Industrializing Additive Manufacturing: Proceedings of AMPA2023 (Springer Tracts in Additive Manufacturing) [1st ed. 2024] 3031429826, 9783031429828

Citation preview

Springer Tracts in Additive Manufacturing

Christoph Klahn Mirko Meboldt Julian Ferchow   Editors

Industrializing Additive Manufacturing Proceedings of AMPA2023

Springer Tracts in Additive Manufacturing Series Editor Henrique de Amorim Almeida, Polytechnic Institute of Leiria, Leiria, Portugal

Editorial Board Members Abdulsalam Abdulaziz Al-Tamimi, Riyadh, Saudi Arabia Alain Bernard, Ecole Centrale de Nantes, IRCCyN UMR CNRS 6597, Nantes Cedex 03, France Andrew Boydston, University of Washington, Seattle, USA Bahattin Koc, Maltepe, Sabanci University, Istanbul, Türkiye Brent Stucker, Louisville, KY, USA David W. Rosen, Atlanta, GA, USA Deon de Beer, Bloemfontein, South Africa Eujin Pei , College of Engineering, Design and Physical Sciences, Brunel University London, London, UK Ian Gibson, University of Twente, Enschede, Overijssel, The Netherlands Igor Drstvensek, University of Maribo, Maribor, Slovenia Joaquim de Ciurana, University of Girona, Girona, Spain Jorge Vicente Lopes da Silva, CTI Renato Archer, Campinas, São Paulo, Brazil Paulo Jorge da Silva Bártolo, Nanyang Technological University, Singapore, Singapore Richard Bibb, Loughborough University, Leicestershire, UK Rodrigo Alvarenga Rezende, Uniara, Araraquara, Brazil Ryan Wicker, University of Texas at El Paso, El Paso, TX, USA

The book series aims to recognise the innovative nature of additive manufacturing and all its related processes and materials and applications to present current and future developments. The book series will cover a wide scope, comprising new technologies, processes, methods, materials, hardware and software systems, and applications within the field of additive manufacturing and related topics ranging from data processing (design tools, data formats, numerical simulations), materials and multi-materials, new processes or combination of processes, new testing methods for AM parts, process monitoring, standardization, combination of digital and physical fabrication technologies and direct digital fabrication.

Christoph Klahn · Mirko Meboldt · Julian Ferchow Editors

Industrializing Additive Manufacturing Proceedings of AMPA2023

Editors Christoph Klahn Karlsruhe Institute of Technology Eggenstein-Leopoldshafen, Germany

Mirko Meboldt Product Development Group Zurich pd|z ETH Zürich Zürich, Switzerland

Julian Ferchow Design for New Technologies inspire AG Zürich, Switzerland

ISSN 2730-9576 ISSN 2730-9584 (electronic) Springer Tracts in Additive Manufacturing ISBN 978-3-031-42982-8 ISBN 978-3-031-42983-5 (eBook) https://doi.org/10.1007/978-3-031-42983-5 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland Paper in this product is recyclable.

Preface

The 3rd AMPA Conference (Additive Manufacturing in Products and Applications) focused on the ongoing opportunities and challenges of the industrialization of Additive Manufacturing (AM). During the three years after the 2nd AMPA Conference, AM has emerged as an industrial production technology in many business areas. Therefore, AM is facing new challenges in terms of cost models, mass production quality, end-to-end digital value chains and process chain integration. The industry has faced many challenges in recent years. Disruptions in global supply chains, rising energy and raw material costs, lack of skilled workers and economic uncertainty were dominant topics. In addition, the need to reduce CO2 emissions leads to new regulations in the field of industrial sustainability. AM has shown that it has the potential to tackle the challenges by fast and automated production and digital value chains. Some industries already implemented AM on high technology readiness levels, while others are still exploring the value opportunities for their application. The exchange of knowledge between industries helps to promote AM industrialization and provide unprecedented opportunities for further applications. To further advance the emerging fields of AM industrialization, new challenges need to be addressed scientifically. The topics of the AMPA2023 Conference cover all fields necessary to develop and produce innovative end-user products: Design tools & methods, emerging technologies, manufacturing process chains, teaching and training and business cases of AM applications.

All scientific contributions are double-blind peer-reviewed by the members of the industry committee for their relevance to industry and by the members of the scientific committee for their scientific quality. In a two-staged process, 28 contributions were selected out of 55 submitted abstracts.

vi

Preface

We thank everyone who contributed to the success of the Additive Manufacturing in Products and Applications Conference: Thanks to the authors for their valuable papers and talks, to the members of the industrial and scientific committees for their hard but fair reviews and for chairing sessions, to the participants of the sessions for the fruitful discussions and last but not least to all those who supported the conference in the background.

Contents

Design for AM 1 Novel Design Method for Manufacturing Internal Cooling Channels Additively with a Functional Support Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bhupesh Verma, Johannes Willkomm, and Johannes Henrich Schleifenbaum Comparative Evaluation of Optimization Algorithms for Automatic Build Orientation for Powder Bed Fusion of Metals Using a Laser Beam . . . . . . . . . . . . Leonie Pauline Pletzer-Zelgert, Sebastian Dirks, Corinna Müller, and Johannes Henrich Schleifenbaum Review and Development of Design Guidelines for Additive Tooling of Injection Molds Using PolyJet Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stefan Junk, Steffen Schrock, and Nico Schmieder

3

19

35

Design for AM 2 Design of Additively Manufactured 3D Lattice Cores of Sandwich Panels . . . . . Hussam Georges, Christian Mittelstedt, and Wilfried Becker Designing Variable Thickness Sheets for Additive Manufacturing Using Topology Optimization with Grey-Scale Densities . . . . . . . . . . . . . . . . . . . . . . . . . . Felix Endress and Markus Zimmermann Sustainability-Oriented Topology Optimization Towards a More Holistic Design for Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Klaus Hoschke, Konstantin Kappe, Sankalp Patil, Sebastian Kilchert, Junseok Kim, and Aron Pfaff

49

63

77

Process Chain 1: Digital Process Chain Integration of the Whole Digital Chain in a Unique File for PBF-LB/M: Practical Implementation Within a Digital Thread and Its Advantages . . . . . . . . . Konstantin Poka, Benjamin Merz, Martin Epperlein, and Kai Hilgenberg

91

Approach to an Automated Method for Load-Optimized Design of Multimaterial Joints for Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . 115 Christoph Leupold and Maren Petersen

viii

Contents

Uncoupling Development Time from the Size of a Library of AM Parts Through Complexity Reduction and Modeling of Topology Optimization Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Guilain Lang, Gerald Perruchoud, David Novo, and Stephane Brun Process Chain 2: Physical Process Chain Fabrication Forecasting of LPBF Processes Through Image Inpainting with In-Situ Monitoring Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Hans Aoyang Zhou, Song Zhang, Marco Kemmerling, Daniel Lütticke, Johannes Henrich Schleifenbaum, and Robert H. Schmitt The Influence of Nozzle Size on the Printing Process and the Mechanical Properties of FFF-Printed Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Joakim Larsson, Per Lindström, Christer Korin, Jens Ekengren, and Patrik Karlsson Systematical Assessment of Automation Potential in Additive Manufacturing Process Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Julian Ulrich Weber, Hanna Jörß, and Mirco Jankowiak Emerging AM Technologies Numerical Modeling of Part Formation in Volumetric Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 Roozbeh Salajeghe, Daniel Helmuth Meile, Carl Sander Kruse, Deepak Marla, and Jon Spangenberg Cost Saving Potential of a Shell-Core Strategy of Combined Powder Bed Fusion of Metals with Laser Beam and Hot Isostatic Pressing . . . . . . . . . . . . . . . . 198 Lukas Bauch, Leonie Pauline Pletzer-Zelgert, and Johannes Henrich Schleifenbaum Automated Design of 3D-Printed Silicone Parts: A Case Study on Hand Rehabilitation Gloves for Stroke Patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Felix Weigand, Julia Föllmer, and Arthur Seibel Investigation of the Feasibility to Process NiTi Alloys with Powder Bed Fusion for Potential Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 Rico Weber, Adriaan B. Spierings, and Konrad Wegener

Contents

ix

Simulation Design for Additive Manufacturing and Finite Element Analysis of Fe-Mn Biodegradable Fracture Fixation Plate with Varying Porosity Levels . . . . . . . . . . 239 Mustafiz Shaikh, Fadi Kahwash, Zhilun Lu, Mohammad Alkhreisat, and Islam Shyha Laser Powder Bed Fusion Processing Simulation of Simple Geometries in Inconel 738 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 Raphael Gloor, Matthias Fankhauser, Jakob Benz, Matthias Hoebel, and Kaspar Löffel Toward Automated Topology Optimization: Identification of Non-Design Features of CAD Models Using Graph Neural Networks . . . . . . . . . . . . . . . . . . . . 267 Michael Jasinski, Fabian Schöfer, and Arthur Seibel Teaching and Training Investigating the Use of Augmented Reality Head-Mounted Displays to Teach Design for Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 Gustavo Melo, Rohit Ravi, Lucas Jauer, and Johannes Henrich Schleifenbaum Skills Requirements of Additive Manufacturing - A Textual Analysis of Job Postings Using Natural Language Processing . . . . . . . . . . . . . . . . . . . . . . . . 299 Gustavo Melo, Melisa Chaves, Moritz Kolter, and Johannes Henrich Schleifenbaum Evaluating Digital Assistance in Form of Augmented Reality for Manual Processes in the Metal Binder Jetting Process Chain . . . . . . . . . . . . . . . . . . . . . . . . 317 Johannes Helmholz and Maximilian Vogt Process Innovations for Applications Effect of Targeted Porosity in Additively Manufactured Heat Pipes . . . . . . . . . . . 337 Stefan Reich, Daniel Bold, and Johannes Henrich Schleifenbaum On the Effectiveness of Triply-Periodic Minimal Surface Structures for Heat Sinks Used in Automotive Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 Martin Czekalla and Arthur Seibel

x

Contents

Soft Magnetoactive Morphing Structures with Self-Sensing Properties, Using Multi-Material Extrusion Additive Manufacturing . . . . . . . . . . . . . . . . . . . . 365 Somashree Mondal, Michał Kwa´sniowski, Antonia Georgopoulou, Bogdan Sapi´nski, Thomas Graule, and Frank Clemens Use Cases Additive Manufacturing in Gas Cleaning Applications . . . . . . . . . . . . . . . . . . . . . . 389 Felix Reinke, Christian Straube, Jörg Meyer, and Achim Dittler Evaluation of the Ultra-High Vacuum Suitability of Laser Powder Bed Fusion Manufactured Stainless Steel 316L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407 Pascal Romanescu, Daniel Omidvarkarjan, Julian Ferchow, and Mirko Meboldt Exploring the Integration of Additive Manufacturing: Lessons Learned and Success Factors of Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423 Christopher Gustafsson Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441

Design for AM 1

Novel Design Method for Manufacturing Internal Cooling Channels Additively with a Functional Support Structure Bhupesh Verma(B) , Johannes Willkomm , and Johannes Henrich Schleifenbaum Digital Additive Production (DAP), RWTH Aachen University, Campus Boulevard 73, 52074 Aachen, Germany [email protected]

Abstract. Laser Powder Bed Fusion (PBF-LB/M) is a popular additive manufacturing (AM) technology for producing complex metal components. This layerby-layer manufacturing process enables the production of intricate internal structures that conventional manufacturing (CM) processes cannot manufacture. Heat exchangers and heat sinks represent an area of application that can benefit significantly from novel internal structures. Due to the geometric freedom offered by PBF-LB/M, new solutions are possible for designing heat sinks, such as complex flow paths or increased surface-to-volume ratio, which can be used to increase heat transfer efficiency. Despite the design freedom offered by PBF-LB/M, design restrictions also need to be considered. One example of these restrictions is the requirement for support structures while manufacturing overhang features. Internal structures are an example of such features. The support structures must be mechanically removed after manufacturing, which becomes complicated or sometimes impossible, particularly for internal structures. To fully exploit the design freedom, this work focuses on a novel design method to enable support-free production of complex internal cooling structures via PBF-LB/M. The proposed solution in this work is to use the necessary support structures during manufacturing as the cooling structures, which can be adapted to the requirements of the heat sinks or heat exchangers to increase thermal efficiency. In this work, five cooling structures are investigated using conjugate heat transfer (CHT) simulation to maximize the ratio of heat transfer to pressure loss. Test samples are manufactured from AlSi10Mg via PBF-LB/M and examined using optical measurements to validate the manufacturability and dimensional accuracy. Finally, a cylinder head is used to demonstrate how a cooling structure suitable for a real-life application can be designed using the proposed design methodology in this work. Keywords: Laser powder bed fusion · Internal cooling channels · Support structures · Heat Exchanger

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 3–18, 2024. https://doi.org/10.1007/978-3-031-42983-5_1

4

B. Verma et al.

1 Introduction Additive manufacturing (AM) has the potential to facilitate significant improvement for the next generation of efficient heat exchangers. Heat exchangers have previously, and in most cases still, relied on conventional manufacturing (CM) methods such as milling, die-casting, drilling, brazing/welding, or a combination of processes to mass-produce cost-efficient products [1]. Conventional compact heat exchangers, such as microchannel heat exchangers, use fins to augment heat transfer and are manufactured using stamping or folding techniques [2]. These methods restrict the geometry alteration, size, and thickness of features, such as fin thickness, that can be manufactured. AM technology can overcome these design restrictions. ASTM has defined additive manufacturing (AM) as “a process of joining materials to make objects from 3D model data, usually layer upon layer, as opposed to subtractive manufacturing methodologies” [3]. Among various AM technology, manufacturers are particularly taking advantage of laser-based powder bed fusion of metals (PBF-LB/M), an AM technology for manufacturing metallic components [4]. Since the components are built by adding successive layers in PBF-LB/M, designers can prepare innovative monolithic designs with complex internal geometries, which, together with the available variety of materials, can facilitate the production of heat exchangers that use less material, have lower volume, and have increased thermal performance and reliability [5]. Various studies have been performed to distinguish the effect of different geometries on heat transfer and pressure drop performance [6–8]. Wong et al. [1] experimentally demonstrated the advantages of using AM technology for novel heat sink designs by using structures such as a staggered elliptical array for both the pressure drop and heat transfer capabilities of a heat sink. More recently, a simulative study exploited the advantages of AM for designing heat sinks using eight different geometrical shapes of heat sink fins, and their characteristics in air flow, convection, and pressure drop were studied. NACA airfoil and square fin geometries provided a 63% and 64% decrease in pressure drop per diverted energy compared to control plate fins [9]. Kim et al. [10] developed a new design method for generating compact heat exchangers (CHX) consisting of triply periodic minimal surfaces (TPMS) core structures. The innovative CHX was manufactured using AM, and the experimental studies showed that including TPMS structures could create an ultra-efficient CHX while maintaining allowable pressure drop. Despite the general design freedom offered by the PBF-LB/M technology, certain design restrictions and guidelines must be considered while designing the components. For instance, the PBF-LB/M process typically manufactures components with higher surface roughness than CM techniques, which can enhance heat transfer. According to Ventola et al. [11], introducing artificial surface roughness increased convective heat transfer by 63% for flat surfaces and 35% for finned characters. Another example is the need for support structures to manufacture components with overhang features, as illustrated in Fig. 1, which is especially relevant for internal cooling channels. Support structures have a massive effect on required material and post-processing steps, and in the case of internal cooling channels, removing the support structures after manufacturing is nearly impossible due to their poor accessibility. Some other considerations include the orientation of manufacturing via PBF-LB/M and powder removability from the internal channels.

Novel Design Method for Manufacturing Internal Cooling Channels

5

part support structure

substrate

Fig. 1. Schematic representation of support structures for PBF-LB/M

Reducing or avoiding the support structures required for internal structures is advantageous as it can radically reduce the required effort for post-processing components [12, 13]. Although many studies have been conducted for improving heat exchangers using PBF-LB/M, only a few have concentrated on support structure removal and designing heat exchangers focused on manufacturability. In an attempt to contribute to this research gap, this study aimed to develop a design methodology for structures with dual functionality, i.e., that serve as both support structures during manufacturing and as heat sink fins in the final part. The dual objective was to enhance heat transfer while ensuring manufacturability via PBF-LB/M. The paper is organized into four sections. Section 2 starts by describing the design and selection of the functional support structure. After this, the setup used for conjugate heat transfer (CHT) simulation and the manufacturing of the samples is explained. Section 3 compares the designed structures’ performance using the CHT simulation results. Additionally, it evaluates the manufactured samples’ manufacturability and implements the developed methodology to construct a cooling jacket for a cylinder head. Finally, Sect. 4 summarizes the main findings of this work and offers potential future research topics.

2 Methodology This section discusses the methods to design, evaluate, and manufacture internal cooling channels’ cooling structures. The design starts with identifying requirements to ensure a targeted and systematic arrangement of the internal cooling structures, which depends on various factors such as available design space, pressure drop, temperature limitations, and manufacturability using PBF-LB/M. In the next step, cooling structures are selected for further investigation, identifying critical geometric features and creating a parametric design. Following the design, geometries are prepared for numerical analysis via CHT simulation. These simulations subsequently allow the evaluation of the cooling performance and pressure drop of interior cooling channels using the selected cooling structures. Finally, the manufacturing of the samples with AlSi10Mg via PBF-LB/M is discussed briefly.

6

B. Verma et al.

2.1 Design The design of the internal cooling structures starts with determining the requirements for the cooling structures. For the design of a heat sink, such as in this study, heat transfer efficiency has a higher priority than minimizing the pressure drop. Additionally, current work focuses on manufacturing the component via PBF-LB/M, so manufacturability and dimensional accuracy are critical parameters. This focus means the design restrictions must be considered in the structure design process.

5 mm

6 mm

Symmetry

b) Rounded rectangle

a) Cylinder

d) NACA Airfoil

c) Ellipse

e) Lattice Structure (f cc ) 2

z

Fig. 2. Selected cross-sections/geometries for designing functional support structures

For this study, five inner structures are chosen to analyze heat transfer capabilities, shown in Fig. 2. To evaluate the structures, the comparability of the analysis results must be ensured. This is achieved by analyzing structures with similar geometric parameterizations, meaning structures are designed with the same heat-transferring cross-sectional area. Since cellular structures (Fig. 2e) do not have a uniform cross-sectional area, the average cross-section area is kept constant while designing. This procedure is equivalent to determining a designed structure’s relative density (volume fraction). For each geometry, three relative densities – 10%, 15%, and 20% – are considered to analyze the effectiveness of cooling structures, as this can help optimize the required material for support structures to minimize the deformation during PBF-LB/M. Relative density (ratio of inner structure’s cross-section area to the area of the unit cell Aref ) is calculated

Novel Design Method for Manufacturing Internal Cooling Channels

7

w.r.t a 6 × 5 mm unit cell (Aref = 30 mm2 ) as shown in Fig. 2a. The length of the rectangular and NACA airfoil geometry is fixed to be 5 mm. The details of the geometric dimensions are presented in Table 1. Table 1. Dimensions calculated for the selected geometries and relative density Relative density/ cross-section

10%

20%

30%

Cylinder (radius) [mm]

0.977

1.200

1.380

Rounded rectangle (width) [mm]

0.616

0.938

1.269

Ellipse (eccentricity)

1.210

1.480

1.710

NACA airfoil (thickness ratio)

0.176

0.264

0.352

Lattice structure (strut thickness) [mm]

0.494

0.624

0.739

2.2 Numerical Study Conjugate Heat Transfer (CHT) simulations were prepared for calculating the pressure drop and heat dissipation using ANSYS Fluent®. To minimize the computational effort required for simulating all the considered geometries, periodic boundary conditions are used to reduce the computational domain size. This boundary condition ensures that the side walls will not influence the pressure drop and heat transfer. Additionally, an inlet and outlet area are extended on both sides of the cooling structure to ensure that a fully developed flow enters the heat sink and that the outlet does not affect the flow within, as illustrated in Fig. 3. The Reynolds number is calculated at the inlet (shown in Fig. 3) using Eqs. 1 and 2, to ensure that the results can be compared for various structures. A sand grain roughness model with a typical roughness value for PBF-LB/M components [14, 15] is used to consider the surface roughness for the additively manufactured components. The relevant information for the simulation methodology is summarized in Table 2. After setting up the domain and boundary conditions, the mesh independence study is performed with a rounded rectangle cross-section (rel. Density 20%) to determine the sensitivity of a computational fluid dynamics (CFD) solution to the mesh size and to minimize the influence of the mesh on the numerical solution. A mesh independence study was not performed for each geometry, and the same mesh size is used for all other geometries to reduce the computational requirements. The average and maximum face temperature of the top interface face between fluid and solid is considered for this study. A mesh size of 0.175 mm was finally chosen. The results from the mesh-independence study are illustrated in Fig. 4.

here, ρwater is the density, and μwater is the dynamic viscosity of water, vin inlet velocity and dh,in is the hydraulic diameter calculated using the formulation by Huisseune

8

B. Verma et al.

et al. [16]. Ac is the area of cross-section, L is the length, and As is the wetted perimeter of the inlet channel.

Fig. 3. Geometry used for CHT simulation, including the symmetry plane and boundary conditions

2.3 Manufacturing The LPBF system used in this work to produce the test specimens for manufacturability and the demonstrator is an EOS M 290 from the company EOS GmbH, equipped with a 400 W ytterbium fiber laser. The material used is AlSi10Mg. The samples are manufactured with the standard process parameter supplied by EOS for 30 μm layer thickness. To be able to evaluate the manufacturability of the functional support structures and the support effect within a cooling channel, test specimens were designed, built up, and measured. Figure 5 shows the sample design. These samples replicate an internal cooling channel and consist of channel top (1), channel bottom (2), investigated fin structure (3), and side walls (4). For evaluating the manufacturability and suitability of support structures of the considered structures, the specimens produced are 3D scanned to compare them to the original CAD model and analyzed for any deviations. This comparison is conducted with the software GOM Inspect V8. The produced samples distorted the side walls and the channel top. This distortion was used as a quantifiable deviation variable. With the help of this analysis, the support effect of the cooling structures and, thus, the manufacturability are evaluated.

