and Physics of Solids 53(8), 1827–1854. Huang, Z. P., P. Thomson, and S. Di (2007). Lattice contractions of a nanoparticle due to the surface tension: A model of elasticity. Journal of Physics and Chemistry of Solids 68, 530–535. Jonnalagadda, K., N. Karanjgaokar, I. Chasiotis, A. Mahmood, and D. Peroulis (2008). Strain rate, thickness, and processing effects on the mechanical behavior of nanocyrstalline au films. Acta Materialia to be submitted. Lamber, R., S. Wetjen, and N. Jaeger (1995). Size dependence of the lattice parameter of small palladium particles. Physical Review B 51, 10968–10971. Latapie, A. and D. Farkas (2003). Effect of grain size on the elastic properties of nanocrystalline alpha-iron. Scripta Mater 48, 611–615. Ma, F. and K. Xua (2006). Size-dependent theoretical tensile strength and other mechanical properties of [001] oriented au, ag, and cu nanowires. Journal of Materials Research 21(11), 2810– 2816. Shen, T. D., C. C. Koch, T. Y. Tsui, and G. M. Pharr (1995). On the elastic moduli of nanocrystalline fe, cu, ni and cu-ni alloys prepared by mechanical milling/alloying. Journal of Materials Research 10(11), 2892–2896. Silva, E. C. C. M., L. Tong, S. Yip, and K. J. V. Vliet (2006). Size effects on the stiffness of silica nanowires. Small 2(2), 239 – 243. Wang, Z. (2004). Mechanical properties of nanowires and nanobelts. In J. Schwarz (Ed.), Dekker Encyclopedia of Nanoscience and Nanotechnology (Second ed.). New York: Marcel Dekker. Yu, M. F., B. S. Files, S. Arepalli, and R. S. Ruoff (2000a). Tensile loading of ropes of single wall carbon nanotubes and their mechanical properties. Physical Review Letters 84(24), 5552–5555. Yu, M. F., O. Lourie, M. J. Dyer, K. Moloni, T. F. Kelly, and R. S. Ruoff (2000b). Strength and breaking mechanism of multiwalled carbon nanotubes under tensile load. Science 287(5453), 637–640.

5

Be faithful in small things, because it is in them that your strength lies. Mother Theresa

Plastic Deformation of Nanomaterials This chapter discusses the much more complex mechanical behaviors associated with inelastic deformations. Although the grain size, volume fraction and composite approaches discussed in the last chapter are still relevant, the interactions between the size scales and the deformation mechanisms involved in plasticity do not always allow the simple use of continuum approximations to predict the overall behavior. The chapter will first use a materials approach to discuss mechanisms, and then attempt to link the mechanics to these mechanisms and thus define the associated size scales.

5.1 Continuum Descriptions of Plastic Behavior Once the stress (or strain) in a material has increased to the elastic limit, plastic behavior or fracture occurs. Plasticity is a behavior that is generally only observed in metals rather than in ceramics or polymers (except under exceptional circumstances such as very high temperatures or pressures), and so we focus on bulk nanometals in this chapter. The plastic behavior of nanomaterials is not generally relevant to the applications of discrete nanomaterials and nanodevice materials. We focus therefore on the plasticity of bulk nanomaterials. There are essentially four parameters that describe the simple plastic behavior of a material (assumed isotropic, and following the typical von Mises type of behavior, see Chapter 2.): • The yield strength σy

ε ∂σ σ ∂ε σ ε˙ ∂ σ strain rate hardening index m, where m = ∂∂ ln ln ε˙ = σ ∂ ε˙ σ thermal softening coefficient ν , where ν = ∂∂ ln ln θ , with

• The strain hardening index n, where n = • The

∂ ln σ ∂ ln ε

=

• The temperature

K.T. Ramesh, Nanomaterials, DOI 10.1007/978-0-387-09783-1 5, c Springer Science+Business Media, LLC 2009 

θ being the absolute

121

122

Nanomaterials

All of these are generally known for conventional grain size metals. It is known that all four parameters are strong functions of the homologous temperature Th = TTm (where Tm is the melting temperature), but for the purposes of this discussion let us limit ourselves to considering low homologous temperatures (Th ≤ 0.3). The early work on nanomaterials was confined to evaluating the yield strength, and indeed, much of the early apparent promise of structural nanomaterials was in the development of very high strengths at small grain sizes. The other three parameters are more difficult to measure, since they relate to the derivatives of the strength with respect to the strain (in the case of n), the strain rate (m), and the temperature (ν ). It is only since 2004 or so that attention has become focused on the strain hardening and strain rate hardening behaviors (in part because synthesis and characterization techniques have achieved sufficient reliability to make such evaluations possible). There is as yet little data on the thermal softening of nanomaterials, largely because of the competition with grain growth mechanisms. In this chapter, we will address each of these parameters in turn.

5.2 The Physical Basis of Yield Strength By definition, the onset of plastic deformation in metals is associated with the permanent rearrangement of atoms and atomic bonds (because irreversible deformations must result). The physical basis of the yield strength is thus completely different from the physical basis of the elastic modulus. What is the stress required to cause a permanent rearrangement of the atoms of a given crystalline metal? As with most questions involving irreversible deformations, there are many ways of thinking about this question from a physical viewpoint. First, of course, we must define what “permanent rearrangement of the atoms” means. Consider the perfect crystal shown in Figure 5.1a and subjected to a shearing stress as shown. If we imag-

Fig. 5.1 Schematic of slip in a crystalline solid under shear. (a) Original crystal subjected to shear stress. (b) Crystal after deformation (slip) along the shearing plane shown in (a). The final atomic arrangement within the deformed crystal remains that of the perfect crystal, except near the free surface.

5 Plastic Deformation of Nanomaterials

123

ine the specific permanent rearrangement consisting of the sliding of one layer of atoms over another along the dashed line in the figure, we see that this rearrangement involves the motion of the top half of the crystal over the bottom half. At the end of the sliding process, the deformed crystal (Figure 5.1b) has two ledges on the free surfaces corresponding to the sliding. Within the crystal however, even after deformation, the atomic arrangement looks identical to that in a perfect crystal. Thus merely examining the interior of a crystal after deformation does not allow one to determine whether a permanent rearrangement of the atoms has occurred. In this case the signature of the permanent rearrangement is actually at the surfaces of the crystal. Another signature of the permanent rearrangement is that a significant quantity of non-recoverable work must have been done in order to accomplish the sliding. We see here the two key signatures of plastic deformation. First, the observable permanent deformations are macroscopic, not at the atomic scale; the net result is a continuum permanent strain. Second, plastic deformation involves the expenditure of energy, known as the work of plastic dissipation. The two classical ways of examining plastic deformation therefore involve the examination of the development of plastic strain, and the examination of the development of plastic dissipation. The former involves a close examination of the kinematics of the deformation process, and the latter involves a close examination of the energetics of the process. How does the crystal actually go from configuration (a) to configuration (b) in Figure 5.1? It turns out to be energetically unfavorable to move an entire half of a crystal over the other half (put another way, the stress required to perform such µ where µ is the shear an operation would be prohibitively large, of the order of 10 modulus of the material (Asaro and Lubarda, 2006), whereas the measured yield strength of most metals is of the order of 10−3 µ ). It is far cheaper, in terms of energy, to cause the overall sliding of the top over the bottom (Figure 5.1) by moving small defects along the slip plane until the entire block has slid across (Figure 5.3). The nearest everyday analogy is that it is easier to move a large carpet a short distance by generating a ripple in the carpet and then moving the ripple across the carpet, rather than moving the entire carpet at one time. The corresponding ‘ripples’ in the crystal are line defects (i.e., defects in which the atomic arrangement deviates from the perfect crystal along a line) called dislocations, and were first hypothesized in the 1920s by G.I. Taylor. The existence of such line defects has since been established through direct visualization in a transmission electron microscope (Figure 5.2). Figure 5.3 shows a schematic of one such line defect, called an edge dislocation, which amounts to an extra plane of atoms in the otherwise perfect lattice (the line at the bottom of the extra plane is essentially the line defect). The motion of this line defect is energetically favorable to the motion of the entire block of atoms above the slip plane at one time. This turns out to be a general principle: the inelastic deformation of crystalline metals is controlled by the motion of defects, because this motion is energetically much cheaper than large scale simultaneous breaking and rearranging of atomic bonds. Almost every crystal contains defects, many of which are developed during the process of nucleation and growth of the crystal. The defects in crystals occur at four

124

Nanomaterials

Fig. 5.2 Dislocations are visible in the transmission electron microscope. The picture shows dislocations (the lines marked by the black arrow) in a magnesium alloy (ZK60). Note the length scale in the picture. This micrograph was taken by Bin Li.

levels of dimensionality, all of which are of interest to the mechanics of nanomaterials: • 0D or point defects are either atoms missing from the lattice (therefore called vacancies), atoms of the wrong type in a given lattice position (substitutional defects) or atoms that are in located in the interstices between normal lattice positions (interstitial defects). Such point defects play a very important role in the high temperature deformation of materials, and in particular in the inelastic deformation process known as creep. • 1D or line defects are the dislocations previously discussed, and are of two broad types called edge dislocations (shown in Figure 5.3b) and screw dislocations (as the name suggests, a helical line defect). Dislocation motion dominates the plastic deformation of metals. • 2D or planar defects are primarily defects in the stacking sequence in a lattice, and so are called stacking faults. The boundaries of a stacking fault may be described in terms of line defects (dislocations). One particular consequence of stacking faults can be the development of regions of a crystal of different orientation called twins. • 3D or volume defects are essentially either pores or inclusions in a crystal. Such defects are typically immobile, but may be the sites for the nucleation of dislocations, as well as damage mechanisms such as cracks and voids, which can then lead to macroscopic failure.

5 Plastic Deformation of Nanomaterials

125

Fig. 5.3 Expanded version of Figure 5.1 showing the primary deformation mechanism that leads to plasticity in metals: the motion of line defects called dislocations. Edge dislocations are shown gliding along the slip plane. The inset shows an expanded view of an edge dislocation, amounting to an extra plane of atoms in the lattice (with the trace of the extra plane on the slip plane defined as the line defect). The motion of many dislocations results in the macroscopic slip step shown in (c).

Dislocations are the defects that control the plastic deformation of metals, and the onset of the motion of these line defects determines the yield strength. A thorough treatment of the theory of dislocations is beyond the scope of this book; excellent references include the classical work of Hirth and Lothe (1992). For our purposes, it is sufficient to identify dislocations in terms of two vectors: the line vector (a vector tangent to the dislocation line) which we shall call ξ , and a unit Burgers vector b which represents the disregistry in the crystal lattice when a circuit around the line

126

Nanomaterials

defect (Figure 5.3c) is completed (this is, in effect, one measure of the size of the defect in the lattice). In the case of edge dislocations, the line vector and the Burgers vector are perpendicular to each other, while in the case of screw dislocations, the line and Burgers vectors are parallel to each other (Figure 5.4). Thus for an edge dislocation we have ξ · b = 0, while for a screw dislocation we have ξ · b = ±b, where b is the magnitude of the Burgers vector (the line vector ξ is a unit vector). The slip plane or glide plane is that plane along which slip occurs and is defined as the plane with normal b × ξ . For a screw dislocation, we have b × ξ = 0, which implies that the glide plane is nonunique or indeterminate for a screw dislocation, so that a screw dislocation can move on a variety of planes at any time. The combination of a slip plane normal and a slip direction defines a slip system. In general, ξ (x,t), that is the line vector, varies with position in the crystal, but the Burgers vector for the dislocation is determined by the particular slip system that is active.

ξ b

bξ

Fig. 5.4 Schematic of slip with an edge dislocation (top figure) and screw dislocation (bottom figure), showing the Burgers and line vectors in each case.

5 Plastic Deformation of Nanomaterials

127

By definition, as a line defect, a dislocation line cannot begin or end inside the crystal (but it can begin or end at a free surface). Thus dislocation lines generally exist within crystals as dislocation loops or as line segments ending in free surfaces. A dislocation loop within the crystal will have, in general, both edge and screw configurations, as well as regions where the dislocation is of mixed type (both edge and screw). In such regions, the Burgers vector can be decomposed into two parts in terms of its relationship with the line vector: b = be + bs , where be · ξ = 0 (corresponding to the edge part of the mixed dislocation) and bs × ξ = 0 (corresponding to the screw part of the mixed dislocation). Note that for the mixed dislocation the pair be and ξ define the glide plane, since that plane must contain both the Burgers vector and the line vector. Since the dislocation represents a defect in the crystal, there is a stress field within the crystal in the region surrounding the dislocation (because the atoms are not in their ideal positions with respect to the interatomic potential). The stress field around a dislocation can be calculated using the theory of linear elasticity (note, however, that linear elasticity theory breaks down as one approaches within one or two atoms of the center of the defect, a region which is defined as the dislocation core). The stress field around a dislocation (both edge and screw) is observed to have a finite range (that is, the stress decreases rapidly with distance r, as 1r , as one moves away from the dislocation core). This finite range of the stress fields means that dislocations do not interact with each other until they get within some critical distance. Similarly, a dislocation that gets sufficiently close to a free surface effectively feels a force (called an image force) due to the existence of the free surface. The stresses that we have discussed in this paragraph are all self stresses, in that they are developed simply because of the existence of the defect, not because of the application of an external force. The consequence of the existence of these self stresses is that assemblies of dislocation lines can rearrange themselves to minimize the internal stresses, a process known as recovery that is particularly important at high homologous temperatures. What does all of this have to do with the yield strength? Recall, from the chapter on mechanics, that the plastic deformation of materials is primarily the result of shear stresses, and that the plastic deformations do not involve volume change. The movement of dislocations in metals results in the development of deviatoric (distortional) strains, but not of volumetric strains (e.g., look at Figure 5.3d). In crystalline metals, the plastic deformation turns out to be developed by the collective motion of large numbers of dislocations. When an external stress field is applied to a crystal containing dislocations, each dislocation feels a corresponding local force per unit length that results in motion of the dislocation if a critical condition is reached, corresponding to a critical value of the applied stress. The yield strength of the material is defined by the critical applied stress at which the dislocation motion begins. The direction of subsequent motion of the dislocation depends on the applied stress tensor and the character of the dislocation (as defined by the Burgers vector and the line vector). It has been experimentally observed that the movement of the dislocations in glide occurs only along specific combinations of slip planes and slip directions that

128

Nanomaterials

are defined for each crystal type. This specificity of the slip planes and slip directions arises from such features as the spacing of the planes and the packing of the atoms within a plane; close-packed planes are preferred for energetic reasons.

5.3 Crystals and Crystal Plasticity A discussion of crystal structure and crystallography is beyond the scope of this book, but the reader will benefit greatly from a knowledge of basic crystallography and in particular of Miller indices for describing planes and directions inside crystals; an excellent reference is Courtney (2005). The majority of metals of interest have crystal structures that fall into three categories (Figure 5.5): 1. Body centered cubic or bcc structures, example of which are iron (Fe), vanadium (V), tantalum (Ta), tungsten (W) and niobium (Nb) 2. Face centered cubic or fcc structures, example of which are aluminum (Al), copper (Cu), gold (Au), and silver (Ag) 3. Hexagonal close-packed or hcp structures, examples of which are titanium (Ti), zinc (Zn), zirconium (Zr) and magnesium (Mg)

Fig. 5.5 Examples of basic crystal structures that are common in metals: body centered cubic (one atom in the center of the cube), face centered cubic (atoms at the center of each cube face), and hexagonal close packed structures.

Directions in a cubic crystal are typically defined through Miller indices, which are sets of integers which can be viewed in terms of an orthonormal basis consisting of unit vectors e1 , e2 , e3 which are aligned with the axes of the cube. The notation [h, k, l] then defines a direction parallel to a unit vector with com√ ponents ( hs e1 + ks e2 + sl e3 ), where s = h2 + k2 + l 2 essentially normalizes the vector to unit length. Thus the set of Miller indices [1, 1, 2] represents the direction ( √16 e1 + √16 e2 + √26 e3 ). Negative directions are denoted by an overbar, e.g., ¯ represents the direction ( √1 e1 + √1 e2 − √2 e3 ). Families of similar di[1, 1, 2] 6

6

6

rections are conventionally denoted by < h, k, l >. Now, planes are essentially

5 Plastic Deformation of Nanomaterials

129

defined when the normal to the plane is defined, and the normal is just another direction, so a plane in the crystal can be defined in similar terms using Miller indices. In order to distinguish between planes and directions, the notation (p, q, r) using parentheses rather than square brackets  is used to define a plane with unit normal n = ( ps e1 + qs e2 + rs e3 ), where s = p2 + q2 + r2 . Thus the notation (1, 1, 1) represents a plane with unit normal n = ( √13 e1 + √13 e2 + √13 e3 ). Examples of such planes are shown in Figure 5.6. Families of similar planes (similar in the sense of crystal symmetry) are conventionally denoted by {p, q, r}. (0,0,1)

(0,1,0)

x

(1,0,0)

x

Z

x

Z

Z

Y

Y (1,1,0)

Y (2,0,0)

(1,1,1)

x

x

x

Z

Z

Y

Z

Y

Y

Fig. 5.6 Typical planes in a cubic crystal defined using Miller indices. The (111) plane and similar {111} planes are close-packed planes in a face-centered-cubic (fcc) crystal and therefore are part of typical operating slip systems (together with < 110 > type directions).

Hexagonal close packed or hcp metals do not have cubic symmetry, and therefore cannot be described using just the three integers defined in the last paragraph; instead, sets of four numbers are used, e.g., [h, k, l, m]. The typical approach here is to use three vectors defining directions in the basal plane of the crystal, and one vector defining the normal to the basal plane. Clearly only two vectors are needed to define any direction in the plane, so only two of the first three vectors are independent. The reader should find a discussion of hcp notation in any textbook for materials scientists that discusses crystal structure, e.g., Courtney (2005). Now that we can define planes and directions in a crystal, we can list the various slip planes and slip directions in typical crystal structures. The combination of a slip plane (defined by the normal n) and a slip direction (defined by the vector s) is

130

Nanomaterials

Table 5.1 Slip systems with typical numbering scheme in typical fcc metals. System Primary system

Number Designation Normal Slip direction ¯ 1 B4 (111) [101] ¯ 2 B5 (111) [011] ¯ ¯ 3 B2 (111) [110] ¯ Conjugate system 4 C1 (111) [110] ¯ 5 C5 (111) [011] ¯ ¯ 6 C3 (111) [101] ¯ Cross-glide system 7 D4 (11¯ 1) [101] ¯ 8 D1 (11¯ 1) [110] ¯ ¯ 9 D6 (11¯ 1) [101] ¯ Critical System 10 A3 (111) [011] ¯ 11 A6 (111) [011] ¯ 12 A2 (111) [110]

called a slip system. The slip systems that are commonly observed in fcc structures are presented in Table 5.1. Since dislocation motion only occurs within specific slip systems, the stress associated with the onset of plastic deformation (i.e., yield) is related to the stress on those slip systems. The various slip systems in a given crystal structure are typically presented in some specific order, and then numbered (thus one speaks of the second slip system in fcc metals, and so forth). There are 12 slip systems in fcc metals, and as many as 48 in bcc metals (Table 5.2). Let us consider using these concepts together with the mechanics concepts that we learned in Chapter 2. When an external stress σ (x,t) is applied to the crystal, specific tractions (Equation 2.41) will be developed on every slip plane within the crystal. Consider a slip plane with normal n; the traction vector (Figure 2.1) at a point on the plane is given by the operation of the applied stress tensor on the normal vector, t = σ (x,t)  n. The resulting traction vector is essentially a vector made up of stress components, see Chapter 2, and will in general have components normal to the plane (defining the normal stress on the plane) as well as in the plane. The component of the traction vector in the slip direction s will be the shear stress in the slip direction, and is given by

τ = t · s = (σ n) · s = σi j n j si ,

(5.1)

where the last term represents the component form of the expression (see Chapter 2). Equation (5.1) provides the resolved shear stress on a given slip system; changing the slip plane normal and the slip direction changes the slip system, and so will result in a different resolved shear stress. That is, for a given applied stress σ , the α th slip system (which has slip plane normal nα and slip direction sα ) will have a resolved stress τ α . For a given crystal orientation and given applied stress, some particular slip system will have the maximum resolved shear stress in comparison to all of the available slip systems – we will refer to this as the maximal slip system. It has been observed experimentally that in most metals dislocation motion begins when the resolved shear stress on the maximal slip system reaches a critical value, known as the Critical Resolved Shear Stress (CRSS) C: that is, the onset of dislocation motion corresponds to

5 Plastic Deformation of Nanomaterials

131

τ α ≥ Cα .

(5.2)

This is sometimes called the Schmid law. The critical resolved shear stress Cα in Equation (5.2) is the slip system equivalent of yield strength, and defines the onset of plastic deformation. Although the convention is to call Cα the critical resolved shear stress, we shall refer to it as the critical resolved shear strength, because we need to distinguish applied stresses from the material property that is the strength. When a crystal is subjected to an externally applied stress, the slip system with the largest resolved shear stress is the one on which dislocation motion is likely to start, and the external stress at which this motion begins essentially defines a yield strength. It is worth noting that bcc metals seem to show significant non-Schmid effects, i.e., they do not necessarily follow equation 5.2, but these concepts continue to be used for bcc metals as a simplifying assumption. Note that, by the definition above, the yield strength is a function of crystal orientation, reflecting the fundamental anisotropy of plastic deformation at the single crystal level. Simultaneously, note that the plastic deformation occurs along specific slip planes in specific slip directions, reflecting the fundamental heterogeneity of plastic deformation at the single crystal level. Single crystal plasticity is inherently anisotropic and heterogeneous, and in that sense the “strength” of a single crystal is not a well-defined single quantity. The plasticity concept of a single yield strength number for a material is a result of the averaging of a variety of local orientationdependent yield strengths for the very large number of crystals within a polycrystalline metal, and the plasticity concept of a plastic strain at a point is a result of the averaging of a very large number of heterogeneous dislocation motions over a number of slip systems. Experiments on single crystals are difficult to do, not least because the development of slip on a given slip system results in an effective rotation (because shear involves a rotation), and consequently the resolved shear stress on a given slip system changes with continuing deformation if the loading axis remains fixed. Thus even in single crystals, the strength values that are obtained from experiment are usually averages over a number of slip systems. To summarize, the physical basis of yield strength is the stress required for the onset of the motion of dislocations in crystals along specific slip planes and in specific slip directions, combined with averaging over a large number of crystals within a polycrystalline material. Two consequences of this physical basis are of interest: • First, the apparent yield strength of a material is a convolution of the fundamental behavior of single crystals of the material and the arrangement of the grains within the material. For example, the yield strength of a material where all of the grains are oriented in one direction (these oriented polycrystalline structures are said to be textured) may well be different from the yield strength of the material when the grains are randomly oriented. • Second, the yield strength of a material can be controlled by making it difficult to move the dislocations, for example by placing obstacles in the path of dislocations. Obstacles of interest can include dispersoids, particles, and grain boundaries.

132

Nanomaterials

Table 5.2 Known slip systems in a variety of metals, from a table put together by Argon (Argon, 2008). Crystal Structure Cu, Au, Ag, Ni, CuAu, α -CuZn, AlCu, AlZn Al

Lattice Type FCC

α -Fe

BCC BCC BCC BCC BCC HCP c/a>1.85

Mo, Nb, Tz, W, Cr, V Mo, Nb, Tz, W, Cr, V Cd, Zn, ZnCd

FCC

Mg

HCP c/a = 1.623

Be

HCP c/a = 1.568

Ti

HCP c/a = 1.587

Ge, Si, ZnS As, Sb, Bi

Diamond cubic Rhombohedral

Slip plane Slip direction {111} 110 111 100 110 112 123 111 110 112 (0001) ¯ (0010) ¯ (1122) (0001) ¯ 1011 ¯ 1010 (0001) ¯ 1010 ¯ 1010 ¯ 1011 (0001) 111 (111) ¯ (111)

Reference Seeger (1958)

110 110 111 111

Seeger (1958)

111 111 ¯ (21¯ 10) ¯ [1120] ¯ [1¯ 123] ¯ 21¯ 10 ¯ 21¯ 10 2110 ¯ 21¯ 10 ¯ 21¯ 10 ¯ 21¯ 10 ¯ 21¯ 10 ¯ 21¯ 10 101 ¯ [101] [101]

Seeger (1958)

Seeger (1958)

Seeger (1958)

Seeger (1958)

Seeger (1958) Seeger (1958)

Seeger (1958) Schmid and Boas (1935)

In the next two sections, we first discuss the strengthening mechanisms in single crystal metals, and then examine the strength of polycrystalline metals. We will then describe ways of determining the overall response of engineering materials.

5.4 Strengthening Mechanisms in Single Crystal Metals There are multiple ways of increasing the strength of metals, not the least of which is decreasing the grain size. We discuss the various mechanisms within single crystals here, leaving aside the interaction of multiple crystals for a later section. There remains the very interesting question of whether it is appropriate to assume that the final yield strength of the material is obtained by simply adding the strength increments provided by each individual mechanism, or whether it makes more sense to add up the inelastic strain increments provided by each mechanism for a fixed applied stress. From a conceptual viewpoint, this question is identical to the one of whether we should assume that the strains are uniform or the stresses are uniform in a polycrystalline material. For the moment, we will follow the popular approach of adding the strength increments from each mechanism, and consider other options later. An excellent and very detailed discussion of strengthening mechanisms is provided in the recent book by Argon (Argon, 2008).

5 Plastic Deformation of Nanomaterials

133

5.4.1 Baseline Strengths The most fundamental resistance to dislocation motion through a crystal lattice comes from the lattice itself, and is called the Peierls stress or Peierls-Nabarro stress, which we shall call σPN , the Peierls-Nabarro strength. The magnitude of the Peierls-Nabarro strength varies from one material to another, but also varies substantially on the basis of crystal structure. The Peierls-Nabarro strength is quite high for bcc crystals and quite low for fcc crystals, and so the strength of pure bcc metals is typically higher than that of pure fcc metals. In many ways the Peierls-Nabarro strength (also called the friction stress) of the pure metal represents a baseline strength for a material, and so this quantity is sometimes viewed as a fundamental strength. However, this stress can be very low in fcc metals (≈ 10−6 µ , where µ is the shear modulus), so much so that it is essentially ignored from an applications perspective. Modifications to the material, such as the addition of alloying elements, will increase the strength above this fundamental strength.

5.4.2 Solute Strengthening The addition of solute atoms (by alloying) can make it more difficult to move dislocations and thereby increase the strength of a material. It is typically found that a slightly impure metal has a much higher strength than the very pure metal, and that different solute atoms provide different levels of strengthening. For most solute atoms, the additional strengthening ∆ σss provided by a solute can be described in terms of a power law function:

∆ σss = Kcn¯ ,

(5.3)

where c is the concentration of the solute atom and K and n¯ are solute-dependent parameters. Thus, for example, the addition of small amounts of magnesium can greatly strengthen aluminum, leading to a class of Al-Mg alloys (including, with some additional compositional variations, the structural aluminum alloy Al 5083). The addition of 4.8 atomic % of magnesium to aluminum raises the strength by 67 MPa. This strengthening occurs largely because the solute atoms diffuse towards the core of dislocations because of the local stress state there, and this essentially pins the dislocations, making it difficult for them to begin to move. There is no length scale associated with this strengthening mechanism except in that the spacing of the solute atoms decreases as the concentration increases. Since the operation of the strengthening mechanism is essentially through a cloud of solute atoms, this length scale is essentially irrelevant, and this strengthening mechanism should be observed over nearly all specimen sizes down to the 100 nm scale.

134

Nanomaterials

5.4.3 Dispersoid Strengthening Another way to strengthen materials is to put obstacles in the path of the dislocations, such as small particles called dispersoids (typically 10–100 nm in diameter). When a dislocation line moves through the crystal and bumps up against such a particle, it has three options: (a) stop, (b) cut through the particle, or (c) move around the particle through the rest of the crystal (e.g., climb over the particle). If the particle is too strong to cut through, and if the dislocation does not have the character (combination of b and ξ ) to move around the particle, then that segment of dislocation line behind the particle comes to a stop. However, parts of the dislocation line farther away from the particle continue to move, resulting in the dislocation taking on a curved shape. If the particles have some spacing L, then the dislocation takes on the bowed shape shown in Figure 5.7.

Fig. 5.7 Schematic of gliding dislocation bowing around dispersoids that are periodically spaced a distance L apart along a line.

This is called the Orowan bowing mechanism, and the corresponding strengthening is called Orowan strengthening. The additional strengthening obtained through the Orowan mechanism is (Arzt, 1998)

∆ σds = A1

µb , L − 2a

(5.4)

where A1 is a prefactor of order 1, L is the particle spacing (which depends on the dispersoid volume fraction), and a is the particle radius. Note that there is a relationship between the particle spacing, the particle size, and the particle volume fraction (for the same volume fraction, smaller particles must be closer together). This strengthening mechanism is the dominant one in many engineering materials, and can lead to strength increases of the order of 102 MPa. Note that a length scale defined by the particle spacing now exists in the material; this particle spacing is typically much less than the grain size in conventional materials.

5 Plastic Deformation of Nanomaterials

135

5.4.4 Precipitate Strengthening A variety of particles can be generated inside materials, often by precipitating out another phase within the primary crystal. Some (typically larger) particles can be sheared through by the dislocation, in which case the strengthening that is attained through the Orowan mechanism is limited by the particle strength. The effective precipitate strengthening obtained through a distribution of these particles becomes (Arzt, 1998) 3 Fm 2 , (5.5) ∆ σ ps = A2 √ µ Lb2 where A2 is a prefactor of order 1, and Fm is the force required to shear the precipitate particle. If the particle has a shear strength τ p , then this force is Fm = τ p π a2 , and so we have 3 3 π 2 τ p 2 a3 ∆ σ ps = A2 √ . (5.6) µ Lb2 This mode of strengthening has an optimal particle size for fixed particle volume fraction, as described by Arzt (1998). Strengthening increments on the order of 102 MPa can be generated in this way.

5.4.5 Forest Dislocation Strengthening Most crystals contain large numbers of dislocations. A measure of this is the dislocation density, ρd , defined as the total dislocation line length in a crystal divided by the crystal volume. The units of dislocation density are m12 , and therefore √1ρ d represents a length scale in the crystal – it is the average spacing between dislocations. One viewpoint that we can take is to imagine that a dislocation line sweeping through a crystal perceives a forest of other dislocations around it, and these other forest dislocations act as obstacles. Thus other dislocations can cause strengthening of the material. The strength increment ∆ σ f d provided by forest dislocations acting as obstacles is given by √ (5.7) ∆ σ f d = AT µ b ρd , where AT is a constant of order 0.1 − 1. Equation (5.7) is often called the Taylor equation, and provides a measure of the increase in strength of the single crystal because of the presence of some non-zero dislocation density. Now the dislocation density itself evolves during deformation, because there are a number of sources of dislocations inside most crystals – e.g., Frank-Read sources (Courtney, 2005). Indeed, in modeling material behavior, one must often write an evolution equation for the dislocation density:

∂ ρd = Ac fc (ρd ) − Aa fa (ρd ), ∂t

(5.8)

136

Nanomaterials

where fc (ρd ) and fa (ρd ) are functions that represent the rate of creation and annihilation of dislocations respectively, and Ac and Aa represent scaling parameters. For most materials under most conditions, ρd increases with strain, and so Equation (5.7) is often said to represent a strain-hardening behavior (but note that the dislocation density must evolve as in Equation (5.8) for this truly to represent such a hardening behavior). A much more detailed understanding of this process can be obtained by considering multiple slip systems that are simultaneously active within a crystal: one expects that dislocations sliding along one slip plane will intersect with the traces of dislocations sliding along another slip plane, resulting in obstacles that cause hardening on the first slip plane. This process is sometimes called latent hardening (Wu et al., 1991). Equation (5.7) represents a generalization of such ideas over all slip systems in the crystal. Adding all of these strength increments due to all of these mechanisms together, we obtain the strength of the crystal, averaged in some way over all slip systems, as 3

σsc = σPN + Kcn + A1

√ Fm 2 µb + AT µ b ρd . + A2 √ 2 L − 2a µ Lb

(5.9)

Other strengthening mechanisms may be considered in a similar way. Later, when we consider dislocation dynamics, we will return to this issue and suggest another way to add these mechanisms together. Note that the third, fourth, and fifth terms on the right hand side all contain length scales (particle size, particle spacing, and dislocation spacing) that may generate size effects (for example, behavior that depends on sample size).

5.5 From Crystal Plasticity to Polycrystal Plasticity The development of models that explain the transition from single crystal plasticity to polycrystalline plasticity involves the implementation of a number of averaging concepts that are similar to those we have discussed earlier in this chapter. One issue is that of translating an external stress applied to an assemblage of many crystals to the local stress experienced by each crystal in the polycrystalline mass. Once we know the local stress in each crystal, we can use the crystal plasticity approach described above to determine when yield would occur in any given crystal. But does yield in one crystal out of a large polycrystalline mass amount to yield of the polycrystalline material? In principle, of course, the answer is yes. In practice, we define yield strength by when a measurable amount of permanent deformation has taken place (a common number is 0.2% of permanent strain, leading to what is sometimes called a 0.2% yield strength), where this measure is an average measure over the entire polycrystal. That is, by the time that a measurable amount of plastic strain has developed in the polycrystalline material (Figure 5.8), some of the crystals may well have been deformed a great deal. The behavior of

5 Plastic Deformation of Nanomaterials

137

Fig. 5.8 Schematic of a polycrystalline microstructure with a variety of plastic strains (represented by color or shading) within each individual grain.

the overall material is defined by the volume average stresses and the volume average strains, rather than by the stresses and deformations within each individual crystal. This is another version of the problem encountered with multiple crystals in the elasticity case: we do not typically know the local stresses and strains in the polycrystal without doing full blown computations of the crystal plasticity. A number of assumptions are commonly made to make analysis feasible. The most common is known as Taylor averaging, where it is assumed that the strains are uniform in the polycrystal, that is, that all of the crystals see the same strain, which is equal to the average strain computed from the boundaries of the polycrystalline mass. In this case, the stress will of course vary from one crystal (grain) to the next. An alternative but less common assumption is sometimes called static averaging, and states that the stresses are the same everywhere, while the strains vary from crystal to crystal. These two assumptions are analogs to the two bounding cases that we discussed with respect to elastic moduli earlier in this chapter. Clearly neither of these assumptions represents reality, and real behavior will be somewhere in between the two extremes. A number of crystal plasticity codes have been developed that allow for the analysis of the deformation of a polycrystal while simultaneously resolving deformations in individual crystals. This is particularly important for textured materials or low-symmetry materials (such as some of the hcp metals). For our purposes, we will simply assume that the averaging can be done, and ask what some of the consequences of such an averaging will be on the macroscopic response. If we know the behavior of the single crystal, what is the simplest way to modify this to obtain the polycrystal behavior? This raises a different question: how do

138

Nanomaterials

we quantify the polycrystalline character of the material? The primary variables defining the polycrystalline structure are: 1. 2. 3. 4.