Novel Design Method for Manufacturing Internal Cooling Channels Table 2. Details of the numerical model and the material properties for CHT simulation Turbulence model

k- Turbulence model with 3% inlet turbulence and scalable wall functions

Thermal boundary conditions

Top of the cooling channel: 373 K The fluid temperature at the inlet: 298 K

Reynolds number

500, 2000, 5000, 9000

Fluid properties

Material: water Thermal conductivity = 0,6 Wm−1 K −1 Density = 998,2 k gm−3 Specific heat capacity = 4182 J kg −1 K −1 Viscosity: temperature-dependent [17]

Solid properties

Material: AlSi10Mg Thermal conductivity = 173 ± 10 Wm−1 K −1 Density = 2670 k gm−3 Specific heat capacity = 890 ± 50 J kg −1 K −1 [18]

Mass flow rate (calculated according to Re)

0.75 g/s, 3.009 g/s, 7.52 g/s, 13.54 g/s

Surface roughness

117 μm

Fig. 4. Results of the mesh-independence study

9

10

B. Verma et al.

Fig. 5. Dimension of the sample (in mm) for analyzing manufacturability

3 Results and Discussion This section compares the numerical results for the pressure drop and heat transfer for the five inner structures. This is followed by manufacturing the considered structures via PBF-LB/M and comparing them to the CAD model. Finally, a test case for designing a cylinder head is presented. 3.1 Functional Support Structure Performance The result of the temperature distribution within the fluid (along with streamlines) and the solid region from the CHT simulation for the cylindrical inner structure with Re = 500 is shown in Fig. 6 and Fig. 7. In these results, it can be seen that the temperature of the fluid increases as it flows toward the outlet. Heat flux enters the fluid at the fluid-solid interface and is transported away, cooling the interface. The turbulence the support structure generates can improve the heat transfer at a higher Reynolds number. The increased surface area due to the support structure also enhances the heat transfer between the fluid and the solid. To compare the various inner structures in this work, the heat dissipation and the pressure drop for all the structures with a relative density of 10% are plotted in Fig. 8 and Fig. 9. It can be seen from these results that the cooling capacity of the inner structures increases with Reynolds number, as hypothesized previously. Furthermore, it can be observed that cellular structure (f2 ccz ) has the highest heat dissipation, on the other hand, the classical structures have a lower pressure drop. The most notable, rounded rectangle and NACA airfoil show the lowest pressure drop. In order to find the best structure that enhances the heat dissipation with low-pressure drop, the ratio of the heat dissipation to pressure drop is tabulated in Table 3. The heat dissipation was increased by a factor of approx. 2.5 as compared to a cooling channel with no support structure while pressure drop was within the allowed limits. From these results, it can be concluded that the NACA airfoil internal structure offers the best trade-off between heat dissipation and pressure drop. Finally, the results show that using the presented design and simulation

Novel Design Method for Manufacturing Internal Cooling Channels

11

methodology, a suitable structure can be chosen depending on the requirement of the real-life application. Results for the other structures can be found in Appendix.

Fig. 6. Temperature distribution within the fluid region; along with streamlines

Fig. 7. Temperature distribution within the solid and fluid domain at mid-plane

3.2 Manufacturability The quantifiable parameters - distortion of the side walls and the channel top - are used to investigate the manufacturability as described in Sect. 2.3. To measure the distortion of the side walls and the channel top, the CAD model and 3D scan of the specimen with a rounded rectangle inner structure (20% relative density) are compared, as illustrated in The measured average distortions for all the manufactured structures are shown in Fig. 11 and Fig. 12. These results show that the cooling structure’s relative density affects the cooling channel’s manufacturability, as lower-density structures showed relatively higher distortion. High-density structures show less distortion due to their higher stiffness as well as better dissipation of energy during PBF-LB/M. When analyzing the different geometries, it was observed that the classical structures have higher distortion of the side walls than cellular structures. A possible explanation for this could be the lower stiffness of the structure, as, unlike cellular structure, there is no connection between individual

12

B. Verma et al.

Fig. 8. Heat dissipation vs Reynolds number for all structures at 10% rel. density

Fig. 9. Pressure drop vs Reynolds number for all structures at 10% rel. density

support structures. The f2ccz lattice structure with 20% relative density is found to have the overall lowest deviation (Fig. 10).

Novel Design Method for Manufacturing Internal Cooling Channels

13

Table 3. The ratio of heat dissipation to pressure drop for all structures Re

500

f2 CCz

Ellipse

Cylinder

NACA airfoil

Rounded rectangle

10% 15% 20% 10% 20%

20%

10% 15% 20% 10% 15% 20% 10% 15% 20%

2.1

1.7

11.9

9.8

8.5

0.15

1.8

1.6

1.5

1.2

1.04 0.9

4.5

3.6

2.7

2.9

2.4

1.9

5000 0.08 0.06 0.04

0.7

0.62

0.5

0.43 0.36 0.3

1.6

1.3

0.9

1.1

0.9

0.8

9000 0.04 0.03 0.02

0.4

0.3

0.27 0.21 0.17 0.14

0.8

0.7

0.5

0.6

0.5

0.4

2000 0.24 0.2

1.3

13.8 12.9

7.5

19.8 18.3 15.8 18.6 17.3 15.1

Fig. 10. Distortion of the side wall (red) and top wall (orange) after manufacturing for rounded rectangle

Distortion [mm]

Distortion of the top wall

f2ccz

10 % relative density

Ellipse

Cylinder

15 % relative density

NACA

Rounded rectangle

20 % relative density

Fig. 11. Average distortion of the top wall for all the structures

3.3 Case Study To demonstrate the transferability of the results generated in this study to an actual engineering application, a test case to re-design and manufacture a cylinder head is chosen in this study. Figure 13 shows the considered part of the cylinder head for re-designing, highlighted in red. As illustrated in Fig. 14, manufacturing the cylinder head via PBFLB/M is impossible without support structures in the area designed for passing the cooling fluid. Therefore, following the results of this study, functional NACA airfoil support

14

B. Verma et al.

Distortion [mm]

Distortion of the side wall

f2ccz

10 % relative density

Ellipse

Cylinder

15 % relative density

NACA

Rounded rectangle

20 % relative density

Fig. 12. Average distortion of the side wall for all the structures

structures were designed to ensure increased heat transfer capability and manufacturability via PBF-LB/M without a significant increase in pressure drop. Figure 15 illustrates the CAD model for the cylinder head, including the functional support structures and the successfully re-designed and manufactured cylinder head via PBF-LB/M.

Fig. 13. Part of the cylinder head considered for re-designing (highlighted in red)

Fig. 14. Challenging feature present in the cylinder head; cannot be manufactured without support structure via PBF-LB/M

Novel Design Method for Manufacturing Internal Cooling Channels

30 mm

15

Z Y X

Fig. 15. CAD model of the re-designed cylinder head highlighting re-designed regions (top left); 3D scan of the cylinder head manufactured with AlSi10Mg via PBF-LB/M (top right) and the manufactured component at the bottom.

4 Conclusion and Outlook The critical findings of this study, as well as future research possibilities, can be summarized as follows: 1. A design methodology is presented for the components - with overhang features by including the support structure to enhance the functionality (heat dissipation) and ensure the manufacturability via PBF-LB/M. 2. The design methodology presented in this work is especially relevant for highperformance heat exchangers with internal cooling channels which can be manufactured via PBF-LB/M. 3. The proposed design methodology reduces the amount of required post-processing for the removal of support structures after manufacturing. 4. The design methodology is implemented to re-design a part of a cylinder head with cooling jackets and manufacture it successfully. 5. In this work, CHT simulations are compared qualitatively. These should be validated using experiments to further improve the simulation model and prediction. 6. The design method presented in this work can be extended further to include various other structures, such as gyroid and BCC/FCC lattice structures, for designing high-performance heat exchangers while ensuring their manufacturability using AM technology. Acknowledgment. The authors gratefully acknowledge the financial support of the Federal Ministry of Education and Research (BMBF) as this work was performed within the scope of the project MAYFEST – Maßgeschneiderte Legierungsentwicklung einer hochwarmfesten Aluminiumlegierung für die hybride additive Fertigung von Komponenten im Antriebsstrang (03XP0168E).

16

B. Verma et al.

We also thank our colleagues, Dr. Sebastian Bold, for providing valuable feedback on the script and Mr. Jonas-Bitter Davidts and Mr. Sreekar Sajjala Reddy, for helping with the simulation setup and experiments.

Appendix (See Figs. 16, 17, 18 and 19).

Heat dissipation [W]

15% relative density

Reynolds number [-] f2ccz

Ellipse

Cylinder

NACA

Rounded Rectangle

Fig. 16. Heat dissipation vs Reynolds number for five structures at 15% rel. Density

Pressure drop [Pa]

15% relative density

Reynolds number [-] f2ccz

Ellipse

Cylinder

NACA

Rounded Rectangle

Fig. 17. Pressure drop vs Reynolds number for five structures at 15% rel. Density

Novel Design Method for Manufacturing Internal Cooling Channels

Heat dissipation [W]

20% relative density

Reynolds number [-] f2ccz

Ellipse

Cylinder

NACA

Rounded Rectangle

Fig. 18. Heat dissipation vs Reynolds number for five structures at 20% rel. Density

Pressure drop [Pa]

20% relative density

Reynolds number [-] f2ccz

Ellipse

Cylinder

NACA

Rounded Rectangle

Fig. 19. Pressure drop vs Reynolds number for five structures at 20% rel. Density

17

18

B. Verma et al.

References 1. Wong, K.K., Ho, J.Y., Leong, K.C., et al.: Fabrication of heat sinks by Selective Laser Melting for convective heat transfer applications. Virtual Phys. Prototyp. 11, 159–165 (2016). https:// doi.org/10.1080/17452759.2016.1211849 2. Arie, M.A., Shooshtari, A.H., Dessiatoun, S.V., et al.: Performance characterization of an additively manufactured Titanium (Ti64) heat exchanger for an air-water cooling application. ASME 2016 Heat Transfer Summer Conference collocated with the ASME 2016 Fluids Engineering Division Summer Meeting and the ASME 2016 14th International Conference on Nanochannels, Microchannels, and Minichannels (2016). https://doi.org/10.1115/HT20161059 3. Alcisto, J., Enriquez, A., Garcia, H., et al.: Tensile properties and microstructures of laserformed Ti-6Al-4V. J. Mater. Eng. Perform. 20, 203–212 (2011). https://doi.org/10.1007/s11 665-010-9670-9 4. Schneck, M., Gollnau, M., Lutter-Günther, M., et al.: Evaluating the use of additive manufacturing in industry applications. Procedia CIRP 81, 19–23 (2019). https://doi.org/10.1016/ j.procir.2019.03.004 5. Kaur, I., Singh, P.: State-of-the-art in heat exchanger additive manufacturing. Int. J. Heat Mass Transf. 178, 121600 (2021). https://doi.org/10.1016/j.ijheatmasstransfer.2021.121600 6. Sahiti, N.: Thermal and fluid dynamic performance of pin fin heat transfer surfaces. PhD (2006) 7. Khan, W.A., Culham, J.R., Yovanovich, M.M.: The role of fin geometry in heat sink performance. J. Electron. Packag. 128, 324–330 (2006). https://doi.org/10.1115/1.2351896 8. Behnia, M., Copeland, D., Soodphakdee, D.: A comparison of heat sink geometries for laminar forced convection: numerical simulation of periodically developed flow. In: ITherm’98. Sixth Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (Cat. No.98CH36208). IEEE (null) 9. Storfeldt, J.: Design benefits with Additive Manufacturingfrom a convective heat transfer perspective (2019) 10. Kim, J., Yoo, D.-J.: 3D printed compact heat exchangers with mathematically defined core structures. J. Comput. Des. Eng. 7, 527–550 (2020). https://doi.org/10.1093/jcde/qwaa032 11. Ventola, L., Chiavazzo, E., Calignano, F., et al.: Heat transfer enhancement by finned heat sinks with micro-structured roughness. J. Phys. Conf. Ser. 494, 12009 (2014). https://doi.org/ 10.1088/1742-6596/494/1/012009 12. Obeidi, M.A.: Metal additive manufacturing by laser-powder bed fusion: guidelines for process optimisation. Results Eng. 15, 100473 (2022). https://doi.org/10.1016/j.rineng.2022. 100473 13. Ceccanti, F., Giorgetti, A., Citti, P.: A support structure design strategy for laser powder bed fused parts. Procedia Struct. Integrity 24, 667–679 (2019). https://doi.org/10.1016/j.prostr. 2020.02.059 14. Thomas, D.: The Development of Design Rules for Selective Laser Melting. PhD, University of Wales, National Centre for Product Design & Development Research (2009) 15. Adams, T., Grant, C., Watson, H.: A simple algorithm to relate measured surface roughness to equivalent sand-grain roughness. Int. J. Mech. Eng. Mechatron. 1, 66–71 (2012) 16. Huisseune, H., T’Joen, C., de Jaeger, P., et al.: Performance enhancement of a louvered fin heat exchanger by using delta winglet vortex generators. Int. J. Heat Mass Transf. 56, 475–487 (2013). https://doi.org/10.1016/j.ijheatmasstransfer.2012.09.004 17. Korson, L., Drost-Hansen, W., Millero, F.J.: Viscosity of water at various temperatures. J. Phys. Chem. 73, 34–39 (1969). https://doi.org/10.1021/j100721a006 18. EOS GmbH EOS Aluminium-AlSi10Mg: Data Sheet

Comparative Evaluation of Optimization Algorithms for Automatic Build Orientation for Powder Bed Fusion of Metals Using a Laser Beam Leonie Pauline Pletzer-Zelgert(B) , Sebastian Dirks , Corinna Müller, and Johannes Henrich Schleifenbaum RWTH Aachen University, Digital Additive Production, Campus Boulevard 73, 52074 Aachen, Germany [email protected]

Abstract. For additive manufacturing (AM) using Powder Bed Fusion of Metals using a Laser Beam (PBF-LB/M), digital preparation steps (part design, part orientation in the chamber, supporting, etc.) are required. In the current state of the art digital process chains, these steps are done manually by an engineer, utilizing specialized AM preparation software, e.g., Autodesk Netfabb or Materialise Magics. Especially the part orientation has a significant impact on part quality and costs. However, setting up a suitable orientation algorithm is a complex and multi-dimensional engineering problem, so computationally intensive evaluation procedures are necessary. The orientation selection is usually done by comparing the pre-evaluated part orientations relative to each other. The selection strategy thus significantly influences the quality of the orientation schemes and the computational effort. The selection strategy often follows brute force logic to investigate the entire orientation space. This approach divides the solution space into equidistant steps that limit the search resolution. Therefore, this paper conducts a comparative evaluation of different new strategies to replace the brute force algorithm with a suitable optimization algorithm that selects the calculated partial orientations based on a priori decision logic. Three optimization algorithms Covariance Matrix Adaption Evolution Strategy (stochastic method), Multilevel Coordinate Search (direct search method), and Efficient Global Optimization (surrogate model optimization) are evaluated regarding their orientation evaluation results. The number of objective function evaluations required is compared to brute force. Each orientation optimization algorithm is applied to 42 parts. All algorithms reduce the number of orientation evaluations compared to the discrete method of brute force, but differ in the evaluated orientation quality. The Covariance Matrix Adaption Evolution Strategy produced the best-evaluated orientation but also required the highest number of orientation evaluations. In the future, the work should help to find a suitable optimization algorithm for the automatic part orientation finding for PBF-LB/M. Keywords: Powder Bed Fusion of Metals using a Laser Beam · orientation algorithm · optimization algorithm

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 19–34, 2024. https://doi.org/10.1007/978-3-031-42983-5_2

20

L. P. Pletzer-Zelgert et al.

1 Introduction Powder Bed Fusion of Metals using a Laser Beam (PBF-LB/M) is one of the most common Additive Manufacturing (AM) processes for high-quality metal parts [1]. Because of the significant investment requirements for PBF-LB/M machines, many industrial customers make use of AM service providers for manufacturing. In the typical digital process preparation chain in this scenario, multiple steps by the two participants (customer and AM service provider) are required to prepare a part geometry, as shown in Fig. 1. First, the part must be designed with a Computer-aided Design (CAD) program by the customer. This might include additional simulation and design iteration steps to optimize the geometry for AM. The orientation approach presented in this paper concentrates on the automatic orientation of a finished design, the use case from the perspective of the AM service provider. The part design is not automatically adapted to optimize the results or consider limitations specific to the AM manufacturing technology, such as stable heat conduction during the build process or the accessibility of supported part areas [2]. This approach is necessary in many scenarios where there is no design freedom, for example in spare part production. In any scenario where design freedom can be utilized, the orientation optimization presented can instead be repeatedly used as one step in an iterative, partially manual design improvement process.

Customer

AM service provider

CAD

Tesselated Export

Orientation

Support Generation

Nesting

Slicing

Additive Manufacturing

Post Processing

Fig. 1. PBF-LB/M digital process chain

After the design process is finished the geometry is typically transferred to the AM service provider, which requires an export of the geometry. At the AM service provider site, the part geometry is imported into a specialized AM build preparation software (e.g. Autodesk Netfabb, Materialise Magics). The part is oriented in the chamber manually or by using the algorithms provided in the software. For PBF-LB/M, additional support structures are required for overhanging surfaces. Supports dissipating heat from the Laser

Comparative Evaluation of Optimization Algorithms for Automatic

21

source into the build plate as well as sustaining strain induced by temperature gradients during cooldown to prevent part warpage [3]. After orientation and support generation, the required number of parts needs to be nested into build volumes and scheduled for printing. After printing, the parts need to be cut off the build plate and supports must be removed in post processing. [4]. Software assisted automation of the digital process steps has the potential to increase the consistency of the printing result quality and reduce manual labor costs. Here, we focus on algorithmic optimization of the orientation step in the process chain. In an automated solution, the different influences of the part orientation on the process are evaluated based on calculated evaluation criteria from the literature (as shown in Table 1). These criteria are combined by using a weighted sum to form a single value representing the solution quality. The weighted sum thus represents the so-called objective function of the optimization. Current solutions that are available in commercial software are driven by a half manual process, where orientation suggestions are computed using brute force logic on demand and multiple suggestions are presented to the AM engineer for selection [5, 6]. The engineer selects one orientation and generates support for it. Additional simulation tools are available based on Finite Element Methods (FEM) for detailed thermal analysis of single orientations, which have a typical running time of several hours. However, industry experiences have shown that the criteria used in orientation selection might yield insufficient support quality, depending on the part geometry [7]. Thus, to automate the orientation process, the consequences of the orientation for the thermal interaction between the laser energy, part geometry and support structures must be considered. The number of orientations to evaluate using the simple brute force approach is spanning a two-dimensional search space for the two relevant rotation axes of 180° × 360° (y [−90°, 90°), × [0°, 360°))). Therefore, the number of evaluations scales quadratically with the angular resolution of the search. Since typical printing times are on the order of hours to days, it is not feasible to perform FEM thermal simulations for all orientations. In this paper, we evaluate three optimization approaches that replace simple brute force to reduce the number of objective function evaluations. This allows for an orientation optimization based on novel, approximate and faster to calculate thermal criteria. These criteria are based on the layered vector data generated in the slice step. They are introduced in detail in Sect. 3.1 and have a high correlation with thermal properties of the process. Their mathematical characteristics as an objective function are comparable to FEM simulations. They are non-differentiable and show a discontinuous behavior for varying input angles. Thus, the aim of the work is to investigate different optimization algorithms for the concrete application case of automatic orientation finding. This should enable to find a relative good evaluated orientation for both continuous and non-continuous objective functions compared to the brute force method. In addition, fewer orientation evaluations should be necessary to find the optima, so that in the future more computationally expensive evaluations (such as thermal evaluations of the orientations) can also be integrated into the objective function.

22

L. P. Pletzer-Zelgert et al.

2 State of the Art An overview of different optimization algorithms for the automatic orientation of additive manufactured parts was presented in 2020 by Di Angelo et al. [8]. In the article, 41 studies were considered and classified in terms of the optimization algorithms used and the integrated evaluation criteria of the orientation algorithms. Regarding the evaluation criteria, 15 studies use a weighted sum of different objective functions in their orientation algorithm. Seven use genetic algorithms, which classify as a stochastic method [9]. Five studies use decision-making algorithms specifically adapted to their respective evaluation criteria. In three studies, the optimization toolbox of Matlab is used [10]. Studies after 2020 also apply only genetic algorithms (such as [11]). All studies focus on the evaluation of orientation and do not systematically benchmark different optimization algorithms regarding their applicability in automatic part orientation and the number of objective function evaluations. Since different evaluation criteria with different mathematical models in the orientation algorithm are considered, the comparability between different studies is limited. Table 1 shows an overview over the evaluation criteria used in orientation algorithms in literature and the input data used for the algorithm. A differentiation is made between evaluation criteria based on the layer data or the 3D geometry of the part which is discretised by triangular forms (so-called mesh). Commercial software solutions (such as Autodesk Netfabb [5], Materialise Magics [6], etc.) take only a fraction of the evaluation criteria into account. Table 1. Overview of Evaluation Criteria Evaluation Criteria

Input of the geometry of the part

Included in commercial Sofware

Continues and Differentiable?