The mean grain size, d The grain size distribution, g(d) The grain orientation distribution function, gθ (θ ) The grain shape (e.g., the grain aspect ratio α )

One should expect the properties of the polycrystal to depend on all of these variables. However, most of these variables are not reported by most experimentalists who present data in the literature on the yield strength.

5.5.1 Grain Size Effects The one variable that is routinely reported in the literature is the mean grain size, d, and thus the most common approach in terms of yield strength is to add to the strength in Equation (5.9) an additional term ∆ σgs that captures the effects of polycrystallinity through the mean grain size. Experimental measurements of grain size effects on the yield strength of metals has shown that it is often well described by an inverse square-root law: 1 ∆ σgs = ky d − 2 . (5.10) This behavior is called the Hall-Petch behavior, and ky is a material parameter called the Hall-Petch coefficient or sometimes the Hall-Petch slope. For most materials, the “Hall-Petch equation” is written as 1

σy = σi + ky d − 2 ,

(5.11)

where the left-hand side of the equation represents the material’s strength, and the term σi is called the intercept stress and represents the net strength of the material from all of the other strengthening mechanisms put together. This Hall-Petch behavior can be rationalized in a number of ways. The most commonly cited model is that of dislocations piling up at grain boundaries, with the grain boundary thus acting as an obstacle to dislocation motion. This behavior may break down at very small values of d in the range of tens of nm, although such breakdowns remain controversial because of possible variations in materials created by different processing routes. Of course, it is expected that there will be a breakdown of such behavior: after all, if the grain size becomes sufficiently small, it is no longer possible to retain a dislocation loop within the grain (Arzt, 1998).

5.5.2 Models for Hall-Petch Behavior The dependence of yield strength on grain size described by Equation (5.10) was first developed simply to explain experimental observations in steel (Hall, 1951;

5 Plastic Deformation of Nanomaterials

139

Petch, 1953). A wide range of experimental measurements since then have demonstrated similar behavior in a wide range of metals. Efforts to describe Hall-Petch behavior on a theoretical basis have evolved over the last five decades. Since the early 1990s, modeling efforts have also focused on identifying possible reasons for the breakdown of Hall-Petch behavior at very small grain sizes. We discuss some, but not all, of these models in this section. The majority of these models fall into three classes.

5.5.2.1 The Dislocation Pile-up Model This was the first (and for many years the most popular) way of rationalizing the observed √1d behavior. In its essence, the model states that a dislocation gliding on a slip system stops when it arrives at a grain boundary, because the boundary represents an obstacle and the next grain is not oriented for easy slip along the same direction for the current applied stress. The dislocation gliding behind the first dislocation, on the same slip system, bumps up against the stalled dislocation, and this process continues, creating a dislocation pile-up at the grain boundary. The result of this pileup is the generation of a force per unit length on the leading (first) dislocation: π Nbτ π (1 − ν )τ 2 d F = = , (5.12) L 4 4µ where F is the force, N is the number of dislocations involved in the pileup, τ is the resolved shear stress on the system, ν is the Poisson’s ratio for the material (assumed isotropic), µ is the shear modulus, and d is the grain size. Note that this result (Equation 5.12) was originally derived for a symmetric system, that is dislocation loops gliding through the crystal, so that there is an equal and opposite array of dislocations on the other side of the crystal. Because of this force, the dislocation pile-up generates an increased stress in the adjacent grain, across the boundary. As the pile-up continues to grow, the stress in the adjacent grain also grows, until it reaches some critical value τ ∗ at which dislocation motion can begin on another slip system in the adjacent grain. This critical ∗ value corresponds to the development of a critical force per unit length FL on the lead dislocation. The external or applied stress at which this critical force is reached can be obtained by solving Equation (5.12) for the critical resolved stress and then scaling by a factor A that relates the external stress to the critical resolved shear stress in the adjacent grain:  1 4µ F∗ 1 ∗ √ = K1 √ , (5.13) σ = π (1 − ν ) L d d  4µ F ∗ where K1 = π (1− ν ) L is a scaling factor that clearly depends on the elastic properties but also depends on the critical force F ∗ . Further plastic deformation is now accommodated by slip in the adjacent grain. In real materials, it appears that the

140

Nanomaterials

total external stress cannot be described through Equation (5.13), but rather, this equation describes a strengthening increment due to the grain boundaries (like the second term in Equation 5.11). A very complete discussion of the effects of pileups is found in Hirth and Lothe (1992). In simplified terms, when the grain size is decreased, the number of dislocations within a pileup that can be accommodated within a given grain decreases. Consequently a greater applied stress is needed to cause slip to begin in the adjacent grain. In this model, this is the source of increased strengthening at smaller grain sizes. The major drawback of this model is that dislocation pileups are not always (or even commonly) seen in materials, and thus there is not a great deal of experimental evidence to support the primary mechanism postulated by the model. Further, similar grain size strengthening is observed of the flow stress of a material at finite plastic strain (i.e., well beyond initial yield), and the dislocation pileup argument does not make much sense at large strains when a significant network dislocation density already exists inside each grain. Finally, it is not clear that such a pile-up model is relevant to nanocrystalline materials where the grain size may be too small to accommodate full dislocation loops (Arzt, 1998), the number of dislocations in a pileup may be too low, or the number of dislocation sources may be too small.

5.5.2.2 The Grain Boundary Ledge or Network Dislocation Model This approach was first introduced by Li (1963), and invokes the grain boundaries as sources of dislocations rather than merely as barriers. The concept is that grain boundary ledges (Figure 5.9) act as dislocation sources, and if the ledge density (i.e., number of ledges per unit grain boundary area) is essentially constant (i.e., independent of grain size), then the density ρntwk of the dislocation network that is generated is inversely proportional to grain size (because of the area/volume ratio).

Fig. 5.9 Schematic of grain boundary ledges in a polycrystalline material. In the grain boundary ledge model, these ledges are believed to act as dislocation sources and generate a dislocation network.

5 Plastic Deformation of Nanomaterials

141

Consequently, using a strengthening increment of the form of Equation (5.7), we have √ 1 (5.14) ∆ σgs = AT µ b ρntwk = K2 √ , d where K2 is proportional to AT µ b, and AT represents the polycrystalline averaging and is of order 1. This has a form identical to that of Equation (5.10), and does not require the existence of dislocation pile-ups. A refinement of this model is provided by Scattergood and Koch (1992). Note the dependence of the coefficient K2 on the shear modulus, as in the case of K1 from Equation (5.13). 5.5.2.3 Geometrically Necessary Dislocation and Strain Gradient Models The mismatch between grains that occurs at grain boundaries will lead to the development of high local dislocation densities. These dislocations are generated to accommodate the mismatch (rather than to accommodate externally driven plastic flow), and are therefore said to be geometrically necessary; the corresponding dislocation density is called the geometrically necessary dislocation density or GND density (Ashby, 1970). Associated with the development of this GND density is the development of a strong strain gradient at the grain boundaries (Shu and Fleck, 1999). The GND density is also inversely proportional to grain size, so a relationship exactly like that in Equation (5.14) follows. Strain gradient models are somewhat more complex (Shu and Fleck, 1999), in that it is not immediately clear that the length scale of interest is the grain size, but the essential behavior remains the same as with the GND case. Related models are of the core-boundary type, such as the Meyers-Ashworth (Meyers and Ashworth, 1982) and Fu et al. (Fu et al., 2001) models, postulating a hardened grain boundary layer. Many of these models differ in terms of their predicted behaviors at very small grain sizes; we will return to this issue later. A summary of the three basic models for the observed Hall-Petch behavior is presented in Figure 5.10 in schematic form. The schematic in Figure 5.10a shows a dislocation pileup at a grain boundary, while the schematic in Figure 5.10b shows grain-boundary ledges at one of the boundaries in a polycrystalline material. The schematic in Figure 5.10c shows (on the left), the misfits between many grains as a result of deformation along their individual slip systems, while on the right the grains have been put back together in a compatible form with the misfits now accounted for by a (geometrically necessary) dislocation distribution near the grain boundaries. 5.5.2.4 Observations of Hall-Petch Behavior A very wide range of materials has been studied in terms of the grain size dependence of the yield strength, and there is a very substantial literature on the Hall-Petch strengthening of materials. Most of the literature presents Hall-Petch √ coefficients in terms of Equation (5.11), where the units of ky would be MPa m. However, this

142

Nanomaterials

a

Fig. 5.10 A summary of the three basic models for the observed Hall-Petch behavior. (a) A dislocation pileup at a grain boundary. (b) Grain boundary ledges in a polycrystalline material. (c) A schematic of the geometrically-necessary dislocation model. On the left side of this subpart of the figure, the individual crystals that make up this polycrystalline material are assumed to have slipped along their respective slip systems, resulting in misfits between the grains. However, the polycrystalline material must remain compatible at the grain boundaries, assuming that voids do not open up at the boundaries. These misfits can be accommodated by creating a new set of dislocations (the so-called geometrically-necessary dislocations) that then result in a dislocation distribution near the individual grain boundaries. Only one such GND distribution is shown in one grain on the right.

5 Plastic Deformation of Nanomaterials

143

dimensional form can lead to some confusion, so in this book we choose to examine ˜ where material behavior in terms of the normalized Hall-Petch coefficient K, ky K˜ = √ , µ b

(5.15)

so that the traditional Hall-Petch coefficient is normalized by the shear modulus and the square root of the Burgers vector. Other normalizations are also possible, of course. For most materials, the normalized coefficient K˜ turns out to be in the range of 0.1–0.6. Hall-Petch coefficients (both traditional and normalized) for a number of materials are presented in Table 5.3, with the caveat that the reliability of some of these numbers may be low because of variations in materials processing and in the accuracy of the mechanical tests. A particular concern arises when Hall-Petch coefficients are extracted from measures of hardness variation with grain size rather than yield strength variation with grain size, since hardness measures do not correlate well with yield strength for all specimen geometries (consider, for example, the errors introduced by a substrate interaction just below an indent in a ˜ while most thin film). Note that most fcc metals (Cu, Al, Au, Ni) have small K, ˜ bcc metals (Fe, Mo, W, Ta) have high K. Table 5.3 Hall-Petch coefficients for a variety of materials. Material 1. Copper 2. Aluminum 3. Gold 4. Nickel 5. Zinc 6. Magnesium 7. Titanium 8. Cobalt 9. Armco Iron 10. Molybdenum 11. Tantalum 12. Tungsten

ky, MPa 0.14 0.04 0.25 0.16 0.22 0.28 0.40 0.29 0.58 1.77 0.31 3.1

√

m Shear modulus, MPa Burgers vector, m 45,000 2.56E-10 26,000 2.86E-10 27,000 2.88E-10 79,000 2.49E-10 36,285 2.60E-10 17,300 3.25E-10 45,000 2.95E-10 82,000 2.51E-10 81,400 1.43E-10 126,000 2.73E-10 69,200 2.83E-10 160,600 2.70E-10

K˜ Reference 0.19 0.09 0.55 0.13 0.38 0.89 0.52 0.22 0.60 0.85 0.27 1.17

Hansen (2005) Hansen (2005) Chew et al. (2007) Hansen (2005) Armstrong (1970) Armstrong (1970) Armstrong (1970) Karimpoor et al. (2007) Courtney (2005) Armstrong (1970) Koo (1962) Wei and Kecskes (2008)

An example of Hall-Petch behavior is presented in Figure 5.11, which shows the variation of the flow stress (the strength of a material at fixed strain) as a function of grain size for consolidated ball-milled iron. The flow stress is used because these particular experiments were unable to accurately capture the yield strength of the iron. This material was produced by ball-milling a nanopowder, followed by coldpressing and then hot-pressing to √achieve nearly full density. The corresponding Hall-Petch slope is ky = 0.69 MPa m, which is significantly higher than observed in some pure irons (Table 5.3), perhaps as a result of the ball-milling that was used. Notice that if the Hall-Petch plot is made in terms of hardness, defined here as a third of the Vickers hardness, the slope is slightly different.

144

Nanomaterials 3500

Consolidated Iron (Jia et al., 2003) Quasistatic Flow Stress (MPa)

3000 y = 139.38 + 0.69478x R = 0.99685 2500 2000 1500 1000 Flow Stress (MPa) Hardness (MPa)

500 0

0

500

1000

1500

2000

2500

3000

3500

4000

1/Sqrt (grain size (m)) Fig. 5.11 The variation of the flow stress (measured at 4% strain) with grain size for consolidated iron, taken from the doctoral dissertation of Dexin Jia at Johns Hopkins. The variation of hardness (defined here as H3v , where Hv is the Vickers hardness) is also shown.

In considering the use of an equation such as Equation (5.9) in conjunction with a supposed strengthening increment such as in Equations (5.10) or (5.11), i.e., when multiple mechanisms are active simultaneously, the question of coupling immediately arises. That is, should one expect, for example, that the parameters associated with Orowan strengthening are independent of the solute content of the material? This question must be addressed for each strengthening mechanism. In the case of ˜ is ingrain size strengthening, the question amounts to asking whether ky (or K) dependent of solute content, precipitate content, and so forth. That is, would the K˜ measured (or computed from a model) for a pure polycrystalline metal also be appropriate for use with an alloy of the metal? It is known that the elastic moduli of alloys do not change dramatically with respect to the pure metal, so Equations (5.13) and (5.14) suggest that K˜ for the polycrystal should be about the same as K˜ for the alloy. Unfortunately, this turns out not to be the case: the ky and K˜ are functions of substructure: ky = f (solutes, precipitates, dispersoids, dislocation substructure...).

(5.16)

Some of this is of course to be expected, because it is known that the CRSS is itself a function of solute concentration; for a discussion, see Argon (2008). However, the degree of coupling at the the level of ky is difficult to predict, as we see now for the specific case of aluminum. Table 5.4 presents Hall-Petch coefficient data for a variety of aluminum-based materials (Han et al., 2003), including pure aluminum.

5 Plastic Deformation of Nanomaterials

145

Table 5.4 Hall-Petch coefficients for pure aluminum and select aluminum alloys. Data taken from the work of Witkin and Lavernia (2006). √ Material ky , MPa m Reference Pure Aluminum Al (99.999%) 0.13 Al (99.5%) 0.08 Al and Al-Al2 O3 composites Cryomilled Al Ball-milled Al ECAP Al SAP Al

0.09 0.012 0.12 0.03-0.07

Witkin and Lavernia (2006) Perez (1998) Robinson (1953) Perez (1997); Phillips (1953); Rodriguez (2003)

Al-Mg alloys Cryomilled Al 5083 MA IN905XL (Al-4Mg-1.5Li-0.4O) Al-3.5Mg Al 5052 (Al-2.6Mg) Al 5754 (Al-3Mg) Al-5Mg Al-5Mg-0.2Mn ECAP Al-1Mg ECAP Al-3Mg Al-6Ni

0.28 0.32 0.12–0.25 0.21 0.27 0.15 0.22 0.13 0.15 0.14

Witkin and Lavernia (2006) Rosler (1992) Semiatin (2001) Siegel (1994) Siegel (1994) Siegel (1994) Siegel (1994) Robinson (1995) Robinson (1995)

Mechanically alloyed Al composites Al plus C, O (14.1 vol.%) Al plus C, O, Cu (12.8 vol.%) Al plus C, O, Y (5.9 vol.%) Al plus C, O, Nb (16.1 vol.%) Al plus C, O, Y, Nb (15.9 vol.%) Al plus C, O, Ti (31.6 vol.%)

0.23 Sun (2000) 0.31 0.20–0.27 0.16 0.22 0.25

One notes from Table 5.4 that the Hall-Petch coefficients of aluminum-based materials vary over a wide range. This is because strengthening mechanisms such as solid solution strengthening and precipitate strengthening appear to have an effect not just on the strengthening increments presented in Equations (5.3)–(5.8), but also on the effective Hall-Petch coefficient in these materials. A significant part of the data on aluminum in Table 5.4 is derived from the work of Lloyd and co-workers, who conducted a comparative study of several aluminum alloys of controlled composition, and we can discuss these coupling effects in the context of their work. For example, the Hall-Petch coefficient for cryomilled Al 5083 is apparently much larger (Witkin aluminum alloy √ and Lavernia, 2006) than that for the coarse-grained√ (0.068 MPa m) and that for cryomilled pure aluminum (0.09 MPa m) (Han et al., 2003). By comparing the ky values of different aluminum and Al-Mg based aluminum alloys (Han et al., 2003), it is observed that the effective Hall-Patch coefficient is related to the solid solution, the distribution of dispersoids, and the oxygen and nitrogen contents in these alloys. Parts of this behavior have been predicted in

146

Nanomaterials

terms of the grain boundary ledge argument by J.C.M. Li for the specific case of solute concentrations. Lloyd and Court (2003) discussed the effects of Mg and Mn content on the intercept stress and slope of the Hall-Petch equation for these Al alloys. Their work shows that Mg and Mn contribute to both the intercept stress and the Hall-Petch coefficient as solute atoms and through precipitates. Examining this data, it appears (on a heuristic basis) that the effective k¯y for an Al-Mg-Mn alloy can be obtained from k¯y = kyPure Al + kyMg addition + kyMn addition , (5.17) where kyPure Al is the Hall-Petch coefficient for pure Al, and kyMg addition and kyMn addition are the contributions of Mg and Mn to the effective coefficient. From experimental data, √ the term kyMg addition in a cryomilled Al 5083 can be approximated as Al-5Mg (0.16) and cryomilled pure 0.07 MPa m (the difference of ky between √ Al (0.09)), and kyMn addition is 0.09 MPa m (the difference of ky between Al-5Mg0.5Mn(0.25) and Al-5Mg (0.16)). Equation (5.17) allows us to compute the effective √ Hall-Petch coefficient k¯y for the√cryomilled Al 5083 as 0.25 MPa m. This value is close to the value of 0.28 MPa m estimated from the experimental data of Lavernia et al. (Witkin and Lavernia, 2006). This approach explains why the Hall-Petch coefficient for cryomilled Al 5083 is so much larger than that of the conventional grain size version of the same Al alloy, and demonstrates that when the Hall-Petch equation is used to estimate grain size strengthening in an alloy, the solution and precipitation mechanisms may couple into the grain size strengthening mechanism. Since the moduli of alloys generally do not vary dramatically from the moduli of the pure polycrystalline metal, models such as Equations (5.13) and (5.14) suggest that the critical force F ∗ or the network dislocation density ρntwk must be affected by these alloying additions (e.g., through precipitation at grain boundaries, or solute segregation to grain boundaries).

5.5.2.5 Limits of Hall-Petch Behavior The grain-size dependence of yield strength that is described by the Hall-Petch behavior appears to suit experimental data very well for metallic materials with grain sizes that are larger than about 500 nm. The approach breaks down in two limits. first, at very large grain sizes, the observed yield stress does not necessarily approximate that of the single crystal. This is of course to be expected, given the averaging issues involved. Second, and of greater interest to us in this book, the Hall-Petch dependence of yield strength on grain size appears to be inaccurate for small grain sizes. This is sometimes known as Hall-Petch breakdown or deviations from HallPetch behavior. Indeed, at very small grain sizes, it has been claimed that the yield strength of materials decreases with grain size, sometimes called the inverse HallPetch effect. This type of behavior was first observed by Chokshi et al. (1989), when they examined the hardness of copper and palladium at grain sizes ≈ 10 nm. Most if not all of the deviations from Hall-Petch that are observed at small grain sizes

5 Plastic Deformation of Nanomaterials

147

are in the direction of lower than expected strength (although most measurements in this grain size range are of hardness, (e.g., Zhou et al. (2003), not strength, see, e.g., Figure 5.12). These deviations are important because, from a technological perspective, they imply that the strength of a nanocrystalline solid cannot be increased indefinitely by decreasing the grain size. d (nm) 9

28.9

17.1

7.1

4.1 650

550 7 450

6

5 0.1

HV

H(GPa)

8

0.2

0.3

0.4

350 0.5

1/d0.5

Fig. 5.12 Evidence of reduction of hardness with decreasing grain size in a Ni-P material, from Zhou, Erb et al., Scripta Materialia, 2003 (Zhou et al., 2003). The right-hand axis represents the Vickers Hardness number HV. Reprinted from Scripta Materialia, Vol. 48, Issue 6, page 6, Y. Zhou, U. Erb, K.T. Aust and G. Palumbo, The effects of triple junctions and grain boundaries on hardness and Young’s modulus in nanostructured Ni-P. March 2003, with permission from Elsevier.

A very large number of experiments have demonstrated a deviation of the strengths of nanocrystalline materials from the classical Hall-Petch behavior at very small grain sizes. However, the data remains somewhat inconsistent. This is partly because of the difficulty of making materials with very small grain sizes, and partly because the differences between materials may be larger than just the grain size (for instance the material with the grain size of 50 nm made by electrodeposition may be very different in character from a material of identical grain size made by severe plastic deformation). An example of the degree of variability in measured material properties can be seen in Figure 5.13, which is taken from the excellent review article by Meyers et al. (2006) and which examines only the properties that have been measured for copper. Even in this one material, very different yield strengths have been measured at similar grain sizes in the range of grain size below 25 nm. In some cases the yield strength is perceived to become independent of grain size, while in other cases the yield strength is perceived to decrease with decreasing grain-size. However, what is certainly apparent is that the behavior deviates from the classical Hall-Petch behavior at grain sizes of less than 25 nm. Similar deviations from the classical Hall-Petch behavior have been observed in other materials, including iron, nickel and titanium (Meyers et al., 2006). In all of these cases, however, the scatter in the behavior at these very small grain sizes remains substantial, so much so that it

148

Nanomaterials

Fig. 5.13 Compendium of data on copper in a Hall-Petch-style plot by Meyers et al. (2006), showing the scatter in data at small grain sizes but the general deviation from classical Hall-Petch behavior. Reprinted from Progress in Materials Science, Vol. 51, Issue 4, Page 130, M.A. Myers, A. Mishra, D.J. Benson, Mechanical properties of nanocrystalline materials. May 2006, with permission from Elsevier.

is difficult to use the data to distinguish between different possible mechanisms for the deviations from Hall-Petch. An examination of the literature on nanocrystalline materials indicates that the majority of experiments demonstrate not so much an inverse Hall-Petch effect (a decrease of the yield strength of decreasing grain size), but rather a reduction in the degree of strengthening that is obtained as the grain size is decreased (e.g., Figure 5.14). It is difficult to draw conclusions about material behavior given the scatter evident in the current dataset (as of 2008), with most of the variability related to either materials synthesis or measurement difficulties. The difficulties associated with making measurements of material behavior at very small grain sizes arise primarily from the difficulty of obtaining samples of sufficient size for traditional testing, as discussed in the chapter on experimental techniques. The rapid evolution of these techniques suggest that this controversy will be resolved in the near future. What are the possible causes of Hall-Petch breakdown? A number of theories have been advanced to explain this behavior. These “Hall-Petch breakdown” models can generally be separated into two categories. • First, models which attempt to show that the primary mechanism for Hall-Petch strengthening (whether viewed as dislocation pileup or GND localization at the boundaries) will break down at small grain sizes. An example of such a model would be one that states that a minimum grain size is required to develop a dislocation loop (Nieh and Wadsworth, 1991).

5 Plastic Deformation of Nanomaterials

149

(a)

(b)

(c)

(d)

Fig. 5.14 Hall-Petch plots for copper, iron, nickel and titanium compiled by Meyers et al. (2006), showing the deviation from classical Hall-Petch at small grain sizes. Note the strength appears to plateau but not decrease with decreasing grain size. Reprinted from Progress in Materials Science, Vol. 51, Issue 4, Page 130, M.A. Myers, A. Mishra, D.J. Benson, Mechanical properties of nanocrystalline materials. May 2006, with permission from Elsevier.

• Second, models which attempt to account for the increasing contributions of the grain boundary and triple junctions as the grain size is decreased e.g., Zhou et al. (2003). Such models often introduce new mechanisms such as grain boundary diffusion (Chokshi et al., 1989), grain boundary sliding (Hahn and Padmanabhan, 1997) and grain rotation. Of critical importance here are the grain boundary and triple junction volume fractions, which we have discussed in the text following Figures 4.3 and 4.6. Such breakdown models will not be examined in detail in this chapter, in part because the experimental dataset remains too sparse to provide solid distinctions between possible models (this is an excellent example of an area in which it is far easier to develop a plausible model than it is to develop a good experiment). A number of these mechanisms will, however, be discussed in other contexts later in the book.

150

Nanomaterials

5.5.2.6 Summary of Grain Size Effects on Strength We have now spent a great deal of time discussing the effects of grain size on the yield strength of materials, with specific reference to pure nanocrystalline materials. This is perhaps appropriate, because it is the strength of nanomaterials that has been best studied (in terms of mechanical properties) with respect to grain size. However, this does not imply that other features of the polycrystalline materials do not have an impact on nanocrystalline material behavior. In the next few paragraphs, we explore briefly the consequences of grain size distributions, grain orientation distributions and grain shape on the plastic behavior of nanocrystalline materials.

5.5.3 Other Effects of Grain Structure 5.5.3.1 Effects of Grain Size Distributions All of the preceding discussions on the effects of grain size on yield strength have essentially assumed that the polycrystalline metal had a single grain size. This is, of course, a fallacy. It is nearly impossible to make materials with a single grain size; instead, there is always a grain size distribution. An example of the grain size distribution in a near-nanocrystalline metal is provided in Figure 5.15, which shows a size distribution derived from TEM analysis of a consolidated iron sample. From a mathematical viewpoint, the grain size distribution can be described in terms of a probability density function, using for example, a Gaussian distribution, a lognormal distribution, or a Weibull distribution. For any given distribution function, we can define the mean grain size and the variance of the grain size (a measure of the spread of the distribution – the square root of the variance is the standard deviation). Other distribution parameters such as the skewness may also be of interest. Our discussion up to this point has simply examined the effects of the mean grain size. How important is the spread of the grain size distribution? Consider two materials that have identical mean grain sizes but very different standard deviations in the grain size distribution. Material A is essentially monodisperse, that is, all of its grains are essentially the same size. Material B has a large standard deviation in its grain size distribution, so that some of the grains are much larger than the mean grain size and some are much smaller. In both cases, we assume that all of the grains are randomly oriented and have essentially the same shape. Imagine that both materials are subjected to a slowly increasing external load. Since the grains are randomly oriented, there is an equal probability that deformation will begin in any given grain at any given load. Thus the onset of plastic deformation (that is the yield) should be essentially the same for the two materials, assuming that yield can be perceived at very small strains. There is an important subtlety here. It is certainly the case that the strength of polycrystalline metals is a function of the grain size. However, the strength of any given crystal is not a function of the size of the crystal (aside from pathological

5 Plastic Deformation of Nanomaterials

151

200nm

0.16

Number Fraction

0.14

d =138 nm 392 grains

0.12 0.10 0.08 0.06 0.04 0.02 0.00

40

440

840

1240

Grain Size (nm) Fig. 5.15 Measured grain size distribution in a consolidated iron sample, based on 392 grains measured from TEM images such as that shown above (Jia et al., 2003). While some of the grains are clearly in the nanocrystalline range, others are more than 100 nm in size. Reprinted from Acta Materialia, Vol. 51, Issue 12, page 15, D. Jia, K.T. Ramesh, E. Ma, Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. July 2003, with permission from Elsevier.

cases where free surfaces dominate the behavior of the crystal, so that dislocations can easily run out to the free surface). In spite of this, a number of polycrystalline plasticity simulations essentially assume that the strength of each crystal is proportional to the inverse square root of the crystal size (this is done in an effort to account for the Hall-Petch behavior). The results of such simulations should be viewed with caution for initial yield of the polycrystalline metal, since they essentially enforce a grain boundary constraint that has not been established for single crystals. The approximation is less of a concern for materials that have been subjected to a prior

152

Nanomaterials

plastic deformation, because the development of geometrically necessary dislocations will provide an effective strengthening mechanism, resulting in a distribution of effective yield strengths in crystals of various sizes. The big effect of grain size distributions is subsequent to yield. The interaction of the dislocations with the grain boundaries (or equivalently, the development of geometrically necessary dislocation densities) is certainly a function of the grain size. Thus the effective hardening of the material will be a function of the grain size distribution, and a greater hardening rate should be expected for the material with the widest spread in the grain size distribution. This approach is used to some benefit in the development of nanomaterials with greater ductility than is typically obtained in materials with narrow grain size distributions. The addition of the larger grains in the wider grain-size distribution does not substantially decrease the yield strength, for the reasons discussed previously, but does improve the overall hardening and therefore the ductility. Indeed, bimodal grain size distributions appear to hold great promise for nanostructured materials systems (Wang et al., 2002).

5.5.3.2 Effects of Grain Orientation Distributions We have essentially assumed that the material has randomly oriented grains within the polycrystalline structure. If the grains are not randomly oriented, then the overall material behavior will reflect the anisotropy of the crystals to some degree. The effective orientation distribution of the grains leads to what is called the texture of the polycrystalline material. The degree to which this is important obviously depends on the anisotropy of the single crystal. The influence of texture is dramatic in low symmetry crystal structures such as hexagonal-close-packed structures, and must be accounted for in the modeling of such materials. It is not clear at this point that the texture effect is significantly different in nanocrystalline materials as compared to conventional grain size materials, but it is not immediately apparent that there should be a coupling between grain size and texture. Of perhaps greater influence are issues such as the development of specific kinds of grain boundaries in materials in which texture is important after severe plastic deformation processes.

5.5.3.3 Effects of Grain Shape The effect of grain shape on the yield strength can be very simply viewed in the following manner. Imagine a material with the microstructure consisting of hexagonal grains of the morphology shown in Figure 4.5, and consider three very different aspect ratios for the grains: α = 10, corresponding to rods, α = 1 corresponding to nearly equisized structures, and α = 0.1 corresponding to plates. As we have demonstrated earlier, the grain boundary and triple junction volume fractions are very strong functions of these shapes. Further, it is not clear what mean grain size one should use in estimating the response of the rod and plate morphologies. If one

5 Plastic Deformation of Nanomaterials

153

looks along and perpendicular to the axis of the hexagonal prism, one would estimate two different sizes (or equivalently, different distances between grain boundaries). If we assume that Hall-Petch behavior is operative in both directions, it should be expected that the material behaves like an anisotropic plastic material, with different yield strengths in the two directions. To avoid the complexity of anisotropy, some authors use as an effective grain size the size of a cube of equivalent volume. For example, consider the microstructure of the cryomilled Al 5083 alloy shown in Figure 5.16. This material was first cryomilled, then HIPped (Hot Isostatic Pressed, i.e., subject to high pressure while at high temperature) and extruded, resulting in an ultra-fine-grain (but not quite nanocrystalline) material. The resulting microstructure shows grain anisotropy between the extrusion and transverse directions (Figure 5.16). Using multiple TEM observations, the linear dimensions of the ultra-fine grains can be measured in each direction. The grains are elongated along the extrusion direction with a mean length of 390 nm (Figure 5.16a). In the transverse direction, equiaxed grains are observed with a mean diameter of 150 nm (Figure 5.16b). If each of these independent grain sizes were to be used with a Hall-Petch strengthening increment, the transverse direction would give a strengthening increment that is 60% larger than the extrusion direction. However, clearly the entire grain participates in the deformation even if the loading is only in the transverse direction, so it is not obvious that the grain diameter in Figure 5.16b should be used for loading in the transverse direction. Thus using the grain size in each cross-section to compute the anisotropic strengthening is not necessarily correct (indeed, in this particular material the strength in the transverse direction was found to be smaller than the strength in the extrusion direction). What effective grain size

Fig. 5.16 TEM micrographs showing microstructure of a cryomilled 5083 aluminum alloy in (a) extruded and (b) transverse directions (transverse to the extrusion axis). Note that the two images were taken at slightly different magnifications. However, elongated grains are evident in the direction of extrusion (Cao and Ramesh, 2009).

154

Nanomaterials

d¯ should be used for this material within the Hall-Petch relation (Equation 5.11)? An effective grain size can be calculated by assuming that the grains are cylin2 ders with a volume given by V = π d4 h (d being the diameter of the equiaxed grain in the transverse direction, and h being the length of each elongated grain in the extruded direction). The side of a cube of equivalent volume can be used to de1

1 2 3 fine the effective grain size d¯ through d¯ = V 3 = ( π d4 h ) , which works out to d¯ = 190 nm.