Studies

Reduction of the production costs

Mesh

Yes

Yes

[12–14]

Reduction of build time

Mesh

Yes

Yes

[11, 13–17]

Reduction of support

Mesh

Yes

Partly

[13, 16, 18]

Increase in mechanical properties

Mesh

No

Yes

[13]

Include special approaches for increasing the quality especially for functional areas and assemblies

Mesh

No

Partly

[19]

Increasing the surface quality

Mesh

No

Yes

[13–15, 17]

No

[20]

Layers

Comparative Evaluation of Optimization Algorithms for Automatic

23

Data related to the build-up process or layer geometries are necessary for estimating the distortions due to thermal induced stresses of a part. As in the study by Ahsan et. al. (see [20] of Table 1), it is therefore to be expected that the mathematical characteristics of the formulated objective functions changes when using layer-based data. Continuity and differentiability of the objective function can no longer be assumed. The table also shows that the objective functions of the orientation optimization formulated so far are mostly continuous and differentiable when the evaluations are based on the part geometry discretized by a mesh. Thus, it was not yet investigated whether the algorithm finds the global optimum, especially for non-continuous and differentiable algorithms. In conclusion the literature alone is not sufficient to select a suitable optimization algorithm for automatic orientation finding.

3 Methodology Taking the review of the state of the art in Sect. 2 into account, different optimization algorithms are compared and evaluated in this chapter. The benchmark is carried out using objective functions that behaves mathematically in a similar way to evaluation criteria based on FEM simulations. First, the objective function is constructed, and its mathematical behavior is analyzed. Afterwards, the selection of the different optimization algorithms is discussed, and their performance benchmarked in a case study. The aim of the work is to select the most suitable optimization algorithm from a preselection of existing optimization algorithms for objective functions with different mathematical properties (continuity, non-continuity). The goal is to reduce the number of required function evaluations and not impair the orientation quality found in comparison to the brute force. 3.1 Formulation of Evaluation Criteria In addition to the evaluation criteria used in literature to date (shown in Table 1), we incorporate 5 additional evaluation criteria listed in Table 2. All of them have an impact on the distortion due to the thermal stress according to literature and are not yet considered in existing orientation algorithms. The evaluation criteria used for this study are quantified in five different mathematical terms that can be combined to objective functions as seen in Table 3. The objective function “Thermal Critical” is the only one that is based on layer data and represents a non-continuous function (see Fig. 5 in the appendix). Thus, it is the function that tests the quality of the optimization algorithms for non-continuous objective functions. In the following, different optimization algorithms are compared for an orientation algorithm that uses a weighted sum that also takes non-continuous objective functions into account, as they are also expected in thermal simulation. Thus, the objective functions used in the orientation algorithm represent a suitable example for the investigation in this paper.

24

L. P. Pletzer-Zelgert et al.

Table 2. Overview of literature for different influences of the reduction of distortion due to thermal stress. Evaluation Criteria

Studies

Reduction of downskin surfaces (surfaces that are visible in z-axis direction from the [22, 23] build plate [21]) Reduction of branches (part areas that are not connected in layered printing)

[22]

Reduction of the laser return time (exposure pause time between two adjacent applied layers)

[24, 25]

Increasing the homogeneity of the energy input in each layer

[24, 26]

Increasing the strength of the binding to the substrate plate

[27]

Table 3. Objective functions of the orientation algorithm. Considered Evaluation criteria

Name of Objective Function

Description

Input of the geometry of the part

Continuity

Evaluation Thermal Critical Criteria of Table 2

Relative quantifying of the property that distortion due to thermal stress occurs

Layers

No

Reduction of support

Support Area

The surface area which must be supported in the print

Mesh

Partly

Reduction of support

Support Volume

Estimation of the support volume of the part

Mesh

Partly

Reduction of build time

Build Height

Build the height of Mesh the oriented part

Yes

Surface Quality

Quantifying the roughness of the oriented part

Yes

Reduction of the production costs Increasing the surface quality

Mesh

3.2 Selection of Optimization Algorithms In this paper, non-differentiable objective functions are used. Therefore, only optimization algorithms that do not require derivatives from the objective function can be used. Kochenderfer et al. define three categories that fulfill this requirement: direct search

Comparative Evaluation of Optimization Algorithms for Automatic

25

methods, surrogate model optimization and stochastic methods [9]. From each category, one optimization algorithm is selected. In the following, the reasons for selection and the functional principle of each algorithm are briefly described. Stochastic Method: In the category of stochastic methods, a distinction is made between the subcategories of evolutionary and population methods. According to Rios et. al. and Szynkiewicz et. al., from the stochastic methods the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) proved to be the most promising with a success rate of 33% on 502 optimization problems [28]. The CMA-ES algorithm is based on the principle of biological evolution. A set of points in the search space (population) changes according to the principles of stochastic mutation, deterministic selection and recombination over different iterations of the algorithm (generation). The population is thereby defined by a multivariate normal distribution density. The optimization is started locally at a defined point in the search space (so called starting point). The implementation details are e.g. described by Auger and Szynkiewicz [29, 30]. Direct Search Method: According to the analysis of Rio et. al., direct search methods often locate global extrema more reliable than stochastic methods [28]. The Multilevel Coordinate Search (MCS) method is widely adapted and thus selected for this category. In MCS, the search space is recursively divided into hyperrectangles. Each hyperrectangle is assigned to an evaluated function value. In the process, the new hyperrectangles resulting from a box are given a higher level than that of the original box. In this way, a hierarchy of hyperrectangles is created by repeated division: a tree structure in which the complete search space forms the root. All other nodes represent subspaces of the entire search space. A detailed description of the algorithm is found in source [31]. Surrogate Model Optimization: An algorithm from the category of surrogate model optimization is tested. These methods generate an approximation of the actual objective function to reduce computationally expensive function evaluations. For non-continuous and differentiable objective functions such as in this work (see Sect. 1), the replacement model thus provides a good solution [9]. In this work, Efficient Global Optimization (EGO) is therefore chosen as a surrogate model optimization. The EGO algorithm consists of several phases. First, an initial rough approximation model of the objective function is created in the initialization phase, which is validated in a subsequent phase. The model then is adjusted based on the maximum expected improvement until a termination criterion is met (see description in [32]).

3.3 Case Study To benchmark the performance of the chosen optimization algorithms, a case study is conducted. Geometries used for the benchmark are taken from the open ABC dataset [33]. 42 part geometries (see Table 4 in the appendix) are selected by their suitability for printing using the AI based Part Identification module of the KEVIN technology processor [7]. The part set is divided into two batches, one of which (see Table 4 in the appendix) examines the orientations only according to the mesh-based objective functions (“mesh batch”), whereas the other batch is additionally examined by the layer-based objective

26

L. P. Pletzer-Zelgert et al.

functions. The mesh batch represents a series of parts that were geometrically too small to obtain enough layer-based information for a meaningful evaluation of the layer-based objective function. The selected optimization algorithms are implemented in C#. The implementations are tested against well-known mathematical test functions to validate their correctness. To evaluate the criteria and resulting objective function described in Sect. 3.1, a generic scorer interface is implemented. The optimization algorithms select a pair of angles (x/y) as input parameters to the scorer. The scorer rotates the geometrical mesh by the given angles. Then all mesh based objective functions are calculated for the current angle between each triangle’s normal vector and the z-axis vector. Finally, the layer-based objective functions are calculated by slicing the mesh into layers. The contour and hatch vectors of each resulting sliced layer are virtually executed using the DAP machine emulator [34]. To construct the objective function, the relative importance of the different objective functions is given by individual weights. However, since the absolute values of the objective functions differ significantly by geometry, the weights are not applied directly to the absolute objective function values. Instead, the objective function values are normalized before computing the weighted sum. The value range that is used to normalize each objective function is statically calculated by typical ranges of minimum and maximum values found in previous studies. The study determined the objective functions for 1022 printed parts from an AM service provider’s database. The maximum values were used to normalize the objective functions originating from the parts of the ABC database. [7]. The optimization algorithm benchmark uses the performance of a brute force search with 10° angle increments as reference. The performance metrics aim to measure three characteristics: the runtime of the optimization, the consistency of the result convergence with different geometries, and the quality of the found minimum. Since the observed runtime is implementation and hardware dependent, the number of objective function evaluations is used to increase comparability instead of the unfiltered runtime. The consistency of convergence and solution quality are both measured by comparing the best minimum found by the optimization algorithm relative to the best minimum of the brute force approach across the various geometries. Two different termination conditions are used to limit the benchmark runtime. The first termination condition is specific to each optimization algorithm and determined by convergence measurements at optimization runtime. Additionally, the number of allowed function evaluations is cut to 500 for MCS and CMA, and to 200 for EGO due to the longer internal computation time of the EGO algorithm. Preliminary tests showed that the median for objective function evaluation in the implementation used is 4 s, 5 s and 21 s for MCS, CMA-ES and EGO, respectively. Tests showed as well that the CMA-ES algorithm performs better when executed multiple times with different starting points. Therefore, CMA-ES is run with 3 different starting points. Each of the 3 runs is limited to 167 objective function evaluations. The performance for only one starting point is reviewed in the discussion in Sect. 4.2.

Comparative Evaluation of Optimization Algorithms for Automatic

27

4 Results and Discussion 4.1 Results  i  found by each optimization algorithm (i ∈ {CMA − The minima f xmin ES, MCS, EGO}) are compared relative to the minimum found by brute force   bruteforce for 10° discrete steps in Fig. 2. The entire range of values is shown in f xmin the insert. The rectangular box shown for each optimization algorithm covers the data range from the first to the third quartile and is separated by a line that marks the median of the data. The whiskers extend to one and a half times the interquartile range (the range from the first to the third quartile). If no value in the data set lies outside this range, then the furthest point is used as the boundary of the whisker. Values outside the whiskers are represented as outliers by black circles. Each data point represents the evaluation of a part using the respective optimization algorithm set. This corresponds to 42 data points per algorithm. A value of 1 indicates that the same orientation score was determined using the brute force method.

Fig. 2. Comparison of the orientation scores calculated by the chosen optimization algorithms and the brute force method.

The results in Fig. 2 show that for all three optimization algorithms the first quartile is smaller than one, i.e. at least 25% of the orientation optima found by the algorithm shows smaller values. For the CMA-ES algorithm, the distance between the first and third quartiles of the orientation evaluation results is the smallest compared to the other two algorithms. The evaluation results are also closer to one and therefore deviate less from the brute force results. For five parts, the algorithm was able to obtain  better orientation  i  bruteforce < 1 in CMA-ES). scores than brute force (see five points with f xmin /f xmin The range of values for the outliers is also lower compared to the other two algorithms. For MCS two outliers at 4.4 and 6.3 and the EGO algorithm at 3.8 and 4.0 can be identified. The CMA-ES showed the outlier with the lowest deviation to brute force with 1.4 compared to the other two algorithms.

28

L. P. Pletzer-Zelgert et al.

For all algorithms, the median lies between 1.002 and 1.020. The upper limit of the whiskers is 1.03 for CMA-ES, 1.40 for MCS, and 1.23 for the EGO algorithm. For the 19 parts of the mesh batch the same best-evaluated orientation could be found for 18 parts compared to brute force (94.7%) (see green marked parts in Fig. 6 in the appendix). The algorithm CMA-ES found the same optimum as a brute force for all 19 parts. It can also be determined that the defined outliers only occurred with parts that also take layer-based objective functions into account in the objective function. Figure 3 shows the number of function evaluations until a termination criterion is  i   bruteforce  . The maximum number of evaluations is reached over the ratio f xmin /f xmin marked by a horizontal line. For brute force with discrete 10° steps in the solution space of [360°,180°), the number of function evaluations are 36 × 18 = 648. The limits are each marked by an orange vertical and horizontal line.

Fig. 3. Plot of the number of function evaluations and the ratio between the minima of brute force over the optimization algorithms.

For the CMA-ES algorithm, the number of function evaluations required for all parts is between 400 and 501. For the other two algorithms, the number of function evaluations is between 50 and 200 for all parts. The optimization by using MCS and EGO is stopped for one and three parts respectively due to the maximum number of function evaluations being exceeded, while the CMA-ES is stopped for 21 parts for this reason. Figure 3 shows that the CMA-ES algorithm also requires the highest number of function evaluations to find the optimum. Therefore, the quality of the evaluation results when performing the calculation with only one starting point is investigated additionally. The same starting points are chosen as before but as a separate run. The results can be found in the appendix (see Fig. 4). The distance between the first and the third quartile of the evaluation results increases but is still smaller than in MCS and EGO.

Comparative Evaluation of Optimization Algorithms for Automatic

29

4.2 Discussion The results show that the CMA-ES algorithm most reliably finds orientations whose scores are close to the same brute force optimum. The optimization search of the algorithm can be improved by the number of starting points and still superior to MCS and EGO. The CMA-ES algorithm stops for half of the tested parts due to the termination criterion of maximum objective function evaluations. Since the calculation results perform still better on average than with the other two algorithms, it should be investigated in the future whether the termination criteria of the maximum number of function evaluations can be further reduced without reducing the evaluation quality. The higher median of the computation time for one function evaluation (computations that are make in scorer (see description in Sect. 3.3) and the computation of the a priori logic for determining the next evaluative orientation) for the EGO algorithm lies that the algorithm performs computationally intensive operations for the model fits (see Sect. 3.2). Thus, the necessary function evaluations have a greater impact on the overall computational time balance than on the other two algorithms. The agreement or even better evaluated orientations of 94.7% for MCS and EGO for mesh batch and 100% for CMA-ES with the orientations of brute force shows that objective functions with larger continuous domains in the solution space and lower variance or sensitivity of the range of values lead to more reliable evaluation results. Thus, the quality of the computational results is significantly dependent on the mathematical properties of the evaluation function. Overall, all three optimization algorithms provide reliable orientation results for the mesh batch. The CMA-ES algorithm produces the most reliable orientation evaluation results for both batches (mesh and layer-based). The number of function evaluations significantly depends on the number of starting points and thus can be reduced dependent on the requirements for the calculation results as well as the mathematical behavior of the objective function. For the other two algorithms, i.e. the EGO algorithm and the MCS algorithm, reliable calculation results were also obtained for the mesh-based objective functions. This shows that the CMA-ES algorithm finds well-evaluated orientations compared to brute force and requires fewer function evaluations. The aim of this work was to find an optimization algorithm that can replace brute force for the orientation algorithm without getting a less good evaluated orientation. This goal could be fulfilled for both discontinuous and continuous objective functions.

5 Conclusion and Outlook In this paper, different orientation algorithms from the category of derivative-free optimization algorithms were investigated for the use case of orientation algorithm for the technology PBF-LB/M. The orientation algorithms were deliberately tested on objective functions that are not continuous and differentiable in order to postulate evaluation criteria that are mathematically similar to evaluations based on FEM simulations. Then, three different optimization algorithms are selected, implemented, and verified using

30

L. P. Pletzer-Zelgert et al.

test functions. The performance of the optimization algorithms is benchmarked relative to the number of objective function evaluations and solution quality of the brute force method. The algorithms were applied to 42 parts to find their optimal orientations. The CMA-ES algorithm found the optimum for the five objective functions more reliably compared to the other EGO and MCS algorithms, but required the highest number of function evaluations. However, the quality as well as the number of function evaluations is affected by the choice of starting points. The EGO and MCS algorithms also show reliable computational results for objective functions with larger continuous ranges and lower variance as well as sensitivity, but performed comparatively worse than the CMAES algorithm due to longer computation times (in case of the EGO algorithm) as well as for less continuous objective functions. In conclusion, this work evaluated and suggested an algorithm for finding the optimal part orientation in the PBF-LB/M build chamber, which significantly reduces the amount of needed orientation evaluations compared to traditional brute force methods, while providing the same reliability. The work helps to find a better fitting optimization algorithm depending on its objective functions for the orientation algorithm. In future research, the calculation results should be further investigated depending on the starting points as well as the convergence behavior of the optimization. This can improve the optimization regarding to the formulated requirements. Likewise, the calculation results of the optimization algorithms are dependent on the setting parameters of the optimization algorithm and should therefore also be investigated in further studies. These studies should rule out that the higher dispersion of the computational results of the EGO and MCS algorithms is not due to an inappropriate choice of parameters for the optimization problem. It should also be investigated how the algorithms behave for other sets of parts. In particular, the results for parts specifically designed for AM should be investigated in more detail in the future. Acknowledgements. The authors would like to thank the Federal Ministry of Economic Affairs and Climate Action for funding the research in the project “Ressourcenschonende Prozessroute für hochintegrierte Hydrauliksysteme am Beispiel einer elektrifizierten mobilen Arbeitsmaschine” (HyRess, grant number 03LB3030G).

Comparative Evaluation of Optimization Algorithms for Automatic

31

Appendix

Fig. 4. Comparison of the orientation scores of CMA-ES with different starting points and MCS and EGO.

Table 4. Datasets of Parts. Parts Declaration After Nomination of Dataset [34] (mesh batch)

00000021 00000033 00000039 00000045 00000086 00000120

00000122 00000129 00000130 00000248 00000321 00000391

00000406 00000421 00000436 00000464 00000467 00000583

00000586 00000606 00000644 00000689 00000775 00000799

00000928 00000987 00001117 00001165 00001250 00001412

00001483 00001571 00001629 00001670 00001693 00001699

00001700 00001701 00001709 00001730 00001847 00001910

Fig. 5. Investigation of mathematical behavior of three objective functions.

32

L. P. Pletzer-Zelgert et al.

Fig. 6. Orientation Results of each Part

References 1. DIN Deutsches Institut für Normung e. V. DIN EN ISO/ASTM 52900:2022-03, Additive Fertigung_- Grundlagen_- Terminologie (ISO/ASTM 52900:2021); Deutsche Fassung EN_ISO/ASTM 52900:2021. Beuth Verlag GmbH, Berlin. https://doi.org/10.31030/3290011 2. Kumke: Methodisches Konstruieren von additiv gefertigten Bauteilen. Springer Fachmedien Wiesbaden (2018) 3. Gibson, I., Rosen, D., Stucker, B., Khorasani, M.: Additive Manufacturing Technologies. Springer International Publishing, Cham (2021) 4. Zeyn, H. (ed.): Industrialisierung der Additiven Fertigung: Digitalisierte Prozesskette - von der Entwicklung bis zum einsetzbaren Artikel, Neuerscheinung, p. 226. VDE Verlag, Berlin, Offenbach (2017) 5. Autodesk, 2016. Autodesk Netfabb 2017 Handbuch 6. Materialise Software. Materialise Magics Software Manual (2017) 7. Ziegler, S., Dirks, S.: Schlussbericht zum Teilvorhaben Technologieprozessor - KEVIN (Kognitive effiziente validierte intelligente Numerik) im Verbundprojekt Industrialisierung und Digitalisierung von Additive Manufacturing (AM) für automobile Serienprozesse (IDAM) (2022). https://opac.tib.eu/DB=1/XMLPRS=N/PPN?PPN=1844410633. Accessed 1 Feb 2023 8. Di Angelo, L., Di Stefano, P., Guardiani, E.: Search for the optimal build direction in additive manufacturing technologies: a review. J. Manufac. Mater. Process. 4(3) (2020)

Comparative Evaluation of Optimization Algorithms for Automatic

33

9. Kochenderfer, M.J., Wheeler, T.A.: Algorithms for Optimization. The MIT Press, Cambridge (2019) 10. Holmström, K.: The TOMLAB optimization environment in Matlab. Adv. Model. Optim. 1(1), 47–69 (1999) 11. Abdulhameed, O., Mian, S.H., Moiduddin, K., Al-Ahmari, A., Ahmed, N., Aboudaif, M.K.: A multi-part orientation planning schema for fabrication of non-related components using additive manufacturing. Micromachines 13(10) (2022) 12. Alexander, P., Allen, S., Dutta, D.: Part orientation and build cost determination in layered manufacturing. Comput. Aided Des. 30(5), 343–356 (1998) 13. Brika, S.E., Mezzetta, J., Brochu, M., Zhao, Y.: Multi-objective build orientation optimization for powder bed fusion by laser. Indust. Eng. Manage. 6, 1–9 (2017) 14. Byun, H.-S., Lee, K.H.: Determination of the optimal build direction for different rapid prototyping processes using multi-criterion decision making. Robot. Comput. Integrat. Manufac. 22(1), 69–80 (2006) 15. Cheng, W., Fuh, J., Nee, A., Wong, Y.S., Loh, H.T., Miyazawa, T.: Multi-objective optimization of part- building orientation in stereolithography 1(4), 12–23 (1995) 16. Chowdhury, S., Mhapsekar, K., Anand, S.: Part build orientation optimization and geometry compensations for additive manufacturing process. J. Manufac. Sci. Eng. 140 (2017) 17. Khodaygan, S., Golmohammadi, A.: Multi-criteria optimization of the part build orientation (PBO) through a combined meta-modeling/NSGAII/TOPSIS method for additive manufacturing processes. Int. J. Interact. Des. Manufac. (IJIDeM) 12 (2018) 18. Mele, M., Campana, G., Lenzi, F., Cimatti, B.: Optimisation of build orientation to achieve minimum environmental impact in Stereo-lithography. Procedia Manufac. 33, 145–152 (2019) 19. Moroni, G., Syam, W.P., Petrò, S.: Functionality-based part orientation for additive manufacturing. Procedia CIRP 36, 217–222 (2015) 20. Ahsan, N., Habib, A., Khoda, B.: Geometric analysis for concurrent process optimization of AM. Procedia Manufac. 5 (2015) 21. DIN Deutsches Institut für Normung e. V. DIN EN ISO/ASTM 52911–1:2020–05, Additive Fertigung_- Konstruktion_- Teil_1: Laserbasierte Pulverbettfusion von Metallen (ISO/ASTM 52911–1:2019); Deutsche Fassung EN_ISO/ASTM 52911–1:2019. Beuth Verlag GmbH, Berlin. https://doi.org/10.31030/3060962 22. Ranjan, R., Ayas, C., Langelaar, M., van Keulen, F.: Fast detection of heat accumulation in powder bed fusion using computationally efficient thermal models. Materials (Basel, Switzerland) 13(20) (2020) 23. Seyda, V.: Werkstoff- und Prozessverhalten von Metallpulvern in der laseradditiven Fertigung, p. 250. Springer, Berlin Heidelberg, Berlin, Heidelberg (2018) 24. Liu, Y., Yang, Y., Wang, D.: A study on the residual stress during selective laser melting (SLM) of metallic powder. Int. J. Adv. Manuf. Technol. 87(1–4), 647–656 (2016) 25. Pohl, H., Simchi, A., Issa, M., Dias, H.C.: Thermal Stresses in Direct Metal Laser Sintering (20010 26. Xiao, Z., et al.: Study of residual stress in selective laser melting of Ti6Al4V. Mater. Des. 193, 108846 (2020) 27. Mohr, G., Scheuschner, N., Hilgenberg, K.: In situ heat accumulation by geometrical features obstructing heat flux and by reduced inter layer times in laser powder bed fusion of AISI 316L stainless steel. Procedia CIRP 94, 155–160 (2020) 28. Rios, L.M., Sahinidis, N.V.: Derivative-free optimization: a review of algorithms and comparison of software implementations. J. Global Optim. 56(3), 1247–1293 (2012) 29. Auger, A., Hansen, N.: Tutorial CMA-ES: evolution strategies and covariance matrix adaptation 30. Szynkiewicz, P.: A comparative study of PSO and CMA-ES algorithms on black-box optimization benchmarks. J. Telecommun. Inform. Technol. 8, 5–9 (2019)