5.6 Summary: The Yield Strength of Nanomaterials Like conventional materials, the yield strength of nanomaterials is determined by the ease with which dislocations can be moved through the material to develop plastic strain. All of the strengthening mechanisms that are active in conventional grain size materials are also active in nanocrystalline materials. The key characteristics associated with the yield strength of nanomaterials, and specifically nanometals, are the following: • The primary additional strengthening mechanism that arises in nanocrystalline materials is a result of the interaction of grain boundaries with dislocation generation and motion. • The grain morphology of a nanocrystalline material can be usefully characterized in terms of its grain size, grain size distribution, grain orientation distribution and grain shape. • The grain size has the dominant influence on the yield strength, determining the strengthening through a Hall-Petch type response (Equation 5.11). Very high strains can be obtained in nanocrystalline metals. • At very small grain sizes (of the order of 25 nm or less in most materials) the strength increment deviates from the Hall-Petch behavior, typically resulting in less strengthening than would be expected through Hall-Petch. • The so-called inverse Hall-Petch effect, where the strength of the material actually decreases with decreasing grain size, remains controversial and is not commonly seen. • The breakdown of Hall-Petch behavior at small grain sizes can be described in terms of grain boundary volume fractions, triple junction volume fractions, grain boundary sliding, and grain boundary diffusion. • The grain-size distribution appears to have at most a weak effect on yield strength but may have a significant role to play in the hardening and the stability of nanocrystalline systems. • The grain orientation distribution leads to texture effects, and is particularly important in low-symmetry systems such as hexagonal-close-packed systems. • Grain morphology effects can include variations in grain boundary volume fractions and the development of plastic anisotropy of the yield strength.

5 Plastic Deformation of Nanomaterials

155

5.7 Plastic Strain and Dislocation Motion Now that we have an understanding of yield strength, let us consider the relationship between the development of plastic strain and the motion of dislocations. Remember that dislocations are line defects that move along crystallographic slip planes with normal n in specific slip directions s. Consider the special case of slip along only one slip system in a crystal, and let us suppose that the dislocation line of length l moves the distance δ x. The corresponding increment of shear strain that is developed in the crystal will be bl δ x , (5.18) γ= V where V is the volume of the crystal and b is the Burgers vector. Of course, this is the shear strain increment along the direction of slip. Since the dislocation does not move back, this is a permanent (plastic) strain increment. How does this plastic shear strain increment relate to the tensorial plastic strain increment δ εipj that we have discussed in the mechanics chapter? This relationship must depend on the slip plane normal n and the slip direction s, and can be shown to be given by

δ εipj = γ µi j =

bl δ x bl δ x (si n j + ni s j ), µi j = V 2V

(5.19)

where

1 (5.20) µi j = (si n j + ni s j ) 2 is called the transformation tensor. In tensorial notation, the transformation tensor is 1 (5.21) µ = (s ⊗ n + n ⊗ s). 2 In practice, there will be many dislocations on many different slip systems. Given that multiple slip systems may be operative at one time, it follows that the net plastic strain increment δ ε˜ipj will be α

blm (δ x) α δ ε˜ipj = ∑ µ i j, V

(5.22)

where the summation is carried out over the α separate slip systems. The dislocation line length involved here is lm , which is the total length of the mobile dislocation line segments (note that not all dislocations will be mobile, some being pinned by various mechanisms). We define the mobile dislocation density as

ρm =

lm , V

(5.23)

so that the net plastic strain increment is

δ εipj = bρm (δ x)α µ α i j ,

(5.24)

156

Nanomaterials

where the summation symbol has been dropped for clarity of the dependence of the plastic strain increment on the dislocation slip. The distinction between the mobile dislocation density and the total dislocation density can be very important in some materials under some conditions. In terms of the scalar shear strain, the analog to Equation (5.24) would be γ = bρm (δ x), a notation in common use in materials science but somewhat confusing in that the mobile dislocation density is not welldefined over a single slip system.

5.8 The Physical Basis of Strain Hardening

Engineering Stress

Strain hardening in a material is, by definition, the increase of the strength with the strain (when this is thought of as increase of strength with plastic work, the process is called work hardening). Recall from Chapter 2 that after a material has been loaded plastically to a given strain, unloading is elastic, and reloading is along the elastic line (Figure 5.17) until the former flow stress has been achieved (which is thus a new yield strength). Why would the yield strength increase with increasing strain? Examining the list of strengthening mechanisms that we have discussed (solutes, dispersoids, precipitates, other dislocations, grain boundaries), we see that only one of these changes with increasing strain, and that is the dislocation density. All of the other strengthening obstacles are predetermined and do not themselves change with strain (e.g., the solute concentration does not evolve with strain). Thus, the primary mechanism for strain hardening is the increase of dislocation density with increasing plastic strain, and the subsequent interaction of a moving dislocation with this increased array of dislocation-derived obstacles. Strain hardening in single crystals is normally thought of as occurring in three stages (called Stages I, II, and III) as slip continues beyond yield; some authors

σ σ σ

Engineering Strain Fig. 5.17 Stress strain curve for a material, showing the elastic unloading response and the increased yield strength upon reloading (this is called strain hardening, since the material has become harder because of the increased plastic strain).

5 Plastic Deformation of Nanomaterials

157

also define Stages IV and V. Stage I covers very little strain and shows very little hardening (corresponding to single slip at the CRSS). The resolved shear strength of a single crystal typically increases nearly linearly with strain in Stage II, and this can occur for significant strains. For large strains the strength increases less than linearly with strain, with a decreasing rate of strengthening – this is Stage III. For the polycrystalline and in particular nanocrystalline systems of interest to this book, the early stages of strain hardening are difficult to observe (except in some pathological highly textured cases), and so we will ignore this nomenclature. Most of what is observed in nanocrystalline materials is already effectively Stage III. Very sophisticated theories of strain hardening have been developed for fcc metals, and can be found in a number of reference works, e.g., Chapter 7 in Argon (2008). Strain hardening in bcc and hcp materials is less well understood. The emphasis on strain hardening in this book is through the evolution of the dislocation density (Equation 5.8) and its consequences on the strengthening of the material (Equation 5.7). Whether or not the dislocation density grows depends on the relative rates of creation and annihilation of dislocations, which is a strong function of temperature. Indeed, it is possible to dramatically reduce the overall dislocation density by holding the material at a fixed high temperature for some time (a process often referred to as recovery). Conversely, it is possible to increase greatly the dislocation density by severely deforming the material (for example by rolling). An additional point to recognize is that although Equation (5.7) uses an average dislocation density, at high dislocation densities, it is energetically favorable for the dislocations to pattern themselves into a dislocation substructure that may involve dislocation cells or subgrains, and this may require special treatment (Mughrabi, 1983). It is known that the dislocation density in the material cannot increase indefinitely, and instead tends to saturate at some level corresponding to the development of specific dislocation substructures (this correlates well with the effect of saturation of strain hardening observed in many materials). How does strain hardening evolve in nanocrystalline metals? Viewing nanocrystalline metals as simply special cases of polycrystalline metals in which the grain size is very small, the contributions to strain hardening that are specific to nanocrystalline metals will arise from Equations (5.7) and (5.10), with the latter being involved in that the normalized Hall-Petch coefficient K˜ may well be a function of the averaged dislocation density (Equation 5.16). Briefly, still assuming that strengthening mechanisms can be added in terms of strength increments, the overall strength of the material may be written as √ 1 √ σnet = σi + AT µ b ρd + K˜ µ bd − 2 ,

(5.25)

where the evolution of the dislocation density is given by Equation (5.8). Note that in this way of thinking the magnitude of the strength increment due to the dislocation substructure itself does not change as a result of grain size. It is certainly possible that the normalized Hall-Petch coefficient K˜ does change with increasing dislocation density, although this appears to be a small effect (e.g., see Table 5.4 for aluminum). The consequences of this in terms of overall constitutive

158

Nanomaterials

function are considerable. Since a considerable degree of strengthening has already been obtained in a nanocrystalline material using the reduction in grain-size, the additional increment of strengthening generated through strain-hardening is relatively small, and as a consequence the apparent strain hardening of nanocrystalline materials tends to be relatively small. In addition, many nanocrystalline materials are made by processes that involve severe plastic deformation (see Chapter 1), and so the dislocation substructure may have evolved into an equilibrium state, where as many new dislocations are being generated as are being annihilated within the substructure so that the substructure no longer evolves. This implies that the strengthening increment provided by the dislocation substructure no longer changes, resulting in the near-vanishing of the strain hardening. This has important implications for the stability of deformation in nanomaterials, as we shall see later. Consider, for example, the commonly used power law expression (Equation 2.95) for describing the tensile stress-strain curve for a material:  n ε σnet = σ0 , (5.26) ε0 where ε0 = σE0 is a reference strain and the material parameter n is called the strain hardening index (the stresses and strains in this equation are the true stresses and strains (see Chapter 2). Let us define a new variable q, called the normalized hardening, which is a normalized measure of the strain hardening: q=

1 ∂ σnet . σnet ∂ ε

(5.27)

This measure has the advantage that at any given strain it is proportional to the strain hardening index n, and in addition it is normalized by the strength of the material (so that it is possible to compare two materials with very different strengths in terms of their relative hardening with strain). The term ∂ ∂σεnet is simply the tangent to the stress-strain curve at any given strain (the slope of the stress-strain curve, a more intuitive measure of the hardening).

5.8.1 Strain Hardening in Nanomaterials When one considers nanocrystalline materials, the normalizing factor σnet becomes very large because of the grain size strengthening (Equation 5.25, assuming small grain sizes) and so the normalized hardening becomes small. That is, the apparent strain hardening of the nanocrystalline material is small even though quantitatively the additional strengthening provided due to the dislocation density may be the same as in the conventional grain sized version of the material. Indeed, because σnet is dominated by the grain size term for nanocrystalline materials, for such materials we will have

5 Plastic Deformation of Nanomaterials

159

q∝

√

(5.28)

d.

This is in fact observed in many nanocrystalline materials, as demonstrated in Figure 5.18, which presents the q (at a plastic strain of 3%) plotted against the square root of grain size for various materials reported in the literature. Reduced values of q can be seen with decreasing grain size in the figure, where all the data is from uniaxial compression (thus avoiding the problems associated with the apparent reduction of hardening during necking in tension). The reduction of q with decreasing grain size is similar to the reduction of observed ductility with decreasing grain size shown in Koch et al. (1999). The influence of strain hardening on the ductility of materials is dramatic. A fairly simple way to think about this is to consider the simple tension test discussed in Chapter 2. Simple tension of a perfectly plastic material leads to instabilities at very small strains, because any perturbation in the geometry of the specimen should result in a local increase in the true stress because of the reduction in the area, but the inability of the material to sustain a higher stress implies that the strain in that area has to increase dramatically (and is essentially bounded only by the finite amount of energy available for the deformation). Thus the necking instability begins very

Fe - Current Study

Pd [10]

Fe - Khan et al [2]

Cu [10]

Fe - da Silva & Ramesh [3]

Cu [11]

Alpha-Fe - Ostwaldt et al. [4]

Cu [12]

Fe-10%Cu - Carsley [5]

Cu [13]

Fe28Al2Cr - Jain & Christman [6]

Ni [13]

Ti - Current Study

Cu [14]

Ti - Chichili et al. [8]

Cu [15]

Ti - Nemat-Nasser et al. [9]

Normalized Strain Hardening, q

20

15

10

5

0

–5

2

4 6 Grain Size, d 1/2 (d in microns)

8

10

Fig. 5.18 Variation of normalized strain hardening with the square root of grain size for a variety of nanocrystalline and ultra-fine-grain materials (from Jia’s doctoral dissertation, Jia et al., 2003). The reference numbers correspond to those in Jia’s dissertation.

160

Nanomaterials

rapidly in a perfectly plastic material, leading to an effective loss of ductility. The presence of some strain hardening allows one to delay the onset of the necking instability, effectively providing some additional ductility. As a consequence, unless a strain-hardening mechanism is incorporated in the material, nanocrystalline materials tend to have very little ductility in tension. This is discussed further in the section on failure mechanisms.

5.9 The Physical Basis of Rate-Dependent Plasticity The material behaviors that have been discussed up to this point have essentially been relationships between the stress, the strain, and the microstructure. However, the behavior of most materials is also a function of the rate of deformation, rather than simply of the deformation alone (that is, the behavior is also a function of the strain rate rather than just of the strain). This rate-dependence arises from three sources: • The force required to move a dislocation depends, in a fundamental way, on the velocity of the dislocation. Moving the dislocation at a higher velocity requires a higher force, or a higher stress. Since the velocity of dislocation motion is directly related to the strain rate, a higher stress is required to achieve a higher strain rate in the material. This is essentially a result of dislocation dynamics. • Dislocations in materials are constantly in motion as a result of available thermal energy. At higher temperatures, they oscillate to a greater degree around a mean state. This results in a relationship between the motion of the dislocation and the temperature, resulting in both easier deformation at higher temperatures but fixed rate of deformation, and more resistance to deformation at higher velocities but fixed temperature. This is essentially a result of thermal activation. • The rate of increase of the dislocation density is a function of the rate of deformation because the mechanisms associated with the annihilation and creation of dislocations depend on the strain rate. Since the dislocation density affects the strength of the material, the strengthening of the material depends implicitly on the strain-rate. This is essentially a result of dislocation substructure evolution.

5.9.1 Dislocation Dynamics Many experimental observations on single crystals have demonstrated that the dislocation velocity is a strong function of the applied stress. Many of these experimental results are collected in Figure 5.19, taken from the review article by Clifton (1983). For small stresses at small dislocation velocities, the relationship between the stress and the velocity is not clear and may vary substantially from material to material. However, at sufficiently high stresses and sufficiently high velocities, the velocity first becomes linearly dependent on the applied shear stress (a domain known

5 Plastic Deformation of Nanomaterials

161

Fig. 5.19 The dependence of dislocation velocity on shear stress for a variety of materials, from Clifton (1983). Note that a limiting velocity is expected, corresponding to the shear wave speed. There is a domain in the figure where the dislocation velocity is linear with the applied shear stress, and this is called the phonon drag regime. Reprinted from Journal of Applied Mechanics, Vol. 50, p. 941–952, R.J. Clifton, Dynamic Plasticity. 1983, with permission from original publisher, ASME.

as dislocation drag) and then becomes highly nonlinear, with the stress becoming very large  as the dislocation velocity approaches the limiting speed of a shear wave (cs = µρ , where µ is the shear modulus and ρ is the mass density) in the material (this is called the relativistic domain). To connect this behavior with the rate-dependence of the stress, we use the rate form of the equation defining plastic strain in terms of dislocation motion (Equation 5.24), so that we can relate the plastic strain rate to the dislocation velocity:

ε˙ipj = bρm x˙µi j = bρm vd µi j ,

(5.29)

where vd is the dislocation velocity and the tensor µi j relates to the slip system and was defined in Equation (5.20). A scalar strain rate measure can be obtained from this equation by defining a scalar called the effective plastic strain rate (analogous to our definition of the von Mises stress in Chapter 2) as  (5.30) ε¯˙ p = 2ε˙ipj ε˙ipj ,

and using Equation (5.29) we obtain

ε¯˙ p = bρm vd .

(5.31)

This equation is sometimes called the Orowan equation, although in materials science textbooks it is usually written in terms of a shear strain rate γ˙ which is

162

Nanomaterials

identically equal to our effective plastic strain rate ε¯˙ p , i.e, γ˙ = ε¯˙ p . For a tensorial stress state σi j , we can also extract a scalar shear stress measure, which is the von Mises stress σ¯ (Chapter 2). Now Figure 5.19 gives us a relationship between the shear stress and the dislocation velocity for a given material: vd = f (σ¯ ). Using this in Equation (5.31) we obtain ε¯˙ p = bρm f (σ¯ ). (5.32) This can be used to obtain a measure of the strengthening due to dislocation dynamics (note that in obtaining this equation we have not considered any of the other strengthening mechanisms). For example, if we assume that the applied stress is such as to put us in the linear portion of Figure 5.19, we have vd = f (σ¯ ) = Bσ¯ , where B is a constant called the drag coefficient, and it follows that in this case

ε¯˙ p = bρm Bσ¯ ,

(5.33)

which corresponds to a strengthening increment due to dislocation drag that shows up as a rate-dependent strengthening that is linear in the strain rate:

∆ σ¯ =

1 ¯p ε˙ . bρm B

(5.34)

The rate-dependent strengthening of materials due to dislocation drag is generally observed only at very high strain rates (conventionally assumed to be rates larger than 106 s−1 ). At lower rates, the thermal activation and substructure evolution are traditionally considered to dominate the behavior.

5.9.2 Thermal Activation The mechanisms discussed up to this point have ignored the effects of temperature. In practice, atoms do not stay in fixed positions within the lattice, but rather vibrate around a mean position with the degree of vibration determined by the current temperature (eventually the temperature becomes high enough that the atoms are able to move freely, corresponding to melting). Similarly, defects within the crystal are in constant motion at nonzero temperature, oscillating gently around the mean position that is defined by a minimum energy state. One way to visualize deformation is to imagine that the units of deformation (in this case dislocations) are moving from one minimum energy state to another, and that between these two minima is an energy barrier that represents the difficulty of getting around a specific obstacle. Work is required to move the dislocation over the energy barrier, and this work is typically supplied by the applied stress. However, in any given state, the energy of the dislocation is actually oscillating around the minimum because of the available thermal energy, and so the effective height of the energy barrier must be compared with the available thermal energy.

5 Plastic Deformation of Nanomaterials

163

Consider the dislocation that is attempting to cross an energy barrier. The frequency with which it attempts to cross this barrier is designated ν0 . Using the Arrhenius approximation, the frequency ν with which it is successful in crossing the barrier is −G , (5.35) ν = ν0 exp kB T where G is the height of the barrier (sometimes referred to as ∆ G in the literature to emphasize the fact that it is a difference of energies), kB is the Boltzmann constant, and T is the absolute temperature. In general, the dislocation perceives the identical energy barrier moving in either direction, i.e., the barrier height is G whether the dislocation is moving forwards or backwards. Since the probability of moving forwards or backwards is identical, the average motion of the dislocation is zero, and this corresponds to the expected thermal vibration around the mean position. However, when a shear stress τ is applied in a specific direction, the stress essentially provides additional work that can be used to overcome the energy barrier. In effect, the dislocation perceives a lowered energy barrier in the direction of the applied stress. How does the barrier height depend on the applied stress τ ? In general, this will be some function G(τ ) that will change depending on the specific obstacle that defines the barrier (an example of such an obstacle might be a precipitate particle). The standard approach to handling functions that are poorly known is to use a Taylor series expansion of the function about the point at which the value of the function is known (in this case, that point is τ = 0, and G = G0 when τ = 0): G(τ ) = G0 +

∂G | τ + higher order terms, ∂ τ τ =0

(5.36)

where only the first term has been written down because that τ is small 

it is assumed (this is called linearizing the function). Identifying ∂∂Gτ |τ =0 as a constant (−Ω ), the barrier height becomes G0 − τΩ in the direction of the applied stress, where Ω is a parameter (with the dimensions of volume) that is called the activation volume.:

Ω =−

∂G | . ∂ τ τ =0

(5.37)

For the moment, let us assume that the activation volume is a constant. Note that the dislocation perceives an increased activation barrier G0 + τΩ in the direction working against the applied stress. Thus because of the applied stress, the frequency ν f with which the dislocation is successful in crossing the barrier in the forward direction, i.e., the direction of the applied stress, is

ν f = ν0 exp

−G0 + τΩ , kB T

(5.38)

while the frequency νb with which the dislocation is successful in crossing the barrier in the backward direction is

164

Nanomaterials

νb = ν0 exp

−G0 − τΩ . kB T

(5.39)

Given the arguments of the exponentials, the forward frequency will be greater than the backward frequency, so the net motion of the dislocation is in the forward direction and has a success frequency

νnet = ν0 sinh

−(G0 − τΩ ) . kB T

(5.40)

If the dislocation moves the distance λ (that is, the slip distance) as a result of crossing the barrier, the corresponding dislocation velocity is vd = λ νnet = λ ν0 exp

τΩ −G0 sinh . kB T kB T

(5.41)

This is called a thermally activated motion. The corresponding effective plastic strain rate is then (using Equation 5.31) given by

ε¯˙ p = bρm vd = bρm λ ν0 exp

τΩ σ¯ Ω −G0 −G0 sinh = bρm λ ν0 exp sinh , (5.42) kB T kB T kB T kB T

where the shear stress τ has been replaced by the effective stress σ¯ for consistency of the definitions (otherwise, using τ , we would have to have γ˙ on the left hand side of Equation 5.42). This provides a relationship between the applied stress and the strain rate that results from the thermally activated motion of the dislocations. Now, the definition of the hyperbolic sine function is sinh x =

ex − e−x , 2

(5.43)

and when the argument x of the function is large, we have ex ≫ e−x , so that sinh x → ex σ¯ Ω 2 . In our case we have x = kB T , so that for reasonably large stresses the forward rate is much higher than the backward rate (which can therefore be ignored), and Equation (5.42) then reduces to

σ¯ Ω σ¯ Ω −G0 exp = ε˙0 exp , ε¯˙ p = bρm λ ν0 exp kB T kB T kB T

(5.44)

0 where we have defined ε˙0 = bρm λ ν0 exp −G kB T . This equation can be solved for the stress in terms of the strain rate, providing a relation for the strengthening increment due to thermally activated motion of the dislocations (remembering that only this mechanism is being considered):

∆ σ¯ =

kB T ε¯˙ p ln , Ω ε˙0

(5.45)

5 Plastic Deformation of Nanomaterials

165

where it should be noted that ε˙0 is typically much greater than ε¯˙ p . Note that this dependence of the stress on the strain rate is logarithmic, and much weaker than that obtained from the dislocation drag argument of Equation (5.34). Although this is a very common approach to estimate the effects of thermal activation on rate-dependent response, note that we have made two possibly contradictory approximations to obtain this relationship. First, we assumed that τ was sufficiently small in order to linearize the barrier height G(τ ) (Equation 5.36); second, we assumed that τ was sufficiently large in order to reduce the sinh function to an exponential function (Equation 5.44). Is there a range of τ where both of these conditions are satisfied? It is not obvious that there is, because the acceptable range depends on the specific obstacle being considered. This caveat is worth recalling as specific models are built for any given thermally activated deformation mechanism, but is not a concern if it is true that (for that material) G(τ ) = G0 − τΩ for all τ . An example of the rate-dependent behavior of a material is shown in Figure 5.20, which presents a range of strength data for A359 aluminum (a cast aluminum alloy). All of the strengths were measured at a fixed strain but varying strain rates; because of the nature of high strain rate experiments, it is not generally possible to measure accurately the yield strength of materials at high strain rates. A logarithmic dependence like that in Equation (5.44) appears to be a reasonable description of the rate-dependent behavior for strain rates less than 104 s−1 .

3 102 A359

Flow Stress (MPa)

2.5 102 2 102 1.5 102 1 102 5 101

0 –4 10

10–2

100 102 Strain Rate (s–1)

104

Fig. 5.20 Rate-dependence of the flow stress of A359 aluminum alloy over a wide range of strain rates. All of the flow stresses are plotted at a fixed strain of 4% (because of the nature of high strain rate experiments, it is not generally possible to measure accurately the yield strength of materials at high strain rates). This result is due to the work of Yulong Li, and includes compression, tension and torsion data.

166

Nanomaterials

5.9.3 Dislocation Substructure Evolution There is one significant concern that arises when rate-dependence of the material behavior is presented in the form of a figure like Figure 5.20: is it reasonable to compare flow stresses at fixed strain? The difficulty is that some materials, particularly fcc metals, appear to have strain hardening behavior that is itself a function of strain rate (higher strain rates leading to higher strain hardening). Thus the increase in strength (at fixed strain) with strain rate can be viewed as a strain hardening behavior rather than a “truly rate-dependent” behavior. Strain hardening is essentially a consequence of the evolution in dislocation density. It turns out that in some materials the rate of change of the dislocation density is a function of the rate of deformation: the dislocation density appears to grow faster at higher rates of deformation. The evolution of the dislocation density ρd is given by Equation (5.8): ∂ ρd (5.46) = Ac fc (ρd , ε˙ ) − Aa fa (ρd ), ∂t and it appears that (for fcc metals) the rate of creation of dislocations fc is a function not just of ρd but also of the strain rate ε˙ . As a consequence, ρd grows faster at higher strain rates, and then because of Equation (5.7) the strength of the material also increases with strain rate (because the strain hardening has increased). This kind of rate-dependence is said to be the result of dislocation substructure evolution. This particular situation brings out one of the classic differences between the mechanics and materials perspectives in this area: when is a material considered to be rate-dependent? From a mechanics perspective, the material is described in terms of a constitutive relation such as σ = f (ε , ε˙ , T ) where ε and ε˙ are the current strain and current strain rate, and T is the current temperature. The constitutive function f () has a specific form and specific values of parameters for each material, but the function typically does not contain microstructural parameters such as the dislocation density. If the strength of a material changes with strain rate at fixed strain and temperature, the material is rate-dependent from the mechanics perspective. From a materials perspective, the material behavior is typically described in terms of microstructural parameters such as the dislocation density, which are viewed as evolving with the deformation, and the material is typically said to be rate-dependent if the behavior is sensitive to strain rate at fixed (dislocation) structure. Thus from a materials perspective the increase of strength with strain rate as a consequence of the increase of dislocation density with strain rate may not be considered to be a ratedependent behavior. The identical behaviors, therefore, are viewed differently by individuals with mechanics and materials backgrounds. Materials scientists would prefer to look at figures like Figure 5.20 with the flow stress measured at constant structure. This can be accomplished, for example, through strain rate jump tests, where the strain rate is changed suddenly and the immediate response of the stress is measured, assuming that the dislocation substructure has not had time to evolve (Figure 5.21).

5 Plastic Deformation of Nanomaterials

167

490 480

Stress (MPa)

470 460 450 440 430 420

Cold Deformed Cu ECAP+Cold Deformed Cu 0.02

0.024

0.028

0.032

0.036

Strain Fig. 5.21 Results of strain rate jump tests on severely plastically deformed copper (both coldworked and ECAPed). Note the jump in the stress associated with the step increase in the strain rate on a specimen. One of the curves has been shifted to the right for ease of discrimination. Step increases in strain rate (jump from a lower strain rate to a higher strain rate) are used because a step decrease will exacerbate the effect from the machine compliance, and the interpretation of experimental data becomes more involved. In this case the strain rate was increased by a factor of 2 between consecutive rates. Reprinted from Materials Science and Engineering, Q. Wei, S. Cheng, K.T. Ramesh, E. Ma, Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Sep. 2004, with permission from Elsevier.

This effect of the evolution of substructure would be captured from a mechanics perspective by considering the effects not just of the current strain rate but of the strain rate history. It is clear that if the dislocation substructure evolves differently at different strain rates, then the strength of the material is determined not just by the current strain rate but also by the history of the strain rate, because this controls the dislocation density and thus the strength through Equation (5.7). In this viewpoint, fcc metals are not just rate-dependent but also strain rate history dependent.

5.9.4 The Rate-Dependence of Nanomaterials Although experiments employing strain rate changes are very useful in revealing deformation mechanisms, there is very limited experimental data on the strain-rate sensitive mechanical properties of nc/UFG metals. The primary material parameter of interest is the strain rate sensitivity, which is related to the activation volume defined earlier (Equation 5.37). A common definition of the strain rate sensitivity (common in engineering) is in terms of a rate sensitivity parameter m defined as m=

∂ ln σ ε˙ ∂ σ = . ∂ ln ε˙ σ ∂ ε˙

(5.47)

168

Nanomaterials

This measure of rate sensitivity is valid for any constitutive function (for examm ple, it is directly the strain rate exponent in the power law function: σ = σ0 ε˙ε˙0 ), and is an indicator of the strain-rate response that is useful for technological purposes. The “engineering” strain rate sensitivity m and the activation volume Ω are related as follows (in the particular case of thermal activation): m=

kB T . σΩ

(5.48)

Experiments of the type shown in Figure 5.21 are typically used to estimate m, using the slopes of linear fits on a log-log plot of stress vs strain rate. The activation volume can then be computed using Equation (5.48). The dependence of rate-sensitivity on grain size can be obtained by performing such experiments on materials with a range of grain sizes.

5.9.4.1 FCC Nanometals Carreker and Hibbard (1953) first discussed the grain size dependence of the rate sensitivity for tensile deformation of high-purity Cu. They observed that m increased slightly from 0.004 (large d) to 0.0072 (small d), but the grain size range studied was very small (12 − 90 µm). Early work on Ag and Cu also showed that polycrystalline samples had larger values of m than single crystals (Conrad, 2003). Coarse-grained Cu (d = 40 µm) has a rate sensitivity m of the order of 0.006 according to Follansbee and Kocks (1988). More recent studies of ultrafine-grained Cu also showed elevated m (Gray et al., 1997). A very large m (0.14) for Cu was reported by Valiev et al. (2002) on a material obtained using 16 passes of ECAP. The rate-sensitivity (m = 0.03) of a UFG Cu processed by ECAP+CR has been measured at large plastic strains (Wang et al., 2002b) using load-reload compression tests. The tensile jump tests shown in Figure 5.21 give m = 0.02 for a similar ECAPed Cu (d = 200 nm). Thus the available data indicate a substantial increase in m when the grain size is reduced into the UFG and nc regime, regardless of the technique used to process the material. This behavior is summarized in Figure 5.22, showing the change in the rate sensitivity of copper over a wide range of grain sizes. Similar behavior is observed in other fcc metals, such as nickel and gold. Note that all of these rate-sensitivities are for deformations that are at nominal strain rates from 10−4 to 103 s−1 . The apparent rate-sensitivity is much higher (m  0.5) for all grain sizes of fcc metals in the creep domain, because of diffusional processes. What causes the increase in rate-sensitivity with decreasing grain size in fcc metals? There is no clear explanation for this at present, but one plausible explanation is that the evolution of dislocation substructure (which is known to be important in fcc metals) is controlling the behavior. For coarse-grained fcc metals, the primary obstacles to the motion of glissile dislocations are known to be forest dislocations, leading to the rate dependence of flow stress through thermal activation. According to the work of Conrad (1964), the activation volume can be written as

5 Plastic Deformation of Nanomaterials

169

Rate-sensitivity, m

0.1 0.08 0.06

[47,48]

[41]

[37]

[49]

[21]

Cu, this work

[22]

Cu, this work

[25]

Cu, this work

0.04 0.02 0 101

102

103 Grain size, nm

104

105

Fig. 5.22 The rate-sensitivity of copper as a function of the mean grain size, including the nanocrystalline, ultra-fine-grain and coarse-grain domains. Data is presented for materials made through a variety of processing routes, some of which involve severe plastic deformation. The figure is taken from the work of Wei et al. (2004a). Reprinted from Materials Science and Engineering, Q. Wei, S. Cheng, K.T. Ramesh, E. Ma, Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Sep. 2004, with permission from Elsevier.

Ω = bxl ∗ ,

(5.49)

where x is the distance swept out by a gliding dislocation during one activation event and l ∗ is the length of the dislocation line segment involved in the thermal activation (this is also the Friedel sampling length, which scales with the average contact distance between two obstacles, so it is the obstacle spacing). For large-grain fcc metals with forest cutting as the dominant mechanism for plastic deformation, l ∗ is roughly the average forest spacing which scales with the forest density √1ρ f , or following Taylor’s relation (Equation 5.7), l ∗ with µτb . The sliding distance x may also decrease slightly with increased applied stress (Kobrinsky and Thompson, 2000). It thus follows that the activation volume would decrease with applied stress. As an example, for heavily deformed Cu, Ω decreases to about 100 b3 (Dalla Torre et al., 2004). Combining Equations (5.48) and (5.49) we have m=

kB T kB T = , σΩ σ bxl ∗

(5.50)

in which x is roughly approximated to be constant. Therefore, the variation of m has to be understood from the combined variation of σ and l ∗ through the product σ l ∗ . We expect that the flow stress has contributions arising from both grain boundaries and dislocations, and it can be written as 1 √ σ = σ0 + AT µ b ρ f + ky d − 2 ,

(5.51)

170

Nanomaterials

where the first term accounts for all of the other strengthening components, the second arises from the Taylor contribution (Equation 5.7), and the third arises from the Hall-Petch relation. The obstacle spacing or “sampling length” l ∗ is assumed to have two possible limits, l ∗ 1 and l ∗ 2 : 1 l ∗ 1 = α1 √ ρf

(5.52)

l ∗ 2 = α2 d,

(5.53)

where α1 and α2 are proportionality factors. These limits are chosen because they represent, respectively, the two classes of obstacles of interest: forest dislocations (l ∗ 1 ) and grain boundaries (l ∗ 2 ). Of the two possible limits for the obstacle spacing, l ∗ 1 is likely to be the controlling length scale when the grain sizes and dislocation densities are large, and l ∗ 2 is likely to be the controlling length scale at very small grain sizes (and small dislocation densities). In this model, as the grain size is decreased there will be a transition grain size when the controlling sampling length changes from l ∗ 1 to l ∗ 2 . A crude estimate of the transition size can be obtained by asking when l ∗ 1 and l ∗ 2 are of the same order of magnitude, and assuming that α1 and α2 are of the same order. This gives 1 dtransition ≈ √ . ρf

(5.54)

Using typical dislocation densities in metals of 1012 − 1017 m−2 , one estimates that dtransition is of the order of 10–500 nm. What are the perceived material behaviors in the two limits? At conventional grain sizes we know that the forest cutting is dominant, so that we have (using Equations 5.51 and 5.52) 1 1 1 √ σ l ∗ = σ l ∗ 1 = (σ0 + AT µ b ρ f + ky d − 2 )α1 √ = AT µ bα1 + ky α1  , (5.55) ρf dρ f

where the initial strengthening term has been dropped since it is usually small for pure fcc metals (note that this is not a good approximation for alloys). Using Equation (5.55) in Equation (5.50), we obtain m=

1 kB T bx AT µ bα1 + ky α1 √1

,

(5.56)

dρ f

Equation (5.56) indicates that when the dislocation density is increased (for example by large deformations) in conventional grain-sized metals, the strain rate sensitivity should increase. This explains the observations reported by Zehetbauer and Seumer (1993). Consider now the case of nanocrystalline fcc materials, where the grain size is refined into the UFG/nc regime, the controlling sampling length becomes l ∗ 2 , and we now have

5 Plastic Deformation of Nanomaterials

171

√ 1 √ √ σ l ∗ = σ l ∗ 2 = (σ0 + AT µ b ρ f + ky d − 2 )α2 d = AT µ bα2 d ρ f + ky α2 d, (5.57) and now m=

1 kB T √ . √ bx AT µ bα2 d ρ f + ky α2 d

(5.58)

Equation (5.58) indicates that when the grain size is refined into the UFG/nc regime, the rate sensitivity parameter m should increase with reduced grain size, as was observed in Figure 5.22. Further, as d is reduced to below the transition size, the forest dislocation density in the grain interior may become very low, whereas the obstacle density associated with grain boundaries becomes very high. It is thus plausible that the controlling obstacles are the grain (or subgrain) boundaries.

5.9.4.2 BCC Nanometals Very different rate-dependent behavior is observed in nanostructured bcc metals. In these materials it is observed that the effective rate-sensitivity decreases with decreasing grain size. Such a decrease has been observed in iron, tantalum, tungsten and vanadium, and a summary of these data is presented in Figure 5.23. 0.07

Rate Sensitivity, m

0.06 0.05 0.04

Consol. Fe [16] ECAP Fe, C [34] nc Fe, [51] nc Fe, [53] ECAP Ta, C [35] Consol. V [52] ECAP Fe ECAP+Roll Fe ECAP+Roll Ta JMR model

0.03 0.02 0.01 0 1 10

102

103 Grain size (nm)

104

105

Fig. 5.23 The rate-sensitivity of bcc metals as a function of grain size from a variety of sources (the figure is a variant of one published by Wei et al. (2004a). The decrease of rate sensitivity with decreasing grain size is the opposite behavior to that observed in fcc metals. Reprinted from Materials Science and Engineering, Q. Wei, S. Cheng, K.T. Ramesh, E. Ma, Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Sep. 2004, with permission from Elsevier.

172

Nanomaterials

The source of this behavior is the deformation mechanism that dominates bcc metal plasticity, that of kink nucleation and propagation, and is discussed in detail in the next section. In brief, the explanation is that the plastic deformation of nanocrystalline bcc metals is controlled by the Peierls-Nabarro stress and the temperatureand rate-insensitive grain boundary controlled Hall-Petch mechanism, and for the very high-strength nanocrystalline bcc metals these terms are much larger than the rate-dependent terms, resulting in a reduction in the effective rate-sensitivity.

5.10 Case Study: Behavior of Nanocrystalline Iron As a case study, let us work through understanding the plastic deformation of nanocrystalline iron (we discussed the elastic behavior of single crystal iron in Chapter 2). Iron is a bcc metal and has been studied extensively both for fundamental science reasons and as the primary constituent of steel. Plasticity in metals is a consequence of dislocation motion, and the nature of dislocation slip in bcc metals is somewhat different than that in fcc systems. One of the major differences is that in bcc metals the edge and screw dislocations (or edge and screw components of a dislocation loop) have vastly different mobilities. Screw dislocation cores are dissociated into a non-planar configuration in bcc structures, which makes it difficult for them to move (Duesbery and Vitek, 1998). In contrast, edge dislocations move freely, and the mobility of the edge components of dislocation loops is much higher than the mobility of the screw components. As a consequence TEM images of dislocations in bcc metals tend to show lots of screw dislocations and few edge dislocations, because the edges have all run out to the free surfaces. Since the edges are so mobile, plastic deformation in bcc metals is generally controlled by the motion of screw dislocations, which means the plasticity is controlled by the barriers to the motion of the screws. The primary barriers to the motion of screw dislocations in bcc metals turn out to be lattice-related, rather than forest dislocations: the Peierls-Nabarro stress is relatively high. This high friction stress derives in part from the fact that the screw dislocation core tends to be nonplanar. These lattice barriers essentially define the energy landscape in the thermal activation theory discussed previously. Moving the entire dislocation line across the lattice costs a significant amount in energy terms and, nature being as resourceful as she is, the dislocation finds an energetically cheaper way to move. Essentially, rather than move the entire screw dislocation at one time, one small segment of the screw collapses back into a linear configuration and jumps across the barrier, resulting in a kinked configuration (Figure 5.24). The sides of the kink pair (which are called the kinks) turn out to have an edge orientation and have high mobility, so they fly out to the sides, causing effectively a propagation of the screw across the barrier. This is the mechanism of motion of the screws in bcc metals, and the kink mechanism determines the mobility of the screw dislocations.

5 Plastic Deformation of Nanomaterials

173

Fig. 5.24 Schematic of motion of screw dislocations by kink pair nucleation and propagation. The dislocation is visualized as a line that needs to go over the energy barrier. Rather than move the entire line over the barrier, the dislocation nucleates a kink, which jumps over the barrier. The sides of the kink pair have an edge orientation, and so fly across the crystal because of their high mobility, resulting in an effective motion of the screw dislocation.

The kink mechanism involves two steps: kink nucleation and then the propagation of kink pairs. Kink propagation is rapid, so (at low homologous temperatures) the screw dislocation mobility is controlled by kink nucleation. The kink pair nucleation rate νk is given by −Gk νk = νk0 exp , (5.59) kB T where νk0 is the attempt frequency and Gk is the activation energy for kink-pair nucleation. The corresponding average dislocation velocity is vd = λ νk = λ νk0 exp

−Gk , kB T

(5.60)

where λ is the distance between the barriers (of the same order as the Burgers vector b in bcc metals). The plastic shear rate corresponding to this mechanism is then, using the Orowan equation,

ε˙ p = ρm bλ νk0 exp

−Gk , kB T

(5.61)

where a Taylor factor of the order of 3 has been absorbed into the assumed value for the mobile dislocation density in Equation (5.61), since ρm is usually only known to within an order of magnitude in any case. The key remaining question is the dependence of the activation energy on the stress (e.g., we had assumed a linear dependence in Equation (5.36). A common approach to this definition of the barrier shape is to use an empirical function first proposed by Kocks et al. (1975):   e a b σˆ Gk = Gk0 1 − , σ0

(5.62)

where a and b are parameters that define the barrier shape, σˆ e is the effective stress (which we shall call a Peierls stress) for this thermally activated mechanism, and σˆ 0 is a material parameter. Various values for a and b are used in the literature, with particularly common values (for bcc metals) being a = 12 , b = 32 . Using Equation (5.62) in (5.61) and solving for the stress, the effective thermally assisted “Peierls stress” σˆ e is then given by

174

Nanomaterials

σˆ e = σˆ 0

  1  1a b kB T ρm bλ νk0 1− . ln p ˙ Gk0 ε





(5.63)

The collection of microscopic terms in the numerator of the logarithmic function is sometimes referred to as a “reference” strain rate ε˙0 = ρm bλ νk0 . Considering typical values for these parameters for an annealed metal, we see that ε˙0 ≥ 105 s−1 (although usually many orders of magnitude larger). The effective Peierls stress σˆ e is the component of the flow stress σ related to thermally activated motions over short-range barriers (corresponding to the Peierls barriers in this case). To develop the constitutive function for the material we need to add to this the “athermal” stress corresponding to that needed to overcome the long-range barriers perceived by the moving dislocations, i.e., σ = σathermal + σˆ e , so that we obtain the flow stress as e

σ = σathermal + σˆ = σathermal + σˆ 0

  1  a1 b kB T ρm bλ νk0 ln . 1− p Gk0 ε˙





(5.64)

The primary part of the athermal stress (i.e., the most important long-range barriers) for the materials that we are considering is that arising from the grain boundaries, and this brings in a grain size dependence that we assume to be of the HallPetch form: √ σathermal = σa0 + ky d + g(ε p ), (5.65) where we have also included a strain hardening function g(ε p ) in an additive fashion, since experimental evidence for bcc metals generally indicates that the strain hardening function is not affected by strain rate or temperature in the lowtemperature regime, e.g., Ono (1968). We now have an effective response function that should be applicable to our bcc iron:    2 2  3 √ kB T ε˙0 p ln σ = σa0 + ky d + g(ε ) + σˆ 0 1 − , (5.66) p Gk0 ε˙ using the common bcc values for a and b in Equation (5.66). We have thus obtained a constitutive formalism that we can use to compare with experimental data. The strain hardening of BCC metals is typically described fairly well by a power law, p and so we may write g(ε p ) = σg0 ( εε0 )n , where σg0 and ε0 are a reference stress and a reference strain used to fit a quasi-static stress-strain curve and n is the material parameter known as the strain hardening index. Using this function for the strain dependence, we have    p n   2 2 3 √ kB T ε ε˙0 σ = σa0 + ky d + σg0 + σˆ 0 1 − . (5.67) ln ε0 Gk0 ε˙ p The key feature of Equation (5.66) for our purposes arises from the assumed mechanism of plastic deformation in bcc metals – we should expect that the change

5 Plastic Deformation of Nanomaterials

175

of grain size will not affect the additive stress term corresponding to the ratesensitive response of the material. In terms of experimental data, Equation (5.66) tells us that it is useful to examine the variation with strain rate of the rate-dependent term alone as in Equation (5.63) (i.e., we examine the flow stress difference σ − σ f 0 , where σ f 0 is the average flow stress for each grain size measured at the low rates) since the effect of grain size and strain represent additive terms for the assumed deformation mechanism. The model is able to capture the primary features of the observed behavior using physically reasonable parameters for this strain rate and temperature regime. Although the absolute increases in strength as a result of changes in strain rate are consistent across these grain sizes in accord with the model predictions, the relative rate-dependent strengthening of the nc-Fe is much smaller than that observed at conventional grain sizes, so that the total flow stress is less sensitive to strain rate in ultra-fine-grain and nanocrystalline iron (again as predicted by the model). This is illustrated in Figure 5.23, which plots the rate-sensitivity as a function of grain size for a number of bcc metals. The effective rate-sensitivity of these materials (as described, for example in the index of a power-law rate-dependence function) decreases as a function of grain size because the dominant part of the strength is the grain boundary strengthening at small grain sizes.

5.11 Closing The elasticity and plasticity of materials discussed in the last two chapters are examples of the constitutive behavior of materials. The corresponding constitutive equations (Chapter 2, Section 2.4) relate the stresses, strains and strain rates at every point in the material, and each of these constitutive equations contain material parameters (such as the Young’s modulus, Poisson’s ratio, and the yield strength) that must be obtained from experimental measurements, or from calculations (such as atomistic or first principles calculations) that are derived from a lower scale, as we will discuss in a subsequent chapter. An equation such as Equation (5.67) describes the dependence of the material’s strength on the strain, the strain rate, the temperature, as well as an internal variable (the grain size). The application of Equation (5.67) for the development of a full constitutive equation for a material (in this case iron) will be shown in a later chapter. First, in the next chapter, we discuss the processes of mechanical failure in nanomaterials because these processes define the limits of application of such materials.

5.12 Suggestions for Further Reading 1. R.W. Hertzberg, Deformation and Fracture Mechanics of Engineering Materials. John Wiley, New York, Fourth Edition, 1995.

176

Nanomaterials

2. A.S. Argon, Strengthening Mechanisms in Crystalline Metals. Oxford University Press, Oxford, 2008. 3. R.J. Asaro and V.A. Lubarda, Mechanics of Solid Materials. Cambridge University Press, Cambridge, 2006.

5.13 Problems and Directions for Research 1. In fcc systems, the Miller indices essentially define the components of the vector. Consider a face centered cubic single crystal. The slip systems available are listed in Table 5.1. Define the slip system normal vector m and the slip direction s for each slip system using that table. 2. Now compute the “transformation tensor” µ , defined as µi j = 21 (si n j + ni s j ) for each slip system (Equation 5.20). Each µ that you compute corresponds to one of the µiαj tensors in Equation (5.26), i.e., the transformation tensor associated with that slip system. 3. Consider now a single crystal of a face centered cubic material (say copper) that is being pulled in tension along the [100] type direction. Calculate the resolved shear stress on each slip system in this case. 4. Search the literature to obtain a critical resolved shear stress for copper along slip system 10 in Table 5.1. Assuming that the crystal is being pulled in simple tension, calculate the applied stress at which slip will begin in the crystal.

References Argon, A. (2008). Strengthening Mechanisms in Crystal Plasticity. Oxford Series on Materials Modelling. Oxford: Oxford University Press. Armstrong, R. (1970). The influence of polycrystal grain size on several mechanical properties of materials. Metallurgical Transactions 1, 1169–1176. Arzt, E. (1998). Size effects in materials due to microstructural and dimensional constraints: a comparative review. Acta mater 46(16), 5611–5626. Asaro, R. and V. Lubarda (2006). Mechanics of Solids and Materials. New York: Cambridge University Press. Ashby, M. (1970). Deformation of plastically non-homogeneous materials. Philosophical Magazine 21(170), 399. Cao, B. and K. T. Ramesh (2009). Strengthening mechanisms in cryomilled 5083 al. Scripta materialia 60, 619–622. Carreker, Jr., R. and W. Hibbard Jr. (1953). Tensile deformation of high-purity copper as a function of temperature, strain rate, and grain size. Acta Metall. 1, 654–663. Chew, Y., C. Wong, F. Wulff, F. Lim, and H. Goh (2007). Strain rate sensitivity and hall-petch behavior of ultrafine-grained gold wires. Thin Solid Films 516, 5376–5380. Chokshi, A., A. Rosen, J. Karch, and H. Gleiter (1989). On the validity of the hall-petch relationship in nanocrystalline materials. Scripta Materialia 23(10), 1679–1683. Clifton, R. (1983). Dynamic plasticity. Transactions ASME: Journal of Applied Mechanics 50, 941–952. Conrad, H. (1964). Thermally activated deformation of metals. Journal of Metals 16, 582–588.

5 Plastic Deformation Nanomaterials

177

Conrad, H. (2003). Grain size dependence of the plastic deformation kinetics in cu. Materials Science and Engineering a-Structural Materials Properties Microstructure and Processing 341 (1–2), 216–228. Courtney, T. (2005). Mechanical Behavior of Materials. Long Grove: Waveland Press. Dalla Torre, F., E. V. Pereloma, and C. H. J. Davies (2004). Strain rate sensitivity and apparent activation volume measurements on equal channel angular extruded cu processed by one to twelve passes. Scripta Materialia 51(5), 367–371. Duesbery, M. and V. Vitek (1998). Plastic anisotropy in b.c.c. transition metals. Acta Materialia 46, 1481–1492. Follansbee, P. and U. Kocks (1988). A constitutive description of the deformation of copper based on the use of the mechanical threshold stress as an internal state variable. Acta Metallurgica 36, 81–93. Fu, H. H., D. J. Benson, and M. A. Meyers (2001). Analytical and computational description of effect of grain size on yield stress of metals. Acta Materialia 49(13), 2567–2582. Gray, G. T., T. C. Lowe, C. M. Cady, R. Z. Valiev, and I. V. Aleksandrov (1997). Influence of strain rate & temperature on the mechanical response of ultrafine-grained cu, ni, and al-4cu-0.5zr. Nanostructured Materials 9(1–8), 477–480. Hahn, H. and K. A. Padmanabhan (1997). A model for the deformation of nanocrystalline materials. Philosophical Magazine B-Physics of Condensed Matter Statistical Mechanics Electronic Optical and Magnetic Properties 76(4), 559–571. Hall, E. O. (1951). The deformation and ageing of mild steel .3. discussion of results. Proceedings of the Physical Society of London Section B 64(381), 747–753. Han, B., Z. Lee, S. Nutt, E. Lavernia, and F. Mohamed (2003). Mechanical properties of an ultrafine-grained al-7.5 pct mg alloy. Metallurgical Materials Transactions A 34, 603. Hansen, N. (2005). Boundary strengthening in undeformed and deformed polycrystals. Materials Science & Engineering A 409, 39–45. Hirth, J. and J. Lothe (1992). Theory of Dislocations (Second ed.). Malabar, FL: Krieger. Jia, D., K. T. Ramesh, and E. Ma (2003). Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. Acta Materialia 51(12), 3495–3509. Karimpoor, A., K. T. Aust, and U. Erb (2007). Charpy impact energy of nanocrystalline and polycrystalline cobalt. Scripta Mater 56, 201–204. Koch, C. C., D. G. Morris, K. Lu, and A. Inoue (1999). Ductility of nanostructured materials. MRS Bulletin 24(2), 54–57. Kobrinsky, M. J. and C. V. Thompson (2000). Activation volume for inelastic deformation in polycrystalline ag thin films. Acta Materialia 48(3), 625–633. Kocks, U., A. Argon, and M. Ashby (1975). Thermodynamics and kinetics of slip. Progress in Materials Science 19, 1–288. Koo, R. (1962). Grain size effects on the deformation of tantalum at low temperatures. Journal of the Less-Common Metals 4(2), 138–144. Li, J. (1963). Petch relations and grain boundary sources. Transactions Metallurgical Society of AIME 227(1), 239. Lloyd, D. J. and S. A. Court (2003). Influence of grain size on tensile properties of al-mgalloys. Materials Science and Technology 19(10), 1349–1354. Meyers, M. and E. Ashworth (1982). A model for the effect of grain size on the yield stress of metals. Philosophical Magazine A 46(5), 737–759. Meyers, M. A., A. Mishra, and D. J. Benson (2006). Mechanical properties of nanocrystalline materials. Progress in Materials Science 51(4), 427–556. Mughrabi, H. (1983). Dislocation wall and cell structures and long-range internal stresses in deformed metal crystals. Acta Metallurgica 31(9), 1367–1379. Nieh, T. G. and J. Wadsworth (1991). Hall-petch relation in nanocrystalline solids. Scripta Metallurgica Et Materialia 25(4), 955–958. Ono, K. (1968). Temperature dependence of dispersed barrier hardening. Journal of Applied Physics 39(3), 1803, A8822. Perez, R. J. (1997). Ph.D. Thesis, University of California.

178

Nanomaterials

Perez, R. J. (1998). Grain growth of nanocrystalline cryomilled Fe-Al powders. Metallurgical and Materials Transactions A 29, 2469. Petch, N. J. (1953). The cleavage strength of polycrystals. Journal of the Iron and Steel Institute 174(1), 25–28. Phillips, V. A. (1953). Yield-point phenomena and stretcher-strain markings in aluminium magnesium alloys. Journal of the Institute of Metals 81, 625. Robinson, J. M. (1995). In-situ deformation of aluminium alloy polycrystals observed by highvoltage electron microscopy. Materials Science and Engineering A 203, 238. Rodriguez, R. (2003). Tensile and creep behavior of cryomilled Inco 625. Acta Materialia 51, 911. Rosler, J. (1992). Microstructure and creep properties of dispersion-strengthened aluminum alloys. Metallurgical Transactions A 23, 1521. Scattergood, R. O. and C. C. Koch (1992). A modified-model for hall-petch behavior in nanocrystalline materials. Scripta Metallurgica Et Materialia 27(9), 1195–1200. Schmid, E. and W. Boas (1935). Kristallplastizit¨at. Berlin: Springer. Seeger, A. (1958). In: A. Flugge (Ed.). Encyclopedia of Physics. Vol. 7, p. 1. Berlin: Springer. Shu, J. and N. Fleck (1999). Strain gradient crystal plasticity: size-dependent deformation of bicrystals. Journal of the Mechanics and Physics of Solids 47(2), 297–324. Semiatin, S. L. (2001). Plastic flow and fracture behavior of an Al-Ti-Cu nanocomposite. Scripta Materialia 44, 395. Valiev, R. Z., I. V. Alexandrov, Y. T. Zhu, and T. C. Lowe (2002). Paradox of strength and ductility in metals processed by severe plastic deformation. Journal of Materials Research 17(1), 5–8. Siegel, R. W. (1994). What do we really know about the atomic-scale structures of nanophase materials. Journal of Physics and Chemistry of Solids 55, 1097. Sun, X. K. (2000). Preparation and mechanical properties of highly densified nanocrystalline Al. Metallurgical and Materials Transactions A 31, 1017. Wang, Y. M., M. W. Chen, F. H. Zhou, and E. Ma (2002). High tensile ductility in a nanostructured metal. Nature 419(6910), 912–915. Wei, Q., S. Cheng, K. T. Ramesh, and E. Ma (2004a). Effect of nanocrystalline and ultrafine grain sizes on the strain rate sensitivity and activation volume: fcc versus bcc metals. Materials Science And Engineering A-Structural Materials Properties Microstructure And Processing 381(12), 71–79. Wei, Q. and L. Kecskes (2008). Effect of low-temperature rolling on the tensile behavior of commercially pure tungsten. Materials Science & Engineering A 491, 62–29. Witkin, D. and E. Lavernia (2006). Synthesis and mechanical behavior of nanostructured materials via cryomilling. Progress in Materials Science 51(1), 1–60. Wu, T. Y., J. L. Bassani, and C. Laird (1991). Latent hardening in single-crystals .1. theory and experiments. Proceedings of the Royal Society of London Series a-Mathematical Physical and Engineering Sciences 435(1893), 1–19. Zehetbauer, M. and V. Seumer (1993). Cold work-hardening in stage-iv and stage-v of fcc metals .1. experiments and interpretation. Acta Metallurgica Et Materialia 41(2), 577–588. Zhou, Y., U. Erb, K. T. Aust, and G. Palumbo (2003). The effects of triple junctions and grain boundaries on hardness and young’s modulus in nanostructured ni-p. Scripta Materialia 48(6), 825–830.

Stone cracks from a hard enough blow. Steel shatters. The oak fights the wind and breaks. The willow bends where it must and survives. Robert Jordan, The Wheel of Time

6

Mechanical Failure Processes in Nanomaterials The definition of the failure of a material is of course determined by the intended application. For example, in the majority of traditional mechanical design problems, the material is intended to operate in the elastic range. Therefore, in that application, the material may be said to fail when it begins to deform plastically (that is, when the yield strength is reached). We have discussed the onset of plasticity (yield, or the elastic limit) in the previous chapter. Beyond yield, the material continues to carry load in a nonlinear manner, corresponding to the plastic deformation; the nonlinear response is something that designers often avoid because of the complexity associated with the constitutive behavior in that domain. In this chapter, our interest is primarily in the applications of nanomaterials beyond the elastic limit, assuming a modicum of plastic deformations. Situations where there is no nonlinear deformation after the elastic limit is reached, as in ceramics, will be discussed as well. As any good mechanical designer knows, once the material goes plastic, it is desirable to have the component continue to carry load for at least some time so as to provide an additional safety margin in the application. How much deformation can be sustained in the nonlinear range before a complete loss of load-carrying capacity occurs? This determines how gracefully the component will fail and is a key part of effective design. From a mechanical perspective, the complete loss of load-carrying capacity is the ultimate definition of mechanical failure. If the amount of nonlinear deformation that can be sustained before complete loss of load-carrying capacity occurs is nearly zero, we say that the material is effectively brittle. This is a condition that is generally to be avoided in safety critical applications (equivalently, this is a condition under which the designer must incorporate a much larger factor of safety in her design calculations). If a significant amount of nonlinear deformation can be sustained before complete loss of load-carrying capacity, the material is said to be ductile. We shall develop more precise definitions of brittleness and ductility within this chapter. It is important to recognize that the loss of load-carrying capacity of a component or structure is fundamentally a structural response that is modulated by the material

K.T. Ramesh, Nanomaterials, DOI 10.1007/978-0-387-09783-1 6, c Springer Science+Business Media, LLC 2009 

179

180

Nanomaterials

behavior. That is, mechanical failure by this definition occurs because of the failure of the structure, which depends on the failure of the material but also involves the geometry and boundary constraints provided by the structure. As an example, the collapse of a steel-frame building can be triggered by the failure of the steel within one of the columns, but the collapse process itself is dominated by the structural features of the building – the relative mass distribution, the locations of other columns and the nature of the joints between the columns. Similar considerations are relevant to mechanical components and small engineering structures, which are after all the eventual applications of nanomaterials. With this caveat, in this chapter we focus on the failure of materials rather than on the failure of structures.

6.1 Defining the Failure of Materials

Engineering Stress

How is the failure of a material defined after yield? This is a complicated question which has caused much handwringing in the mechanical design and materials literature. Most authors define material failure by the loss of the load-carrying capacity of a specimen of the material subjected to a specific kind of load (e.g., the failure of a specimen in a uniaxial tensile test, see Chapter 2). This approach amounts to defining a canonical structure (such as a tensile test specimen) and defining material failure as corresponding to the failure of the canonical structure made of the material. The most common definition of material failure of this type is that of final fracture in the simple tension test (Figure 6.1). In this experiment, a specimen is subjected to a state of nearly uniaxial tensile stress, and the stress is plotted against the strain along the direction of tension (the axial strain). Beyond yield, the stress-strain curve is nonlinear and has a maximum when looked at in terms of the engineering

σ σ σ

Engineering Strain Fig. 6.1 Stress strain curve for a material obtained from a standard tensile specimen tested in uniaxial tension, showing the final fracture of the specimen in uniaxial tension.

6 Mechanical Failure Processes in Nanomaterials

181

stress and engineering strain (the conventional approach). Note that the engineering stress is defined in terms of the load P per unit initial area A0 , while the engineering strain is defined in terms of the change in length δ l divided by the initial gauge length l0 (see Chapter 2). This maximum engineering stress is called the ultimate tensile strength, and is often cited as a material property. After the maximum engineering stress is reached, the engineering strain continues to increase while the engineering stress is decreasing until specimen fracture occurs and the specimen is no longer capable of carrying a load. Truly brittle materials such as ceramics have no apparent plastic deformation after the elastic limit, and the elastic limit in tension is essentially also the fracture strength. Since the simple tension test involves the measurement of two quantities (the uniaxial stress and the axial strain), this definition of material failure produces two parameters that define the failure: the stress at the time of fracture (called the fracture stress, σ f = AP0 ) and the strain at the time of fracture (called the fracture strain,

ε f = δl0l ). Most material scientists ignore the fracture stress and use the fracture strain as the measure of failure. There are several variations on the fracture strain that are in common currency in the materials science literature, including the elongation to failure, the reduction in the area, and the so-called ductility. When two materials are compared, the material with the larger fracture strain is said to be more ductile. The implicit assumption is that the fracture strain is in fact a material property. However, the fracture strain is actually a property of the specimen rather than of the material alone, and changing the specimen dimensions can lead to a change in the measured fracture strain for the same material. Consider two tension specimens (of one material) that are identical in gauge length (say 30 mm) but have very different diameters: one has a diameter of 10 mm, and the other has a diameter of 100 µm. The fracture strain that is measured with the small diameter specimens will be considerably smaller than the fracture strain measured with the large diameter specimen (we will see why this is so once we have understood the failure process). Thus the fracture strain should only be used as a material property if it is agreed, as a matter of convention, that only a standard specimen will be used with a standard tensile test protocol. This is not a problem for conventional materials, where there are agreed-upon standards for tensile testing that include a specification of specimen shape and dimensions. Thus it is perhaps reasonable to compare the fracture strains obtained on 5083-H131 aluminum alloy and 6061-T6 aluminum alloy within an ASTM standard tensile test. Specific ASTM standards have been defined for tensile testing of steels (ASTM A370), aluminum alloys (ASTM B557) and so forth, but these are typically for conventional materials in terms of grain size and microstructure. There is a specific ASTM standard for high-strength materials, but that involves a different specimen geometry. The use of this geometry can result in a different failure process within the material, so that different fracture strains will be measured for the same material using the two different ASTM standards. One immediately sees a problem: nanocrystalline materials are typically very strong, so the measurement of the elastic modulus and the yield strength would require the use of the ASTM

182

Nanomaterials

standard for high-strength materials. However, the measured fracture strain of the nanocrystalline material cannot then be compared to that of the conventional-grainsize version of the same material using the other ASTM standard. Further, nanomaterials generally cannot be obtained in quantities large enough to perform a standard tension test, and so they tend to be subject to a variety of specialized mechanical testing techniques. Given the dependence of the failure process on the specimen geometry and testing protocol, it is generally not reasonable to compare the fracture strains of different nanomaterials tested in tension using different specimen geometries. However, the literature is replete with comparisons of the fracture strain (the so-called ductility) of materials as a function of grain size. These comparisons should be viewed with a very jaundiced eye. It is useful to remember why fracture strains are measured in the first place. The intent is to compare one material to another, and to be able to determine which of these materials would be best suited for a specific application in which failure is to be avoided or controlled. If the fracture strain is unable to distinguish between the usefulness of materials in a specific application, then it is not a useful material failure parameter for that application. It is straightforward to find applications for which the fracture strain is not a useful parameter. Consider two materials A and B, for both of which standard tensile tests have been performed resulting in the measurement of fracture strains ε f A and ε f B . The two materials are now used to make identical short columns which carry only an axial compressive load. Let us suppose that ε f A > ε f B . In general, it is not possible to conclude that the column made of material B will lose its load carrying capacity before the column made of material A. The failure processes are generally different in the two stress states of tension and compression, even though the yield strengths are typically the same in tension and compression for conventional metallic materials. The failure process in a shear-dominated problem may be different again. We must, therefore, develop an understanding of failure processes that we can apply to a variety of stress states, because a variety of stress states will be observed in application. This is obviously an issue in conventional materials, but becomes much more important in nanocrystalline materials for reasons that we shall see in this chapter. Another important difference between constitutive behaviors (such as elasticity and plasticity) and the failure process within the material is that failure processes tend to localize (so that all of the action occurs, for example, at the crack tip, rather than everywhere within the loaded solid). This is the primary reason why failures tend to be so sensitive to structural geometry and boundary constraints, since the latter may have a strong effect on the localization process. A related consequence is particularly significant: the localization associated with failure means that failure processes are very defect sensitive, with localization (and the associated failure) occurring much more easily in regions of high defect density. The control of failure processes is therefore often tied to the control of defect distributions in materials (this is particularly important in ceramics, but is an issue even in ductile metals). Two nanomaterials that are otherwise similar can show very different failure behaviors because of differences in defect density.

6 Mechanical Failure Processes in Nanomaterials

183

6.2 Failure in the Tension Test When a tensile test specimen (Figure 6.2) is loaded beyond yield along the specimen axis, increasing plastic strains are developed while the stress increases because of strain hardening (for the moment we only consider the strain hardening case). If the specimen is properly designed so that the stress concentrations at the grips have been minimized, the plastic strains are first developed within the gauge section of the specimen. As loading continues, the plastic strain continues to increase in the gauge section of the specimen and spreads to some degree into the region of the grips. A simple finite element calculation of the development of plastic strains inside the tensile specimen shows that the strains are not homogeneous outside the gauge section. The inhomogeneities at the ends of the specimen arise due to the changing geometry near the grips, and these inhomogeneities can propagate into the gauge section to some degree. This results in several complications. The gauge length is typically used to calculate the strain in the specimen, on the basis of the assumption that the deformations within the gauge section are uniform. However, calculations show that the fraction of the specimen that has uniform plastic strains depends on material properties such as strain-hardening. h

P

neck

P

Fig. 6.2 The development of a neck during plastic deformation of a specimen within a simple tension experiment. The position and length of the neck are determined by geometric and material imperfections in the specimen.

What is the effective gauge length le of the material in the plastic range? This length is needed for calculating the strain in the specimen, but the number itself depends on the material properties, which are unknown (indeed that is why the tension test is being conducted). If the original gauge length (based on elastic deformations of the specimen) is used to calculate the engineering strain, then the strain measure will be inaccurate to some degree. This problem is recognized by most experts in mechanical testing, and so local (true) strains are measured on the specimen using strain gauges or digital image correlation. However, plastic deformations in the tension test may not be stable, and even this local measure of the strain may not be useful once the deformations become unstable. Consider the effect of a small geometric imperfection in a metal tension specimen, so that at some location in the gauge length there is a slightly smaller crosssection area (for example, as a result of machining-induced surface roughness). Since force equilibrium requires that the force be the same at every cross-section in the specimen, the local stress at the imperfection will be higher because of the smaller cross-section area. Because the local stress is higher, the local strain must increase to keep pace with the true material behavior (the true stress vs true strain

184

Nanomaterials

curve). However, since plastic deformation in metals occurs at constant volume, an increase in the local strain (which implies an increase in length at the imperfection) results in a decrease in the local cross-section area and therefore now a further increase in the local stress. This process is unstable, and leads to the development of a local region of much smaller cross-section area called a neck. Eventually the neck becomes so small that the specimen separates (breaks) into two pieces, and that results in a complete loss of load-carrying capacity. In a tension test, the load carried by the specimen has a maximum at the point corresponding to the ultimate tensile strength. Why is there a maximum in the load? This is at the heart of the question of the stability of plastic deformations in tension. Consider the derivative of the relationship P = σ A between the load P, the true stress σ along the direction of tension, and the local cross-section area A: dP = Ad σ + σ dA.

(6.1)

The load has a maximum when dP = 0, at which point dσ dA =− . σ A

(6.2)

However, the assumption that plastic deformation occurs at constant volume implies that Al = constant =⇒ Adl + ldA = 0 =⇒

dA dl = − = −d ε , A l

(6.3)

so that a maximum in the force that can be carried by the specimen will occur when dσ = σ. dε

(6.4)

The maximum strength of the material will therefore occur when Equation (6.4) is satisfied, implying that the maximum strength that can be achieved depends on the material behavior, and we discuss this next.

6.2.1 Effect of Strain Hardening The constitutive law for the material, that is the stress-strain curve for the material, provides a relationship between the axial stress and the axial strain: σ = f (ε ), so that dσ ′ d ε = f (ε ) for the material. Thus a maximum in the load carried by the specimen will occur when f ′ (ε ) = σ . (6.5) In the particular case of a power-law relationship between stress and strain in the uniaxial tensile test,

6 Mechanical Failure Processes in Nanomaterials

σ = f (ε ) = σ0

ε n = αε n , ε0

185

(6.6)

which is sometimes called the Hollomon equation. We now have f ′ (ε ) = εn σ , so that a maximum in the load will occur when (using Equation 6.5)

ε = n.