34

L. P. Pletzer-Zelgert et al.

31. Huyer, W., Neumaier, A.: Global optimization by multilevel coordinate search. J. Global Optim. 14(4), 331–355 (1999) 32. Jones, D., Schonlau, M., Welch, W.: Efficient global optimization of expensive black-box functions. J. Global Optim. 13, 455–492 (1998) 33. Koch, S., et al.: ABC Dataset Chunk 0000 (2019). https://archive.nyu.edu/handle/2451/44309 34. Dirks, S.H.G., Schleifenbaum, J.H.: Adaption of Cost Calculation Methods for Modular Laser-powder Bed Fusion (L-PBF) Machine Concepts. Metal Additive Manufacturing Conference 2019 : Monday, 25 November 2019-Wednesday, 27 November 2019, Örebro Castle : proceedings/ASMET - the Austrian Society for Metallurgy and Materials, pp. 79–87 (2019)

Review and Development of Design Guidelines for Additive Tooling of Injection Molds Using PolyJet Modelling Stefan Junk(B)

, Steffen Schrock , and Nico Schmieder

Laboratory of Rapid Prototyping, Offenburg University of Applied Sciences, Campus Gengenbach, Klosterstr. 14, 77723 Gengenbach, Germany [email protected]

Abstract. Due to globalization and the resulting increase in competition on the market, products must be produced more and more cheaply, especially in series production, because buyers expect new variants or even completely new products in ever shorter cycles. Injection molding is the most important production process for manufacturing plastic components in large quantities. However, the conventional production of a mold is extremely time-consuming and costly, which creates a contradiction to the fast pace of the market. Additive tooling is an area of application of additive manufacturing, which in the field of injection molding is preferably used for the prototype production of mold inserts. This allows injection molding tools to be produced faster and more cheaply than through the subtractive manufacturing of metal tools. Material Jetting processes using polymers (MJT-UV/P), also called Polyjet Modeling (PJM), have a great potential for use in additive tooling. Due to the poorer mechanical and thermal properties compared to conventional mold insert materials, e.g. steel or aluminum, the previously used design principles cannot be applied. Accordingly, new design guidelines are necessary, which are developed in this paper. The necessary information is obtained with the help of a systematic literature research. The design guidelines are mapped in a uniform design guide, which is structured according to the design process of injection molds. The guidelines do not only refer to the constructive design of the injection mold or the polymer mold insert, but to the entire design process and describe the four phases of planning, conception, development and realization. Particular attention is paid to the special geometric designs of a polymer mold insert and the thermomechanical properties of the mold insert materials. As a result, design guidelines are available that are adapted to the special requirements of additive tooling of molds inserts made of plastics for injection molding. Keywords: Design guidelines · Material Jetting · Additive Tooling · Injection Molding · mold inserts

1 Introduction Globalization and the resulting increase in market competition mean that products have to be produced at ever lower cost, especially in series production. Industrial customers expect new variants or even completely new products in ever shorter cycles. Accordingly, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 35–45, 2024. https://doi.org/10.1007/978-3-031-42983-5_3

36

S. Junk et al.

the factors of time and cost are playing an increasingly important role in the development of new products in companies. Efficient product development is responsible for the success of a company, especially in the injection molding industry [1]. Injection molding is the most important production process in the manufacture of plastic components (molded parts) and takes place in an injection mold on an injection molding machine, which can produce large quantities of molded parts until failure [2]. However, the conventional production of a mold is extremely time-consuming and costly, which creates a contradiction with the fast-moving market [2]. Therefore, it is obvious that additive manufacturing processes are also used in the injection molding industry in recent years. In particular, the production of injection mold inserts, in which the molded part is injected and which represent the most important component of a mold, is being intensively researched [1]. This area of application is referred to as additive tooling. In this process, prototyping of the mold inserts takes place, which can reduce the production time and the manufacturing costs of the molds. Material Jetting (MJT) processes using curing of polymers by means of UV light (MJT-UV/P), also called PolyJet Modelling (PJM), and vat photopolymerization (VPP) processes, e.g. Stereolithography (SLA) and Direct Light Processing (DLP) processes have the greatest potential of all additive manufacturing processes for use in additive tooling. Various types of plastics can be used as the starting material. Due to poorer mechanical and thermal properties compared to conventional mold insert materials, e.g., steel or aluminum, the previously used design principles cannot be applied [3]. In recent years, the specifics of additive tooling have been intensively studied and the available technologies have been continuously developed. However, it must be noted that the topic is only dealt with to a limited extent in the literature and there are currently no works that describe and summarize additive tooling comprehensively. These findings raise the question, which design and manufacturing guidelines should be applied to the additive tooling of injection mold inserts using polymerization processes? In order to answer this question, this paper first deals with a systematic literature review. Subsequently, the necessary information for the design guidelines and influencing factors is collected, organized by topic and analyzed. The results of this investigation provide an overview of important guidelines for the design of the tools and the process.

2 Procedure of Literature Research The systematic literature research is an elementary component of scientific work and represents a systematic procedure, which aims at an almost complete determination of the current information within a research topic. The greatest challenge usually consists of filtering the relevant information from a multitude of existing references. In order to obtain the most qualitative results possible, the systematic approach is essential and must be consistently followed [4]. There are various approaches to systematic literature searches in the literature. However, the basic structure of the literature search always remains the same. The following procedure is mostly based on the systematic literature review according to Fink, A. [5], but is extended in some places. This procedure is chosen because it deals with the whole process in a clear and very comprehensive way. In total, 82 references containing information on additive tooling are identified after abstract and full text analysis. The statistical analysis shows that besides gray literature,

Review and Development of Design Guidelines for Additive Tooling

37

only scientific publications contain relevant information, although any kind of specialized literature, e.g., monographs, reference books, or reference works, are stored in the databases. Thus, this it can be deduced that although a lot of research is being done in this subject area, no summary works exist yet. There is a clear research gap. In addition, the evaluation shows that of the total of eight available polymerization processes, only the three processes based on material jetting using of polymers curing by UV light (MJT-UV/P), vat photopolymerization curing by laser beam (VPP-UVL) or light from LED diodes (VPP-LED) are currently used for additive tooling [6].

3 Design Guidelines for Additive Tooling This chapter covers design guidelines or special features on the topics of geometric structures and materials of the mold inserts (cavity), gating system, tempering system, modular mold concept and demolding system. It is the largest chapter of the guideline and contains, as far as possible, specific numerical values for the described geometric features. The geometric features are demonstrated using a demonstrator component from a case study. This case study involves a scratcher that was developed in a workshop with Master’s students (see Fig. 1). All product development phases were run through, from the design of the component and the tools to the additive manufacturing of the tool inserts to the use on a professional injection molding machine for the production of a small series [7]. The design rules developed in this case study were illustrated and put into practice using the MJT-UV/P process (also called PJM) for Additive Tooling as an example.

Fig. 1. Demonstrator component: scratcher a) CAD model and b) Molded parts based on [7]

3.1 Geometric Structures in Additive Tooling First, the design of geometric structures is considered, which represent all the features that a designer should consider when creating a CAD model. To illustrate, Fig. 2 represents a pair of polymer mold inserts from a case study with particularly distinctive geometric structures. After analyzing the guidelines from literature research [8–13], the geometric structures are classified into the following categories.

38

S. Junk et al.

Fig. 2. Polymer mold inserts representing distinctive geometric structures based on [7].

Lowered Features Lowered features as shown in Fig. 2 are depressions in a cavity, e.g. notches, elongated holes or especially bores. Holes with a diameter smaller than 0.5 mm should not be printed during additive manufacturing, but should be re-drilled afterwards in post-processing [8]. Holes for ejector pins should always be re-drilled due to the recommended fits [9]. Raised Features Raised features are elevations in a cavity, e.g. pins, ribs or walls (see Fig. 2). If the aspect ratio of pins is greater than 3 to 1, they should not be printed but manufactured as replaceable pins from other materials, e.g. metal, and then inserted in the cavity. Appropriate holes must be drilled for this purpose [9]. Ribs should not be designed thinner than 1.6 mm [10]. In addition, direct positioning of the pins and perpendicular positioning of the ribs to the melt flow is not recommended. Walls in the direct melt flow or in the vicinity of the gate should always be reinforced [11]. Undercuts, Corners and Edges Due to the poorer thermomechanical properties, undercuts should be avoided, since the mold insert materials generally cannot withstand forced demolding [12]. All edges and corners should be rounded with a large radius as demonstrated in Fig. 2 [13]. Small chamfers should be applied when in direct contact with the printer build platform, facilitating removal after additive manufacturing [8]. Symmetry and Separation Plane The mold insert halves should be secured against horizontal displacement during injection molding. This behavior occurs when the parting line is wedge-shaped. Securing or guiding can be achieved, for example, by trapezoidal structures (see Fig. 3a) or centering pins outside the cavity or by a special course of the parting line, curved curves being preferred [14]. The cavities should be designed as symmetrically as possible in terms of length, width, height and geometric structures [15]. In order to reduce manufacturing

Review and Development of Design Guidelines for Additive Tooling

39

times, the cross-sections of the mold insert halves should be optimized at points which do not support the cavity or are carriers of the clamping force [8]. Venting Additional venting is necessary in polymer mold inserts to counteract air entrapment and surface burn. This can be achieved via surfaces, ejector pins or vent holes. The most versatile option is surface venting via the parting line in the form of a vent channel, scratching of the parting line or increasing the surface roughness by grinding [13]. Other measures to avoid air entrapment include changing the wall thickness, the gate position or the injection speed [16]. In the case of complex cavity geometries, it is possible to divide geometric structures into different inserts and then incorporate them into the basic structure of the polymer mold insert, thereby generating automatic surface venting [10]. 3.2 Materials for Mold Inserts Polymers used as the mold insert material have poorer thermomechanical properties than conventional materials, which results in a significantly shorter service life. This results not only in lower resistance to stresses, but also in greater susceptibility to expansion. In order to achieve the longest possible service life, special knowledge of the properties of the mold insert material is necessary, whereby these can be controlled and generally better injection molding conditions can be achieved. Thermosetting materials are mainly used as mold insert material, but the development of thermoplastics and elastomers is progressing steadily. Thermoplastics and elastomers can be integrated, especially through the use of multi-materials. Individual features can be manufactured from these materials and, if necessary, additionally coated with thermosets. Furthermore, reinforcement through the use of fibers, e.g. carbon or glass fibers, is possible. Multi-materials improve or combine the material properties of polymers [17]. The materials Accura Bluestone, Somos PerForm, Digital ABS Plus and Tough achieve the longest average service lives and generally the best results in terms of molded part quality [18]. 3.3 Runner System in Additive Tooling Next, the gating system is considered, which is divided into the following categories after analyzing the existing guidelines: Runner Bushing and Runner Taper Direct contact between the hot barrel nozzle and the mold insert material should be avoided. Thus, a conventional sprue bushing made of metal is used (see Fig. 3b). The sprue bushing can be integrated either directly in the parent mold or in the polymer mold insert. When integrating the sprue bushing in the mold insert, the bushing diameter should be as large as possible for better force distribution, and the required bore diameter should be 0.2 mm smaller. The bore can should be reamed mechanically [19].

40

S. Junk et al.

Fig. 3. Design guide lines for the tooling concept: a) parting line, b) nozzle and c) tool frame, based on [7].

Runner and Runner Manifold Apart from avoiding hot runner systems, the general design guidelines for conventional injection molds can be applied. However, special features of additive tooling, e.g., generous edge rounding, must be taken into account [13]. Gate Technology While bar, film, tape, ring and screen gating can be used, point and tunnel gating should be avoided. The gate diameter should be two to three times larger than in conventional mold inserts from 5 mm to 8 mm is recommended [9]. 3.4 Tempering System After the runner system, the temperature control system is considered. The temperature control system plays an essential role in additive tooling, since the cooling time is the decisive factor in volumetric mold shrinkage and temperature fluctuations in the polymer mold insert and accounts for by far the largest proportion of the cycle time. Due to the poor thermal conductivity of polymers, the cooling time is five to seven times as long as in conventional mold inserts and should therefore be set as short as possible [19]. Since conventional cooling channels have only a minor influence on the cooling time, they are only recommended to a limited extent in polymer mold inserts. Cooling channels close to the contour are not recommended [20]. Besides the cooling time, the control of the mold insert temperature is the most important task of the temperature control system. The aim of this control is to extend the service life by cooling the mold insert temperature to 35 °C before each shot. Since this is generally not possible during the actual injection molding process, e.g. through cooling channels, a sufficient opening time should be set after demolding. The most efficient solution to this problem is to blast the cavity surfaces with compressed air, which can more than halve the opening time [13]. 3.5 Tooling Concept and Demolding System Modular tooling concepts can be mold frames or steel plates, although the mold frame concept is recommended for most applications. The mold frame is made of conventional

Review and Development of Design Guidelines for Additive Tooling

41

materials as shown in Fig. 3c. It is used to position the polymer mold insert and to absorb the greatest mechanical loads during injection molding [18]. The mold insert should be 0.2 mm (maximum 0.5 mm) thicker than the mold frame in the closing direction. This procedure achieves both complete closure of the mold inserts halves by compression of the polymer and distribution of the greatest loads over the mold frame. In addition, the width and length of the mold insert should be oversized by 0.5 mm and then optimally adapted to the mold frame by mechanical post-processing. To withstand the stresses occurring in the cavity, the distance between the cavity wall and the mold insert wall should be 20 mm to 25 mm at any point (see Fig. 2). A greater wall thickness is not suitable due to longer pressure times and greater material consumption [13]. Finally, the demolding system is considered. Demolding is usually carried out with conventional ejector pins. The holes required for this should always be re-drilled in the already manufactured polymer mold insert and designed 0.2 mm to 0.3 mm smaller. The recommended clearance fit H7 can be achieved by subsequent reaming. General design guidelines for injection molds can be applied when placing the ejector pins on the molded part [9]. Due to the higher volumetric molding shrinkage and the layering process of the 3D-printers using MJT-UV/P process, there is greater surface adhesion between the molded part and the cavity, requiring greater demolding force, which shortens tool life. There are several ways to reduce surface adhesion, with the demolding system offering two options explored to date. First, release agent should be sprayed on the cavity surfaces before each injection molding cycle, and second, the draft angles should be designed with a 5° angle [8, 21]. If undercuts in the cavity are unavoidable, the current state of the art allows the use of slide technology or the use of multi-materials in the form of elastomers (preferably for the demolding of threads) [12].

4 Process Parameters for Pre- and Post-processing The next step is to determine and optimize the process parameters of the polymerization process, which include the pre-processing, polymerization and post-processing steps. The aim of this determination is to influence the surface roughness, dimensional deviation, strength and surface hardness of the polymer mold insert to be manufactured, thus extending the service life and producing a negative image of the molded part that is as accurate as possible [22]. 4.1 Pre-processing During preprocessing, a number of important parameters are defined. This is done on by parameter settings in the 3D printer software and by the placement of the component in the build space of the 3D printer. 3D-Printing Accuracy The accuracy depends on the respective 3D-printer and its performance indicators. This refers primarily to the layer thickness, the XY (build platform) resolution and the 3Dprinting material. To achieve the lowest possible dimensional deviations, the mold insert should be printed with the glossy print mode. The layer thickness should be less than

42

S. Junk et al.

100 µm, whereby the best printing result is achieved with the smallest possible layer thickness. The advantage of a very small layer thickness is the minimization of the stairstep effect, which leads to dimensional deviations and poor surface quality, especially with curved surfaces or curves [23]. Currently, cavities should not have a cavity smaller than 100 µm (in terms of length, width or height), since otherwise the molded part will not be completely formed during injection molding [24]. Alignment of the Additively Manufactured Part The additively manufactured parts (polymer mold inserts) should always be aligned on the build platform in such a way that the cavity points upwards or that there are as few overhangs as possible and the pressure lines run in the same direction as the subsequent melt flow in the cavity. This generates a better surface finish and better filling behavior [24]. With a small gap dimension, the gap direction should be aligned in the printing direction (often the X-axis of the printer), so that the printing material does not stick before solidification. Support Material The use of support material should be avoided, especially in direct contact with the cavity surface. If this is not possible, the polymer mold insert can be separated into several parts and aligned accordingly on the build platform. This procedure does not create any indentations on the surface. 4.2 Post-processing Mechanical and thermal post-processing are used in additive tooling. Mechanical postprocessing refers to material-removing finishing processes, e.g. grinding, polishing, sandblasting, drilling or reaming, which are used to create a better surface finish and for exact dimensional adjustment. If very high demands are placed on the surface or the dimensional accuracy of the molded part, the cavity surfaces should be mechanically reworked. Otherwise, polymerization processes generate sufficient results for use in injection molding [9]. If burrs occur on the molded part during injection molding, either the clamping force should be adjusted or the parting line should be polished [26]. Thermal post-processing refers to material-changing post-processing, e.g. UV or thermal post-curing, which is used to positively change the thermo-mechanical properties of the mold insert material. Possible properties that can be positively changed are heat resistance, tensile strength, flexural strength and modulus of elasticity. Thermal postprocessing on a polymer mold insert results in measurable expansion, which is why this should be considered during the design of the CAD model and adjusted accordingly [25].

5 Process Parameters for Injection Molding Finally, the process parameters of the injection molding cycle are determined or optimized, with the mold insert temperature, injection pressure, injection speed, holding pressure and cooling time being particularly relevant. Polymer mold inserts can withstand significantly lower loads than mold inserts made of steel or aluminum, which is

Review and Development of Design Guidelines for Additive Tooling

43

why the conventional method of parameter determination (determination of the highestpressure settings of the injection molding machine with subsequent optimization or step-by-step minimization of the parameters) cannot be used. In order not to destroy the polymer mold inserts during the first injection molding cycle, initially low input variables should be set with subsequent optimization or stepwise increase of the parameters [27]. It should be noted that there is currently no uniform procedure for parameter determination, since the parameters are in most cases application-specific (material, molded part size, design, etc.) and the subject area is too little researched.

6 Conclusion and Outlook Through the systematic literature search, a total of 82 relevant references were found, of which the most important are mentioned in this contribution. From the content analysis, it was possible to deduce that the complexity of polymer mold inserts and the performance indicators of AM systems in particular have been further developed over the past two decades. Although complex polymer mold inserts can now be additively manufactured with high accuracy, the average service life is still relatively short. Based on this, the design guide was developed from the collected design guidelines. The structure of these design guidelines is based on the design process developed for additive tooling and should be worked through by designers according to the specified structure. In particular, the design of the four most important technology systems of an injection mold, the different molded part and mold insert materials and the parameter settings for the additive manufacturing process and the injection molding process should be considered when designing a polymer mold insert. In conclusion, additive tooling should generally be used today for small and complex molded parts, as this is where the advantages of additive manufacturing come into play more strongly. The design guidelines presented here are only a first draft based on the current state of the art. They should therefore be further developed in future research work. First of all, they should be validated by practical field tests and expert discussions. This should be followed by a regular literature search, which will maintain the current state of research and allow the design guidelines to be developed into a standardized design catalog over time. Particular attention should be paid to methods for parameter determination, since these have hardly been analyzed in previous research work, but are essential for success in additive tooling. Furthermore, it would be possible to transfer the design guidelines to an online database to enable easy access and additional functions, such as search and compare.

References 1. Lerma Valero, J.R.: Plastics Injection Molding: Scientific Molding, Recommendations, and Best Practices. Hanser Publishers; Hanser, Munich, Cincinnati (2020) 2. Catoen, B., Rees, H.: Injection Mold Design Handbook. Hanser, Munich (2021) 3. Gebhardt, A., Kessler, J., Thurn, L.: 3DPrinting: Understanding Additive Manufacturing, 2nd edn. Hanser Publishers, Munich (2019) 4. Becker, W., Ulrich, P., Stradtmann, M.: Geschäftsmodellinnovationen als Wettbewerbsvorteil mittelständischer Unternehmen. Springer Gabler, Wiesbaden (2018)

44

S. Junk et al.

5. Fink, A.: Conducting Research Literature Reviews: From The Internet to Paper. Sage, Los Angeles, London, New Delhi, Singapore, Washington DC (2014) 6. Schrock, S., Junk, S., Albers, A.: A method for additive tooling in integrated product development. In: How Product and Manufacturing Design Enable Sustainable Companies and Societies, 16th–18th Aug 2022, p. 12 (2022) 7. Maier, A., Herr, C., Hofer, J., Hug, M., Groß, P., Haag, W.: Workshop Additive Tooling Abschlusspräsentation Wintersemester 2021/2022, Hochschule Offenburg (2022) 8. Formlabs. Low-Volume Rapid Injection Molding With 3D Printed Molds (2020). https://3d. formlabs.com/injection-molding 9. Godec, D., Breški, T., Kataleni´c, M.: Additive manufacturing of polymer moulds for smallbatch injection moulding. Teh. glas. (Online) 14(2), 218–223 (2020) 10. Somos® Materials, Ed.: Injection Molding using Rapid Tooling, Covestro Deutschland AG (2021) 11. Bagalkot, A., Pons, D., Symons, D., Clucas, D.: Analysis of raised feature failures on 3D printed injection moulds. Polymers 13(10), 1541 (2021). https://doi.org/10.3390/polym1310 1541 12. Schuh, G., Bergweiler, G., Lukas, G., Abrams, J.A.: Feasibility and Process capability of polymer additive injection molds with slide technology. Procedia CIRP 93(12), 102–107 (2020) 13. Stratasys, PolyJet For Injection Molding (2016). https://www.alphacam.ch/fileadmin/ user_upload/Applikationen/PDFs/Technical_Application_Guide_-_Injection_Molding_-_ PolyJet_For_Injection_Molding_-_English_A4_Web.pdf 14. Zong, X., Ruan, J., Liu, H., Sun, W., Liu, Y.: Rapid injection moulding process of polyether ether ketone based on stereolithography. SN Appl. Sci. 1(11), 96 (2019) 15. Bogaerts, L., et al.: Influence of thermo-mechanical loads on the lifetime of plastic inserts for injection moulds produced via additive manufacturing. Procedia CIRP 96, 109–114 (2021) 16. Saman, A.M., Abdullah, H., Nor, M.A.M.: Computer simulation opportunity in plastic injection mold development for automotive part. In: 2009 Int. Conference on Computer Technology and Development, pp. 495–498. Kota Kinabalu, Malaysia (2009) 17. Hofstätter, T., Mischkot, M., Pedersen, D.B., Tosello, G., Hansen, H.H.: Evolution of Surface Texture and Cracks During Injection Molding of Fiber-Reinforced, Additively-Manufactured, Injection Molding Inserts (2016) 18. Kampker, A., Alves, B., Ayvaz, P.: Technological and economic comparison of additive manufacturing technologies for fabrication of polymer tools for injection molding. In: Almeida, H.A., Vasco, J.C. (eds.) Progress in Digital and Physical Manufacturing: Proceedings of ProDPM’19, pp. 28–39. Springer International Publishing, Cham (2020). https://doi.org/10. 1007/978-3-030-29041-2_4 19. Raz, K., Chval, Z., Habrman, M., Milsimerova, A.: Thermal specification of 3D printed injection moulds made from PA12GB. In: IOP Conference Series: Materials Science and Engineering, vol. 1199, no. 1, p. 12009 (2021) 20. Rodriguez, J.: Use of Additive Manufacturing (AM) for Mold Inserts in Injection Molding. American Society for Engineering Education (2016) 21. Bagalkot, A., Pons, D., Symons, D., Clucas, D.: Categorization of failures in polymer rapid tools used for injection molding. Processes 7(1), 17 (2019) 22. Kampker, A., Kreisköther, K., Reinders, C., Reinders, C.: Material and parameter analysis of the polyjet material and parameter analysis of the polyjet process for mold making using design of experiments. Int. J. Mater. Metallurgic. Eng. 3, 11 (2017) 23. Hopkins, M., et al.: Stereolithography (SLA) utilised to print injection mould tooling in order to evaluate thermal and mechanical properties of commercial polypropylene. Procedia Manufac. 55(1), 205–212 (2021)

Review and Development of Design Guidelines for Additive Tooling

45

24. Surace, R., Basile, V., Bellantone, V., Modica, F., Fassi, I.: Micro injection molding of thin cavities using stereolithography for mold fabrication. Polymers 13(11), 1848 (2021). https:// doi.org/10.3390/polym13111848 25. Whlean, C., Sheahan, C.: Using additive manufacturing to produce injection moulds suitable for short series production. Procedia Manufac. 38(1), 60–68 (2019) 26. Vogeler, F., Verspreet, J., Geyskens, K., Bogaerts, L., Moens, D., Faes, M.: Breakout analysis of plastic material jetted moulds for injection moulding. In: Conference: International Conference on Polymers and Moulds Innovations - PMI2018 (2018) 27. Bagalkot, A., Pons, D., Clucas, D., Symons, D.: A methodology for setting the injection moulding process parameters for polymer rapid tooling inserts. RPJ 25(9), 1493–1505 (2019)

Design for AM 2

Design of Additively Manufactured 3D Lattice Cores of Sandwich Panels Hussam Georges1,2(B) , Christian Mittelstedt2 , and Wilfried Becker1 1

2

Technical University Darmstadt, Institute of Structural Mechanics, Franziska-Braun-Str. 7, 64287 Darmstadt, Germany [email protected] Institute for Lightweight Engineering and Structural Mechanics, Technical University Darmstadt, Otto-Berndt-Str. 2, 64287 Darmstadt, Germany Abstract. This study analyzes 3D truss-based lattice cores in sandwich panels subjected to localized transversal loads using analytical methods. Compared to the common sandwich theories, the presented model may reveal stress concentrations induced by the localized loads. An advanced core design to reduce stress concentrations is provided by varying the core strut diameter through the thickness which leads to a graded core. A mass comparison between the homogeneous and the graded core shows that grading the core reduces the core weight by more than 20%. Based on finite element analysis, it can be demonstrated that the present model with a lowered modeling effort well captures the lattice strut stresses of the graded cores, particularly in load application regions. Keywords: Core design application

1

· lattice core · sandwich panels · load

Introduction

Outstanding mechanical properties offered by sandwich panels have encouraged engineers to use this kind of composite structures in lightweight applications [1–3]. Typically, sandwich panels consist of a low-density core enclosed by two high-strength face sheets. This three-layer design provides high bending stiffness with a relatively low structural mass. However, the sandwich core is vulnerable when localized transverse loads are applied [4,5]. [6,7] observed core failure near the load application region in sandwich structures during 3-point bending tests. Reinforcing the core using inserts may enhance the sandwich’s stiffness and strength, but it requires more material use and manufacturing processes [8,9]. With additive manufacturing, novel approaches to reinforce sandwich cores are enabled since additive manufacturing has revolutionized the design of components and parts due to the flexibility offered by the layerwise manufacturing technology [10]. Thus, the core properties may be customized to fulfill local requirements. The resulting core is called a graded core, as the mechanical properties vary across the core. Using graded structures may significantly enhance c The Author(s), under exclusive license to Springer Nature Switzerland AG 2024  C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 49–62, 2024. https://doi.org/10.1007/978-3-031-42983-5_4

50

H. Georges et al.

the specific properties and the load carrying capacity of the sandwich and avoid over-dimensioned designs [11]. Recent studies considered lattice structures composed of periodic representative volume elements (RVE) as cores in sandwich panels to enhance the sandwich’s mechanical performance [12–14]. The design of these truss-like structures is derived from metallic atom crystals and offers high specific mechanical properties [15,16]. [17] observed that layerwise graded lattices may exhibit higher energy absorption than homogeneous lattices. Since merely experimental and numerical methods are used to investigate these lattice structures, a novel analytical model to analyze graded 3D lattice cores is presented in this work. The analytical model allows an efficient analysis of strutbased cores with many RVEs, easy modeling and sufficiently precise calculation for pre-dimensioning and optimizing sandwich structures with lattice cores. First, an RVE topology that satisfies the core design requirements is selected. An anisotropic homogeneous material with the corresponding elastic properties of the lattice on the macroscale is used to replace the lattice core and reduce the modeling effort. Thus, the lattice is homogenized. The lattice strut stresses are determined by dehomogenizing the core using the sandwich displacements. A higher-order sandwich theory is used to obtain the displacement in the considered sandwich. Due to the employed higher-order approach, stress concentrations near the load application regions are captured. To homogenize the stress distribution through the core thickness and achieve an even load distribution in the core, the lattice strut diameter and, thus, the core properties vary in a layerwise manner through the core thickness.

2

Model and Approach

This work considers a sandwich panel subjected to a 3-point bending load. Due to the applied transversal force, normal bending stresses and transverse shear stresses occur in the face sheets and the core. Furthermore, normal transverse stresses are expected in the load application area. The sandwich face sheets mainly support the normal bending stresses since the core exhibits a relatively low stiffness. While the normal bending stresses decrease with lower core stiffness, the transverse normal stresses increase in the core. Besides supporting the transverse normal stress, the core transfers the transverse shear stresses in the sandwich. Therefore, a core material with a corresponding shear stiffness and vertical stiffness is required. Such core material properties are provided by facecentered cubic RVEs with vertical struts (F2CCZ), as shown in Fig. 1. The 45◦ inclined strut provides a high effective shear stiffness G∗xz in the xz-plane. The vertical strut reinforces this RVE in the z-direction and enhances the vertical ∗ . Thus, the F2CCZ RVE exhibits an orthotropic material behavior stiffness Ezz that can be described by nine indepndent elastic constants. The F2CCZ RVE can be characterized by the ratio between the RVE size a and the RVE strut diameter d, which is called aspect ratio a/d. Depending on the aspect ratio, the

Design of AM Lattice Cores

51

elastic properties and the density ρ∗ of the RVE can be determined by ∗  a −2.17 ∗ Eyy Exx = = 0.51 , Es Es d

G∗xz Gs

= 0.57

 a −2.01 d

 a −2.21 ∗ Ezz = 1.37 , Es d ∗

ρ = 3.51 ρs

,

 a −1.88 d

(1) ,

where Es and ρs are the strut material’s elastic stiffness and the density. Equation 1 is determined based on the analytical model presented by [18] and valid for aspect ratios a/d > 6. This model assumes the lattice struts as beam elements made of isotropic elastic material and allows a precise calculation of the relative density. The relative density obtained by this model is not overestimated since the model considers the overlapping region between the lattice struts.

z

a

y x

π/4 π/2

a

Fig. 1. F2CCZ RVE topology

Figure 2a illustrates the considered sandwich with the F2CCZ lattice core. Depending on the F2CCZ RVE size a and the number of the RVEs in each spatial direction (Nx , Nz , Ny ), the sandwich dimensions can be given by l = Nx × a,

h(c) = Nz × a,

t = Ny × a,

(2)

where l denotes the sandwich length, h(c) the core thickness, and t the sandwich depth. In Fig. 2b, the 3D model is reduced to a 2D model by assuming a plane strain state in the xz-plane (εyy = 0). Furthermore, the face sheets are considered to be made of an isotropic material with the stiffness E (f) and the Poisson’s ratio ν (f) . The sandwich total thickness h results from the sum of the core thickness h(c) and the face sheet thickness h(f) (h = 2h(f) + h(c) ). In addition to the global coordinate system xz at the midplane of the sandwich, each face sheet has a separate coordinate system xm zm at the center of the layer, with m being the number of the face sheet. The considered core is modeled as a homogeneous material representing the mechanical properties of the lattice at the macroscale,

52

H. Georges et al.

which are determined by Eq. 1. Core decomposition into mathematical layers of equal thickness is performed to allow a variation of the core mechanical properties through the core thickness. Through the mathematical layers, merely the strut diameter of the RVEs is varied. Grading the strut diameter of the RVE affects a variation of the core stiffness through the core thickness. In this study, a symmetric graded core is intended as stress concentrations occur in the upper core half near the load application and in the bottom core half near the support points. To enable a symmetric core grading, at least four mathematical layers are required. The mathematical layers may involve several physical layers and have the corresponding mechanical properties of the physical lattice layers (c) (c) (c) (Exx,j , Ezz,j , Gxz,j , etc.), where j denotes the number of the mathematical layer (j = [1, 2, 3, 4]). This modeling approach minimizes the number of degrees of freedom and reduces the modeling effort. Increasing the number of the mathematical layers may allow a smoother grading between the layers and offer more degrees of freedom while designing the core. However, the computational effort increases with each mathematical layer. z2

t

E (f) , ν (f) (c)

(c)

(c)

x2

m=2

(c)

(c)

(c)

h(f)

(c)

Exx,4 , Ezz,4 , Eyy,4 , Gxz,4 , νxz,4 , νxy,4 , νyz,4

z

y h

x

z (c)

(c) (c) (c) Gxz,3 , νxz,3 , νxy,3 , νyz,3

(c) (c) (c) Exx,3 , Ezz,3 , Eyy,3 , (c)

(c)

(c)

x

(c)

(c)

(c)

h(c)

(c)

Exx,2 , Ezz,2 , Eyy,2 , Gxz,2 , νxz,2 , νxy,2 , νyz,2 (c) (c) (c) (c) (c) (c) (c) Exx,1 , Ezz,1 , Eyy,1 , Gxz,1 , νxz,1 , νxy,1 , νyz,1

z1

E (f) , ν (f)

x1

m=1

h(f)

l

(a) Sandwich panel with 3D lattice core

(b) Sandwich with homogenized core

Fig. 2. Considered sandwich model

In previous studies, it has been proven that first-order shear theory displacement presentations are suitable for describing the deformation behavior of the face sheets in sandwich panels. Therefore, each face sheet deformation is repre(f) (f) sented by the angular rotation ψm and the horizontal displacement u0,m and the (f)

vertical displacement w0,m of the face sheet’s mid-axis, where m = 1 refers to the bottom face sheet and m = 2 to the top face sheet. Thus, the displacement field can be represented as (f)

(f) u(f) m (x, z) = u0,m (x) + zψm (x), and (f)

(f) (x, z) = w0,m (x). wm

(3) (4)

Design of AM Lattice Cores

53

Since the vertical displacements in these representations are functions of merely the x-coordinate, transverse strains εzz cannot be considered in the face sheets. The horizontal strain εxx and the shear strain γxz are determined by the derivatives (f)

ε(f) xx,m (f) γxz,m

(f) ∂u0,m (x) ∂ψm (x) +z , and = ∂x ∂x (f) ∂w0,m (x) (f) + ψm = (x). ∂x

(5) (6)

With the decomposition of the core into four mathematical layers, new degrees (c) (c) ˜q ) are used to describe the horizontal and vertical displaceof freedom (˜ uq , w ment along the interfaces between the mathematical layers, where q indicates the number of the interface (q = [1, 2, 3, 4, 5]). The displacement function in each mathematical layer is represented by a linear term that corresponds to the linear interpolation between the interface displacements. This displacement representation enables the determination of layerwise-constant transverse strain through (c) (c) (c) (c) `j , w ˆj , w `j ) the core thickness. Introducing new degrees of freedom (ˆ uj , u (c) (c) with corresponding higher-order functions (fˆj (z), f`j (z)) in each mathematical layer allows capturing linear and cubic strain variation through the core thickness. The core displacement functions are given as (c)

(c)

(c)

(c)

uj (x, z) = [(j − 2)˜ uj (x) + (3 − j)˜ uj+1 (x)] +

(c)

˜j (x)) 4(˜ uj+1 (x) − u h(c)

z

(c) (c) (c) (c) + fˆj (z)ˆ uj (x) + f`j (z)` uj (x),

(7) (c)

(c) wj (x, z)

= [(j −

(c) 2)w ˜j (x)

+ (3 −

(c) j)w ˜j+1 (x)]

+

(c)

˜j (x)) 4(w ˜j+1 (x) − w h(c)

(c) (c) (c) (c) + fˆj (z)w ˆj (x) + f`j (z)w `j (x). (c)

z (8)

(c)

The shape functions fˆj (z), f`j (z) vanish at the interfaces to avoid displacement discontinuities 16(−2j + 5) 43 2 (c) z − fˆj (z) = − 4(3 − j)(2 − j) − 2 z , and h(c) h(c) (c) (c) f`j (z) = fˆj (z) z.

(9)

(10)

54

H. Georges et al.

Differentiating the displacement functions in Eq. 7 and Eq. 8 yields the core strains in each core mathematical layer (c)

∂uj , (11) ∂x (c) ∂wj (c) , and (12) εzz,j = ∂z (c) (c) ∂wj ∂uj (c) + . (13) γxz,j = ∂z ∂x Derived from the generalized Hooke’s law and concerning the plane strain conditions, the constitutive law for the effective orthotropic core material core can be given as ⎡ ⎤−1 (c) (c) (c) (c) (c) 1−νxy,j νyx,j ν +νxy,j νyz,j ⎡ (c) ⎤ ⎡ (c) ⎤ − xz,j (c) 0 (c) ⎢ ⎥ Exx,j Exx,j ε σxx,j ⎢ ⎥ ⎢ xx,j ⎥ ⎥ ⎢ ⎢ ⎥ ⎥ ⎥ ⎢ ⎢ ν (c) +ν (c) ν (c) ⎥ ⎢ (c) (c) ⎢ (c) ⎥ (c) ⎥ 1−νyz,j νzy,j ⎥ ⎢ xz,j xy,j yz,j . (14) ⎢ ⎢ σzz,j ⎥ = ⎢ ε − 0 ⎢ ⎥ ⎢ zz,j ⎥ (c) (c) ⎥ ⎥ ⎢ Exx,j Ezz,j ⎢ ⎥ ⎣ ⎦ ⎦ ⎣ ⎢ ⎥ ⎣ ⎦ (c) (c) τxz,j γxz,j 1 0 0 (c) (c)

εxx,j =

Gxz,j

The elastic constants may vary layerwise through the core thickness, resulting in a graded core design. The stresses in the isotropic face sheets are obtained from (f) σxx,m =

E (f) 2

(1 − ν (f) )

ε(f) xx,m , and

(15)

(f) (f) τxz,m = G(f) γxz,m .

(16)

The introduced degrees of freedom are unknown functions depending on the horizontal coordinate x. The principle of minimum potential energy is applied to determine these deformation functions in the considered load case. For this goal, the total potential energy Π resulting from the sum of the inner strain energy Πi and the external potential energy Πe is required. Regarding the assumptions made, the total potential energy is determined by (f)

(f)

(c)

Π = Πi,1 + Πi,2 + Πi

+ Πe ,

(17)

with (f) Πi,m

1 = 2 j=1 4

(c) Πi

1 = 2

l 2

− 2l



l 2



− 2l



h(f) 2 (f)

− h2

−h(c) 2 −h(c) 2



(f) (f) (f) σxx,m ε(f) xx,m + τxz,m γxz,m dz dx , and

(c)

+ jh4

+

(18)

(j−1)h(c) 4

(c)

(c)

(c)

(c)

(c)

(c)

(σzz,j εzz,j + τxz,j γxz,j + σxx,j εxx,j ) dz dx . (19)

Design of AM Lattice Cores

55

The considered load case is illustrated in Fig. 3a. The applied force on the top face sheet at the mid of the top face sheet and the corresponding vertical displacement of the top face sheet leads to the following external energy formulation Πe =

−F (f) w (x = 0). t 0,2

(20)

F t

(c)

w(3)

h(f)

E (f) , ν (f)

u(3)

(c)

(c)

Exx,4 , Ezz,4 , Eyy,4 , etc.

z (c) (c) (c) Exx,3 , Ezz,3 , Eyy,3 , etc. (c)

(c)

(c)

(c)

(c)

(c)

x

Exx,2 , Ezz,2 , Eyy,2 , etc.

w

7

(4)

u(4)

Exx,1 , Ezz,1 , Eyy,1 , etc.

w

5

(5)

4

h(f)

E (f) , ν (f)

8

3

h(c)

u(6)

2

w(6)

u(5) 9 6

w(2) u(2)

1

u(1) w(1)

l(k) l

(a) 3-point bending load case

(b) RVE node displacements

Fig. 3. Boundary conditions and obtained lattice node displacements

The resulting equations from the condition δΠ = 0 represent a second-order differential equation system. The procedure for solving this equation system is presented in detail in [19]. Solving the corresponding equation system yields the unknown displacement functions of the sandwich and, thus, the displacement at the lattice nodes (u(r) , w(r) ), where r describes the RVE node number (Fig. 3b). However, the occurring stresses in the lattice struts are not obtained directly. Therefore, the homogenized core has to be dehomogenized by replacing it with the lattice on the mesoscale to determine these lattice strut stresses. The struts in the lattice on the mesoscale undergo a length change caused by the applied load. This length change can be expressed as a relationship between the displacements of the corresponding strut nodes. For this, the node displacements have to be transformed from the global coordinate system to the local coordinate system of the considered strut. The following equations describe the resulting length change of each strut within the F2CCZ RVE Δl(1) = (u(2) − u(1) ) cos(π/4) + (w(2) − w(1) ) sin(π/4), Δl(2) = (u(3) − u(2) ) cos(3π/4) + (w(3) − w(2) ) sin(3π/4), Δl

(3)

= (u

(4)

−u

(3)

) cos(5π/4) + (w

(4)

−w

(3)

) sin(5π/4),

(21)

56

H. Georges et al.

Δl(4) = (u(1) − u(4) ) cos(−π/4) + (w(1) − w(4) ) sin(−π/4) Δl(5) = (w(3) − w(1) ), Δl(6) = (w(5) − w(1) ) sin(π/4) Δl

(7)

= (w

(3)

−w

(5)

(22)

) sin(π/4).

Due to the symmetry, the change of the length in the struts 6 and 9, and the struts 6 and 7 are identical, i.e. Δl(6) = Δl(9) and Δl(7) = Δl(8) . With the assumption that the nodes of the lattice do not transfer moments, the lattice strut stress is given by σ (s) =

Δl(s) (s) E , l(s)

(23)

where the quantity s denotes the number of the strut within the F2CCZ RVE, as indicated in Fig. 3b.