(6.7)

This equation is called the Considere criterion and essentially states that a maximum in the load-carrying capacity of the specimen will be attained when the local strain on the specimen reaches a value equal to the strain hardening index. That is, a specimen made of a material with a low strain hardening index will reach a maximum in the load-carrying capacity at a small strain (e.g., if the strain hardening index for a steel is 0.1, the maximum load will occur when the true strain in the specimen is 0.1). There is a strong temptation, therefore, to conclude that low hardening materials will have small fracture strains, since the specimen usually fails soon after a maximum load is reached. As a consequence, much of the effort to develop nanocrystalline materials with improved fracture strain has focused on the development of some strain hardening within the material together with the high strengths produced by the reduced grain size. However, note that the Considere criterion only addresses the development of the ultimate tensile strength, not the fracture strain itself: what happens after the ultimate tensile strength is reached? Experiments show that necking in specimens typically begins just before the development of the peak load, that is, just before the development of the ultimate tensile strength. Many authors assume that necking begins at the ultimate tensile strength, but this is generally not so. The precise point at which necking begins depends on the uniformity of the specimen, and on the possible presence of defects within the material (that is, on geometric defects and material defects). Once necking begins, the deformation is said to have localized: most of the plastic deformation now begins to occur only in the next region, while the rest of the specimen unloads elastically. As a consequence, the stress-strain curve is no longer meaningful as a constitutive behavior for the material once necking begins, since the stress-strain curve assumes that the stresses and strains are uniform in the gauge section. A correction to the stress-strain curve is still feasible after the necking begins using the Bridgman analysis developed by Bridgman (1952), but this correction is very rarely provided in the literature. The necked region is inherently unstable for the reasons discussed earlier, and so the question of when the final failure will occur is a question of the stability of the necking deformations. An important general question when discussing the stability of deformations is the nature of the loading: is the tensile test being performed at controlled loading rate or controlled displacement rate? This question is intimately coupled to the compliance of the testing machine, but we shall not discuss this complexity here. For our purposes, it is sufficient to note that a test performed at constant loading rate would by definition not be able to maintain a constant loading rate once a maximum load is reached. Assume, therefore, that the simple tension test being discussed here involves a controlled displacement rate (that is, the ends of

186

Nanomaterials

the tension specimen are separated from each other at a controlled velocity). Under this condition, once necking begins, the rate-sensitivity of the material becomes an important factor in the stability of the deformations.

6.2.2 Effect of Rate-Sensitivity To see this in a simplified manner, consider the definition of the strain rate given in Chapter 2 through Equation (2.39), where we see that the strain rate is essentially a velocity gradient. Equation (2.120) shows that the total strain rate is made up of elastic and plastic parts, but once yielding has begun the elastic strain rate is usually relatively small, so that the total strain rate is almost all due to plastic deformation. Suppose the two ends of the specimen are being separated at a constant velocity difference δ V . Then the average strain rate in the specimen is given by ε˙uniform = δV l0 during uniform plastic deformations, where l0 is as before the gauge length. However, once necking begins, all of the plastic deformation concentrates within the region of the neck, which is much smaller than the original gauge length. Let us suppose the necking begins over a region of length h is shown in Figure 6.2; since, once necking begins, the rest of the specimen is only elastically deformed, all of the action is occurring within the necked region and the local strain rate is approximately given by δV . (6.8) ε˙neck = h Since h is smaller than the gauge length, the strain rate in the neck is larger than the average strain rate before necking began: ε˙neck > ε˙uniform . Now the material response within the neck depends on the rate-sensitivity, in that for a rate-sensitive material the stress within the necked region must be increased to account for the increase in the strain rate. Thus maintaining the deformation in the neck now requires a higher load. This effect competes against the effect of the reduction in area, which requires a lower load to develop a given stress. Thus the rate-sensitivity of the material acts to stabilize the deformations, and reduces the tendency of the specimen to further necking. The influence of rate sensitivity is actually pervasive, affecting more than the post-necking behavior (Hutchinson and Neale, 1977). The Considere analysis presented previously is only appropriate for a rate-independent material. In the case of the rate-dependent material, the onset of necking can be much delayed from that predicted by the Considere criterion, and indeed the total elongation of a material can be a strong function of the rate-sensitivity (see Figure 6.3). A more rigorous approach to this problem is to consider a stability analysis of the equations describing plastic deformation in tension. Conceptually, this amounts to asking the question whether perturbations in some variable that is part of the equations will grow or shrink with the deformation. The perturbation is prescribed with some amplitude aN and some wavelength λ , e.g., a = aN cos( 2λπ x ). If the perturbation grows, the system is unstable while if it shrinks the system is stable; growth may occur for

6 Mechanical Failure Processes in Nanomaterials

187

STRAIN RATE SENSITIVITY, m

1

0.1

0.01

Trend line 0.001

1

10

Fe - 1.3% Cr - 1.2% Mo Fe - 1.2% Cr - 1.2% Mo - 0.2% v Ni Mg - 0.5% Zr Pu Pb - Sn Ti - 5%AI - 2.5% Sn Ti - 6%AI - 4%V ZIRCALOY-4 100 TOTAL ELONGATION %

1000

10,000

Fig. 6.3 The influence of rate-sensitivity on the total elongation to failure of materials, from the work of Woodford (1969). Note the strong effect of the rate-dependence, and the similarity in behavior of a wide variety of materials.

some wavelengths and not for others (e.g., a surface scratch may not cause necking, while a wide machining groove might). In the uniaxial tension problem, it would be appropriate to consider perturbations in the geometry (essentially a defect in the geometry such as machining grooves) or perturbations in the material behavior (a defect within the material), and to ask whether these perturbations would result in the growth of necks. There are two versions of such stability analyses, known as linearized stability analysis and nonlinear analysis. Such stability analyses become particularly important when both strain hardening and rate-sensitivity are considered in the material behavior: σ = f (ε , ε˙ ). The Considere criterion discussed above corresponds to the long wavelength instability in strain hardening materials. Hart (Lin et al., 1981) and others have identified the onset of stability in rate-sensitive materials using linear stability analysis, resulting in criteria such as

γ = 1 − m, (6.9)  for onset of instability, where γ = σ10 ∂∂ εf 0 is a work-hardening coefficient and m = ε˙0 ∂ f σ0 ∂ ε˙ |0

is the rate-sensitivity (the subscript 0 indicates the uniform behavior far from a neck, i.e., the behavior without a neck). This is called the Hart criterion for instability. Hutchinson and Neale (1977) conducted a nonlinear analysis of the influence of rate-sensitivity on necking and showed that the linearized stability analyses of the

188

Nanomaterials

type that provide the Hart criterion miss important features of the necking behavior in rate-dependent materials. However, the key behavior remains that rate-sensitivity stabilizes the deformation. For this reason, some researchers attempt to increase the rate-sensitivity of nanomaterials. Note that in the case of fcc materials, going to smaller grain sizes appears to increase the rate-sensitivity, while the reverse is true for bcc materials. This suggests that the plastic deformations of fcc nanocrystalline materials will typically be more stable than those of bcc nanocrystalline materials.

6.2.3 Multiaxial Stresses and Microscale Processes Within the Neck The stress state within the necked region can no longer be simple uniaxial tension, because the axial strains within the necked region result in lateral contractions that are resisted by the material surrounding the neck; this in effect adds a transverse stress to the system. The resulting stress state is said to be triaxial in that there are stresses in three directions (one along the axes and two perpendicular to the axis of loading). A common measure of the degree of triaxiality of the stress state is called the triaxiality χ , and is defined as the ratio of the mean stress σm to the effective stress σ¯ : σm χ= , (6.10) σ¯ where the mean stress relates to the hydrostatic pressure, and the effective stress relates to a shearing stress (see Chapter 2, Equations 2.53 and 2.115). The triaxiality within the necked region increases as necking increases, as shown by full numerical simulations of the developing neck, through the development of an increasing tensile hydrostatic stress. A consequence of this hydrostatic tension may be the development of small voids within the material, through nucleation, growth and subsequent coalescence (which leads to final failure of the specimen). In the case of brittle materials, the fracture surface in a simple tension experiment is developed after very little necking, is oriented essentially perpendicular to the tensile axis (Figure 6.4b), and shows a faceted character. The fracture surface developed after simple tension of a ductile material involves a substantial neck, and typically, a cup and cone fracture, with one side shaped like a cup and the other side shaped like a cone (Figure 6.4c); both surfaces show a dimpled character, demonstrating the growth of voids. Microvoid development is an example of a microscale failure process that is not externally visible until final failure, but the signatures of the voids (appearing as dimples) are quite clear on the fracture surface after failure has occurred. The consequences of the microvoid development are twofold. First, the actual failure of the specimen is determined by the microvoid nucleation, growth and coalescence process, and so this is what determines the fracture strain (as distinct from the onset of necking, which does not depend on the voiding process). Second, once the voids have nucleated, the effective constitutive behavior of the material (that is the average

6 Mechanical Failure Processes in Nanomaterials

189

Fig. 6.4 Form of the fracture developed in a simple tension test. (a) Original specimen configuration before loading. (b) Tensile failure of a brittle material, showing that the fracture surface is essentially perpendicular to the loading axis. (c) Tensile failure of a ductile material, showing the cup and cone failure morphology (in section).

material behavior that would be observed) changes dramatically from the classical plasticity description discussed previously: the deformation is no longer volume preserving. Thus it is not a priori reasonable to use classical plasticity descriptors to examine the development of mature necks in materials.

6.2.4 Summary: Failure in the Simple Tension Test Our considerations of failure in the simple tension test allow us to identify several key concepts that will be useful throughout the rest of this chapter: 1. Failure in simple tension is an unstable process that involves both the material response and the structural response of the tensile specimen and the testing machine. 2. The instability is observed as the rapid growth of a neck within the tensile specimen. The necking begins just before the development of the ultimate tensile strength. 3. The onset of the instability (and associated necking) depends on material properties such as the strain hardening and rate sensitivity, both of which must be defined in terms of the true stress and true strain.

190

Nanomaterials

4. High strain hardening and high rate sensitivity both delay the development of the instability. Criteria such as the Considere and Hart criteria (Equations 6.7 and 6.9 respectively) may underestimate the strain developed before the onset of necking. 5. The fracture strain is larger than the strain associated with the ultimate tensile strength, but how much larger depends on many factors including microscopic failure processes within the material. 6. The stress state within the neck is no longer uniaxial but rather triaxial, and void nucleation and growth may dominate the evolution of the neck. 7. Final failure, which defines the fracture strain, is determined by the interaction of microvoid processes within the neck. 8. The fracture strain is not itself a material property – it is a property determined by the combination of material behavior and test conditions such as geometry. Comparisons of fracture strains made with different specimen geometries should be viewed with great caution.

6.3 The Ductility of Nanomaterials The effective ductility of a material is loosely defined as the tensile strain that can be carried by a material before complete loss of load-carrying capacity occurs in a simple tension test. Initial work on the mechanical behavior of nanomaterials focused primarily on the effect of increasing strength that could be obtained by changing the grain size of bulk nanomaterials. Once it was apparent that high strains could be achieved, it became important to identify the conditions under which these nanomaterials could be used in applications, and in particular to define when mechanical failure would occur. The first bulk nanomaterials that were developed had significant strengths, but failed immediately after the yield strength was reached in tension. A significant amount of effort has been directed in the last decade towards developing nanostructured and nanocrystalline metals that can withstand significant tensile strains before failure. It is generally true, even in conventional materials, that an increase in strength is usually associated with a reduction in ductility. Thus very high strength steels are usually much less ductile than mild steel. This reduction in ductility has two proximate causes. First, high strengths in conventional materials are typically obtained at the expense of strain hardening and rate-sensitivity, i.e., most high-strength materials have low strain hardening and low rate sensitivity. Second, given a standard specimen geometry, the high-strength material in the tension test is more sensitive to geometric and material defects because the growth rate of microscale failure processes is often related to the applied stress. Both of these issues also result in reduced ductility in many nanostructured materials as the strength is increased. As a specific example, consider the total elongation to failure of copper specimens made of copper materials of decreasing grain size (corresponding to increasing yield strength) as shown in Figure 6.5. Similar behaviors have been observed in a broad range

6 Mechanical Failure Processes in Nanomaterials

191

E

Fig. 6.5 The total elongation to failure of copper materials of varying grain sizes, as presented by Ma et al. (Wang et al., 2002). It is apparent that the ductility decreases dramatically as the yield strength is increased. The point labeled E in the figure corresponds to a special copper material produced by Ma and coworkers that included both nanocrystalline and microcrystalline grain sizes, with the larger grains providing an effective strain hardening in the material. Reprinted by permission from Macmillan Publishers Ltd: Nature, Vol. 419, Issue 6910, pages 912–915, High tensile ductility in a nanostructured metal, Yinmin Wang, Mingwei Chen, Fenghua Zhou, En Ma. 2002.

of nanocrystalline FCC metals, as shown in Figure 6.6. The elongation to failure (which is essentially the ductility) of the copper materials drops drastically with increasing yield strength, which corresponds to decreasing grain size. Ma et al. (Wang et al., 2002) showed that it was possible to obtain increased ductility together with high strength in a copper material by providing a bimodal grain-size distribution of both nanocrystalline and microcrystalline sizes, with the larger grains allowing the development of some effective strain hardening in the overall material. Thus the importance of strain-hardening in improving ductility of nanostructured metals is clearly established. Unfortunately, our understanding of strain hardening mechanisms in truly nanocrystalline materials remains weak, and so improving the overall strain hardening of the fully nanocrystalline structure remains difficult. Special cases seem to abound, such as apparent hardening due to stress assisted grain growth and hardening through the incorporation of deformation twinning (discussed in a later chapter). The influence of internal defects on the apparent ductility of nanostructured and nanocrystalline materials is clearly evident if one considers the scatter of the data that is observed in measurements of the ductility as compared to that observed in data on the yield strength of materials. A much larger scatter is observed in the ductility, and correlates quite well with the variety of fabrication procedures used for making both the nanomaterials and the specimens used for the testing. The larger scatter in ductility as compared to strength is a consequence of the fact that the

192

Nanomaterials 80

nc Cu [31. Sanders PG. Acta, 1997;45:4019] nc Cu [62. Legros M. et al, Phil. Mag. A 2000;A80:1017] nc Cu [63. Nieman GW. et al, JMR 1991;6:1012] nc Cu [64. Wu XJ, Nanostrust Mater., 1999;12:221] nc Cu [65. Cheng S et al, Acta. 2005;53:1521] nc Cu [66. Wang YM et al, APL, 2003;83:3185] nc Cu [67. Conrad H. Yang D., Acta Mater. 2002;50:2851] nc Cu [41. Lu et al, Science, 2004] nc Cu [68. Youssef KM, et al, APL, 2005;87:091904] nc Ni [69. Schwaiger R. et al, Acta mater., 2003;51:5159] nc Ni [70. Torre FD. et al, Acta Mater., 2002;50:3957] nc Co [71. Karimpoor AA, et al, Scripta mater, 2003;49:651] nc Pd [72. Nieman GW et al, JMR, 1991;6:1012]

CG Cu [65] CG Ni [70] CG Pd [72] CG Co [71]

70

Elongation (%)

60 50 40 30 20 10 0

0

200

400

600

800

1000 1200 1400 1600 1800

Yield strength (MPa) Fig. 6.6 The total elongation to failure of fcc materials of nanocrystalline and microcrystalline grain sizes, as presented by Dao et al. (2007). Reprinted from Acta Materialia, Vol. 55, Issue 12, page 25, M. Dao, L. Lu, R.J. Asaro, J.T.M. DeHosson, E. Ma, Toward a quantitative understanding of mechanical behavior of nanocrystalline metals. July 2007, with permission from Elsevier.

ductility is sensitive to defects in the materials, while the strength is not as sensitive to as wide a range of defects. A fairly simple example of such a defect influence can be identified in the porosity. A few large pores may have a small effect on the scatter of the yield strength of a material, particularly if the yield strength is measured using hardness testing. However, those few large pores can have a dramatic impact on the apparent ductility of the material in tension. The overall ductility of nanocrystalline materials has improved as the procedures for fabricating these materials have become better defined and more capable of generating high-quality and consistent material. The influence of defects on the ductility is nicely illustrated by the work of Youssef et al. (2005), who were able to make a relatively defect-free nanocrystalline (mean grain size of 23 nm) copper material by using special fabrication procedures. The use of the special procedures resulted in the development of a material that had a total elongation to failure on the order of 15%, which is substantially higher than had been observed in copper of similar nanocrystalline grain-sizes in the past. It is important to recognize that some of the apparent lack of ductility of nanomaterials is a consequence of how they are tested. Most nanostructured materials are typically subjected to tension in experimental configurations that deviate very far from the ASTM standards even for high-strength materials, because it is difficult to get sufficiently large quantities of these materials to perform standard tension tests. The much smaller tensile specimens that are being evaluated are typically much more sensitive to geometric defects, and therefore generally produce lower elongations to failure. In addition, some of the data in the literature are almost certainly

6 Mechanical Failure Processes in Nanomaterials

193

compromised by experimental artifacts associated with testing machine compliance and inaccurate measures of strain. Further, new mechanisms may become possible in some specimen geometries (such as grain growth in thin films) that are not possible in others, and that therefore color the results. Improvements in processing, specimen fabrication and experimental design are all required to develop a clearer sense of the ductility of nanocrystalline and nanostructured materials. That said, recent improvements in the apparent tensile failure strains shown by new nanomaterials are very promising. A number of specific approaches have been suggested to improve the ductility of nanomaterials, while retaining their high strengths (Ma, 2006). Most of these involve techniques for improving the strain hardening or rate sensitivity of the material (for example, through the use of bimodal grain-size distributions or through the addition of alloying phases), while some address the issue of the defect sensitivity by reducing the number of defects generated by the processing. Other approaches include the incorporation of special boundaries within the material through twinning. It is likely that each nanomaterials system will require a different cocktail of such mechanisms in order to achieve the desired ductility. What is the acceptable ductility within a nanomaterial system? Much of the literature assumes that a total elongation to failure of ≥10% will be acceptable, but the precise number of course depends on the specific application. Product designers learn to live with the limitations of the materials that are available to them, but the market for bulk structural nanomaterials will improve substantially if ductilities comparable to that of low carbon steel (ε f = 0.25) can be obtained.

6.4 Failure Processes There are several failure processes within materials, each of which tends to occur under a specific stress state, and some of which are more likely to occur in certain kinds of materials. This section describes the microscopic failure processes that lead to macroscopic failure: 1. Crack nucleation, propagation and coalescence. These processes are what most people think about in terms of mechanical failure. 2. Void nucleation, growth and coalescence, leading to what is typically called ductile failure. 3. Shear band initiation and evolution. These are processes of localization of shearing deformations. 4. Relatively specialized processes such as crazing and kink band formation in polymers and composites respectively. We shall not discuss these in detail. These failure processes may be coupled: for example, voids are often formed within shear bands in materials, and cracks may follow along shear band planes. In the next few sections, each of these failure processes is examined in terms of the failure of nanomaterials, with an emphasis on identification of phenomena rather than on the modeling of the failure process.

194

Nanomaterials

There are three fundamental steps in all of these failure processes. First, the microscopic failure process (for example a micro-crack) must be nucleated. Second, the microscopic failure must propagate or grow. Finally, multiple instances of the failure must coalesce and cause a macroscopic failure (or a microscopic failure must grow to become a macroscopic failure).

6.4.1 Nucleation of Failure Processes The nucleation process usually depends on the achievement of a critical stress, such as at a stress concentration. Since defects are major stress concentrators, microscopic failure nucleation typically occurs at a defect (whether a geometric defect – like a surface scratch – or a material defect, like an inclusion). In some steels, for example, microvoid nucleation typically occurs at manganese sulfide inclusions, while the initiating defect is a carbonaceous inclusion in many hot-pressed ceramics. A greater density of such defects (inclusions) will lead to a larger number of nucleation sites, and therefore more nucleation of the failure at a give stress state. A classical nucleation site is a pre-existing pore. Given this dependence on nucleation site density, we now understand why processing-induced defects play such a big role in the apparent ductility of nanostructured materials – they provide an array of nucleation sites. Careful processing that avoids the development of initial porosity and minimizes the number of inclusions therefore improves the tensile behavior of nanomaterials, delaying macroscopic failure, as demonstrated by Youssef et al. (2005). Note that the typical size of defect that is involved in failure process nucleation is of the order of the grain size or larger in conventional grain-sized materials, because most of these failure processes do not develop easily within single crystals. Failure processes in single crystals can be initiated by smaller stress concentrators such as dislocation jogs, but these require much higher stresses before cleavage facets or significant microvoid nucleation can occur. What are the critical stresses associated with nucleation? This is not a question with a clear-cut answer, because the critical stress and the kind of stress will vary both with type of failure process (cracks, voids or shear bands) and with the type of nucleating defect (e.g., initial porosity vs hard inclusion). In a generic sense, however, cracks nucleate when critical tensile stresses are attained, and voids nucleate when critical hydrostatic tensile stresses are reached. The critical conditions for shear band initiation are not understood. Since nanocrystalline and nanostructured materials are generally very strong, the stresses that they carry are relatively high, and so there is often enough stress available to nucleate failures (this is a general argument that can be used for high-strength materials). A comparison between the nucleation of failure processes for nanomaterials and conventional materials may be useful. In conventional materials, the nucleation sites are typically defects such as inclusions and pores. Depending on the process used to make the nanomaterial, these defects may be more common in nanomaterials,

6 Mechanical Failure Processes in Nanomaterials

195

or may be more widely distributed (it is not clear, for example, whether the inclusions will also change in size scale when the overall material is nanostructured). On the other hand, it is typically the case that (in nano metals) homogeneous plastic deformation has become more difficult, and so the overall stresses associated with deformation are higher, and as a consequence there is more stress available to nucleate the failure process. Thus, both tendencies work towards making the nucleation of failure processes more common in nanomaterials.

6.4.2 The Growth of Failures Whereas the nucleation of failure processes is typically stress controlled, the growth of failure processes is controlled by the available energy. This is most easily seen in the case of cracks. Propagating cracks require the generation of new surface area, and the generation of new surface area requires additional energy in the form of surface energy. Thus crack propagation requires the delivery of energy to the crack tip (the amount of energy that must be delivered is related to the fracture toughness of the material). If the energy is not available, the crack will not grow. If the energy is not provided fast enough, the crack will be forced to grow slowly (this controls the speed of crack growth, and is an inherent problem in dynamic fracture mechanics). The energy for crack growth (and similarly for other failure processes such as void growth) typically comes from strain energy locally stored in the structure that is undergoing the failure process. The magnitude of the energy that is required for the growth of failure processes arises depends on the processes involved. For example, in the case of cracks, the energy is required for developing new surface area; in the case of voids in a metallic material, the energy is required to develop the plastic work that must occur at the surface of the voids as the voids grow; in the case of shear bands, the energy is required to develop plastic work within the shear bands. The growth of failures can therefore be controlled by limiting the available energy. Conversely, if the available energy is very large, the growth of failures is nearly impossible to control. An excellent example of this is observed in the testing of ceramics. Once microcracks have nucleated in ceramics within a tension or compression experiment, it is typically very difficult to control the growth of the cracks because so much strain energy is stored in the testing machine and is available for crack growth, while the energy needed for crack growth is relatively small in ceramics. Thus it is very difficult to recover ceramic specimens that contain only small amounts of microcracking, although metallic specimens containing small amounts of microcracking can often be recovered after a mechanical test. The availability of energy is also related to the question of whether the application involves load control or displacement control, or equivalently whether the application has fixed available energy or fixed available load. If the application involves fixed available energy, the growth of the failure processes will be limited to that possible with the available energy.

196

Nanomaterials

6.4.3 The Coalescence of Cracks and Voids Final macroscopic failure is the result of the coalescence of microscopic failure processes. The process of coalescence is extremely difficult to model, since it is essentially a runaway instability. Consider, for example, two voids that are growing within a material after nucleating at nucleation sites that are a distance l apart. In the early stages of void growth, both the stress fields and the available energy around each void is essentially unaffected by the presence of the other void. Thus each void grows as though it is alone within an infinite medium, and this has been modeled quite carefully in the mechanics literature (Wu et al., 2003). A simple model of a void growing in a uniform external hydrostatic tensile stress field shows that most of the deformation occurs right near the void surface in a region that is heavily deformed plastically, and as one moves radially away from the void surface the deviatoric stresses decrease until plastic deformation can no longer occur, so that there is an effective plastic zone around each void. Once the voids get to be sufficiently large, however, they begin to interact with each other through their stress fields (a quick but relatively inaccurate measure of when interactions begin is to consider the time when the two plastic zones of the voids begin to overlap). The result of the interaction is typically to accelerate the growth of the voids, thus leading to an instability. As the voids continue to grow, the space between the voids eventually shrinks to a mere ligament. From here on, the interaction between the voids is occurring through the plastic extension of the ligament, and finally the ligament breaks, resulting in the coalescence of the two voids. Such a process repeated across the entire specimen results in the macroscopic ductile failure of the specimen, and the holes that remain within the specimen after the ligaments are broken give the fracture surface a dimpled appearance. Although the coalescence of failure processes is difficult to model, it is also true that the runaway instability associated with coalescence usually results in macroscopic failure that occurs very soon after coalescence begins (that is, there is very little global deformation – or strain – accommodated once coalescence begins). For practical purposes, therefore, it may be sufficient to consider the onset of coalescence and use that to define the onset of microscopic failure.

6.4.4 Implications of Failure Processes in Nanomaterials The preceding discussion has demonstrated that in order to control failure in nanomaterials one must control the nucleation, growth, and coalescence of the failure processes that develop within each material. Controlling nucleation is largely a question of controlling the available defects that are the nucleation sites, that is, one requires excellent control of the processing so as to reduce the available defects. It is useful to recognize here, however, that some microstructures inherently contain the defects that act as nucleation sites for failure, and that these “defects” may in fact be important in producing the strength or the strain hardening that is required

6 Mechanical Failure Processes in Nanomaterials

197

for the application. For example, the incorporation of a weak phase within the microstructure may result in enhanced strain hardening, but the weak phase may also result in the easier nucleation of voids within the material. More broadly, this is a recognition of the fact that the microstructural features that control plastic deformation may have significant implications on failure processes, so that a trade-off must sometimes be considered. Once the failure has nucleated, the growth of the failure is typically not controllable in microstructural terms, although it may be controllable from an applications standpoint (for example by controlling the applied loading). Inertial effects ensure that the growth of failures is typically rate dependent (Wright and Ramesh, 2008). In a broad sense, however, the relative roles of nucleation and growth dominate the failure process, and so the control of the failure process can be viewed in terms of controlling the relative rates of nucleation and rates of growth. For example, if there are relatively few nucleation sites, the failure process may begin at the same stress as in the material with many nucleation sites, but the failure process will remain stable for a longer time because the interaction of the failures (and eventual coalescence) does not occur until later in the material with few nucleation sites.

6.5 The Fracture of Nanomaterials As is generally the case in this field, the fracture of nanomaterials has primarily been studied in FCC materials. Recall that failure processes in materials are highly localized, and are generally sensitive to the presence of geometric and material defects. In order to be able to compare the resistance to fracture of one material with the resistance to fracture of another, one must perform experiments in which the localization occurs in a specific place through the use of a controlled defect (either a notch or a crack). From a mechanics perspective, fracture should be studied through the use of specialized fracture mechanics experiments which evaluate the likelihood of the propagation of a pre-existing crack within the material. The result of such an experiment is the evaluation of the fracture toughness KIC of the material (see Chapter 2). This involves two distinct steps. First, a fracture mechanics experiment needs to be designed so that the appropriate quantities can be measured. Second, a pre-crack must be generated in the specimen. Standardized experimental techniques exist for the measurement of the fracture toughness of materials (for example, a standardized fracture mechanics experiment for plane strain fracture toughness measurement uses ASTM E399), but these generally require fairly large specimens (e.g., Figure 6.7), and such specimens typically cannot be made with current quantities of most nanomaterials. Few if any standard fracture mechanics experiments have been conducted on nanomaterials, and thus there are no fracture toughness data on nanometals or nanoceramics, although there are some data on polymer nanocomposites. Such data will require the development of larger quantities of bulk nanomaterials, or the development of specialized experimental techniques for small-scale fracture.

198

Nanomaterials

Fig. 6.7 A standard compact tension (CT) specimen used for fracture toughness measurements. The specimens are several centimeters in size.

Recall, from Chapter √ 2, that materials with low values of fracture toughness (typically KIC ≤ 10 MPa m) are considered √ to be brittle, but materials with high fracture toughnesses (typically KIC ≥ 20 MPa m) are said to be tough (sometimes, they are said to be ductile). Tougher materials have a wider range of application, since they are less sensitive to minor cracks. It turns out that there is a very strong correlation between the fracture toughness of a material and the character of the fracture surface after a fracture mechanics experiment. Brittle materials show faceted fracture surfaces, with the facets representing typically crystalline cleavage (fracture along crystallographic planes) or fracture along the grain boundaries, while ductile materials show dimpled fracture surfaces, with the dimples being the signatures of void growth and coalescence leading to separation of the two fracture surfaces. It appears to take less energy to grow the faceted fracture surfaces as compared to the dimpled fracture surfaces, so that the fracture toughness of brittle materials is lower than that of ductile materials. There is also a strong correlation between the brittleness of the material and the form of the fracture surface developed in a simple tension experiment. From a materials perspective, therefore, fracture processes are typically studied through the examination of the fracture surfaces developed after a standard tensile test. What kinds of fracture surfaces are typically observed in nanocrystalline materials? Nanocrystalline ceramics behave in a brittle fashion in tension, and their fracture surfaces are faceted, as would be expected. Nanocrystalline metals, on the other hand, typically show dimpled fracture surfaces. Thus the local failure processes in

6 Mechanical Failure Processes in Nanomaterials

199

nanometals are of the ductile variety, indicating that the macroscopic low ductility observed in many nanometals can perhaps be mitigated, and that it may be possible to make nanometals reasonably tough. An interesting example of the development of failures within a nanomaterial is provided by the in situ observations of Kumar et al. (2003). These authors examined the propagation of tensile failures in nanocrystalline nickel within a transmission electron microscope, so that they could observe the microscopic process of failure. A sequence of images from their in situ TEM work is presented in Figure 6.8. They

Fig. 6.8 A sequence of in situ TEM micrographs obtained by Kumar et al. (2003) during the loading of nanocrystalline nickel using multiple displacement pulses. Images a–d show the microstructural evolution and progression of damage with an increase in the applied displacement pulses. The presence of grain boundary cracks and triple-junction voids (indicated by white arrows in (a)), their growth, and dislocation emission from crack tip B in (b–d) in an attempt to relax the stress at the crack tip as a consequence of the applied displacement, can all be seen. The magnified inset in (d) highlights the dislocation activity (Kumar et al., 2003). Reprinted from Acta Materialia, Vol. 51, Issue 2, page 19, K.S. Kumar, S. Suresh, M.F. Chisholm, J.A. Horton, P.Wang, Deformation of electrodeposited nanocrystalline nickel. Jan. 2002, with permission from Elsevier.

200

Nanomaterials

observe the development of grain boundary cracks, void nucleation at triple junctions, and the simultaneous movement of dislocations within the material. Thus at the microscopic scale, both fracture processes and plastic deformation processes are active within the nanocrystalline nickel. If, on the other hand, the fracture surface of the nanocrystalline nickel (typical grain size of 40 nm) studied by Kumar et al. (2003) is examined after loading, what is observed is the jagged fracture surface shown in Figure 6.9. Such a post mortem observation leads one to underestimate the importance of plastic deformation during the failure process simply because dislocations are not immediately evident. This work therefore points out the difficulty of inferring processes (and in particular assuming the lack of specific processes such as dislocation motion) from post mortem observations. Dimpled fracture surfaces are often observed in nanocrystalline metals after tensile failure (again, most of this work has been in FCC nanometals). Jonnalagadda and Chasiotis (2008) have examined the tensile behavior of thin films of nanocrystalline fcc metals (gold and platinum) using a microtension set up that is capable of a range of strain rates. An example of the failure surface morphology observed in nanocrystalline gold (with an average grain size of 38 nm) after the tensile failure of thin films is presented in Figure 6.10. Large (≥100 nm) dimples are seen on the fracture surface, with the dimple size many times larger than the average grain size. Significant plastic deformation is therefore occurring before macroscopic failure. It is not clear what the nucleation sites were for the voids that produced the dimples (note that these are essentially pure nanocrystalline metals produced by electrodeposition, and therefore would appear to contain few processing defects).

Fig. 6.9 Postmortem microscale fracture morphology observed by Kumar et al. (2003) after the loading of nanocrystalline nickel in tension. There is no direct evidence of the presence of dislocations in this image, although some of the grains appear to have necked before separation. Reprinted from Acta Materialia, Vol. 51, Issue 2, page 19, K.S. Kumar, S. Suresh, M.F. Chisholm, J.A. Horton, P. Wang, Deformation of electrodeposited nanocrystalline nickel. Jan. 2002, with permission from Elsevier.

6 Mechanical Failure Processes in Nanomaterials

201

Fig. 6.10 Postmortem fracture surface morphology of nanocrystalline (average grain size of 38 nm) gold thin film tested in tension till failure by Jonnalagadda and Chasiotis (2008). Dimples are seen on the fracture surface, with an average size of more than 100 nm. The void size just before coalescence is therefore much larger than the grain-size, and the growth of the voids must involve plastic deformation in many many grains.

Dimpled fracture surfaces have been observed in a number of fcc nanometals, including copper, gold, platinum, and nickel. There are few if any observations of fracture surfaces in bcc and hcp metals.

6.6 Shear Bands in Nanomaterials Shear banding is a plastic shear flow instability in which initially uniform shearing deformations localize into a narrow region called the shear band. This is a common mode of failure in a variety of materials, particularly under shear-dominated and compressive loadings (in tension-dominated loadings, the crack and void growth processes are typically favored from an energetic viewpoint). The shear localization process is illustrated in Figure 6.11. An initially uniform shearing deformation (in the plastic range) can begin to become nonuniform, with most of the plastic deformation occurring over a small region that becomes increasingly narrow, until all of the plastic deformation occurs within this narrow region (called the shear band) with only elastic deformations occurring outside the shear band.