3

Results

In this study, sandwich panels with a F2CCZ lattice core are analyzed. Two core types are compared: a homogeneous lattice core with a constant strut diameter (Fig. 4a), and a layerwise-graded lattice core with a variable strut diameter through the core thickness (Figure 4b). Since the cell size a does not vary through the thickness, changing the strut diameter leads to a varied aspect ratio and, thus, a varied stiffness and relative density. In the graded core, the strut diameter shows a symmetrical variation through the core thickness and affects a corresponding distribution of the effective core stiffness. The strut diameter within a mathematical layer remains constant. Each mathematical layer in the considered core includes four physical lattice layers, resulting in 16 physical layers (Nz = 16). The graded core’s maximum strut diameter conforms to the homogeneous core’s strut diameter. The cell size a of the graded lattice and the homogeneous core are identical (a = 3 mm). In the considered graded core, the ratio between the maximum transverse stiffness and minimum transverse (c) (c) stiffness is assumed to be (Ezz,max /Ezz,min = 2). The ratio between the maximum strut diameter and the minimum strut diameter √ of the graded core can be determined according to Eq. 1 by dmax /dmin = 2.21 2. Thus, the graded core has approx. 22% lower mass than the homogeneous core. Since both cores are composed of 16 physical layers, the core total thickness is 48 mm. Compared to the core thickness, the face sheets are thin and have a thickness of 2 mm. While the core consists of 176 RVEs along the sandwich length (Nx = 176), merely one RVE is considered in the y-direction (Ny = 1). The results generated by the derived model are compared to a FE solution obtained from a 3D analysis using the FE software ABAQUS. In the finite element analysis (FEA), quadratic 3D solid elements are used to mesh the face sheets, and beam elements represent the struts of the 3D lattice core. Additional to the boundary conditions presented in Fig. 4, no displacements in the depth direction are permitted in the FEA to fulfill the plane strain state conditions. Three paths along the core length are specified

Design of AM Lattice Cores

57

to assess the stresses in the core using the normalized strut stress σ (s) = σ (s) /F . The first path goes along the bottom core layer (highlighted in red in Fig. 4b), the second path along the mid core layer (blue), and the third one along the top core layer (green). Due to the symmetry, merely the struts of the left core half are analyzed. Furthermore, the stress variation through the core thickness at the load application position is evaluated. F tσ

F tσ

j=4

j=3

j=2

j=1

(a) Homogeneous core (HC)

(b) Layerwise graded core (GC)

Fig. 4. 2D view of the FE model with the corresponding boundary conditions

Figure 5 shows the stresses in the vertical struts through the thickness of the graded core (GC) and the homogeneous core (HC) at the load application position x = 0. The occurring stresses in this area are mainly induced by the localized transverse loads. Thus, the struts there are under compression load. It can be demonstrated that the maximum stress occurs at the same position in the upper half of both cores and exhibits the same value. Due to the grading approach, the struts in the graded core show higher stresses in the mid core layers compared with the homogeneous core. However, the stresses there are below the maximum value. The stresses in the core bottom and top layers show similar distributions in both cores since these layers have the same strut diameter in the graded and homogeneous core. Moreover, it can be shown that the present model is in good agreement with FEA. The stress in the vertical struts along the pre-defined paths is illustrated in Fig. 6. For better visualization, merely the region beyond the load application and the support point is shown. Compared with the struts near the load application area, the struts are hardly loaded. Along these paths, the local effects caused by the transverse load subside. The stress distribution shows a compressive load in the upper core half and a tensile load in the bottom core half. No significant stresses are observed in the mid core layers. This distribution is because the vertical struts transfer the effective bending stress in the core, exhibiting a linear shape through the core thickness.

58

H. Georges et al.

Due to the relatively low stiffness of the core compared to the face sheet stiffness, the face sheets mainly support the normal bending stress, and the strut stresses induced by the sandwich bending are negligible compared to the strut stresses caused by the concentrated load. The stress along the bottom and top layer in the graded core presents a similar shape as in the homogeneous core. In the mid layer, a stress increase in the graded core is identified due to the smaller strut diameter in this layer.

Normalized vertical position z/h(c)

0.50

FEA Present model

0.25

0.00

−0.25

−0.50

0.5

1.0

1.5

1 Absolute normalized strut stress |σ (s) | in mm 2

Fig. 5. Absolute normalized stresses in the vertical struts through the core thickness at x = 0

The stress distribution along the core length in the inclined struts of the xz-plane is presented in Fig. 7. It can be seen that the inclined struts are higher loaded than the vertical struts beyond the stress concentration areas. While the stresses there show no variation along the mid layer, the stress in the top and bottom exhibits a linear distribution beyond the stress concentration areas. It can be observed that the stress in the inclined struts consists of two components. The constant effective shear stress causes the first stress component. The second stress component is induced by the sandwich bending and shows a linear increase with respect to the horizontal axis of the sandwich. Since the strut diameter of the bottom and top layers in both cores is identical, no stress changes along these layers caused by the core grading are identified. However, the strut diameter in the mid core layer shows a minimum value compared to the homogeneous core. Therefore, the struts in the graded core are higher loaded than in the homogeneous core. Although the inclined struts are higher stressed in the graded core, the stress there shows lower values than the maximum stress in the stress

Design of AM Lattice Cores

59

concentration area. Stresses in the inclined struts of the yz-plane are not presented as they support marginal stresses.

1 Normalized strut stress σ (s) in mm 2

0.10

FEA GC Present model GC Present model HC

0.05

0.00

−0.05

−0.10 −0.4

−0.3

−0.2

−0.1

Normalized horizontal position x/l(k)

1 Absolute normalized strut stress |σ (s) | in mm 2

Fig. 6. Normalized stresses in the vertical struts along the core length FEA GC Present model GC Present model HC

0.8

0.6

0.4

0.2

0.0 −0.50

−0.25 Normalized horizontal position x/l

0.00 (k)

Fig. 7. Absolute normalized stresses in the inclined struts along the core length

60

H. Georges et al.

Finally, the deflection of the top and bottom face sheets in the sandwiches with the homogeneous core and the graded core are compared, as illustrated in Fig. 8. A core compression is observed near the load application and the support points in both configurations. Core compression is the difference between the top face sheet deflection and the bottom face sheet displacement. Due to the employed grading approach, the sandwich with the graded core exhibits a higher deflection than the sandwich with the homogeneous core. In the sandwich with the graded core, the deflection increases by approx. 14% compared to the sandwich with the homogeneous core. Moreover, an increase in the core compression can be identified. However, the increase of the core compression and the sandwich deflection does not affect the strength of the core. 0.00

Displacement w in mm

Top face sheet Bottom face sheet −0.25

−0.50

core compression −0.75 −0.50

−0.25

0.00

0.25

Normalized horizontal position x/l

0.50 (k)

Fig. 8. Deflection of the sandwich with a graded core (GC) and a homogeneous core (HC)

Based on the results obtained by the present model, it can be concluded that graded cores may significantly reduce the core mass without compromising the core strength. By using the selected grading approach, a mass reduction of 22% is achieved. The fabrication of layerwise graded lattices is enabled by additive manufacturing technologies, for instance, selective laser melting, as shown in [20]. Compared with results obtained from the FEA, the present model yields adequate stresses within one-tenth of the FEA computing time. Moreover, some new knowledge about the load distribution in strut-based lattice cores is acquired. With these gained acquirements about the load distribution across the strut-based lattice core, more suitable grading approaches may be applied to customize the strut diameter distribution in the core. This may be carried out within optimization loops to find an advanced core design. Sublayers may be introduced within the mathematical layers to enable a smoother strut diameter variation through the layers. Thus, the number of the sublayers conforms to the

Design of AM Lattice Cores

61

number of the core physical layers. If the strut diameter varies from physical layer to physical layer, each sublayer can be assigned a corresponding stiffness. This modeling yields good results without introducing additional displacement functions, as shown in [21]. While designing graded cores, the deflection of the sandwich must still be considered since the core stiffness declines with decreasing relative density. The increased deflection may be compensated, for instance, by grading the core along the horizontal axis of the sandwich.

4

Summary and Conclusion

This study introduced an analytical model to analyze graded strut-based lattice cores in sandwich panels. Compared to homogeneous cores, graded cores may minimize stress concentrations induced by localized loads without increasing the structural mass or reducing the core strength. Furthermore, the analytical model determines the strut stresses of the graded core using higher-order displacement representations. With the presented model, the core’s optimization and design procedure can be more efficiently performed since the derived model reduces the modeling and computation time by the factor 10. Compared with the homogeneous core, the graded core is advantageous as the graded core shows a reduced mass but the same strength. While designing graded cores in applications where the sandwich displacement may be critical, the total sandwich deflection should be considered since the core stiffness may decrease with strut diameter grading.

References 1. Chai, G.B., Zhu, S.: A review of low-velocity impact on sandwich structures. Proc. Inst. Mech. Eng. Part L J. Mater. Des. Appl. 225(4), 207–230 (2011) 2. D’Alessandro, V., Petrone, G., Franco, F., De Rosa, S.: A review of the vibroacoustics of sandwich panels: models and experiments. J. Sandwich Struct. Mater. 15(5), 541–582 (2013) 3. Blakey-Milner, B., et al.: Metal additive manufacturing in aerospace: a review. Mater. Des. 209, 110008 (2021) 4. Richert, P., Dafnis, A., Schr¨ oder, K.U.: Design guidelines for load introduction points at the boundaries of sandwich panels. Fatigue Fract. Eng. Mater. Struct. 42(7), 1510–1520 (2019) 5. Thomsen, O.T., Frostig, Y.: Localized bending effects in sandwich panels: photoelastic investigation versus high-order sandwich theory results. Compos. Struct. 37(1), 97–108 (1997) 6. Wei, K., et al.: Mechanical analysis and modeling of metallic lattice sandwich additively fabricated by selective laser melting. Thin-Walled Struct. 146, 106189 (2020) 7. Ghannadpour, S., Mahmoudi, M., Nedjad, K.H.: Structural behavior of 3D-printed sandwich beams with strut-based lattice core: experimental and numerical study. Compos. Struct. 281, 115113 (2022) 8. Bozhevolnaya, E., Lyckegaard, A.: Structurally graded core inserts in sandwich panels. Compos. Struct. 68(1), 23–29 (2005)

62

H. Georges et al.

9. Zenkert, D.: The Handbook of Sandwich Construction. Engineering Materials Advisory Services (1997). http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva263055 10. Li, Y., et al.: A review on functionally graded materials and structures via additive manufacturing: from multi-scale design to versatile functional properties. Adv. Mater. Technol. 5(6), 1900981 (2020) 11. Maconachie, T., et al.: SLM lattice structures: properties, performance, applications and challenges. Mater. Des. 183, 108137 (2019) 12. Mesto, T., Sleiman, M., Khalil, K., Alfayad, S., Jacquemin, F.: Analyzing sandwich panel with new proposed core for bending and compression resistance. Proc. Inst. Mech. Eng. Part L J. Mater. Des. Appl. 237(2), 367–378 (2023) 13. Zhang, J., Ding, S., Yanagimoto, J.: Bending properties of sandwich sheets with metallic face sheets and additively manufactured 3D CFRP lattice cores. J. Mater. Process. Technol. 300, 117437 (2022) 14. Zhou, X., Qu, C., Luo, Y., Heise, R., Bao, G.: Compression behavior and impact energy absorption characteristics of 3D printed polymer lattices and their hybrid sandwich structures. J. Mater. Eng. Perform. 30(12), 8763–8770 (2021) 15. Benedetti, M., Du Plessis, A., Ritchie, R., Dallago, M., Razavi, S., Berto, F.: Architected cellular materials: a review on their mechanical properties towards fatiguetolerant design and fabrication. Mater. Sci. Eng. R. Rep. 144, 100606 (2021) 16. Riva, L., Ginestra, P.S., Ceretti, E.: Mechanical characterization and properties of laser-based powder bed-fused lattice structures: a review. Int. J. Adv. Manuf. Technol. 113(3), 649–671 (2021) 17. Al-Saedi, D.S., Masood, S., Faizan-Ur-Rab, M., Alomarah, A., Ponnusamy, P.: Mechanical properties and energy absorption capability of functionally graded F2BCC lattice fabricated by SLM. Mater. Des. 144 (2018) 18. Souza, J., Großmann, A., Mittelstedt, C.: Micromechanical analysis of the effective properties of lattice structures in additive manufacturing. Addit. Manuf. 23, 53–69 (2018) 19. Georges, H., Großmann, A., Mittelstedt, C., Becker, W.: Structural modeling of sandwich panels with additively manufactured strut-based lattice cores. Addit. Manuf. 55, 102788 (2022) 20. Maskery, I., et al.: A mechanical property evaluation of graded density Al-Si10-Mg lattice structures manufactured by selective laser melting. Mater. Sci. Eng., A 670, 264–274 (2016) 21. Georges, H., Meyer, G., Mittelstedt, C., Becker, W.: 2D elasticity solution for sandwich panels with functionally graded lattice cores. Compos. Struct. 300, 116045 (2022)

Designing Variable Thickness Sheets for Additive Manufacturing Using Topology Optimization with Grey-Scale Densities Felix Endress(B) and Markus Zimmermann Laboratory for Product Development and Lightweight Design, TUM School of Engineering and Design, Technical University of Munich, Boltzmannstrasse 15, 85748 Garching near Munich, Germany [email protected]

Abstract. Topology optimization is a powerful tool to automatically generate optimal geometries for Additive Manufacturing (AM). However, to ensure manufacturability, e.g. by material extrusion-based AM (MEX) or Laser-Beam Powder-Bed-Fusion of Metals (PBF-LB/M), a minimum structure thickness has often to be maintained. In this paper, a simple interpolation scheme for penalizing grey-scale densities in topology optimization is applied. It locally reduces the stiffness-to-weight ratio of elements in the variable thickness sheet problem for densities between zero and a critical density. A cantilever beam is optimized, confirming that less penalization produces stiffer structures. Results for the optimization of an L-shaped bell crank are 12% stiffer (and only 4% less stiff) than the design based on conventional (and no) penalization. Simulating and printing the hinge using MEX and PBF-LB/M confirm enhanced manufacturability. In regions of load concentrations, where stresses vary significantly, the results show a general potential for performance improvement, when switching from conventional designs (e.g. sheet metals) to more complex designs that would require advanced manufacturing methods, such as AM. Keywords: Topology Optimization · Additive Manufacturing Constraints · Variable Thickness Sheet · Grey-Scale Densities

1

Introduction

Manufacturing. Plate-like components, such as sheet metals, are commonly used structural components. However, sheet metal designs are often not optimal, e.g. in bending load cases. Therefore, where applications allow, switching from simple sheet metals to standard profiles (e.g. DIN EN 10055 for T-shaped steel, DIN 1025-5 for I-shaped steel, etc.) is reasonable as the shapes are more appropriate for the specific load case, while still maintaining a low cost of production. Unfortunately, their performance may be limited for complex load cases and c The Author(s), under exclusive license to Springer Nature Switzerland AG 2024  C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 63–76, 2024. https://doi.org/10.1007/978-3-031-42983-5_5

64

F. Endress and M. Zimmermann

design domains. Therefore, predominantly in lightweight design, tailored rolled blanks are used as they are often more appropriate for loadings and functions of the application. They are commonly used, e.g. in the automotive industry. Tailored rolled blanks are manufactured using a combination of technologies, such as rolling, welding, cutting, and others [1]. Unfortunately, tailored rolled blanks are limited with respect to cost for manufacturing and design flexibility. Additive manufacturing overcomes these restrictions and enables the manufacturing of complexly shaped 3D components in various sizes in diverse industries. For example, in [2], a load-carrying sheet-metal design that is replaced by a topology-optimized printed part is the rear elevator bell crank in a Cessna 172. Unfortunately, designs generated by numerical methods, such as topology optimization, usually cannot guarantee manufacturability. Following, Design for Additive Manufacturing (DfAM) approaches were introduced, to consider limitations of AM in numerical optimizations. Well-known challenges include overhang constraints, minimum length scales (e.g. minimum layer thickness), and others. Unfortunately, the realized structures are often not optimal with respect to stiffness when the characteristics and restrictions of AM are taken care of when designing parts. Topology Optimization. The density-based topology optimization approach SIMP (Solid Isotropic Material with Penalization) was introduced in [3] and usually results in black-and-white designs, with only a few intermediate densities in the final results, when a penalization is applied [3]. Alternative approaches, like bi-directional evolutionary structural optimization (BESO), also generate discrete, frame-like designs [4]. Unfortunately, these frame-like structures are, in several contexts, not optimal in stiffness [5]. Relaxing the restriction to framelike structures can be accomplished by (partially) not penalizing intermediate densities in the SIMP approach. Structures result, that include grey-scale densities, i.e. intermediate densities between zero (also white or void) and one (also black or solid), which are up to 80 % stiffer for optimization problems with one load case [5]. However, the resulting density fields are not manufacturable per se. So, in practice, grey-scale densities are usually avoided (full penalization), or must be interpreted: (1) as materials (i.e. as porous material [6], as multiple materials with different stiffnesses [7], or as structures with a reduced volume fraction (e.g. filled with lattice structures (cf. Hashin-Strikhman bound for an isotropic material in [8])), or (2) as a dimension, i.e. the thickness of a sheet in 2D. The latter was introduced as the Variable Thickness Sheet problem (VTS) in [9], and is also used in this paper. Designing Variable Thickness Sheets using Topology Optimization. To compute manufacturable, plate-like structures with variable thicknesses using SIMP, only parts of the total range of densities are penalized by an interpolation scheme. Modifications of the penalization scheme were introduced, applied and extended in several publications (cf. [10–13]). For the VTS problem in [12] a tanh-function was used to restrict the solution space. However, 2.5% of the resulting densities in an example were below the critical density value η = 0.5. To overcome the challenge of the resulting densities in the penalized region, two design fields were

Designing Variable Thickness Sheets Using Grey-Scale Densities for AM

65

introduced in [12]. An auxiliary nodal field s and an element field x were used to cut irrelevant parts of the structure in order to “clean up solutions” and facilitate dehomogenization [12]. Unfortunately, this approach requires extensive changes in popular codes, such as the 99 lines code in [16], and optimization setup, to find structures only containing densities in the non-penalized region. A simple interpolation scheme for grey-scale densities is qualitatively proposed in [14]. The aim was to ensure manufacturable, porous metal structures [6]. However, its implementation is not documented, and the application to the VTS problem is missing. In this paper, the method is fully documented, implemented and applied to a cantilever beam and a real use case. The results of the optimization are manufactured using material extrusion-based AM (MEX) and Laser-Beam Powder-Bed-Fusion of Metals (PBF-LB/M). This paper is organized as follows: In Sect. 2, the general problem statement is introduced. In Sect. 3, the scheme is applied to a cantilever beam (Sect. 3.1) and then used to optimize an aerospace component as a use case (Sect. 3.2). Section 4 discusses the potentials and challenges of the approach, as well as limitations, with respect to the design and manufacturing of plate-like components, analysing the manufacturing of the part using MEX and PBF-LB/M. The paper ends with a conclusion in Sect. 5.

2

Problem Statement

Optimization Problem. The following article addresses the minimum compliance design of plate-like components using topology optimization. The mathematical formulation of the underlying optimization problem reads min c = UT KU = ρe

s.t.:

Ne 

ρ˜e (ρe )E0 uT e ke ue

e=1 Ne 

(1)

ρe = f V0 ,

e=1

KU = F, 0 < ρmin ≤ ρ˜e ≤ ρe ≤ 1,

e = 1, . . . , Ne .

The objective function is the compliance c, U and F are vectors containing the global displacements and applied forces, K is the global stiffness matrix, ue is the element displacement vector, and ke is the stiffness matrix for a single element. A volume fraction f is prescribed as the only constraint in the optimization problem. It is defined by the ratio of the actual volume of the structure (as a function of the design variables) to V0 , the maximal volume of the entire design domain. Following the SIMP approach, a Young’s modulus E0 is multiplied by (and potentially reduced by) a modified density ρ˜e , which again is a function of the design variables ρe . In the SIMP approach, an element’s contribution to the volume of a structure is kept constant while “artificially” reducing its stiffness if

66

F. Endress and M. Zimmermann

ρe < 1 and p > 1. To only penalize a range of grey-scale densities, the following interpolation scheme is used.   p ρc ρρec for ρe ≤ ρc ρ˜e = (2) ρe else. It penalizes intermediate densities up to a critical density ρc . For ρ > ρc , a linear relationship between an element’s stiffness and its density exists. The VTS optimization problem without any penalization of grey-scale densities (p = 1) is convex, i.e. the only minimum in the optimization is the global one. Only filtering approaches and spatial discretization prevent the final topologies from being optimal in stiffness [15], ignoring manufacturing challenges. For the introduced approach ρc should always be positive and non-zero (as defined in Eq. 3). For ρc = 1 the classical SIMP approach results, making the cases in Eq. 2 obsolete. 0 < ρc ≤ 1

(3)

Gradients. Due to the modified interpolation scheme, gradients change below the critical density ρc and remain linear above it. Thereby, the contribution of the penalization scheme reads ⎧   ⎨ ρe p−1 ∂ ρ˜e for ρe ≤ ρc p ρc (4) = ⎩1 ∂ρe else. Implementation. In the well-known 99-lines MATLAB code [16], a simple conditional statement (if-else) can be used to switch between the modified, i.e. penalized, and linear updating scheme and its gradients. Additionally, the maximum value for a gradient was limited to the value of the penalization factor p.

3

Application and Use Case

In the following, two problems are introduced to demonstrate and discuss the effects of the modified interpolation scheme. First of all, a 2D cantilever study serves as an example to demonstrate the effectiveness of modifying the penalization of grey-scale densities (Sect. 3.1). Secondly, in a practical use case, an L-shaped bell crank is optimized and printed using MEX and PBF-LB/M (Sect. 3.2). The method of moving asymptotes (MMA) is applied to solve the optimization problems [17]. Sensitivity filtering with a radius of rmin = 1.2 is used throughout the optimizations. In all optimizations a penalization exponent of p = 3 was used if not stated otherwise. The authors furthermore decided not to apply a continuation method. The termination criterion of the optimization is defined by the greatest absolute change of a design variable in the complete set of design variables ρe in one iteration. For the cantilever study, the termination

Designing Variable Thickness Sheets Using Grey-Scale Densities for AM

67

criterion was set to ct = 0.01, whereas in the real use case, it was reduced to ct = 0.001. Young’s modulus was set as 1 MPa throughout. The colour scale used for the visualizations of topologies between white and black is linearly interpolated. Black corresponds to a density of ρe = 1 (solids), whereas densities of ρe = 0 are visualized in white (voids). The terminology “grey-scale densities” stems from this visualization and matches the term intermediate densities. 3.1

Application: Cantilever Beam

Problem Statements. The design domain of the cantilever beam is discretized into 120 elements in the horizontal direction and 80 elements in the vertical direction. Each node has two degrees of freedom for the displacement in the xand y-direction. A unit load (1 N) was applied to the last node on the bottom right in the negative vertical (y-)direction. All degrees of freedom on the very left of the design domain are fixed in both the x- and y-directions. A volume fraction of f = 0.3 is used. The modified interpolation scheme was applied for eleven different values for the critical density ρc . In addition, an optimization with the SIMP approach was twice conducted with a penalization exponent of p = 1 and p = 3, corresponding to ρc = 0.0 and ρc = 1.0. These results are not shown but serve as a validation of our implementation since no difference in results was observed.