202

Nanomaterials

Fig. 6.11 Progressive localization of a block (a), with initially uniform shearing deformations (b) developing into a shear band (c). The final band thickness depends on the material behavior.

Stress

There are many reasons for the localization of the plastic deformation into a band, but in a broad sense such localizations occur when the material behavior locally reaches a softening domain where increasing deformations can be accommodated at decreasing stresses as in Figure 6.12. Consider, for example, a slab of material (with a constitutive behavior corresponding to Figure 6.12) that is deformed in simple shear as in Figure 6.11, with slowly increasing shearing stresses prescribed on the boundaries. Suppose that some location in the slab has a local stress concentration (e.g., due to the presence of an inclusion). Then the material at that point will reach the peak stress in Figure 6.12 before the material elsewhere in the slab. At that material point, therefore, runaway deformation occurs, because increasing deformations can be sustained at decreasing stress. This then causes neighboring points to reach the peak stress more quickly, and a cascade process begins leading to a localized zone of deformation as in Figure 6.11. Note that the local softening may not be immediately observable at the macroscopic scale, because much of the slab is still in the hardening regime. Once the shear band has run all the way through the slab, macroscopic softening may also be observed. At intermediate states, the

Strain

Fig. 6.12 Stress-strain curve for a material showing softening after a peak stress. This is a curve corresponding to a material undergoing thermal softening, but similar behaviors can arise from other causes such as grain reorientation.

6 Mechanical Failure Processes in Nanomaterials

203

macroscopic behavior may be that of decreasing hardening, or even perfect plasticity (when the hardening and softening regions balance each other).

6.6.1 Types of Shear Bands Two kinds of softening dominate materials problems: softening resulting from local reorientation and texture development (this is often called geometric softening) and softening resulting from temperature rise (thermal softening). Shear bands arising from geometric softening are often simply called deformation instabilities in shear, while shear bands arising from thermal softening are said to be adiabatic shear bands. Adiabatic shear bands are typically observed in dynamic loading conditions, when the heat generated by plastic work does not have enough time to be conducted away (hence the term adiabatic). A very large number of materials will localize into adiabatic shear bands if driven at sufficiently high strain rates. The likelihood of the development of such adiabatic shear bands is determined by the combination of the plastic properties of the material and the thermal properties (thermal conductivity and specific heat). An excellent summary of the mechanics issues associated with such adiabatic shear bands is provided by Wright (2002), and a description with a materials emphasis is provided by Bai and Dodd (1992). Because nanostructures change the plastic properties of materials, the onset of adiabatic shear band development will also be influenced by grain size, but this has not been carefully studied. In this book the discussion is focused on shear bands generated in nanocrystalline materials as result of geometric softening. Localized shearing deformations (shear bands) have been reported in many nanocrystalline metals. Localization of plastic deformation into shear bands has been observed in nc- Fe-10%Cu (Carsley et al., 1998) and nc-Fe (Jia et al., 2003). Malow et al. (1998) observed shear bands in nc-Fe samples after microhardness tests. Shear bands have also been found in nc-Pd (Sanders et al., 1997) under compression, in nc-Cu under fatigue (Vinogradov and Hashimoto, 2001) and in nc-Ni under high-rate tension (Dalla Torre et al., 2002). We discuss shear band development in nanocrystalline metals through the particular example of shear band development in bcc iron, discussed in a series of papers by the author’s group (particularly Jia, Wei and others – Jia et al., 2003, 2000; Wei et al., 2004b).

6.6.2 Shear Bands in Nanocrystalline bcc Metals The observations of Jia et al. (2003) and of others (e.g. Carsley et al., 1998), indicate that there is a transition from uniform to non-uniform deformation as the grain size decreases down to the nanoscale. In the case of iron (Jia et al., 2003), when the grain size is above ≈ 1 µm, the plastic deformation takes place uniformly at least up to a

204

Nanomaterials

plastic strain of about 15% over the entire sample. However, for smaller grain sizes (≤ 300 nm) shear band development was observed to occur immediately after the onset of plastic deformation, correlating to an observed change in apparent strain hardening at those grain sizes. Most of the plastic deformation was localized in the narrow shear bands, often along planes containing the highest shear stresses. Comparative optical micrographs are presented in Figure 6.13 after low-rate compressive deformations to similar plastic strains, showing (a) the uniform deformations in Fe with a mean grain size of 980 nm and (b) the strongly non-uniform deformations in Fe with a mean grain size of 268 nm. Shear band populations similar to Figure 6.13b were observed in all specimens with grain sizes. Figure 6.13c shows the strongly non-uniform deformations observed in the 268 nm Fe after high-rate loading to a similar strain as in Figure 6.13a, b. Thus the shear bands were observed during both quasistatic and high-rate deformations for these grain sizes, suggesting that the shear bands resulted from geometric rather than adiabatic (thermal) softening. The process of development of these shear bands was studied using “progressive” low-rate compression tests performed on the 268 nm-Fe with repeated loading/unloading/reloading at controlled nominal levels of strain. The evolution of the shear band population over an entire specimen as a function of strain is described by Wei et al. (2002) and is presented in Figure 6.14. These authors show that the onset of plasticity in the nanostructured iron corresponds to the appearance of shear bands emanating from one of the edges of the specimen, i.e., there is negligible uniform plastic deformation before localized deformations begin. It was demonstrated in Wei et al. (2002) that additional shear bands appear with increasing strain and

(a(a) )

(b) (b)

(c) (c)

Fig. 6.13 Gross deformation features of coarse-grain and ultra-fine-grain iron samples after compressive deformations to nearly identical strains (Jia et al., 2003). (a–b) represent quasistatic compression, while (c) represents high-strain-rate compression. (a) Homogeneous deformation of coarse-grain (20 µm grain size) Fe. (b) Shear band pattern development in ultra-fine-grain (270 nm grain size) Fe. (c) Shear band pattern development in ultra-fine-grain Fe after dynamic compression at a strain rate of ≈ 103 s−1 . Reprinted from Acta Materialia, Vol. 51, Issue 12, page 15, D. Jia, K.T. Ramesh, E. Ma, Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. July 2003, with permission from Elsevier.

6 Mechanical Failure Processes in Nanomaterials

205

III

III

A

II

II

iv I

I

(a)

(b)

Fig. 6.14 Shear band patterns evolve with strain in compressed ultra-fine-grain Fe (Wei et al., 2002). Note the propagation of existing shear bands, the nucleation of new shear bands, and the thickening of existing shear bands. The development of families of conjugate shear bands is also observed. Reprinted from Applied Physics Letters, Vol. 81, Issue 7, pages 1240–1242, Q. Wei, D. Jia, K.T. Ramesh, E. Ma, Evolution and microstructure of shear bands in nanostructured fe. 2002, with permission from American Institute of Physics.

that the newly generated shear bands have similar orientations (in the four possible shearing planes for these cuboidal specimens). Large numbers of shear bands are observed, rather than single dominant bands that lead to failure. The process of shear banding that develops during a quasistatic experiment can be observed in Figure 6.14, which shows a single area (well away from the edges) of a 268 nm Fe specimen at nominal plastic strains of (a) 3.7% and (b) 7.8%. The loading axis is vertical in these figures. The straight lines produced by the directed polishing described earlier are visible in both figures and may be used to deduce the amount of shear in each band as the nominal strain develops. Three shear bands denoted I, II and III are visible in Figure 6.14a, including one band tip denoted by A (note that band tips are atypical in our data in unconfined regions). By the time the nominal plastic strain has reached the level of 7.8% in Figure 6.14b, a minimum of eight shear bands are visible (i.e., five new bands have appeared). Bands I, II, and III are all considerably broader (Band I, for example, has grown in width by more than 50%). Band III has propagated, including several branches, one of which has apparently merged with Band II. One of the new bands (Band IV) is at least as broad as the current size of Bands II and III, indicating that the deformation within a newly nucleated band can be substantially larger than that in an existing band. Tracing the polishing lines, one observes that the total strain in Band I has increased. Observations with optical microscopy and TEM for the deformed UFG-/nc-Fe specimens show that large plastic deformations are developed within the bands (Wei et al., 2002). Thus, it appears that the process of shear banding in this material involves

206

Nanomaterials

the nucleation of new bands, propagation along the shear plane, increase in width (broadening) and increase in strain (flow) within the band. Figure 6.15a shows a high-magnification view of a single shear band with welldefined boundaries (this band was observed at a nominal plastic strain of 0.3%, just after yield). The shear deformation within the band can be traced by following a polishing scratch across the band. The measured width of the shear band is ≈ 16 µm and the shear strain in the band is 25% (nearly two orders of magnitudes higher than the nominal specimen strain).

~7°

(a)

(b)

30 µm

Fig. 6.15 (a) Shearing deformation across one band and (b) the stable intersection of multiple shear bands in 270-nm Fe after quasistatic compression. The shear offset is clearly visible across the first band. Note that no failure (in terms of void growth) is evident at the intersection of the shear bands. Reprinted from Acta Materialia, Vol. 51, Issue 12, page 15, D. Jia, K.T. Ramesh, E. Ma, Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. July 2003, with permission from Elsevier.

Although no shear bands are present along the conjugate planes in Figure 6.14, the bulk of the specimen develops shear bands along more than one of the conjugate planes. As a result, shear band intersections are common and organized networks of shear bands are observed. Figure 6.15b shows a network of shear bands at about 45◦ to the loading axis in a 268-nm Fe specimen after low-rate loading to 7.8% nominal plastic strain. The relative uniformity of the shear band spacing must be described by models of this failure mode (that is, a model that correctly captures the physics of multiple shear banding should predict this shear band spacing in compression). The intersections of shear bands are observed to remain compatible within optical resolution, which is itself remarkable given the large strains within each shear band. For most of the shear bands, the two boundaries can be clearly identified, so that the width can be accurately measured. The band widths (all measured using the optical microscope) for the 138-nm and 268-nm Fe are the thickest bands observed at low plastic strains ( zone axis. We therefore infer that the texturing involves the activation of the < 111 > /110 slip systems of bcc Fe. Although slip systems of < 111 > /110, < 111 > /112 and < 111 > /123 are all reported for bcc metals such as Ta, Mo, Cr, < 111 > /110 appears to be a much preferred slip system for bcc Fe (Nemat-Nasser et al., 1998). Since the material within the shear band shows this texture, it follows that the large plastic deformation observed is developed through conventional dislocation mechanisms. Thus even at this relatively small grain size, dislocation mechanisms dominate the material behavior. In summary, within the shear band we have elongated and textured grains, suggesting that the grains have reoriented themselves for easy slip. In other words, the grains have experienced rotations and alignment in forming the band with texture, allowing easy slip propagation in certain orientations. This is a geometric softening mechanism. It is conceivable that such cooperative deformation across the band is easier when it involves many tiny grains rather than large grains, possibly explaining the enhanced propensity for shear banding when the grain size is in the nc/UFG regime. An approach to modeling such a softening mechanism is presented in the next chapter. These observations of microstructure within the shear bands are interesting from another perspective. Various investigators (e.g., Meyers et al., 1994; Chichili et al., 1998) have examined the microstructures of shear bands in some conventional grainsized metals in which the shear bands were developed only under high-rate loading. When the interiors of adiabatic shear bands are examined in the transmission electron microscope, significant amounts of recrystallization are typically observed (the

6 Mechanical Failure Processes in Nanomaterials

209

recrystallization is thought to be due to either a significant temperature rise or due to very large plastic deformations). Indeed, one typically observes grain refinement within the center of an adiabatic shear band (rather than grain elongation), and the refined grains tend to be equiaxed and nearly dislocation free. Originally microcrystalline microstructures can become refined into nanocrystalline structures inside an adiabatic shear band; 50-nm grains were obtained in a commercially pure titanium that was originally 30 µm in grain-size by Chichili et al. (1998). In such adiabatic shear bands, the TEM also reveals a gradient of microstructure across the shear band. The shear bands that are observed in nanocrystalline iron are clearly very different. In this case the microstructures are relatively uniform across the band, with relatively sharp boundaries, as shown by the montage presented in Figure 6.17. It appears, therefore, that the shear bands observed in nanocrystalline iron are not adiabatic shear bands, but rather deformation instabilities developed as a result of geometric softening.

Fig. 6.17 TEM micrograph showing the microstructure near a shear band boundary in nanocrystalline iron (Wei et al., 2002). The boundary between the material within the band and that outside the band is shown by the solid line. Note that the transition occurs over a transition width that is about one grain diameter. Reprinted from Applied Physics Letters, Vol. 81, Issue 7, pages 1240–1242, Q. Wei, D. Jia, K.T. Ramesh, E. Ma, Evolution and microstructure of shear bands in nanostructured fe. 2002, with permission from American Institute of Physics.

We can learn much about the mechanisms associated with shear band development and the deformation in the nanocrystalline iron by examining the processes of shear band evolution presented in Figure 6.14. First, the fact that existing bands continue to widen indicates that a saturation condition is present in the system. Thus if the center of the band reaches some saturated condition, so that no further geometric softening can occur (for example, because the optimum orientation has already been achieved), then the band must widen because geometric softening can still occur at the edges of the band. Second, the fact that additional bands are nucleated indicates

210

Nanomaterials

that a hardening mechanism must also exist (for example, local strain hardening due to dislocation accumulation), because otherwise further deformation could simply accommodated by widening of the existing bands. Examination of Figure 6.17 shows that the length scale associated with the softening mechanism is relatively small, only about one grain wide, since the transition between the outside of the band and the saturated orientation condition within the band occurs over this range. There are thus three distinct length scales in this problem at any given time: (i) the width of the band, (ii) the width of the transition zone, and (iii) and the spacing of the shear bands. Presumably the spacing of the shear bands is associated with the competition between the nucleation and growth of the shear bands, with nucleationdominated problems showing smaller band spacings.

6.6.4 Effect of Strain Rate on the Shear Band Mechanism High-strain-rate experiments were also performed by Jia et al. (2003) on the nanocrystalline iron, using controlled strains (by choosing the length of the projectile and the initial velocity). For a similar strain level, the distribution of shear bands was more uniform in the low-rate specimens (Figure 6.13b) than in the high-rate specimens (Figure 6.13c). This arises primarily because of the differences in specimen aspect ratio (note the shear zones in Figure 6.13c) because the high-rate specimens had a length:diameter aspect ratio of ≤1 while the low-rate specimens have a length:diameter aspect ratio of ≥1. Length to diameter ratios of less than one guarantee that a shear band at an angle of 45◦ to the loading direction will intersect one of the loading boundaries, resulting in constrained deformations. The shear band orientations and widths, however, remain similar. Wei’s TEM observations at high rates (Wei et al., 2002) did not differ qualitatively from those at low rates (in the case of the nanocrystalline iron). No macroscopic shear bands were observed in the 20-m Fe or 980-nm Fe loaded at high strain rates. Thus the basic mechanism of shear banding does not appear to change between the low-rate and high-rate loading, and the failure mechanism appears to be mediated by grain size rather than strain rate. This also indicates that the bands observed in nanocrystalline bcc iron are not adiabatic shear bands.

6.6.5 Effect of Specimen Geometry on the Shear Band Mechanism Jia et al. (2003) also tested a number of short specimens (with the same aspect ratio as that used for the high-strain-rate tests) at low strain rates to examine possible effects of the different specimen geometries conventionally used in such tests. In the 20-m Fe and 980-nm Fe, which deformed uniformly like conventional ductile metals, no effects of specimen geometry were observed on the stress-strain behavior. There was also no significant difference in the observed stress-strain curves for the long and short specimens of 268-nm Fe, even though the shear band mechanism is

6 Mechanical Failure Processes in Nanomaterials

211

active at this grain size. The specimens (both long and short) were able to sustain very large deformations without failure. The experiments showed that the measured yield strengths and the shear band orientations were identical for the long and short specimens, but the total strain to catastrophic failure was much larger for the short specimens (as is expected, because of the effective constraint provided by the compression platens in the small aspect ratio case). In summary, the grain sizes at which the shear band mechanism is observed in iron, the stress levels at which the shear bands are formed, the orientations of the shear bands, and the morphology (width, microstructure) of the shear bands all appear to be independent of specimen geometry. However, final failure as a result of this instability is dependent on geometry (as with the tensile instability called necking).

6.6.6 Shear Bands in Other Nanocrystalline Metals Shear bands are not commonly seen in other nanocrystalline materials (other than bcc metals) unless the material is driven to extreme conditions such as through very high strain rates or within highly deformed ligaments in the last stages of necking. The development of shear bands in the final stages of the tensile instability is well known even in conventional grain sized metals, and is also observed in nanocrystalline metals. It is generally also true that conventional grain size hcp metals are prone to shear band development because of the lack of symmetry in their plastic deformations, but nanocrystalline hcp metals have not been studied in great detail.

6.7 Suggestions for Further Reading 1. Bai, Y. and B. Dodd. Adiabatic Shear Localization. Pergamon Press, New York 1992. 2. Wright, T.W. The Physics and Mathematics of Adiabatic Shear Bands. Cambridge University Press, Cambridge, 2002. 3. Anderson, T.L. Fracture Mechanics: Fundamentals and Applications. 3rd Edn. CRC Press, New York 2004. 4. Broberg, K.B. Cracks and Fracture. Academic Press, London 1999.

6.8 Problems and Directions for Research 1. Make a plot of the tensile strain to failure (as quoted in the literature) as a function of grain size, for example fcc and bcc nanocrystalline metals. 2. Using the Considere and Hart criteria, compute the strain to onset of instability in tension for conventional grain-size copper, aluminum, and iron. How does this compare with experimental data on these materials?

212

Nanomaterials

3. It appears that grain size distributions can have a profound influence on the total strain to failure in tension. Identify three reasons why this might be so. 4. The double torsion test has been used to measure the fracture toughness of very small diamond specimens – see the work of Field et al. (Davies, 2004). Can this technique be used to study nanomaterial samples (e.g., electrodeposited samples available in thin sheets)? 5. Relatively thick ( τ f , full dislocations are preferentially generated, and twinning is unlikely. The stresses required to nucleate full and partial dislocations will be equal when 2α µ (b f − b p )b p . (7.10) dc = γ At grain sizes below this critical grain size dc , the nucleation of partial dislocations becomes easier than the nucleation of full dislocations and this partial dislocation generation implies that twinning may become a preferred deformation mode. Note that the Burgers vector for the full dislocation is always greater than the Burgers vector for the partial dislocation. The preferential generation of partial dislocations over full dislocations in nanocrystalline materials has also been observed in molecular dynamics simulations (Yamakov et al., 2002). More sophisticated models of this type have also been developed, which account for the different stresses that are required to generate leading partials, trailing partials and twinning partials (Liao et al., 2003). The potential increased incidence of twinning in nanocrystalline materials will have implications for the multiaxial constitutive functions that must be used to describe the mechanical properties of these materials. An example of twinning in a nanocrystalline aluminum (mean grain size of 50 nm) is shown in Figure 7.10 (Cao et al., 2008). Aluminum has a high stacking fault energy, and does not appear to twin at conventional grain sizes, but develops deformation twins at nanocrystalline grain sizes, as first shown by (Chen et al., 2003a). A second way in which deformation twinning has become of interest in nanocrystalline materials is based on the idea that the twin boundaries represent additional obstacles (akin to grain boundaries) or provide additional degrees of freedom in terms of deformation. Some highly twinned nanocrystalline or ultrafine grained materials have been observed to have both high strength and high ductility (Dao et al., 2006; Chen and Lu, 2007). The precise reasons for this behavior remain unclear as of this writing.

7 Scale-Dominant Mechanisms in Nanomaterials

235

Fig. 7.10 High resolution electron micrograph of deformation twins developed in nanocrystalline aluminum subjected to large shearing deformations (Cao et al., 2008). The diffraction pattern on the right demonstrates the twinned character.

7.2.3 Grain Boundary Motion The fact that the grain boundary volume fraction begins to dominate the behavior of bulk nanocrystalline materials leads one to examine the possible mechanisms that are directly associated with the grain boundary. To this point, we have discussed two different ways in which the grain boundary can affect the behavior of the material. First, the grain boundary may act as an obstacle to dislocation motion, or in general as an obstacle to the development of compatible deformations. Second, the grain boundary may act as a source of dislocations (or twins), and thus may actually aid the development of plastic deformations. However, the existence of so many grain boundaries also makes possible deformation mechanisms that directly invoke movement of (or at) the grain boundary. Idealizing the grain boundary as a plane separating two grains, one sees (Figure 7.11) that two motions are possible: (a) normal to the grain boundary and (b) in the plane of the grain boundary. In vector form, this can be expressed as

δ = δn n + δt t

(7.11)

where δn is the normal displacement of the boundary and δt is the tangential (inplane) displacement of the boundary. The first component results in the growth or the shrinkage of the grain, while the second results in the relative sliding of two grains. In general, thermodynamics does not favor the shrinkage of the grain, unless substantial amounts of energy are provided. In contrast, particularly in nanocrystalline

236

Nanomaterials

δ

δ

Fig. 7.11 The possible grain boundary motions: displacements δn in the direction normal to the GB (grain growth or shrinkage), and δt in a direction tangential to the plane of the boundary (grain boundary sliding). The unit normal vector is n and the unit tangent vector is t, as in Equation (7.11).

materials, thermodynamic processes favor grain growth. Thermally assisted grain growth is familiar to most mechanical engineers and materials scientists, since this is a classical method of grain coarsening and develops routinely during annealing processes. However, grain growth can also be a stress assisted process (Gianola et al., 2006), and this may be of importance to many deforming nanocrystalline solids. Grain boundary sliding is a mechanism that is well known from studies of creep deformations of materials. This mechanism is not discussed in any detail here, but the approach itself is a classical one in materials science (e.g., Ashby and Verrall, 1973), involving diffusive processes.

7.2.4 Grain Rotation As the grain size decreases, the grains in the bulk nanocrystalline materials become harder (in the sense that they do not deform easily). Is there a point at which a nanocrystalline material can be treated like a granular solid? The limiting condition would appear to be one of elastic grains deforming as a granular assembly. However, most bulk nanocrystalline materials cannot be viewed in this way at any length scale, because they generally maintain compatible deformations even at the grain boundaries (except perhaps at triple points or junctions) and because they usually retain the ability to develop some plastic deformations. Plastic deformations have been observed in the TEM for grain sizes as small as 5 nm. Still, the increasing strength suggests that some of the concepts that are relevant to granular materials may

7 Scale-Dominant Mechanisms in Nanomaterials

237

also be important for nanocrystalline materials. We examine one of these concepts in this section. There are two motions of grains (rather than grain boundaries) that may be of interest in nanocrystalline solids: grain translation and grain rotation. The rigid translation of grains is generally not a significant mechanism, although this may be important in the case of very heterogeneous grain size distributions. The rotation of grains, however, plays a significant part in accommodating deformations. For example, grain rotations have been observed in the development of shear bands in nanocrystalline metals (Wei et al., 2002). Joshi and Ramesh (2008b) have demonstrated that grain rotations lead directly to the development of localized shearing deformations in nanocrystalline solids, and suggest that this mechanism may result in the development of limited stable deformations and specific grain size regimes (Joshi and Ramesh, 2008c). The core concepts are discussed in this section. There are few theoretical investigations of shear banding in bulk nanocrystalline materials as a function of grain size. The continuum approach cannot predict such size effects because of the absence of a length-scale in the governing equations. Strain gradient plasticity theories can be used, but these theories do not explicitly account for grain size (Carsley et al., 1997; Zhu et al., 1997). Joshi and Ramesh modeled the growth of shear bands in nanocrystalline and ultrafine-grained materials using an explicit consideration of grain rotation driven geometric softening. The model considers the development of shearing instabilities through a visco-plastic constitutive law for the material response coupled with micromechanics-based softening through grain rotation, motivated by the observations of Jia and co-workers (Jia et al., 2003; Wei et al., 2002). The grain size enters naturally through the grain rotation mechanism in this model. The essence of the model is as follows. Assume that each grain in the material has anisotropic plastic behavior, with a hard orientation and a soft orientation, as shown in Figure 7.12a. The initial state of the material has all of the grains randomly oriented. Now, the average response of the polycrystalline material will clearly be different if the grains are all oriented in a specific direction. For example, a material with all of the grains oriented so that the soft orientation is in the direction of shearing will need a lower shear stress (than the randomly oriented polycrystalline material) to sustain further plastic deformation. Because the shear strength of the material decreases with grain orientation (this is called geometric softening), the grains will tend to rotate so that the soft plastic direction rotates into the direction of shearing. The model of Joshi and Ramesh demonstrates that a consequence of this specific geometric softening behavior is shear localization, that is, the development of the shear band. This is illustrated in Figure 7.12b, where the random orientation of the grains is shown as having changed into a specific orientation within a localized region called the shear band. Consider the one-dimensional shear problem (Figure 7.13) of an infinitely long slab with finite thickness W subjected to a constant nominal shear strain rate γ˙a . In the absence of inertia (i.e., for slow loading), the balance of linear momentum reduces to an equilibrium equation, ∂∂Xτ = 0, where τ is the shear stress and X is the

238

Nanomaterials

s h

Fig. 7.12 (a) Schematic of a grain showing its soft and hard orientations (with respect to plastic deformation, not elastic stiffness). (b) The grains are initially randomly oriented in the material, but begin to orient themselves so that the soft direction is in the direction of shearing, and the process of grain rotation into the soft orientation results in the localization of the deformation into a shear band.

thickness direction. For simplicity, only isothermal problems are considered here (note that this eliminates the consideration of adiabatic shear bands). The material is assumed to be elastic/viscoplastic, so that the total strain rate can be decomposed into an elastic strain rate and a viscoplastic strain rate: γ˙ = γ˙e + γ˙p . From this we obtain ∂τ (7.12) τ˙ = = µ γ˙e = µ (γ˙a − γ˙p ), ∂t where µ is the elastic shear modulus, γ˙e is the average elastic strain rate, γ˙p is the  thickness-averaged plastic strain rate defined by W1 γ˙p dX, and the total applied strain rate is γ˙a = γ˙e + γ˙p .

γ

W

Fig. 7.13 Schematic of simple shearing of an infinite slab, showing the terms used in examining the shear localization process in simple shear (Joshi and Ramesh, 2008b).

7 Scale-Dominant Mechanisms in Nanomaterials

239

A viscoplastic constitutive law is assumed for the material, assuming powerlaw strain hardening and power-law rate hardening, and allowing for the geometric softening that arises from grain rotation:     γ˙p m γp n τ (7.13) = 1+ (1 − c¯φ ), 1+ τ0 γ˙0 γp where φ is an internal variable defined (in the next paragraph) as the number fraction of grains in a given region that have the soft orientation aligned to the shearing direction, and γ p is the plastic strain. The terms γ˙0 , τ0 , γ0 , n, and m in Equation (7.13) represent the characteristic strain rate, yield strength (of the polycrystalline material), yield strain (corresponding to τ0 ), strain hardening index, and strain rate sensitivity, respectively. The factor c¯ is defined as c¯ = 1 − ττs0 where τh0 and τs0 are h0 the yield strengths of the grain in the hard and soft orientations (corresponding to a single crystal). By solving this equation for the strain rate, we obtain the evolution of plastic strain corresponding to this constitutive law:  1

γ p −n m 1 τ γ˙p = γ˙0 { } − 1 , 1+ τ0 (1 − c¯φ ) γ0 where  is defined through f =

(7.14)

( f +| f |) . 2

7.2.4.1 Definition of the Internal Variable Consider a representative volume element (hereafter referred to as an RVE or a bin) that contains a large number of grains Nb (b for bin). We define the internal variable as φ (t) = NNs , where Ns is the number of grains in the bin that have the soft b orientation aligned with the shearing direction, and the evolution of φ represents the evolution of rotation of the grains. Thus, as more and more grains rotate into the soft shearing orientation, φ increases and (through the (1 − c¯φ ) term) a softening mechanism develops. Joshi and Ramesh demonstrated that this softening mechanism leads to localization, with a shear band that is essentially the region where φ ≈ 1. 7.2.4.2 Evolution of the Internal Variable The fraction φ of soft-oriented grains evolves through contributions from two specific mechanisms: φ˙ = φ˙1 + φ˙2 . (7.15) The first component φ˙1 in Equation (7.15) represents the contribution of overall plasticity (Figure 7.14) to grain rotation, while φ˙2 represents the contribution due to interaction of a grain with nearest neighboring grains (Figure 7.15d). φ ceases to grow once it reaches unity (since no further re-orientation is possible), so that φ˙ = 0 when φ = 1.

240

Nanomaterials

Fig. 7.14 Schematic of the rotation of ensemble of nano-grains occupying region ℜ embedded in a visco-plastic sea S subjected to shear. The background image shows the undeformed configuration.

Fig. 7.15 A hierarchical approach to homogenization of grain rotation due to interface traction. (a) Material continuum. (b) Collection of bins in sample space. (c) Grains within a RVE. The colored shading represents the average grain orientation in that bin. (d) Interaction at the grain level. Grains with individual orientations are described by different colors.

7.2.4.3 Contribution of Overall Plastic Deformation We focus on the contribution to rigid rotation of an ensemble of grains (ℜ) due to overall plastic deformations in the viscoplastic sea (S) surrounding the ensemble (Figure 7.14). Strain compatibility at the interface of ℜ and S gives a rotation rate that is proportional to the rate of plastic deformation. Thus, the contribution of overall plastic deformation to the evolution of φ is given to first order by

7 Scale-Dominant Mechanisms in Nanomaterials

¯ γ˙p φ˙1 = ψ (φ , d)

241

(7.16)

where d¯ is the mean grain size within the ensemble. The function ψ indicates the ease with which the ensemble ℜ can rotate rigidly in the sea, and is referred to as a fabric factor (because it represents, in some sense, a measure of the local fabric or microstructure). An ensemble residing in a region where surrounding grains are soft (φ > 0) will rotate with relative ease as the soft surrounding medium deforms plastically. On the other hand, in a region where the grains are randomly oriented (φ ≈ 0), rotation (in the direction consonant with macroscopic shear) is difficult because the ¯ = ψ0 , surrounding grains will not deform easily. For simplicity, let us set ψ (φ , d) where ψ0 is of order 1–10. In practice, ensembles with very small d¯ will tend to rotate to accommodate the global plastic deformations, while those with large d¯ will tend to deform plastically through crystallographic slip. The mean grain size in the ensemble is determined from the grain size distribution, and in this way the grain size distribution may affect φ˙1 .

7.2.4.4 Contribution of Inter-Granular Interaction Grain rotation may also be caused by the local interactions of a single grain with its neighboring grains of similar size. Grain rotation models based on interfacial contact resistance have been successfully implemented in simulating granular flow (Iwashita and Oda, 1998) including elastoplastic contact (Hu and Molinari, 2004). This latter elastoplastic contact mechanism can be included in this model for grain rotation in nanocrystalline materials through a hierarchical scheme. Using a threegrain model, the concept of interfacial interaction at the grain level is used to determine a driving force for rotation; subsequently, an evolution law for grain rotation is prescribed in terms of the driving force. In granular materials, rolling resistance causes neighboring grains to create interfacial traction. Consider the interaction between three adjacent grains as shown in Figure 7.15 (for simplicity, we consider a 2D case). Let the current rotation angles of the three grains (a, b and c) in Figure 7.15d be denoted as θa , θb and θc , respectively. Focusing on the central grain (b) and assuming that θa > θb > θc , the interfacial tractions on grain b are then F|ab = kt |ab δt |ab

(7.17)

due to the interaction with grain a, and F|bc due to the interaction with grain c: F|bc = kt |bc δt |bc ,

(7.18)

where kt is the effective contact stiffness and the tangential displacements are δt (see Equation 7.11). These tangential displacements are of the form

242

Nanomaterials

δt |ab = δt |a − δt |b + Ra (θa ) − Rb (θb ),

(7.19)

with Ra and Rb denoting the radii of grains a and b. The grain boundary sliding mechanism is represented by the term δt |a − δt |b . Since this section is only considering grain rotation, we set the grain boundary sliding to zero, so that δt |a − δt |b = 0. Further, let us assume that all of the grains are the same size, so that Ra = Rb = d2 , where d is the grain diameter. Then Equation (7.19) gives us

δt |ab =

d (θa − θb ), 2

(7.20)

andδt |bc =

d (θb − θc ). 2

(7.21)

The net interface traction (per unit length) on grain b is

[∆ F]b = Fab − Fbc ,

(7.22)

and using Equations (7.20) it follows that kt d [θa − 2θb + θc ]. 2 The corresponding net torque on grain b is then given by [∆ F]b =

(7.23)

kt d 2 d [∆ F]b = [θa − 2θb + θc ]. (7.24) 2 4 But what is the tangential contact stiffness kt ? This can be obtained from elasticplastic contact micromechanics (Stronge, 2000). The contact stiffness of grains undergoing elasto-plastic deformation is given in terms of the radius rc of the contact area by 8rc kt |ab = , (7.25) Cab Tb =

where Cab is called the contact compliance for the two grains in the contact and is given by (Stronge, 2000) Cab =

1 − νa 1 − νb 2(1 − ν ) + = , µa µb µ

(7.26)

using the shear modulus (µ ) and Poisson’s ratio (ν ), and assuming that these moduli are identical for the two grains. In general, the contact area will evolve as grains rotate (Conrad and Narayan, 2000; Van Swygenhoven and Derlet, 2001), and will be higher for plastic contacts than for elastic contacts. Let us, however, assume a fully plastic contact for a polycrystalline material and set the radius of the contact equal to the grain radius, so

7 Scale-Dominant Mechanisms in Nanomaterials

243

that rc = d2 . Using Equations (7.26) in (7.25) and then using the result for kt |ab in Equation (7.24), we obtain the driving force for grain rotation as the torque: Tb ==

µ d3 [θa − 2θb + θc ]. 2(1 − ν )

(7.27)

How will the grain respond to this driving force? The response will clearly depend on the surrounding grains and their grain boundaries as well. To simplify the problem, we postulate a kinetic relation between the evolution of grain rotation and the driving force, with the rate of rotation given by

θ˙b = (mr T )b ,

(7.28)

where mr is a mobility coefficient, representing the resistance provided by the surrounding grains to the rotation of grain b. Equation (7.28) is similar to those adopted in describing grain rotation coalescence (Moldovan et al., 2001). In the latter work the driving force is due to anisotropic grain boundary energies and the plastic deformation is accommodated by grain boundary and lattice diffusion mechanisms, which are believed to be dominant at very small grain sizes (d < 10 nm) and at elevated temperatures. Similarly, in defining the mobility, we account for the viscoplastic resistance offered by the region surrounding the rotating grain.

Fig. 7.16 Enlarged view of ℜ (Figure 7.14) showing intergranular interaction in the region. Rotation of the central grain is accomodated by rotation of the surrounding grains over a length L.

The key concept with respect to grain rotation is the following. Consider the rotation of a specific grain such as the central grain shown in Figure 7.16. If you were to be an observer sitting on another grain that is sufficiently far away from the rotating

244

Nanomaterials

grain, you will not even be aware that the central grain has rotated. However, grains neighboring the central grain will have to rotate to some degree to accommodate the rotation of the central grain. That is, the rotation of any single grain must be accommodated by the surrounding grains (and the grain boundaries), but there is a finite length (L) over which this accommodation occurs (Figure 7.16) and grains sufficiently far away from the rotating grain are unaffected by the rotation. This finite length (L) will depend on the interactions between the grains, and in particular on interactions at the grain boundaries (although the degree of plastic deformation of the grains will also be involved). The length scale for rotational accommodation will be smaller for elastoplastic grains (as in polycrystalline metals) than for purely elastic grains (a limiting example of which would be sand). This rotational accommodation mechanism must be present in all granular solids that contain some dissipative mechanisms (which may be localized to grain boundaries). Using this rotational accommodation concept, we can (from a dimensional analysis viewpoint) define the mobility due to the viscoplastic resistance in terms of the accommodation length scale as (mr )b =

1 1 ( ), L3 ηeff

(7.29)

where ηeff is the effective viscosity of the accommodating region. This can be estimated in some cases using molecular dynamics simulations (Van Swygenhoven and Caro, 1997). From a mechanistic viewpoint, the effective viscosity itself can be derived (through addition of the corresponding strain rates) in terms of the intragranular and grain boundary mechanisms using the following argument. The effective strain rate γ˙eff in the accommodating region is given by

γ˙eff = γ˙p + γ˙gb ,

(7.30)

where γ˙p is the strain rates due to plastic deformation mechanisms and γ˙gb is the strain rate due to deformation at the grain boundaries. Assuming that the stress is uniform across this region, and defining the viscosities in the classical way, we have

from which it follows that

τ τ τ + = , ηeff η p ηgb

(7.31)

1 1 1 + = . ηeff η p ηgb

(7.32)

Thus the effective viscosity of the accommodating region has contributions from dislocation plasticity (η p ) and grain boundary mechanisms (ηgb ). For example, the effective viscosity corresponding to grain boundary diffusion-type mechanisms can be computed using kB T d 3 ηgb = , (7.33) 64δgb Ω Dgb

7 Scale-Dominant Mechanisms in Nanomaterials

245

where kB is the Boltzmann constant, T is the absolute temperature, δgb is the grain boundary thickness, Ω is the atomic volume, and Dgb is the grain boundary diffusion coefficient (Murayama et al., 2002). The contribution to the effective viscosity (η p ) from dislocation plasticity is represented by the viscoplastic resistance provided by the grains themselves, and a lower bound for this is

ηp =

τs0 . γ˙0

(7.34)

What is the magnitude of L in Equation (7.29)? As the grain size increases, the likelihood of global plastic deformation dominates the likelihood of rotation, and eventually this grain rotation mechanism is no longer a significant contributor to the deformation and slip becomes the dominant mode of deformation. For most metals, where dislocation slip is relatively easy, grain rotation is unlikely to be an important contributor for grain sizes larger than about 1 µm, and may not be a significant contributor even in the ultrafine grain range. In ceramics, however, grain rotation may be significant even for much larger average grain sizes, because there is so little plasticity in ceramics (on the other hand, however, grain boundary mechanisms for inelastic deformation in ceramics are relatively few). In general, one can describe L in terms of the number j of other grains involved (in a given linear direction in the plane) in accommodating the rotation of a given grain, so that L = jd. This length scale is a natural length-scale for a bin. Since very large grains will deform by slip rather than rotate, we assume that L is a fixed length scale for a given material (for example, this might be 1 µm for metals). The number of grains involved in accommodating the rotation is j = Ld . In a nano-grained material, many grains ( j > 1) are involved in the accommodation distance L. A quantitative measure of L remains to be established, but L can be viewed in a manner analogous to the microstructural length scales derived in strain gradient plasticity theories (Gao et al., 1999).

7.2.4.5 Homogenized Representation of Grain Rotation An expression for the second contribution (φ˙2 ) to the evolution of grain rotation can be derived by homogenizing the evolution of θ over the RVE (Figure7.15c). The average grain orientation over the RVE is 1 |θk |. θ¯ = Nb ∑

(7.35)

The number of favorably oriented grains in a bin is counted by performing a weighted averaging with a penalty function ζk = δ (|θk | − θs ), where θs is the favorable orientation (aligned with the band orientation). Note this uses a delta function δ (|θk | − θs ), which has value unity when the argument (|θk | − θs ) is zero, that is |θk | = θs , and is zero for all other θk . Further, note that as θ¯ → θs , φ → 1. Thus for the RVE we have 1 φ2 = ζk |θk |. (7.36) Nb ∑

246

Nanomaterials

The average rate of change of φ2 is then (using the kinetic relation, Equation 7.28) 1 1 φ˙2 = ζk θ˙k = ζk (mr T )k . Nb ∑ Nb ∑ Using Equation (7.27), we obtain   1 kt d ˙ φ2 = ζk (θa − 2θb + θc ) . 4 jηe j2 d 2 Nb ∑

(7.37)

(7.38)

Examining Equation (7.38), we identify the terms in the square brackets as ap2 proximating the second derivative ∂∂ Xφ2 in the continuum limit, so that we develop the rate equation kt d ∂ 2 φ ∂ 2φ φ˙2 = = D , (7.39) r 4 jηe ∂ X 2 ∂ X2 where we have identified a new rotational diffusion coefficient Dr as Dr =

1 µ d2 kt d = ( ). 4 jηe jηe 2 − ν

(7.40)

Equation (7.39) may be viewed as characterizing a rotational diffusion mechanism that tends to diffuse variations in φ through inter-granular interactions. Putting together Equations (7.15), (7.16), and (7.40), the net evolution of the internal variable φ is ∂ 2φ (7.41) φ˙ = ψ γ˙p + Dr 2 . ∂X This evolution equation can be integrated together with the momentum balance and constitutive laws to solve the problem of a nanocrystalline material in which grain rotation is a possible mechanism. We now see that the primary consequences of the grain rotation mechanism are the existence of a diffusive term and the corresponding length scale. The governing equations for the problem of simple shear of a nanocrystalline material incorporating grain rotation are summarized below:

∂τ =0 ∂X

 ∂τ = µ γ˙a − γ¯˙p ∂t   1  γ p −n m 1 τ γ˙p = γ˙0 − 1 1+ τ0 (1 − c¯φ ) γ0 ∂ 2φ φ˙ = ψ γ˙p + Dr 2 . ∂X

(7.42) (7.43) (7.44) (7.45)

Boundary conditions and initial conditions must be added to solve these equations. Joshi and Ramesh considered the simple shear of a slab of nanostructured

7 Scale-Dominant Mechanisms in Nanomaterials

247

iron, with initial conditions that are zero everywhere except for the initially uniform shear strain rate. In order to account for typical material variability, an initial defect 2 X − 0.5) ], where φ0 is distribution φi was assumed in the form φi (X) = φ0 exp[−A( W the largest initial value of φ in the distribution and A is a dimensionless parameter which is like a wave-number. Their results on nanostructured iron are of interest because of the experimental observations of shear bands in this material (Jia et al., 2003). The values for the various parameters used by Joshi and Ramesh are listed in Table 7.3 (corresponding to polycrystalline Fe). Given the heavy plastic work during severe plastic deformation processes, the parameter c¯ was obtained at large strains (∼ 1) from the results of Spitzig and Keh (1970) on iron single crystals. Table 7.3 Basic parameters for grain rotation model in polycrystalline iron, as developed by Joshi and Ramesh. Parameter Value Average grain size d, in nm 300 Shear modulus µ , in GPa 76 Poisson’s ratio ν 0.3 Normalizing strain rate γ˙0 , per second 5 × 10−4 Normalizing stress τ0 , MPa 900 Perturbation “wave number” 5 × 104 Strain hardening index n 0.01 Rate hardening index m 0.005 Anisotropy factor c¯ 0.07 Plastic contribution coefficient ψ 10 Maximum initial value of φ , φ0 0.005 Far-field initial value of φ , φ∞ 0 Rotational accommodation length scale L, µm 1.0

For a material with this initial distribution in φ , Figure 7.17 shows the evolution of φ as a function of nominal strain near the band. The internal variable (representing the degree of rotation of the grains into the soft orientation) increases rapidly at the center of the initially perturbed region once the strain reaches 0.02, and stops evolving (φ˙ = 0) once φ reaches the maximum value (φ = 1) in the center. Physically, this represents a stage where following the localization the grain reorientation process saturates, so that no further reorientation occurs in that region. At the onset of severe localization, the plastic shear strain rate is very high (≈ 103 γ˙a ). With increasing nominal strain, γ˙p drops in the band center (Figure 7.18) in this strain hardening case because φ is constant there but the plastic strain continues to grow. Following a fully developed band (defined by a finite thickness over which φ = 1), γ˙p remains high at the edges of the band where φ has not saturated. Subsequent band growth is driven by rotational diffusion and strain hardening. The plastic shear strain in the band center is fully resolved by the simulation (Figure 7.19). In Figure 7.20, the overall stress-strain response for a defect-free material is presented (curve A) and compared with the response obtained from a sample with the perturbation (curve B, solid line). With the perturbation, the sample exhibits

248

Nanomaterials

Fig. 7.17 Evolution of grain orientation fraction (φ ) around the band center. Note the rapid early growth, the spatial localization, and the saturation of φ . Also note the spreading of the band, i.e., the increasing band thickness - this is also observed in experiments.

Fig. 7.18 Evolution of plastic shear strain rate (γ˙p ) around the band center. The greatest activity in this variable is at the band boundaries, where the grains are reorienting into the soft orientation for shear.

Fig. 7.19 Evolution of the plastic shear strain γ p around the band center. After localization, γ p evolves slowly indicating that the plastic flow inside the band develops at the rate of strain hardening at higher nominal strains.

7 Scale-Dominant Mechanisms in Nanomaterials

249

a brief period of strain hardening following yield. The strain hardening is followed by a rapid drop in the stress corresponding to severe localization, with the magnitude of the drop determined by the material hardening and the anisotropy parameters. The nominal strain corresponding to the stress collapse is termed the critical strain γcr , and should be related to (but less than) experimental measures of the critical strain for localization. Following the stress drop, a softening response is observed. Such a macroscopic softening response of a strain hardening material indicates the dominance of geometric softening over material hardening (Yang and Bacroix, 1996). The softening regime exhibits serrated behavior in the presence of strain hardening (this can be seen at a higher magnification on the stress-strain curve) due to the competition between the hardening and softening variables.

Fig. 7.20 Overall stress-strain response for a defect-free sample (curve A) and a sample with an initial defect in φ (curve B). Curve B’ indicates the development of the shear band thickness corresponding to curve B.

We define a fully developed shear band as a region where φ reaches the maximum value and use this definition to also determine the thickness of the shear band. After localization, the material hardening together with rotational diffusion controls the broadening of the shear band. The development of the thickness of the shear band is also shown in Figure 7.20 as a function of overall strain.

7.2.4.6 Effect of Grain Size The grain size explicitly appears in two quantities in the governing equations: the polycrystalline yield strength (τ0 ) and the rotational diffusion coefficient Dr . Consider three grain sizes (100 nm, 300 nm, and 1 µm), with n = 0.01 and m = 0.005 held constant for all the grain sizes under consideration. Using the Hall-Petch

250

Nanomaterials σ

relation for Mises yield in iron, the shear strengths τ0 = √y3 are 1300 MPa, 900 MPa (300 nm), and 290 MPa (1 µm). Figure 7.21 shows the band thickness evolution for different grain sizes. Whether a shear band will develop is strongly grain size dependent. If a shear band does develop, the band thickness is larger for larger grain sizes (at equivalent nominal strains), because the rotational diffusion term depends on the grain size.

Fig. 7.21 Evolution of shear band thickness for different grain sizes, assuming grain rotation mechanism. The material hardening parameters are held constant for all the grain sizes. The applied strain rate is 10−3 s−1 .

7.2.4.7 Relative Contributions of Grain Boundary and Grain Mechanisms to ηe With this definition of the effective viscosity ηe , a transition from dislocation dominated mobility to grain boundary dominated mobility will occur (comparing the two terms in Equation 7.32) at a transition grain size given by

dtr =



64δ Ω Dgb τs0 kB T γ˙0

1

3

.

(7.46)

This transition size depends on grain boundary diffusion and grain boundary thickness variables, and is about 9 nm in the specific case of iron. The grain boundary viscosity ηgb is likely to be dominant for smaller grain sizes (although it will certainly continue to play a role even at much larger grain sizes). Note that

7 Scale-Dominant Mechanisms in Nanomaterials

251

below the transition grain size we will then have the rotational diffusion coefficient increase as the grain size decreases, because in that domain ηe ∼ ηgb , and then Dr ∝ d1 . 7.2.4.8 Comparison with Experiments Joshi and Ramesh considered two quantitative comparisons between the model and specific experiments: (a) bcc-Fe of Jia et al. (2003), and (b) fcc-Al of Sun et al. (2006). Both experiments under consideration involved uniaxial loading whereas the model was developed for the case of simple shear and considers only a single shear band. However, the model was able to distinguish between the two material behaviors. Experiments (Jia et al., 2003; Wei et al., 2002) on nanostructured and ultrafine grain iron show that shear bands occur immediately following yield. Joshi and Ramesh were able to reproduce this behavior in the large using physically reasonable parameters. However, in real bcc structures the hardening parameters may be modified by the grain size due to the variety of processing routes used for generating bulk nanostructured materials. Consolidated Fe (Jia et al., 2003) exhibits low but finite strain hardening (n ≈ 0.01 − 0.04) in the nanostructured regime whereas at larger grain sizes n is high (≈ 0.26) (Malow et al., 1998). Thus the strain hardening increases with grain size in NS-Fe, and the model should account for such variations (similar changes occur in the rate sensitivity, which decreases with grain size). Using the same perturbation parameters as in the bcc-Fe case, the model predictions were compared with the experimental observations of Sun et al. (2006) on equal channel angular extruded commercially pure aluminum (ECAE-CP-Al). Unlike in the case of the iron, the model does not predict shear banding for these three grain sizes in CP-Al. The quasi-static compression experiments (Sun et al., 2006) also showed no shear bands for 350 and 470 nm cases, although apparently anomalous shear bands were observed for the 590 nm case. Our model does not predict shear band development in such rate-sensitive materials as nanostructured and UFG-fcc metals. Indeed, shear bands have not been reported in NS or UFG pure fcc metals in general, although very “diffuse” bands have sometimes been observed in tension. The rate sensitivity effects are considered further by Joshi and Ramesh (2008a) when applying the grain rotation model to different crystal structures.

7.2.5 Stability Maps Based on Grain Rotation Some additional understanding can be obtained by considering a scaled version of the governing equations (Equation 7.42) for the simple shear problem including

252

Nanomaterials

grain rotation. Let us set up the same governing equations in terms of normalized variables Xˆ for position, τˆ for shear stress, and tˆ for time, defining these as follows: X Xˆ = L τm τˆ = τs0 c¯ tˆ = γ˙0t

(7.47) (7.48) (7.49)

where the scaling uses the parameters defined earlier. Using these scaled variables, the equilibrium condition is now

∂ τˆ = 0. ∂ Xˆ

(7.50)

Similarly, the elastic relation from Equation (7.42) (the second equation in that set) becomes  1 ˆ (7.51) τˆ = γ˙a − γˆ˙p , M where the strength index M has been defined as M=

τs0 c¯ , µm

(7.52)

and will be utilized in constructing a stability map. The normalized evolution equation for the internal variable is Dr ∂ 2 φˆ , φˆ = ψ γˆ˙p + γ˙0 L2 ∂ Xˆ 2

(7.53)

where γˆ˙p is the scaled visco-plastic shear strain rate. The solution of this system of equations will tell us how the stress, strain, strain rate, and orientation number fraction will evolve for a given material subjected to simple shear. However, we can learn a great deal about the problem by studying the stability of the equations – that is, by finding out under what conditions a perturbation to one of the variables will grow (corresponding to instability) rather than decay (corresponding to stability). The basic approach is as follows: we assume a perturbation in one or more of the variables (e.g., the plastic strain), insert the perturbed solution into the system of equations, and ask whether the perturbation will grow. If it will grow, the system is unstable in that it cannot survive a perturbation; if it will not grow, the system is stable in that it suppresses perturbations and tends towards homogeneous solutions. In what follows, we analyze the stability of the governing equations using linear perturbation analysis (a limited kind of perturbation analysis, which effectively restricts our attention to early times, i.e., the onset of instability rather than the later stages of unstable growth). With τˆ required to be homogeneous (which is the equilibrium condition), the stability of deformation is investigated by

7 Scale-Dominant Mechanisms in Nanomaterials

253

perturbing the plastic strain γ p and the internal variable φ from their homogeneous solutions, so that ˆ (7.54) γ¯p = γ p h + δ γ p (t˜) sin kX, and ˆ φ¯ = φ h + δ φ (t˜) sin kX,

(7.55)

where, for example, φ h is the homogeneous solution, δ φ (t˜) is the normalized temporal component of the perturbation to the solution (assumed small compared to φ h ), and k is the non-dimensional wave-number of the perturbation (we are essentially perturbing the amplitudes of these variables in space with a perturbation wavelength proportional to the reciprocal of k). Substituting the perturbed solutions in the normalized evolution equation (Equation 7.53) and retaining only the first order terms (linearized perturbation analysis) we find that the perturbations will grow only if

∂ γˆp k2 Dr > , ∂φ ψ γ˙0 L2

(7.56)

where the rotational diffusion coefficient Dr and the rotational accommodation length scale L are the quantities defined previously. Equation (7.56) represents the condition for instability. Letting φ h = 0, the constitutive law gives

∂ γˆp c¯ ≈ γˆph . ∂φ m

(7.57)

Thus the condition for instability is c¯ h k2 Dr γˆp > , m ψ γ˙0 L2

(7.58)

from which it follows that the critical wavelength of a perturbation that will lead to a runaway instability is given by

λcritical =



γ˙0 γˆph ψ c¯ Dr m

 12 −1

.

(7.59)

This instability condition can be rewritten using the strength index defined earlier (Equation 7.52), giving M d2 ηp λcritical = (2 − ν )ψ γˆph j ηe f f 

 21

.

(7.60)

Equation (7.60) should be interpreted as follows. It describes the critical wavelength of a perturbation of the system that will grow. That is, if the internal variable φ has a perturbation (a local variation) over a region with a size greater than λcritical (amounting to a wavelength greater than the critical wavelength), then that perturbation will grow, defining an instability that will become a shear band as shown in

254

Nanomaterials

the last section. Perturbations smaller than the critical wavelength will decay and will thus not be sufficient to initiate a shear band. The grain size dependence of λcritical appears through Dr in Equation (7.59). Note that if the material parameters are such that λcritical ≈ d, the material is inherently unstable in that this perturbation necessarily exists in the material, given the grain size. The condition λcritical < d is mathematically possible but is not physically meaningful unless grain size distributions are included. For most materials the fabric factor ψ will range between 1 and 10. Setting ψ = 10 and γˆph = 1, with γ˙0 = 5 × 10−4 s−1 , we can examine the dependence of λcritical on grain size. Figure 7.22 shows the relationship between λcritical and grain size for three materials (Joshi and Ramesh, 2008c), bcc-Fe, fcc-Cu, and fcc-Al. For all these materials, there are in general three regimes in this log-log plot as the grain size is decreased. In the first regime (d ≥ 100 nm), the critical wavelength decreases with decreasing grain size, the maximum number of participating grains is determined by fixed L, and η p is the rate limiting process. This rate limiting process continues to dominate the second regime (100 nm ≥ d ≥ 10 nm), which is governed by the limited number of participating grains, and λcritical decreases with a different slope. In the third regime (d ≤ 10 nm), the critical wavelength λcritical increases again as the grain size decreases. This is because the dominant rate limiting process changes from being bulk response to grain boundary driven. The region below the line λcritical = d in Figure 7.22 is the inherently unstable domain. As an example, for 300 nm grain size Fe, λcritical is approximately equal to the grain size. A similar calculation for 300 nm grain size Cu yields λcritical ≈ 440 nm, which is somewhat higher than the grain size. Consequently, the Fe is inherently susceptible to instability over

Fig. 7.22 The critical wavelength for instability as a function of grain size for three different metals (note that this is a log-log plot), with the critical wavelength computed using Equation (7.60). This figure has been obtained assuming that j = 10 below d = 100 nm, while for 100 nm < d ≤ 1 µm, we have L = 1 µm and j = Ld . The dashed straight line represents the condition that λcritical = d, which we call the inherent instability line.

7 Scale-Dominant Mechanisms in Nanomaterials

255

a larger range of grain sizes than the Cu. For some materials the entire curve will be above the inherent instability line (e.g., the Al-curve in Figure 7.22). The grain size range over which a material is inherently unstable (if at all) is intimately tied to its elastic properties, plastic anisotropy, and the strain rate sensitivity. Equation (7.60) shows that materials with high crystallographic plastic anisotropy and low rate sensitivity are generally more susceptible to instability.

Fig. 7.23 Stability map, showing the domains of inherent instability in materials as a consequence of the rotational accommodation mechanism. The map is constructed in terms of the strength index and grain size, so that every material of a given grain size represents one point on the map, and a horizontal line represents all grain sizes of a given material. Reprinted figure with permission from S.P. Joshi and K.T. Ramesh, Physical Review Letters, Stability map for nanocrystalline and amorphous materials, 101(2), 025501. Copyright 2008 by the American Physical Society.

This stability analysis allows one to probe the likelihood of instability in the materials and microstructure space. Figure 7.23 presents a stability map where the ordinate is the previously defined strength index M that enables one to compare different materials while the abscissa is the grain size. The curve is the locus of points satisfying the inherent instability criterion (λcritical = d, which was the straight line in Figure 7.22). If the point representing a given material and grain size lies in the saddle (the blue region), then the material is inherently susceptible to the rotational mechanism of shear instability. In general, materials with low strength indices show susceptibility over a smaller range of grain sizes compared to those with higher

256

Nanomaterials

strength indices. Thus, the iron is inherently susceptible to rotational instability over a larger grain size range than the copper, while aluminum does not exhibit any inherent instability. This prediction is qualitatively consistent with the available data on nc-Fe (Jia et al., 2003), nc-Cu (Cheng et al., 2005), and nc-Al (Sun et al., 2006). Note that a point in this figure (a combination of a given material and grain size) lying outside the inherent instability region (the orange region) does not unambiguously guarantee a stable response, as instability can still occur at larger perturbation wavelengths, but such a material would not be inherently unstable. Such maps provide useful guidelines to determine the probability of a material being unstable with respect to this softening mechanism for a range of defect wavelengths.

7.3 Multiaxial Stresses and Constraint Effects It appears that nanomaterials have multiple mechanisms that are sensitive to multiaxial stresses and to constraints (some of which are manifested as pressure effects). Unfortunately these are difficult to characterize experimentally, and so they are discussed in the next chapter (focusing on some modeling approaches).

7.4 Closing There are a variety of deformation mechanisms in nanomaterials that arise from length-scale interactions, with specific mechanisms developing in specific structures and for specific applications. These mechanisms sometimes result in remarkable properties, and a full understanding of the mechanisms may suggest applications that would not be conceived of otherwise. Some of these mechanisms are only discoverable from an experimental viewpoint, unless very large scale simulations are performed to capture the richness of possible phenomena.

7.5 Suggestions for Further Reading 1. Argon, A. (2008). Strengthening Mechanisms in Crystal Plasticity. Oxford University Press, Oxford. 2. Van Swygenhoven, H. and J. R. Weertman (2006). Deformation in nanocrystalline metals. Materials Today 9(5): 24–31. 3. Meyers, MA, A. Mishra, DJ. Benson, (2006). Mechanical properties of nanocrystalline materials. Progress in Materials Science 51: 427. 4. Kocks, U.F. A.S. Argon, M.F. Ashby, (1975). Thermodynamics and kinetics of slip. Progress in Materials Science 19: 1.

7 Scale-Dominant Mechanisms in Nanomaterials

257

7.6 Problems and Directions for Research 1. Examine grain size effects in hard granular materials in terms of macroscopic inelastic deformation and failure processes. This provides a sense of the behaviors that are possible when grain boundary and contact mechanisms are primarily responsible for deformation. 2. The development of twins is often related to the ability to emit and move partial dislocations. What does this say about the maximum speed of twin tips? 3. Let us view a young deciduous tree as a buckling problem (a long thin column with a weight – corresponding to the foliage – concentrated at the top). Assume the tree trunk is circular, with a diameter of 3 cm, and assume the tree is about 3 m tall. Given that the Young’s modulus of the wood is about 10 GPa, compute the maximum allowable weight of the foliage before buckling will occur. Can you climb to the top of this tree (in the winter) without inducing buckling? 4. What is the critical compressive buckling load for a “tree” made of a singlewalled carbon nanotube? Obtain the cross-section geometry from the literature, but assume a length of 2 µm. The Young’s modulus of a thin-walled carbon nanotube is about 1 TPa.

References Alivisatos, A. P. (1996). Semiconductor clusters, nanocrystals, and quantum dots. Science 271(5251), 933–937. Ashby, M. F. and R. A. Verrall (1973). Diffusion-accommodated flow and superplasticity. Acta Metallurgica 21, 149–163. Bruchez, M., M. Moronne, P. Gin, S. Weiss, and A. P. Alivisatos (1998). Semiconductor nanocrystals as fluorescent biological labels. Science 281(5385), 2013–2016. Cao, B., K. T. Ramesh, Li. B and E. Ma (2009). Twinning in nanocrystalline aluminum under high-rate shearing conditions. Submitted for publication. Cao, G. X. and X. Chen (2006). Buckling of single-walled carbon nanotubes upon bending: Molecular dynamics simulations and finite element method. Physical Review B 73(15). 155435. Carsley, J. E., W. W. Milligan, X. H. Zhu, and E. C. Aifantis (1997). On the failure of pressuresensitive plastic materials .2. comparisons with experiments on ultra fine grained fe-10. Scripta Materialia 36(6), 727–732. Chan, S., P. M. Fauchet, Y. Li, L. J. Rothberg, and B. L. Miller (2000). Porous silicon microcavities for biosensing applications. Physica Status Solidi a-Applied Research 182(1), 541–546. Chen, M. W., E. Ma, K. J. Hemker, H. W. Sheng, Y. M. Wang, and X. M. Cheng (2003a). Deformation twinning in nanocrystalline aluminum. Science 300(5623), 1275–1277. Chen, X. H. and L. Lu (2007). Work hardening of ultrafine-grained copper with nanoscale twins. Scripta Materialia 57(2), 133–136. Chen, X. H. Lu, L. Cheng, S., E. Ma, Y. M. Wang, L. J. Kecskes, K. M. Youssef, C. C. Koch, U. P. Trociewitz, and K. Han (2005). Tensile properties of in situ consolidated nanocrystalline cu. Acta Materialia 53(5), 1521–1533. Chichili, D., K. Ramesh, and K. Hemker (1998). The high-strain-rate response of alpha-titanium: Experiments, deformation mechanisms and modeling. Acta Materialia 46, 1025–43. Christian, J. and S. Mahajan (1995). Deformation twinning. Progress in Materials Science 39, 1–157.

258

Nanomaterials

Conrad, H. and J. Narayan (2000). On the grain size softening in nanocrystalline materials. Scripta Materialia 42(11), 1025–1030. Cumings, J. and A. Zettl (2000). Low-friction nanoscale linear bearing realized from multiwall carbon nanotubes. Science 289(5479), 602–604. Cumings, J. and A. Zettl (2004). Localization and nonlinear resistance in telescopically extended nanotubes. Physical Review Letters 93(8). 086801. Dao, M., L. Lu, Y. F. Shen, and S. Suresh (2006). Strength, strain-rate sensitivity and ductility of copper with nanoscale twins. Acta Materialia 54(20), 5421–5432. Dingreville, R., J. Qu, and M. Cherkaoui (2005). Surface free energy and its effect on the elastic behavior of nano-sized particles, wires and films. Journal of the Mechanics and Physics of Solids 53(8), 1827–1854. Gao, H., Y. Huang, W. D. Nix, and J. W. Hutchinson (1999). Mechanism-based strain gradient plasticity - i. theory. Journal of the Mechanics and Physics of Solids 47(6), 1239–1263. Gianola, D. S., S. Van Petegem, M. Legros, S. Brandstetter, H. Van Swygenhoven, and K. J. Hemker (2006). Stress-assisted discontinuous grain growth and its effect on the deformation behavior of nanocrystalline aluminum thin films. Acta Materialia 54(8), 2253–2263. Greer, J. R. and W. D. Nix (2006). Nanoscale gold pillars strengthened through dislocation starvation. Physical Review B 73(24), 245410. Greer, J. R. (2006). Bridging the gap between computational and experimental length scales: A review on nanoscale plasticity. Reviews on Advanced Materials Science 13(1), 59–70. Hu, N. and J. Molinari (2004). Shear bands in dense metallic granular materials. Journal of the Mechanics and Physics of Solids 52, 499–531. Huang, Z. M., Y. Z. Zhang, M. Kotaki, and S. Ramakrishna (2003). A review on polymer nanofibers by electrospinning and their applications in nanocomposites. Composites Science and Technology 63(15), 2223–2253. Iwashita, K. and M. Oda (1998). Rolling resistance at contacts in simulation of shear band development by dem. Journal of Engineering Mechanics-ASCE 124(3), 285–292. Jia, D., K. T. Ramesh, and E. Ma (2003). Effects of nanocrystalline and ultrafine grain sizes on constitutive behavior and shear bands in iron. Acta Materialia 51(12), 3495–3509. Jiang, H., M. F. Yu, J. Q. Lu, Y. Huang, H. T. Johnson, X. G. Zhang, and P. Ferreira (2007). Carbon nanotube electronic displacement encoder with sub-nanometer resolution. Journal of Computational and Theoretical Nanoscience 4(3), 574–577. Joshi, S. and K. Ramesh (2008a). Grain size dependent shear instabilities in bcc and fcc materials. Materials Science and Engineering A 493, 65–70. Joshi, S. P. and K. T. Ramesh (2008b). Rotational diffusion and grain size dependent shear instability in nanostructured materials. Acta Materialia 56(2), 282–291. Joshi, S. P. and K. T. Ramesh (2008c). Stability map for nanocrystalline and amorphous materials. Physical Review Letters 101(2), 025501. Kibey, S., J. B. Liu, D. D. Johnson, and H. Sehitoglu (2007). Predicting twinning stress in fcc metals: Linking twin-energy pathways to twin nucleation. Acta Materialia 55(20), 6843–6851. Kis, A., K. Jensen, S. Aloni, W. Mickelson, and A. Zettl (2006). Interlayer forces and ultralow sliding friction in multiwalled carbon nanotubes. Physical Review Letters 97(2), 025501. Liao, X. Z., F. Zhou, E. J. Lavernia, S. G. Srinivasan, M. I. Baskes, D. W. He, and Y. T. Zhu (2003). Deformation mechanism in nanocrystalline al: Partial dislocation slip. Applied Physics Letters 83(4), 632–634. Malow, T., C. Koch, P. Miraglia, and K. Murty (1998). Compressive mechanical behavior of nanocrystalline fe investigated with an automated ball indentation technique. Materials Science & Engineering A A252, 36–43. Moldovan, D., D. Wolf, and S. Phillipot (2001). Theory of diffusion-accomodated grain rotation in columnar polycrystalline microstructures. Acta Materialia 49, 3521–3532. Murayama, M., J. Howe, H. Hidaka, and S. Takaki (2002). Atomic-level observation of disclination dipoles in mechanically milled nanocrystalline fe. Science 295, 2433–2435.