Fig. 1. Resulting topologies for different critical densities ρc and f = 0.3

Results. In Fig. 1, the resulting topologies are shown with their corresponding compliances. The stiffness of the structures decreases as ρc increases, from the optimum c = 53.62 Nm for ρc = 0.0 to c = 60.42 Nm for ρc = 1.0 (+12.7%). In

68

F. Endress and M. Zimmermann

contrast to the topology of the global optimum, which is hardly interpretable as a part (cf. [5,12]), for ρc > 0, clear shapes of the part become visible in light grey, with local peaks in a darker colour. It can be confirmed, that closed surfaces are preferred for the given load case w.r.t. stiffness when no other objectives exist (cf. [5,18]). Penalized intermediate densities are uneconomical and usually disappear in the final results up to a satisfying, practical level. However, the penalization does not guarantee all densities in the results are at zero or above the threshold (ρc ). As shown in Fig. 2, in the results, densities in the penalized region exist. Densities above zero (plus the termination criterion ct = 0.01) and the critical density (minus ct = 0.01) are coloured in blue. Only at the borders of the structure, densities in the penalized area occur, which is mainly caused by the chosen filter radius. In practice, and for the following use case, a manual redesign follows the topology optimization and thicknesses are manually adjusted to enable manufacturability.

Fig. 2. Distribution of the resulting element densities (grey) of the structure in Fig. 1 for ρc = 0.6, with the corresponding penalization scheme (blue) on the left. Density field for ρc = 0.6 with element densities between ct and ρc − ct in blue on the right.

3.2

Use Case: L-Shaped Bell Crank

The reference part of the use case, an L-shaped bell crank, is usually located in the wing of a glider. As a part of the control system, connecting steering rods from the pilot to the ailerons, the component is critical for operation, and a failsafe design is required and realized in the conventional design as a redundant load path design (two mirror imaged components, i.e. a double sheet metal design). The requirements, and thus objectives of the structural optimization are (1) a weight saving of −20% and (2) a maximization of stiffness while (3) maintaining the attachment areas and dimensions from the reference design.

Designing Variable Thickness Sheets Using Grey-Scale Densities for AM

69

Fig. 3. Sketch of the design domain (black), regions with prescribed densities (grey), boundary conditions and loads for the L-shaped bell crank use case

Problem Statements. Firstly, the design domain of the part is discretized by elements with size 1 mm × 1 mm to ensure a minimum length scale in the x- and y-directions. The dimensions of the original component were used (see sketch in Fig. 3) to ensure interchangeability. For the use case, allowed thicknesses in the range t ∈ [0, 6.5] mm are linearly mapped to the design variables ρe ∈ [0, 1]. For example, attachment areas are modelled with a thickness of 2 mm, i.e. ρe = 2/6.5 = 0.3077. Non-design regions are modelled as void and passive elements with ρpassive = 1.0e−3 . Then, different values for the critical density are chosen and described in Table 1. After the optimization, the resulting density fields are simulated by switching off the penalization (p = 1) to calculate the displacements, compliance and weight (real volume fraction) of the resulting structures for comparison. Results Design. The resulting topologies of the optimization are shown in Fig. 4. The volume fraction after termination and compliance, as well as the relative difference to the optimal topology (for (a) with ρc = 0), are presented in Table 1. As shown in the example of the cantilever beam, the lower ρc was chosen, the lower was the resulting compliance of the structure. In contrast to (a) without any penalization, for structure (b) with ρc = 0.1537 (tmin = 1mm), clear shapes of the structure can be observed in most areas facilitating manual post-processing. It is only 4% less stiff than the optimal topology. A general threshold value,

70

F. Endress and M. Zimmermann

Table 1. Problem statements with identification number ID, description, critical density and corresponding thickness, compliance, realized volume fraction and differences in compliance for the L-shaped bell crank. ID

ρc (tmin in mm)

c in Nm

f

Rel. difference in c in %

Description

(a)

0.000 (0 mm)

8.062

0.805

reference

Optimal structure, no penalization of intermediate densities to find the global optimum with the given optimization setup. VTS interpretation of results.

(b)

0.1537 (1 mm)

8.383

0.805

+4.0

AM constraint for 1 mm thickness in the out-of-plane direction. VTS interpretation of results.

(c)

0.3077 (2 mm)

8.799

0.805

+9.1

AM constraint for 2 mm thickness in the out-of-plane direction, matching the thickness of the attachment structures.

(d)

1.000 (6.5 mm)

9.579

0.805

+18.8

SIMP approach with penalization of all intermediate densities. VTS interpretation of results.

(e)

Constant thickness of t = 2 mm

14.932

0.800

+85.2

Optimization without the VTS interpretation, so 2D only. Black (2 mm thickness) and white (no material) designs preferred (e.g. manufacturable by laser-cutting the original plane sheet metal)

(f)

Constant thickness of t = 2 mm

13.439

1.000

+66.7

Plane sheet as a reference design, no optimization. Finite element modelling as in other cases

where clearly observable shapes start to result was not investigated. Compared to the design of the conventional penalization in (d) (ρc = 1.0), the resulting structure is still 12% stiffer. Comparing structure (b) with a minimum thickness of 1 mm with the original sheet metal design (f) of constant 2 mm thickness, the reference part (f) is 60.3% less stiff and 24.2% heavier. For structure (b) with ρc = 0.1537 (tmin = 1 mm), the resulting topology in several areas is very similar to the optimal one for ρc = 0.0, e.g. for the upper and lower flanges, that mainly transmit bending loads, and for the middle parts distributing shear loads. For higher critical densities, these shear stresses are not transferred by a closed-walled web, but rather by a frame-like structure due to the penalization that is leading to a reduced economic efficiency of low densities. Thereby, sub-optimal structures (frame-like and in the middle of the design domain) also carry bending stresses, i.e. normal stresses, and contribute to the lower compliance of the structures. For structure (c) with ρc = 0.3077 (tmin = 2 mm), a smooth transition to the attachment areas was realized, as expected. The stiffness, thereby, is only 9.1% above the optimal design. It is still 34.5% stiffer than the original sheet metal design (with 19.5% less weight). In general, stiffness potentials of grey-scales can also be seen for the topologies for ρc = 1.0. Intermediate densities still exist when the termination criterion is met, even if a locally reduced stiffness-to-weight ratio is present due to a penalization of all grey-scale densities. The design for (d) is still stiffer than the original sheet with a weight saving of 19.5%, which is mainly due to allowing heights above the initial 2 mm (increased to 6.5 mm).

Designing Variable Thickness Sheets Using Grey-Scale Densities for AM

71

Fig. 4. Resulting topologies (f = 0.8) and original sheet of the L-shaped bell crank use case

The results for the 2D topology optimization (e), without the extrusion in the third (out-of-plane) direction above the original sheet thickness, come with an increase in compliance of 11.1%, compared to the original design, but also 20% in weight-savings.

Fig. 5. Printed parts using (a) MEX, (b) PBF-LB/M on the base plate, and final parts after wire EDM and sandblasting in (c) bottom view and (d) top view.

Manufacturing. After the topology optimization, the results were exported in stl-format in 3D, smoothed and manually post-processed in CAD software. Prototypes were printed using MEX before the parts were manufactured using PBFLB/M. Plastic prototypes of the structure can be seen in Fig. 5, (a). Without a minimum thickness (tmin = 0.0), the part is hardly printable using MEX since, based on experience, a minimum of three layers is recommended to produce robust printing results (Printer used: Raise3d pro2 plus, material: ABS). Built for demonstration purposes, the part without minimum thickness, shown in the picture on the right, is fragile for loads normal to the modelled plane, but stiff

72

F. Endress and M. Zimmermann

Fig. 6. (a) Hinge (grey) with solid support structure (red). (b) Magnitude of deflection after print as a simulation result for Ti6Al4V.

for the load case it was optimized for. The higher the minimum thickness was set, the more robust the part appeared to be with respect to stiffness. Components manufactured using PBF-LB/M are also shown in Fig. 5 for a minimum thickness of tmin = 2.0 mm (cf. structure (c) in Fig. 4). A critical density of 2 mm was chosen, because out-of-plane properties were not modelled, but a critical density corresponding to the height of the attachment structures resulted in designs that are expected being robust also for loads perpendicular to the modelled plane. For the building of the structures, first, manufacturing simulations in Siemens NX 2011 with Simcenter 3D Powder Bed Fusion were used to analyse stresses after print when varying support structure designs and the orientation of the part in the printer (meshing number of elements: 228793). The best results were a distortion after printing of a maximum of 0.0963 mm (abs) (cf. Fig. 6) obtained with solid support printed flat on the (rigidly modelled) base plate. Fillets were added to the support structure (height of 3 mm) to avoid delamination. Then, the parts were printed on a calibrated EOS E290 (material: Ti6Al4V). Due to reasons of confidentiality printing parameters cannot be published. After the print, a heat treatment followed. The heat treatment consisted of four steps, where steps one to three were in vacuum with atmospheric pressure of p ≤ 1 × 104 mbar. Step one comprises the heating of the base plate and the hinges at a rate of 10 ◦ /min. to 835 ◦ C ±10 ◦ C. The second process step kept the temperature at 835 ◦ C ±10 ◦ C for 2 h. Third, a cool down to ≤ 300 ◦ C was applied, before in step 4 a quench with Argon gas applied by a fan was done until temperatures of ≤ 50 ◦ C were reached. Next, wire EDM was applied to remove the parts from the base plate before a sandblasting treatment was applied. The final parts can also be seen in Fig. 5 (c) and (d). Two parts were manufactured, both without failures during print or post-processing.

4

Discussion

Stresses. As discussed, two types of stresses dominate in the bell crank: shear stresses and bending stresses. Whereas the former connect the flanges of the structure, bending stresses normal to the section cuts characterize the stress state within these flanges. As shown in Fig. 7 for ρc = 0.0, the cross-sections

Designing Variable Thickness Sheets Using Grey-Scale Densities for AM

73

resemble the U- or H-profile1 discussed in Sect. 1. However, small deviations between element densities at the inner (position e1 in Fig. 7) and outer (position e20 in Fig. 7) sides of the hinge exist. As the section becomes closer to the ends of the part, it narrows down, as loads were applied point-wise in the optimization setup. Complex, highly individual shapes of the cross-section become optimal (e.g. see dcut = 0 mm in Fig. 7). This emphasizes the need for advanced manufacturing methods, such as AM, to realize structures that are close to optimal in stiffness-to-weight ratios. In areas close to the boundary conditions (in the middle of the bell crank), as well as close to load introductions (at the end of the bell crank close to e20 ), individually shaped cross-sections are required to distribute stresses optimally. So, in an advanced and larger production setting, one could potentially replace sections of the part with pre-products (such as H-shaped standard profiles) and attachment areas by applying AM, in a hybrid manufacturing process.

Fig. 7. Element densities for different sections for the reference topology with ρc = 0.0 (tmin = 0.0, cf. structure (a) in Fig. 4). The first section (dcut = 0 mm) is located in the second row of elements after the prescribed attachment areas.

Manufacturing. In combination with the chosen discretization of the elements, a “minimum printable volume” of 1 mm3 , or minimum thickness approach in the x, y- and z-directions, was established. However, the structures found still require manual post-processing. Sharp edges and corners remaining from the coarse discretization must be smoothed, and, potentially, densities below ρc must be 1

Here, densities (in 2D) were only extruded in one direction, leading to a U-profile. Of course, extrusion in both directions is also feasible (and favourable for introducing stresses into neighbouring elements with different thicknesses), leading to an Hprofile in the presented use case. However, for some 3D printers a flat surface is easily manufacturable and therefore an extrusion to only one side was chosen.

74

F. Endress and M. Zimmermann

reduced to zero or increased to ρc to prepare structures for printing. A volumepreserving approach for this is described in [12]. In the use case, this was done manually in CAD Software, when structures were smoothed and prepared for printing. Within the process of redesigning components using AM, one must decide on the orientation of the part in the printer. With the approach followed in this paper, one orientation is particularly favourable for MEX, as the minimum thickness implicitly incorporates the build direction if 2D densities are extruded in one direction only. For MEX, this may be very efficient, as flat printing on the base plate is usually feasible, accurate and robust. In the use case of the metal hinge, only two parts were printed in a single build job. A horizontal orientation was chosen to reduce print time. Using wire EDM, the parts could easily be removed from the base plate. However, the introduced approach could also be used to ensure a minimum thickness in directions perpendicular to the build direction, for a vertical alignment of the resulting closed walls in the printer. This may enhance process robustness due to several reasons: First, closed surfaces enhance the heat convection within the part. Since heat convection is reduced by powder surrounding a frame-like structure and decreases when adjacent and contacting structures exist, non-uniform strains in the part caused by thermal gradients are reduced [19]. Second, closed-wall structures are often self-supporting in themselves (depending on the orientation in the printer), reducing warping effects and the required support structures. Finally, the closedness helps to ensure dimensional accuracy, e.g. when frames are separated in the build process and later fused again, which often leads to errors in powder-based printing processes for frame-like structures. These advantages come with the increase in stiffness for the given load case in the use case. Limitations. Several limitations of the proposed approach for the redesign of (double) sheet metal designs exist. The shown results all rely on 2D models, only considering the in-plane stresses of a bending load case. However, when element densities are interpreted as the thickness of a sheet (i.e. after the extrusion), large variations in the thickness of neighbouring elements (i.e. steps or jumps) in the final designs may exist. For these large steps, the assumption of only in-plane stresses cannot hold. Stresses may not distribute equally over the full thickness of an element, when building and loading the structures. Also, smoothing cannot be guaranteed to overcome these drawbacks. The assumption that the material is under plane stress is clearly violated locally. A local constraint for the design variables, or a finer discretization, could be applied to overcome this challenge. Alternatively, a density-based filtering could be applied to smooth transitions but also does not guarantee a homogeneous stress distribution in the single elements. In-plane stresses may lead to out-of-plane strains and deformations, which were not considered in the optimization and finite element modelling. A 3D model of hinge (b) in Fig. 4 was simulated in Altair OptiStruct, where outof-plane deformations were found at a maximum of 0.1 mm, for the load the structure was optimized for. At the same time a bending (in-plane) of 1.15 mm

Designing Variable Thickness Sheets Using Grey-Scale Densities for AM

75

occurred (Material: Ti6Al4V, force applied: 1kN). Thus, for the use case the out of plane bending is evaluated as non-critical. Even though topology optimizations in 3D are computationally more intense compared to 2D optimizations, a 3D model may still capture effects that, if applicable, can only be implicitly considered in 2D models (e.g. loads perpendicular to the modelled plane). Moreover, for the results shown, a mesh dependency is always present [5]. However, here, a certain mesh discretization was used as a minimum length scale for AM. A further limitation of the approach is that intermediate densities in the penalized area, below ρc may occur, as discussed above, and therefore post-processing is always required (cf. Sect. 3 or [12]). Furthermore, it was assumed that solid elements could be printed. In fact, porosity and anisotropic material properties characterise the builds of many AM technologies. Fatigue failure, buckling and eigenstresses (e.g. induced by thermal gradients while manufacturing) were not considered in the optimization either.

5

Conclusion

In density-based topology optimizations, stiffness potentials can be used, when intermediate densities are not fully penalized and thus included in the final designs. For flat components, this can be accomplished by topology optimization in 2D, where the resulting densities are interpreted as the thickness of a sheet. To find manufacturable designs, a penalization scheme was investigated that allows a critical density to be chosen, up to which densities are penalized. Results for two test cases are promising: The scheme can be applied as a lower limit, and thus unidirectional minimum thickness constraint, to find manufacturable designs using AM. The lower the critical density was set, the stiffer the structure was found to be for in-plane bending load cases. In a use case, different section types were identified, some similar to standard profiles (e.g. H-profiles) and others individually shaped. The latter, mainly found close to the load introduction and supports, have lightweight potential when employing advanced manufacturing technologies, such as AM. However, it remains questionable if, or up to what thickness, the assumption of planar stress states is generally reasonable. Future work should focus on the testing and validation of the topologies found. Acknowledgements. This research was conducted as part of the PROVING research project in the national aeronautical research program VI-1 funded by the Bundesministerium f¨ ur Wirtschaft und Klimaschutz. We wish to thank our industry partners Oerlikon AM Europe GmbH and RS.aero for the industrial use case, manufacturing of reference parts and inspiring discussions. The authors declare no conflict of interest.

References 1. Merklein, M., Johannes, M., Lechner, M., Kuppert, A.: A review on tailored blanks - production, applications and evaluation. J. Mater. Process. Technol. 214(2), 151– 164 (2014)

76

F. Endress and M. Zimmermann

2. Megan, L., Brian, C., Hubert, L., Sridhar, R., Robert, Y.: White paper: a designvalidation-production workflow for aerospace additive manufacturing. Technical report, DatapointLabs Technical Center for Materials Sridhar Ravikoti and Robert Yancey, Altair Engineering Inc. (2016) 3. Bendsøe, M.P.: Optimal shape design as a material distribution problem. Struct. Multidiscip. Optim. 1(4), 193–202 (1989) 4. Xie, Y.M., Steven, G.P.: A simple evolutionary procedure for structural optimization. Comput. Struct. 49(5), 885–896 (1993) 5. Sigmund, O., Aage, N., Andreassen, E.: On the (non-)optimality of Michell structures. Struct. Multidiscip. Optim. 54(2), 361–373 (2016) 6. Højbjerre, K.: Additive manufacturing of porous metal components. In: Proceedings of the 6th International Conference on Additive Manufacturing. Loughborough, UK (2011) 7. Zuo, W., Saitou, K.: Multi-material topology optimization using ordered SIMP interpolation. Struct. Multidiscip. Optim. 55(2), 477–491 (2017) 8. Bendsøe, M.P., Sigmund, O.: Material interpolation schemes in topology optimization. Arch. Appl. Mech. 69(9–10), 635–654 (1999) 9. Rossow, M.P., Taylor, J.E.: A finite element method for the optimal design of variable thickness sheets. AIAA J. 11(11), 1566–1569 (1973) 10. Groen, J.P., Sigmund, O.: Homogenization-based topology optimization for highresolution manufacturable microstructures. Int. J. Numer. Meth. Eng. 113(8), 1148–1163 (2018) 11. Wang, F., Lazarov, B.S., Sigmund, O.: On projection methods, convergence and robust formulations in topology optimization. Struct. Multidiscip. Optim. 43(6), 767–784 (2011) 12. Giele, R., Groen, J., Aage, N., Andreasen, C.S., Sigmund, O.: On approaches for avoiding low-stiffness regions in variable thickness sheet and homogenization-based topology optimization. Struct. Multidiscip. Optim. 64(1), 39–52 (2021) 13. Larsen, S.D., Sigmund, O., Groen, J.P.: Optimal truss and frame design from projected homogenization-based topology optimization. Struct. Multidiscip. Optim. 57(4), 1461–1474 (2018) 14. Brackett, D., Ashcroft, I., Hague, R.: Topology optimization for additive manufacturing. In: Proceedings of the 22nd Annual International Solid Freeform Fabrication Symposium, pp. 348–362. University of Texas, Austin, Texas, USA (2011) 15. Abdelhamid, M., Czekanski, A.: Revisiting non-convexity in topology optimization of compliance minimization problems. Eng. Comput. 39(3), 893–915 (2022) 16. Sigmund, O.: A 99 line topology optimization code written in MATLAB. Struct. Multidiscip. Optim. 21(2), 120–127 (2001) 17. Svanberg, K.: The method of moving asymptotes - a new method for structural optimization. Int. J. Numer. Meth. Eng. 24(2), 359–373 (1987) 18. Sigmund, O.: On the optimality of bone microstructure. In: Pedersen, P., Bendsøe, M.P. (eds.) IUTAM Symposium on Synthesis in Bio Solid Mechanics, Solid Mechanics and its Applications, vol. 69, pp. 221–234. Springer, Netherlands, Dordrecht (2002). https://doi.org/10.1007/0-306-46939-1 20 19. Kruth, J.P., Froyen, L., van Vaerenbergh, J., Mercelis, P., Rombouts, M., Lauwers, B.: Selective laser melting of iron-based powder. J. Mater. Process. Technol. 149(1– 3), 616–622 (2004)

Sustainability-Oriented Topology Optimization Towards a More Holistic Design for Additive Manufacturing Klaus Hoschke1(B)

, Konstantin Kappe1 , Sankalp Patil1 Junseok Kim1 , and Aron Pfaff1

, Sebastian Kilchert2

1 Fraunhofer EMI, Ernst-Zermelo-Street 4, 79104 Freiburg, Germany

[email protected] 2 University of Freiburg, Emmy-Noether-Strasse 2, 79110 Freiburg, Germany