7 Scale-Dominant Mechanisms in Nanomaterials

259

Nix, W. D., J. R. Greer, G. Feng, and E. T. Lilleodden (2007). Deformation at the nanometer and micrometer length scales: Effects of strain gradients and dislocation starvation. Thin Solid Films 515(6), 3152–3157. Pantano, A., M. C. Boyce, and D. M. Parks (2004). Mechanics of axial compression of single and multi-wall carbon nanotubes. Journal of Engineering Materials and Technology-Transactions of the Asme 126(3), 279–284. Rosakis, P. and H. Y. Tsai (1995). Dynamic twinning processes in crystals. International Journal of Solids and Structures 32(17–18), 2711–2723. Sanvicens, N. and M. P. Marco (2008). Multifunctional nanoparticles - properties and prospects for their use in human medicine. Trends in Biotechnology 26(8), 425–433. Spitzig, W. and A. Keh (1970). The effect of orientation and temperature on the plastic flow properties of iron single crystals. Acta Metallurgica 18(6), 611–622. Stronge, W. (2000). Impact Mechanics. Cambridge: Cambridge University Press. Sun, P., E. Cerreta, G. Gray, and J. Bingert (2006). The effect of grain size, strain rate, and temperature on the mechanical behavior of commerical purity aluminum. Metallurgical and Materials Transactions -A 37A, 2983–2994. Van Swygenhoven, H. and A. Caro (1997). Plastic behavior of nanophase ni: A molecular dynamics computer simulation. Applied Physics Letters 71(12), 1652–1654. Van Swygenhoven, H. and P. Derlet (2001). Grain boundary sliding in nanocrystalline fcc metals. Physical Review-B 64(224105), 1–9. Wei, Q., D. Jia, K. Ramesh, and E. Ma (2002). Evolution and microstructure of shear bands in nanostructured fe. Applied Physics Letters 81(7), 1240–1242. Yamakov, V., D. Wolf, S. R. Phillpot, A. K. Mukherjee, and H. Gleiter (2002). Dislocation processes in the deformation of nanocrystalline aluminium by molecular-dynamics simulation. Nature Materials 1(1), 45–48. Yang, S. and B. Bacroix (1996). Shear banding in strain hardening polycrystals during rolling. International Journal of Plasticity 12(10), 1257–1285. Yu, M. F., B. S. Files, S. Arepalli, and R. S. Ruoff (2000a). Tensile loading of ropes of single wall carbon nanotubes and their mechanical properties. Physical Review Letters 84(24), 5552–5555. Yu, M. F., O. Lourie, M. J. Dyer, K. Moloni, T. F. Kelly, and R. S. Ruoff (2000b). Strength and breaking mechanism of multiwalled carbon nanotubes under tensile load. Science 287(5453), 637–640. Zhang, H., B. E. Schuster, Q. Wei, and K. T. Ramesh (2006). The design of accurate microcompression experiments. Scripta Materialia 54(2), 181–186. Zhu, X. H., J. E. Carsley, W. W. Milligan, and E. C. Aifantis (1997). On the failure of pressuresensitive plastic materials .1. models of yield & shear band behavior. Scripta Materialia 36(6), 721–726.

...In that Empire, the craft of Cartography attained such Perfection that the Map of a Single province covered the space of an entire City, and the Map of the Empire itself an entire Province. In the course of Time, these Extensive maps were found somehow wanting, and so the College of Cartographers evolved a Map of the Empire that was of the same Scale as the Empire and that coincided with it point for point. Jorge Luis Borges, On Exactitude in Science

8

Modeling Nanomaterials 8.1 Modeling and Length Scales The length scales that are of interest to the mechanics of materials were discussed in Chapter 1. The focus there was on the devices and structures that typified the various length scales, and the intent was to define the range appropriate for nanomaterials. In subsequent chapters, typical mechanics approaches to characterizing and describing nanomaterials were considered. As in any field of science, however, we cannot truly understand nanomaterials unless we can develop mathematical models for them. This chapter focuses on the modeling approaches that are appropriate for nanomaterials. It is worthwhile to consider explicitly the modeling process, in order to understand the various approximations being made. This process is represented schematically in Figure 8.1. The process is as follows: • One starts with a physical problem of interest – let us say this is the problem of nanoindentation of a diamond Berkovich tip into a single crystal of copper. • Immediately, the necessity for some form of idealization presents itself: for example, we do not know what the friction is between the nanoindenter tip and the sample, so we will typically idealize the problem as a zero friction problem at the interface. This is a physical idealization made for modeling purposes. A set of such physical idealizations results in an idealized problem that we still think is of interest. • Next, we replace the physical problem with a mathematical problem: we choose the set of equations, with boundary and initial conditions, that we think best represents the idealized physical problem. Thus we may choose to represent the problem through Newton’s equations for the motion of the individual atoms, with the atomic interactions specified through an interatomic potential (e.g., the Lennard-Jones potential). This translation into a mathematical problem represents another set of assumptions: we assume the potential is reasonable for copper (the Lennard-Jones potential is not), and that we can ignore quantumlevel effects. • Now that we have a mathematical problem that in some sense represents our physical problem, we choose a numerical approach to solving the mathematical K.T. Ramesh, Nanomaterials, DOI 10.1007/978-0-387-09783-1 8, c Springer Science+Business Media, LLC 2009 

261

262

Nanomaterials

problem: by choosing, for instance, a specific code with all of the specific algorithms incorporated in the code. There are different levels of accuracy with which specific numerical schemes will solve the same set of mathematical equations, so again we have a set of now numerical approximations. • We then implement the numerical scheme and consider the solutions. Physical Problem

Idealized Physical Problem

Mathematical Problem

Numerical Problem

Numerical Solution

Examine (e.g. visualize) Solutions Fig. 8.1 Typical sequence of steps involved in modeling a mechanics of nanomaterials problem. Note the many layers of approximations involved, pointing out the danger of taking the results of simulations at face value.

It is important to remember when we look at the solutions that we have made (a) physical, (b) mathematical, and (c) numerical approximations, and we should always ask whether we have ended up solving the problem that we set out to solve. This is sometimes stated as the verification and validation problem for complex codes: in the words of Rebecca Brannon, are we solving the right equations, and are we solving the equations right? Eventually, the simulation results must be compared to experiments or to the results of entirely independent simulations. Consider again the range of length scales presented in Chapter 1. Some of the modeling approaches that can be used to describe phenomena over this range of length scales are presented in Figure 8.2, together with the corresponding suite of materials (rather than mechanical) characterization techniques. As an example, we discuss the use of this figure to understand the plastic deformation of a material. At very fine scales below those shown in Figure 8.2, the quantum mechanics problem must be solved, leading to calculations that are variously called ab initio or first principles computations (such calculations account for electronic structure). Slightly larger scale calculations (Figure 8.2) account for every atom (but not the electrons) by assuming an interatomic potential between the atoms and then solving the equations of motion for every atom. Such computations are called molecular dynamics calculations, and are generally limited by length scale and time scale (a mil-

263

Discrete Dislocation Dynamics

Continuum Modeling

Modeling

Molecular Dynamics

10–6 10–8 10–10 m

AFM Observations

Enriched CM

Finite Element Methods

10–2 10–4

SEM TEM

nanomaterials

Optical Microscopy

100

Radar

10+2

8 Modeling Nanomaterials

Fig. 8.2 The typical modeling approaches of interest to the mechanics of nanomaterials, and the approximate length scales over which each approach is reasonable. Note the significant overlap in length scales for several of the modeling approaches, leading to the possibility of consistency checks and true multiscale modeling. An example of the observations that can be made at each length scale is also presented, from a materials characterization perspective.

lion atom computation, with 100 atoms on the side of a cube, would be examining a physical cube that is about 40 nm on a side). Although computations of this size scale can address a number of important issues, they are incapable of accounting for a large number of dislocations. Discrete dislocation dynamics simulations are larger scale computations that no longer have atomic resolution but that can track many thousands of dislocations.

264

Nanomaterials

Such computations are able to capture the development of some dislocation substructures, and to interrogate possible sources and sinks of dislocations. At larger scales still, continuum modeling computations such as crystal plasticity calculations do not attempt to resolve individual dislocations but do resolve individual crystals. Such calculations smear the dislocations into the continuous medium representing each crystal, and replace the dislocation evolution with hardening laws on given slip systems. At even larger scales, the individual crystals are no longer resolved, and instead the computations examine the deformations of a homogeneous continuum that has some prescribed plastic behavior (such as those discussed in Chapter 2). This is the level of classical continuum mechanics, which contains no internal length scales. Some models incorporate ad hoc microstructural length scales into structured continuum calculations, and we refer to these as enriched continuum models (for example, the strain gradient models in vogue in 2008). A computational approach derived from classical continuum mechanics that has been used with some success over a very wide range of length scales is the finite element method, particularly when coupled to computational techniques that are better able to resolve very fine scale physics. We will not discuss this numerical technique in any detail: there are many excellent books on the finite element method and its applications in mechanics. An interesting observation from Figure 8.2 is the overlap between the scales at which each modeling approach is effective. These overlapping regions can be viewed in two ways. First, these are regions where information on the behavior of the material can be obtained using two different approaches, and such situations often greatly enhance our understanding. Second, these are regions in which the ability to resolve behavior at finer scales with one method while computing over larger scales with another method can lead to multiscale approaches that are able to capture phenomena that would otherwise be missed. The range of typical interest in nanomaterials is shown in the big vertical box in Figure 8.2. Larger scales may be of interest for applications of nanomaterials, but the box describes the range where nanomaterials phenomena are determined, i.e., scale-dominant behaviors are typically observed. The microstructural characterization techniques associated with this range have been discussed in Chapter 3. All of the modeling approaches in the figure are useful for modeling nanomaterials, although it is clear that part of this size range is accessible only by molecular dynamics. The range of length scales where a number of modeling approaches overlap demonstrates that there is significant potential for understanding nanomaterials using multiscale approaches. Another way to think about these length scales is to think about them in terms of the features of the material that are important at each length scale. This is illustrated in Figure 8.3, which shows the microscopic features typically associated with each length scale in metallic systems. At very small length scales, the behavior is dominated by atomic motions and so the atom is the critical feature. As the length scales increase, the arrangements of the atoms become important, and so we begin to see the effects of atomic clusters, point defects such as vacancies, and the unusual arrangement of atoms around dislocation cores.

8 Modeling Nanomaterials

Features

Heterogeneities: pores, inclusions, particle Clusters, Point Stacking Faults Defects, Grain ai boundaries,, grain size, e texture Dislocation Dislocation substructure: cells, tangles, cores Boundary conditions, component geometry, walls A constraints, residual stresses Atoms Dislocations Twins 10 –10 m

1 10–8

10–6

10–4

10–2

100

10+2

Fig. 8.3 The length scales associated with various microstructural features in metallic materials. For most crystalline materials, the behavior at larger length scales is the convolution of the collective behavior of features at smaller length scales.

265

266

Nanomaterials

At slightly greater length scales, dislocations themselves become the features that dominate the mechanical behavior, and dissociated dislocations lead to stacking faults. We then begin to see the influence of dislocation substructures, that is, the collective behavior of large numbers of dislocations: this includes the development of dislocation tangles, dislocation walls, cells and sub-grains. At larger scales still, deformation twins begin to become important in some materials. Grain boundary effects, grain size effects, and texture effects can occur over a very wide range of length scales. At similar lengthscales, one begins to see the influence of heterogeneities in the material, such as inclusions, pores and precipitates. Most component features such as boundary conditions, geometry, constraints and residual stresses are at the larger scales in this figure. We have discussed most of these mechanisms and features in previous chapters. From a modeling perspective, what becomes clear is that it is not possible to link directly actions and events at the atomic scale to mechanical behavior at the macro scale without accounting for the specific microstructural features that may develop at intermediate scales. The mechanical behavior at larger scales is modulated by the collective behavior of features and mechanisms at a number of smaller scales – it does not make sense to try to compute the behavior of a steel structure in terms of atomistic calculations unless one is certain that one can also capture all of the features of dislocation motion, dislocation tangling, and dislocation-grain boundary interactions. Microscale and mesoscale interactions are affected by atomistic behavior, but must be specifically accounted for as well in order to predict the material behavior. Now that we have some insight into the physical processes active at each length scale, let us think again about the process of modeling discussed in Figure 8.1. Once we have chosen the physical problem that we would like to understand, we typically need to idealize that physical problem into something that we can actually solve. This process of idealization tends to be discipline-specific, in that the idealizations chosen by materials scientists tend to be different from the idealizations chosen by physicists or mechanical engineers. When we think more clearly about these idealizations, we realize that very often what we are doing is deciding which features of the deformation (Figure 8.3) our model is intended to capture. Thus, for example, we may choose to construct a model that accounts for dislocation density but does not account for individual dislocations or of the distributions of dislocation density associated with dislocation cell walls and dislocation tangles. How do we decide which features we intend to include in the model? Often these are based on experimental observations that indicate that these particular microstructural features are (or are not) observed in the phenomenon or behavior that we wish to describe using the model. For example, suppose we are attempting to describe the deformations of an aluminum alloy. The modeler may know that dislocations are important in plastic deformations, and so she builds a model that accounts for dislocation density. Her model may not be capable of describing the clumping of dislocation density that characterizes dislocation cell walls, and she may believe this to be a reasonable idealization of the physics of deformation. However, TEM observations show that dislocation cells are important in describing the large plastic

8 Modeling Nanomaterials

267

deformations of aluminum, and the model will obviously not account for this feature of the deformation. The modeler cannot, in general, put into the model microstructural features that she does not know exist. There is one modeling approach that should in principle account for all possible microstructural features, even those that the modeler does not consciously put into the model. In this approach, one accounts at a fundamental level for quantum mechanical interactions (some in the literature refer to this as a “first principles” approach), and then computes everything else explicitly from that basis. In principle, an explicit computation of this type that accounts for all of the possible interactions between the atoms should be able to reproduce all the possible behaviors at all possible length and timescales. In practice, of course, one is never likely to have the computing capacity to actually perform such a calculation and so this might be an exercise in futility. However, the advantage of such an approach (even over limited scales) is that one might discover phenomena that one would otherwise never know about, because the fact is that performing experiments over a complete range of length scales and time scales is also prohibitively expensive, and so there are certainly phenomena that we have never observed simply because we have not known to look for them. A computation that handles the true complexity of nature at the fundamental level may thus truly discover the existence of phenomena, and we could then design an experiment to examine the veracity of such a discovery. This complexity-aware approach to modeling is extraordinarily expensive from a computational viewpoint, but has a certain intellectual appeal, and is likely to be of increasing importance as computational capabilities escalate. The major modeling approaches used to study nanomaterials will be described in this chapter, with an emphasis on ultimately providing descriptions that can be viewed in a mechanics context. First, however, we discuss some of the scaling approximations and physical idealizations made by classical continuum mechanics.

8.2 Scaling and Physics Approximations All models must make approximations in order to be useful. The question, of course, is how does one decide which approximations to make? In this section, we attempt to address this question with respect to nanomaterials. Since most of us approach the nanomaterials problem with a prior training in the macroscale sciences, it is important to remember that some phenomena at the nanoscale may have no macroscale counterparts, and we will not naturally think about them in terms of scaling down from the macroscale. Continuum models (such as the mechanics models that we normally use at the macroscale) usually make four core approximations: • It is assumed that classical mechanics (rather than quantum mechanics) represents the behavior. This will clearly be an issue when considering very small particles, such as quantum dots.

268

Nanomaterials

• Material behavior is assumed to hold pointwise in a continuous medium, where every material particle has an infinitesimal size. Thus the atomic or molecular structure is averaged out in this representation. • Thermal effects are averaged out (we do not consider thermal vibrations of atoms or molecules explicitly). • The chemistry is usually ignored, although it can be incorporated in some average sense. The second and third items above both relate also to the concepts of differentiability and localization of equations (that is, when can we write the governing equations as field equations valid at every point?). An interesting example of the influence of such approximations is that of the scaling of thermal effects. In solid mechanics, temperature shows up primarily in terms of thermal conduction as a diffusion term (or as a parameter that modifies material properties, e.g., through thermal softening). Thermal diffusion is represented mathematically in continuum mechanics through a second derivative (as in Equation 2.59). The idea of differentiability shows up here – is the second derivative well-defined (i.e., is the first derivative sufficiently continuous for us to be able to take a second derivative)? Physically, one can only even define the diffusion term if the temperature is smooth over a sufficiently large length scale: thermal energy is moved through the system by atomic or molecular collisions. As we consider smaller size scales, the mean free path of the molecule between collisions can become an issue – if the size scale of interest is comparable to the mean free path, the continuum definition of thermal diffusion is no longer sensible. Further, situations with strong gradients (such as interfaces and surfaces) begin to dominate behavior at small length scales. For these reasons, for example, the definition of thermal conduction along and across a carbon nanotube cannot be the continuum definition. Thus the assumptions implicit in continuum mechanics models must often be challenged on a case-by-case basis when dealing with nanomaterials. The process of integrating information from multiple scales into a model to understand the behavior of a material (or more generally a system) is called multiscale modeling. Traditionally, all length scales (and timescales) below the current scale in a simulation are called subscales, and models are frequently said to represent various levels of subscale information in specific ways. We discuss some of these approaches next.

8.3 Scaling Up from Sub-Atomic Scales There is a significant scientific effort focused towards the goal of being able to simulate the behavior of human-scale objects using atom-scale information. While there is an excellent argument to be made for invoking complexity only when one needs it, there is also the excellent argument that our intuition is so bad at the nanoscale that we do not know what complexity to invoke, and so we may learn a great deal by incorporating the complexity of atomic structure as the foundations of a modeling

8 Modeling Nanomaterials

269

approach. But how do we feasibly scale up from atomistics? There are many ways of approaching this issue, but for the purposes of this book we identify two specific approaches: the enriched continuum approach, and the molecular mechanics approach.

8.3.1 The Enriched Continuum Approach This approach to the problem is often taken by those scientists coming to the problem from a continuum scale, and essentially views the equations of continuum mechanics as core equations. That is, it is assumed that the equations of continuum mechanics (such as the governing equations presented in Section 2.3) will be used for the solution of the problems of interest. This, of course, presupposes that the problems of interest are at ≥ nm scales, which is certainly the case for this book. Given this assumption, atomic-level information primarily enters into the system of equations through the constitutive equation. This approach therefore views the constitutive equation as “informing” other continuum equations of the microscopic degrees of freedom in an averaged sense. Note that this has always been done at the microstructural level (that is our constitutive equations, such as in plasticity, have incorporated microstructural information such as the grain size). However, we recognize that the microscopic degrees of freedom that we choose to incorporate can also come from the quantum (sub-atomic) level. This approach results in the incorporation of quantum-level information into continuum mechanics problems in a well-defined and consistent way. It has the advantage of being a great approach for the modeling of mechanics of materials, but it also has the disadvantage that we must do averaging, and depending on the way we take the average, we may end up averaging out many of the effects of interest to nanoscale phenomena. In terms of the topics of interest to this book, the informed continuum approach may be very effective for bulk nanomaterials but may miss important phenomena in discrete nanomaterials (an excellent example being quantum dots viewed as nanoparticles - it is the quantum effects that are key in this case). An excellent and detailed description of these issues is provided by Phillips (2001).

8.3.2 The Molecular Mechanics Approach This approach gets its name from the idea that the mechanics problem is considered at the molecular level. However, the fundamental concept here is that we must make relevant approximations at very small scales (specifically, sub-atomic scales such as electrons and nuclei) to develop the equations appropriate at the next scale up. We can then calculate interatomic or intermolecular interactions directly, and use these to define interatomic or intermolecular potentials. In a subsequent step, to

270

Nanomaterials

move to the continuum scale, we can use the interatomic potentials to compute the behavior of many-atom systems. The primary benefit of this approach is that it handles the molecular structure directly, and so leads to efficient molecular dynamics calculations. Such an approach has the advantage that it is not tied to continuum dynamics (it does not assume the standard continuum equations), but has the disadvantage that the computations are much more complex, and therefore the physical limitations of the computing system may result in the simulations being scale limited. For example, as of this writing it is difficult to simulate a billion atom system, which amounts to a cube of side ≈300 nm. For us, the primary interest is in interatomic bonds and bond strengths (which arise naturally from electron cloud interactions). It is immediately obvious that the problem quickly becomes intractable if we were to attempt to handle the interactions of every atom with every other atom in terms of electron cloud interactions. However, we can recast this problem in terms of short-range and longrange interactions, and handle each of these separately as approximations. That is, we do not need to consider short range interactions between every atom pair in the simulation, but only those between nearby atoms (within some specified zone of influence defined by the range of the short-range interaction). Even the longrange interactions will typically have some cut-off distance, beyond which they are negligible. How are these short-range and long-range interactions obtained? Typically, one starts with Schrodinger’s equation from quantum mechanics, which is something the reader has probably seen for a single particle interacting with a potential V (r,t): −

∂ ψ (r,t) ℏ2 2 ▽ ψ (r,t) +V (r,t)ψ (r,t) = iℏ , 2m ∂t

(8.1)

where r is the position and m is the mass of the particle, ℏ is Planck’s constant, and ψ (r,t) is the particle wave function. This latter quantity is most easily understood by noting that in one dimension |ψ |2 dx is the probability that the particle is between x and x + dx. The quantity |ψ |2 dx is called the probability density function for the particle. The right-hand-side of this equation reduces to E ψ (r) in the timeindependent case, where E is an energy (really an energy eigenvalue – one obtains different solutions ψα corresponding to energies Eα ). We can write a similar equation to Equation (8.1) for an N-particle system, although it is somewhat more complex, and we would have to invoke the Pauli exclusion principle. The point, however, is that this can be done (writing the equation, that is – solving it can quickly become difficult). In the case of interacting charged particles, for instance, the potential V (r,t) would be qi q j , (8.2) V (r,t) = ∑ 4πε0 di j

8 Modeling Nanomaterials

271

where qi is the charge on the ith particle, and di j is the distance between the ith and jth particles. For example, in the simple 1D case of a particle oscillating within a quadratic potential well, we have V (r,t) = V (x,t) = 21 kx2 , and the now time-independent Schrodinger equation reduces to −

ℏ2 d 2 ψ (x) 1 2 + kx ψ (x) = E ψ (x), 2m dx2 2

(8.3)

which is the equation to a harmonic oscillator and is easily shown to have solutions

ψn (x) =

1 1

2n/2 n! 2



mω πℏ

1 4

    mω mω 2 x Hn x , exp − 2ℏ ℏ

(8.4)

with the corresponding energies being quantized as En = (n + 21 )ℏω , n being an  integer. In the latter two equations, the frequency ω is defined by ω = mk , exactly as in the classical spring-mass system. The point of presenting this solution is to demonstrate that the Schrodinger equation behaves a little differently from the usual continuum spring-mass system, in that the energy states are discrete rather than continuous (only certain energies are allowed). Further, the probability of finding the particle (after all, this is what is defined by ψn ) is non-zero at all x even for the discrete states. That is, you may find the particle anywhere, but it can only have certain discrete energy states. Some of the quantum character of the solution becomes evident. The fact is that Schrodinger’s equation is often written but rarely solved in books on things nano, particularly when it comes to books of this type that attempt to bring together multiple disciplines. This is because while Schrodinger’s equation is easy to write, it quickly becomes difficult to solve for a many-body system, such as, for example, a cluster of 10 aluminum atoms with all of the nuclei and electrons. Thus various approximations become necessary. One typical approximation is called the Born-Oppenheimer approximation, and uses the fact that the nuclei of atoms are much heavier than electrons, so that in most problems of interest the nuclei move slowly in comparison with the electrons. Thus, by separating the problem into the rapid motions of electrons and slower motions of nuclei, one can compute the electron ψ and energy E in a field of initially fixed nuclei. Then one updates the positions of the nuclei, and recomputes ψ and E. In most cases, the problem reduces to finding the total energy ET of a collection of particles as a function of the positions of the nuclei Rn and electrons re : ET = ET (Rn , re ).

(8.5)

Note that the number of nuclei and number of electrons are usually very different in the problems relevant to materials. We can make some further approximations. If, for instance, we decide that we will approximate the total energy of the nuclei and electrons given by Equation (8.5) as

272

Nanomaterials

ET (Rn , re ) ≈ Eion (Rn ),

(8.6)

we are essentially replacing the total energy of the system with an energy that is calculated from the positions of the nuclei alone, and this is like computing an effective interaction between ions. If, on the other hand, we replace the large number of electrons instead with an electron density function ρ (r), we obtain ET (Rn , re ) ≈ Eapprox (Rn , ρ (r)),

(8.7)

and approaches that solve problems using this approximation are called density functional theories. A particularly interesting example of the approximation corresponding to Equation (8.6) is obtained when one sets the energy Eion to be dependent only on the relative distances between particles di j , that is, ignoring directional influences, as Eion =

1 V (di j ), 2∑

(8.8)

where V (di j ) is called a pair potential, and di j is the scalar distance between ion i and ion j. Many such pair potentials are used in the literature, with the canonical example being the Lennard-Jones potential:  12  6  ξ ξ − , V (d) = V0 d d

(8.9)

Energy, U(r)

with V0 defining the depth of the potential well and ξ defining the length scale corresponding to the equilibrium separation of the atoms. This potential represents the interatomic interaction presented in Figure 4.9, which is reproduced here as Figure 8.4 for convenience.

r0

Distance, r

Fig. 8.4 Schematic of the Lennard-Jones interatomic pair potential V (d).

In principle, this pair potential still involves interactions between every pair of atoms in the simulation. In practice, most simulations assume that the potential dies

8 Modeling Nanomaterials

273

off after the nearest neighbors, so that only a small number of interactions need to be included for each atom in the simulation. A very large fraction of the calculations in the literature use such pair potentials. Increasingly sophisticated potentials are also used, incorporating some of the directional dependences of the bonds. Another approximation of interest is the embedded atom method (EAM), which was developed for metals (Daw and Baskes, 1984) and essentially considers each metal atom to be embedded within an electron density field derived from the rest of the metal: Etotal = Einteraction of nuclei + Eembedding of 1 = ∑ φ (di j ) + ∑ F(ρn ), 2

atom in electron gas

(8.10) (8.11)

√ where the electron density is ρ . For example, in the special case of F(ρ ) = ρ , we obtain the Finnis-Sinclair potential. EAM potentials are extensively used in simulations of the deformation of metals, and much of the literature on the atomistic modeling of nanocrystalline metals is based on such potentials. The parameters for such potentials are obtained in a variety of ways, and the usefulness of the simulation is typically determined by the quality of the potential that is used. Good EAM potentials are available for a small number of metals, including copper, nickel and aluminum. An example of such a potential is that developed by Mishin et al. (1999), and the resulting pair potential for aluminum is presented in Figure 8.5. 0.25 Aluminum EAM

0.20

Potential (eV)

0.15 0.10 0.05 0.00 –0.05 –0.10

2

3 4 5 6 Distance between atoms (angstrom)

7

Fig. 8.5 The pair potential corresponding to the EAM potential for aluminum, using the parameters provided by Mishin et al. (1999).

In the next section, we consider computations involving multiple atoms that use such potentials in order to understand material behavior. Such computations that include the dynamics are called molecular dynamics.

274

Nanomaterials

8.4 Molecular Dynamics In a molecular dynamics (MD) calculation, one seeks to determine the motions of all atoms in the simulation by solving Newton’s equations of motion (rather than quantum mechanical equations), with the atoms interacting through assumed interatomic potentials (such as that in Equation 8.8). Aside from length scales, time scales also become an issue. Very small integration time steps are needed in order to obtain an accurate solution, and most MD calculations never go beyond a nanosecond in total time duration. In order to achieve the strain of interest in that time frame, MD computations typically run at equivalent strain rates on the order of 109 s−1 . This section will not attempt to show the reader how to perform MD simulations, but rather to survey some of the issues involved as regards such simulations for nanomaterials. An excellent and relatively concise introduction to molecular dynamics simulations is provided by Rapaport (2004). The basic computational process is as follows: one starts with the positions and velocities of a number of atoms (preferably a large number). The potential energy of the system is used to compute the force on each atom using an assumed interatomic potential. Since the masses of the atoms are known, these calculated forces allow us to compute the accelerations of each atom, and these accelerations can be used to update the velocities of the atoms, and the velocities are used to update the atomic positions. We then begin the process again. These can be very large simulations, depending on the number of atoms involved, given that every atom interacts at least with its nearest neighbors, and depending on the total times (total number of timesteps) that must be computed. There are very sophisticated codes now available in the public domain for solving problems using molecular dynamics. The best known of these is called LAMMPS, and is able to run effectively on a parallel processing platform (many computers working in parallel in terms of the numerical algorithms used for solving MD problems). Given the availability of such codes, molecular dynamics simulations are now in the reach of any scientist willing to make the effort. As always with scientific computations, however, the key is to understand (a) precisely what is the scientific objective of the calculation (what are you looking for?), (b) precisely what is being computed (what equations are being solved numerically?) and (c) when can the computation provide reasonable results (what are the limits of the computational approach?). There is a significant danger in using canned computational packages, in that the user tends to assume that the results are always reasonable, while the developers may not have provided warnings or flags to indicate when the output is inaccurate or even meaningless. There is a significant level of research activity, as of this writing, that uses molecular dynamics simulations to enhance our understanding of nanomaterials. Some of the work has been on the special case of carbon nanotubes, which involve carboncarbon interactions. Since such interactions are covalent and quite directional, a classic pair potential involving only the atomic separation is not appropriate. Modifications of pair potentials that include some directional characteristics (sometimes called bond-angle character in the literature) include the Tersoff (Tersoff, 1988)

8 Modeling Nanomaterials

275

and Brenner (Brenner et al., 2002) potentials. In the case of bulk nanocrystalline materials, most of these simulations have been on metallic nanomaterials, using embedded atom method potentials. EAM potentials have been constructed and validated to various degrees for copper, nickel, and aluminum (e.g., Mishin et al., 1999). As is generally the case in nanomaterials, fcc metals are the best studied. Reviews of MD simulations in nanomaterials appear in the literature every two years or so, a recent example being Van Swygenhoven and Weertman (2006). An example of a three-dimensional nanostructure developed for solution using molecular dynamics is presented in Figure 8.6. The figure shows several tens of grains of nanocrystalline nickel constructed using a Voronoi tessellation approach, and modeled using the LAMMPS software. The colors represent the coordination number associated with each atom, and the grain boundaries are identified by the distinctly different coordination numbers of the atoms compared with the atoms in the face-centered cubic crystals. This collection of atoms can now be subjected to mechanical loading (deformations) and the motions of the individual atoms can be tracked to understand deformation mechanisms.

Fig. 8.6 Nanocrystalline material constructed using molecular dynamics. The material is nickel, and the atoms are interacting using an EAM potential due to Jacobsen. About 30 grains are shown. The color or shade of each atom represents its coordination number (number of nearest neighbors), with the atoms in the crystals having the standard face-centered-cubic coordination number. The change in coordination number at the grain boundaries is clearly visible. This collection of atoms can now be subjected to mechanical loading (deformations) and the motions of the individual atoms can be tracked to understand deformation mechanisms.

276

Nanomaterials

What have we learned, and what can we learn, from MD simulations of nanomaterials? In principle, of course, such simulations have tremendous potential for understanding the physics of these materials, because they give us access to time scales and length scales that are extremely difficult to investigate through experiments. We will discuss some of the advances in our understanding obtained through MD computations in the next paragraphs. The limitations of these calculations arise also from the scales at which they operate, and the continuing limitations of computing capacity. In general, larger simulations (in terms of number of atoms or length scale) are run for shorter total times because of the available computer cycles. Even the largest simulations that are currently feasible are unable to sample a cubic micron of volume for a total time of 1 ns. As a consequence, any physical mechanism that requires significant time to develop or significant space to organize itself cannot be captured using such simulations (although many MD aficionados continue to extrapolate from short times and small scales). Some other distortions of the physics can also occur. To generate significant strains in very short times, extraordinarily high strain rates are applied (almost always higher than 107 s−1 ), and thus the mechanisms that are observed in an MD simulation will not necessarily be observed in a physical experiment at strain rates many orders of magnitude smaller. Further, the dislocations that are developed in MD simulations often have extremely high accelerations, which can generate some pathological situations in terms of mechanics. Thermal equilibration of systems is a continuing challenge, since there is insufficient time for typical thermal diffusion mechanisms; some ingenious approaches have been developed to define a reasonable thermal state (Liu et al., 2004). The small sample sizes that are simulated and the sometimes two-dimensional nature of the simulations can restrict the number of available slip systems and deformation mechanisms. Most MD simulations also do not account for the initial defect structure that exists in real materials, and therefore miss many of the sources and sinks that are active in many nanomaterials (e.g., most of the simulated grains are completely clean and contain no initial defect structures, while most experimentally observed nanocrystals contain large numbers of initial defects such as stacking faults and pre-existing dislocations). Issues with regard to the types of grain boundaries considered must also be addressed. Finally, the nature of the (often periodic) boundary conditions that are applied can couple into the dynamics, causing difficulties with modeling some phenomena that are driven by wave dynamics. Even with all of these limitations, however, MD simulations have been very useful for understanding nanomaterial behavior. For example, such simulations (e.g., Schiotz and Jacobsen, 2003) have indicated that there may in fact be a maximum in the strengthening that can be obtained by decreasing the grain size, so that below a certain critical grain size the strength begins to decrease again as the grain size decreases. This is called the inverse Hall-Petch effect and is somewhat controversial, in that it has been difficult to obtain incontrovertible experimental evidence that the effect exists (because of specimen preparation and testing difficulties). The MD results (such as those of Schiotz and Jacobsen, 2003) demonstrate that there are at least some physical reasons why such an effect might exist. Further, such simula-

8 Modeling Nanomaterials

277

tions have demonstrated that grain boundaries can be significant sources of dislocations (Farkas and Curtin, 2005) and that there are competing dislocation nucleation mechanisms that may become dominant as the grain size changes (Yamakov et al., 2001). The likely dominance of partial dislocations rather than full dislocations in nanocrystalline aluminum has been predicted by MD and subsequently observed through experiments (Chen et al., 2003a). Grain boundary sliding has been shown to be a plausible mechanism for effective plastic deformation in very small grain size metals (Derlet et al., 2003). The simulations also show that dislocations may not survive in the interior of small grains without externally applied stresses, that is, they may be emitted and then reabsorbed by grain boundaries (Yamakov et al., 2001). As a consequence, not seeing dislocations in the interior of nanograins within the TEM does not guarantee that dislocation-mediated plastic deformation did not occur. This is an insight that could not have been obtained definitively without the MD simulation. It is apparent that such simulations continue to have great potential in this field. However, the limitations of strain rate, time scale and length scale must be carefully considered in attempting to relate MD simulations results to experimental measurements and to predictions of engineering behavior and possible applications. In the next section we discuss another modeling approach that examines larger scales in space and time.

8.5 Discrete Dislocation Dynamics We have seen that the length scales that can be simulated with molecular dynamics are relatively small,