Abstract. The Design for Additive Manufacturing of final products needs to target many design objectives, e.g., function, low lead-time, costs, ecological footprint and possibly more. In balancing the latter, results of design changes are frequently counterintuitive, and this happens especially when it comes to sustainabilityrelated qualities. The latter are commonly modeled by a Life-Cycle Assessment (LCA) of the fabrication and use-phase. However, it is likely that related data is not yet available before most design decisions have been fixed. Even though Additive Manufacturing can enable more design freedom than conventional technologies, the process-specific parameters need strong consideration. The earlier this happens in the design process, improvements might have the biggest impact. Topology optimization can be an efficient method for conceptual design and design automation. However, the respective models often need to be simplified, e.g., regarding nonlinear material properties or intricate manufacturing constraints. For this reason, it is typically not possible in topology optimization to deal with all previously mentioned criteria. A Sustainability-oriented Topology Optimization method is proposed within a generative engineering framework. Multimodal analyses of intermediate topology results should enable the computation of intricate measures. For example, regarding Laser Powder Bed Fusion-based Additive Manufacturing expedient build directions and an estimation of support structures are calculated. A predictive LCA model is included that calculates the ecological footprint measures on basis of the intermediate topology results. For that purpose, a simplified product system with representative processes is modeled. The presented approach enables a more holistic Design for Additive Manufacturing that can deal with a multitude of multidisciplinary criteria in a coherent way. Keywords: Generative Engineering · Topology Optimization · Sustainability · Sustainable Design · Lightweight Design · Metal Additive Manufacturing · Design Automation · Life-Cycle Assessment · Design for Additive Manufacturing

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 77–88, 2024. https://doi.org/10.1007/978-3-031-42983-5_6

,

78

K. Hoschke et al.

1 Introduction Lightweight design is essential for parts that are mobile, such as for car or aircraft parts. The environmental impact that results from their use phase can outweigh the contribution of the fabrication. However, in additive manufacturing of lightweight designs also less material is needed as feedstock and needs to be additively welded in the layer-by-layer process. Topology optimization can facilitate extremely effective conceptual design for lightweight components. Based on the model of the available space and loading, the design can be generated automatically considering all optimization goals and boundary conditions. So far, only few studies seek to combine topology optimization with life-cycle assessment (LCA). In [1] topology optimization and LCA for binder jetting production are coupled in a design workflow. It is proposed to iterate the topology optimization with changing the mass constraint until both mechanical as well as LCA-related goals are met. However, the described approach does not concurrently target the multiple objectives within the topology optimization but rather by altering the constraints and as such the optimization parameters. In the very recent publication [2], the coupling problem is solved with a metamodel-based approach for finding the optimized material-fabrication method pair for a given design problem, thereby, altering a multitude of parameters that are inputs or boundary conditions of the topology optimization. In contrast, the method proposed in this paper seeks to find the most effective topology without changing other parameters (e.g., material, processing method, optimization constraints). Optimization regarding the multiple objectives, such as stiffness-to-weight ratio, part quality, and sustainability-oriented measures are made possible by a generative design framework with a 2-model topology optimization. The thereby achieved diversity of design features should facilitate the multi-objective optimization. In addition, a variety of connected analyses of the physical phenomena and domains, such as manufacturing and mechanics, are employed in a multimodal approach to the modeling and simulation. The methodical framework was developed in [3]. This paper provides additional insights regarding life-cycle assessment and how it can be integrated into the multimodal algorithm. The multi-objective formulation is changed accordingly, and the design problem solved in this way gives additional information regarding the life-cycle impact of solutions. In this work, predictive LCA models are used to evaluate preliminary scores for environmental impact categories of intermediate topology optimization results. The collection of reliable data presents the primary obstacle in evaluating the life cycle [4]. However, to calculate meaningful relative measures for the comparison of early designs, also averaged data and simplified models could be sufficient. For the Laser Powder Bed Fusion (LPBF) process, the optimized printing orientation can be identified with calculating related measures on basis of the intermediate topology result. Those are the build time, thermal deformation tendency, support structure volume and post-processing effort. Many orientation options are evaluated using the commercial software Amphyon with its assessment module. The process-related models are discussed in detail in [5], and results with detailed description using the same method for assessing the build orientation are presented in [3]. In this paper, the LCA is integrated with modeling the LCA for the best orientation that is selected for each intermediate topology result.

Sustainability-Oriented Topology Optimization Towards a More Holistic

79

With the generative design including LCA, the sustainability-oriented objectives can for the first time be optimized in topology optimization. The most promising solutions can then be selected for further development. The proposed workflow for the Design for Additive Manufacturing is illustrated in Fig. 1.

Function unction Costs

CAD Design Space

Detailed CAD design

Topology optimization

Shape Optimiza tion

Build orientation & support design

3D printing printin p

Lead ead time Preliminary fabrication assessment

Multi-objective optimization

Sustainability-oriented topology optimization Preliminary LifeCycle assessment

Generate multiple topology results

Multimodal analyses

Fig. 1. Illustration of the complete workflow for the Design for Additive Manufacturing, starting with the design requirements and ending by the final 3D printing with integrating sustainabilityoriented topology optimization in the early stage

The results of the topology optimization frequently require interpretation and clarification to satisfy all fabrication and usage requirements. After selecting the most promising topology, the design can optionally be designed in CAD and shape optimized. In the further progression towards manufacturing, the final build orientation is determined in conjunction with the design of support structures for overhanging surfaces and potentially allowances for milling or other features. A detailed thermal simulation of the deformation tendencies can be carried out, e.g., with the Amphyon software simulation module (see [6] for the modeling description) to validate the configuration. The build direction and support structure design can be altered if excessive deviations are discovered. Manufacturing failures have been shown to be reduced by this strategy, see e.g. [9]. It could happen that the final design steps (detailed CAD, final printing direction, support design etc.) do not give good results for the best selected design from topology optimization and deviate from the predictions. In this case, the generative concept has the advantage that further alternatives have been generated, so that the next best solution could be selected for the detailed design phases. However, further use-case studies will be needed for validating the quality of the predictions with respect to the final design. The sustainability-oriented topology optimization is based on the multimodal topology optimization concept for generative design that has been conceptually presented in [7], published in an early 2D-version in [8] and firstly developed for 3D and in a sustainable design context in [3], however, not yet with LCA integration. The concept of integrating predictive LCA models was first presented in [9] as a foresight of ongoing work. In the meantime, the implementation and formulation have been re-worked and extensively tested. Some results are presented below.

80

K. Hoschke et al.

2 Method Description The workflow of the sustainability-oriented topology optimization is illustrated in Fig. 2. The standard (std.) topology optimization serves as a reference and starting point for the 2-model topology optimization. Both are performed with the OptiStruct® finite element code with the Solid Isotropic Material with Penalization (SIMP) method. The concept for the 2-model topology optimization with inverse-damage models is introduced in [3]. The optimization is iterated with adjusting the weighting between the two inversely damaged models as a design variable. For all generated topology results multiple analyses are performed. Prior to this, each topology result, which is calculated in the finite element domain, is transferred to a geometry representation with a distinct boundary surface using a threshold value for the material utilization in a given area. Inputs: - Design space - Constraints - Paramet ers

Std. Topology optimization (ref erence)

Orient at ion analysis

2-model Topology optimization

Linear analysis

Nonlinear analysis

Life-cycle assessment

Evaluat e multiobject ive

Outputs: - Best Design - Results - Comparison

converged

Updat e design variable

Fig. 2. Workflow of the sustainability-oriented topology optimization with multimodal analyses and integrated LCA

With respect to the mechanical performance, the same static loading case as in the topology optimization is reevaluated by a more detailed linear analysis on a new, more accurate finite element mesh representation of the geometry with its bounding surface. Additionally, the performance regarding an overloading scenario is evaluated by nonlinear analysis. The LPBF-process-specific measures are evaluated in an orientation analysis of multiple possible configurations. The latter is performed with the Amphyon software with the same method described in [3]. For the selected best orientation, the lifecycle impact assessment (LCIA) is performed to evaluate a preliminary score regarding environmental impact categories. For the topology optimization, the two most common objective and constraint configurations are possible. Those are the optimization of stiffness with a mass (or volume) constraint, as well as minimizing mass (or volume fraction) with a stiffness constraint (e.g., local displacement, or the global measure being compliance). They can also be combined in that the reference topology optimization can have a different setup than the 2-model topology optimization, if the constraints can be enforced. The latter could be achieved by adjusting the threshold of material utilization in generating the topology result in the geometry domain. Thereby the topology and effectiveness are not changed, but the values for mass and compliance change within certain limits because the structures become slightly slimmer or thicker. The overall objective is formulated to be largely independent from the topology optimization configuration and is clustered by three categories. The first one is the

Sustainability-Oriented Topology Optimization Towards a More Holistic

81

mechanical performance with respect to the volume of the part. Additionally, to the static stiffness, here represented by the compliance c, also the maximum bearable load before breakage in an overloading scenario is evaluated. The latter is modeled by calculating the maximum reaction force of the structure for a prescribed progressive displacement at the same load application point as in the static scenario. The objective for mechanical performance is formulated as follows   cstat,X VX Fmax,X V0 (1) + wFmax ∗ 2 − FMech. = wstiff ∗ c0 V0 Fmax,0 VX In Eq. (1) the subscript X stands for the value of the current design and 0 for the reference solution. Two factors wstiff and wFmax are introduced to weigh between the contributions. To maximize the static stiffness, the compliance is to be minimized. Contrary, the maximum reaction force Fmax , that counteracts an excessive deformation in overloading, is to be maximized. The objective FMech. takes on the value 1.0 for the reference and is better for topology results with smaller values. A process-related objective is formulated in the same manner with modeling the fractional deviation to the reference design and comparing measures for the deformation tendency def and for the post-processing effort for downward facing surfaces pp, which is strongly correlated with surface quality for the LPBF process. The objective FProcess can be seen as a model of reliability and quality of the processing and is formulated with FProcess = wdef ∗

def X pp + wpp ∗ X def 0 pp0

(2)

The third objective models the impact of the design regarding ecological categories, which are here represented by the climate change potential and metal depletion, both calculated in the LCA. The respective objective FLCA is formulated as follows FLCA = wCC ∗

CC X MDX + wmd ∗ CC 0 MD0

(3)

In Eq. (3), the CC represents the values calculated in the life cycle impact assessment for the climate change potential in kg CO2-eq./unit and the metal depletion MD in kg Fe-eq./unit. All three objectives are combined in one top level multi-objective function being FObj = wM ∗ FMech. + wP ∗ FProcess +wP ∗ FLCA

(4)

The life cycle assessment model for additively manufactured components, which was incorporated into the topology optimization within the scope of this work, is presented in the following section. The AM product system is illustrated in Fig. 3. Most common post-processing steps are included. It is known from [9] and [11] that the AM process with LPBF likely has a high influence, however, this strongly depends on the used machine and setup. The raw material, energy, such as that used to heat the powder, and other utilities, such as compressed air, are the primary inputs to the process chain. Only a small portion of the

82

K. Hoschke et al. Raw mat erials

Elet ricity, utilities

Elet ricity, utilities

Elet ricity, utilities

Elet ricity, utilities

Elet ricity, utilities

Elet ricity, utilities

Powder atomizat ion

AM process (LPBF)

Thermal treat ment

Support removal (Saw ing)

Blast ing

Milling

Compressed Air

Powder reuse

AM machine

Wast e powder

Product ion waste

Support nodes

Scrap aluminium

Consumables

Fig. 3. AM product system used in the study. Common post-processing steps are included. Consumables of the AM process as well as the support nodes that could potentially be recycled after removal are cut-off.

feedstock is used for the additively laser-melted raw structure, and a lot of the powder is used again after each manufacturing cycle. The product system is implemented in Brightway 2, an open-source software for life cycle assessment. The IPCC 2013, GWP 100a, and ReCiPe Midpoint (Hierarchist) V1.13 methods are used to calculate the impacts. For the AM process, a predictive resource model that was developed and experimentally validated in [3] is used to calculate the electricity consumption, compressed air and powder used. For all other processes the ecoinvent 3.8 market data with cutoff is employed to represent averaged data from the industrial system. The integration of the LCA with Brightway 2 into the topology optimization is established and automated via python. An excel file is generated and updated with each new topology as an input for the LCA. The data includes the predicted resources in the AM process and other relevant measures with respect to each intermediate topology. Those are e.g., the electricity used by the AM machine per part, the build time, the volume of the part, the volume of the support structure and the batch size, which depends on the occupied space on the build platform. After creating the life cycle inventory (LCI), as a final step, the life cycle impact assessment (LCIA) regarding the climate change potential and the metal depletion are calculated. In addition to the fabrication, also mobile use-phase modes can be considered. E.g., in the Ecoinvent database, there is data stored for “market for transport, passenger” which can represent the use-phase for a passenger car in a simple manner. The driven kilometers (e.g. 150,000 km) are allocated with the weight ratio between the part and the vehicle and the lifetime of the part is assumed to be the same as for the car.

Sustainability-Oriented Topology Optimization Towards a More Holistic

83

3 Results The sustainability-oriented topology optimization is performed for a cantilever beam problem for a mobile mode of a passenger car and a production of 100,000 parts. The dimensions of the design space are 60 × 60 × 180 mm3 . The beam is loaded at the dropout with a force of 500 N. The finite element model is built with uniform cubic elements with a length of 3 mm. The std. Topology optimization is performed as a reference with a min. Compliance formulation subject to a volume constraint with 15% utilization of the design space. Minimum structural member size control is applied with 10 mm. The iteration count is set to 75. The 2-model topology optimization is performed using a minimum mass with compliance constraint setup. The constraint value is taken from the reference. This represents the design logic of generating alternative solutions to the reference, featuring at least the same stiffness, however, if possible, with lesser mass. If the compliance in the more detailed reanalysis significantly differs from the constraint, the density threshold for generating the geometry is automatically varied such that the compliance is equalized while changing the mass. In doing so, all topology results have the same stiffness, but the more efficient topologies feature a smaller mass. The material used is Scalmalloy© with Young’s modulus 70 GPa, Poisson’s ratio 0.33 and density 2.67 g/cm3 . For the nonlinear analysis, a Johnson-Cook material and damage model is used to model Scalmalloy© in a relatively ductile material state with ca. 12% elongation before breakage. For this study, the EOS M400 system with a single 1 kW laser unit is adopted. The nitrogen inert gas atmosphere for the process is produced within the AM machine, however leading to a relatively high consumption of compressed air. The modeling parameters are the same as stated in [3]. For the example, all weighting factors are set equally and thereby model no preference by the designer. The result of the sustainability-oriented topology optimization run is illustrated in Fig. 4. Design 0 is the reference solution and 10 alternative solutions are generated with the 2-model topology optimization. The individual objective values as well as the overall inclusive objective function Fobj. are stated in the graph. The progression of the design solutions is calculated based on the golden section search. It is clearly visible that this optimization algorithm can find the local minimum represented by the design 7. The design solutions 4, 8, 9 and 10 have the same topology and vary only very slightly in their shape. The design 7 has the best individual value for the LCA objective and is a good compromise for the other two individual objectives leading to the best overall value. The best mechanical performance is shown by design 6, however compromised with less good ratings for the other objectives. The best performance with respect to the process criteria is represented by design 2. While some objectives correlate, like mass reduction in FMech. also improving impact aspects in FLCA due to lower material use, Fig. 4 shows also non-correlating aspects between the two (see e.g. design 2 and design 6) related e.g. to the LCA impact of support structures or differing batch size. The proportions of the individual objectives to the inclusive objective are changing, so that different preferences, expressed by different weightings, would lead to different solutions. This suggests that the objective formulation is appropriate, and the LCA-related objective is significant to the selection. However, this needs further review in use cases.

84

K. Hoschke et al. Design 0 Design 7

Design 1

Design 2

Design 3

Design 4

Design 5

Design 6

Fig. 4. Generated topology results for the design problem (design space and loading: middle, green, red). Individual and inclusive objectives (left) and respective design solutions (right).

Force in [N]

In the following, the reference (design 0) and the best overall solution (design 7) are compared in more detail. The static compliance is equal because it is set as a constraint for the algorithm. However, the mass of design 7 achieving the same stiffness is reduced by ca. 6.4%. In addition, even with using less mass, the reaction force in the overloading scenario is significantly increased with ca. 6.5% higher maximum loading capacity before breakage. The latter is illustrated in Fig. 5 (left). The reason for this is that the load and resulting stresses are distributed more evenly over the component during the overloading scenario, as depicted in Fig. 6 for the displacement (7.2 mm) shortly before the breaking event. The mechanical performance of design 7 is calculated to be ca. 9.9% better than the reference, which also accounts for the reduced mass. Design 7

45000

Design 0 Design 7

37500

30000

22500

15000

7500

0 0

2

4

6

8

10

12

Displacement in [mm]

Design 0

Fig. 5. Reaction force in the overloading scenario of progressive prescribed displacement at the dropout (left). Exemplary support structure design (blue, right) for the two designs, each in their respective best printing orientation.

Sustainability-Oriented Topology Optimization Towards a More Holistic

85

Fig. 6. Plot of the von Mises stress distribution for design 0 (left) and design 7 (right) with a threshold on higher values that could lead to plastic strains for a displacement at the dropout of 7.2 mm, shortly before breakage occurs.

The process performance is modeled from the estimated values for deformation tendency and post-processing effort. Thereby, design 7 achieves ca. 14% less deformation tendency and up to 40% less post-processing effort. The former mainly refers to the size and gradients of the cross-sections in printing direction. Sudden cross-section changes can lead to geometric deviations. The post-processing effort is correlated with the amount of support structures and their accessibility. The Amphyon software estimates a reduction for support structure volume of up to 70% for design 7 in comparison with design 0. Even though the calculated values are rough estimates, they are quite plausible when looking at the topologies with their proposed best building direction. In Fig. 5, the support structure design in the Magics software is illustrated as a type of validation. Here, even a reduction of 85% of support structure volume is achieved. With slight shape adaptions they could be further reduced for both designs, however, an experienced look can comprehend that the overall printability could be much better for design 7. The LCA score is calculated from the contributions of climate change potential and metal depletion. The climate change potential is calculated to be 14.9% less and the metal depletion to be 14.7% less for design 7 in comparison with design 0. The two main inputs that lead to the improvement are the reductions in part mass and support structure mass. In addition, comparing the space taken on the building platform, design 7 occupies much less, leading to an estimation of up to 78 parts that can be printed per batch in comparison to only 36 parts for design 0. In Fig. 7 and Fig. 8, the calculated values and listing of contributions by the constituent processes for the climate change potential and metal depletion categories are depicted. Even if the absolute values are rough estimates and some of the input parameters like the estimation of support structure mass can lead to significant tolerances, the relative importance resulting from the compilation can provide interesting insights. The use-phase contribution is only influenced by the mass of the part in relation to the vehicle mass and scales with the number of parts and the number of kilometers driven. In both categories, the use-phase takes up ca. 25% of the impact. Second most important is the contribution of the AM process and third the production of the Scalmalloy© feedstock. Both AM process and production of feedstock correlate with the amount of material that is printed for the parts and supports. In the metal depletion category, the AM process is more dominant, as the production of the AM machine takes up a significant portion.

86

K. Hoschke et al.

Fig. 7. Listing of contributions by the constituent processes of the product system regarding the climate change potential for the best design 7. Estimation for fabricating 100,000 parts and a use-phase as a passenger car component.

For both categories, the amount of compressed air is quite significant and looking at other machine setups, e.g., with inert gas cylinders, could be promising. For the Scalmalloy© production, the contribution by the alloying elements to the overall impact changes strongly between the two categories. In any case, the choice of alloy and its components have a great influence, and the overall impact of the feedstock production could change significantly for a different aluminum alloy or other lightweight materials like titanium.

4 Discussion and Conclusion Topology optimization and LPBF-based additive manufacturing are potent technologies for lightweight design and sustainability. By incorporating the predictive life cycle assessment (LCA) into topology optimization, a novel workflow for generative design is created that directly targets environmental impact categories as design objectives. It is demonstrated that with the presented implementation of the sustainability-oriented topology optimization a multitude of design criteria from different domains can be evaluated in an automated and coherent fashion. In addition, the results show that significant improvements can be achieved in comparing different design solutions leading to a more holistic design. Both evaluated ecological impact categories and all main contributors to the footprints benefit significantly with a strict lightweight design and an optimized build direction for fewer support structures to reduce the amount of fabricated material. The relevant parameters of the LPBF-based additive manufacturing production are incorporated

Sustainability-Oriented Topology Optimization Towards a More Holistic

87

Fig. 8. Listing of contributions by the constituent processes of the product system regarding the metal depletion for the best design 7. Estimation for fabricating 100,000 parts and a use-phase as a passenger car component.

based on calculating relative, predictive measures with respect to the intermediate topology results. In doing so, the most promising topology results can be selected from the multimodal optimization towards completing the detailed design phases for the final fabrication. The exemplary result showed that without the need for experiments, it is possible to predict the ecological impact. Even though the absolute values can have large tolerances, the relative significance for a specific design problem can be very informative for taking design decisions. The overall optimization took ca. 3 h with each additional generated design taking ca. 20 min. on a desktop PC with 4 CPU cores employed. However, so far, the implementation is only demonstrated for a plain 3D geometry and relatively simple loading conditions and must be demonstrated for more complex design problems, e.g., with multiple loading scenarios. In addition, the comparison of lightweight materials like suitable titanium alloys with aluminum alloys and different fabrication setups like LPBF machines with other print envelope or multi-laser would be interesting. When comparing machine setups or different alloys, the accuracy of the used models for the process and footprint estimation become more important. For such extensions the modeling of resources in the AM process and process-specific measures like the support structures need to be reworked and be validated with a wider scope and more experimental data. In addition, other approaches for generating the design diversity needed in the generative design could be thought of. They could extend the here employed

88

K. Hoschke et al.

inverse-damage model of the 2-model topology optimization. Especially specific implementations that steer the design features towards the chosen target objectives, e.g., with adjusting manufacturing constraints, would be beneficial. Currently no manufacturing constraints are used in the topology optimization. E.g., it could be beneficial to limit the number of overhanging surfaces with constraint functions to further reduce the amount of support structures.

References 1. Tang, Y., Mak, K., Zhao, Y.F.: A framework to reduce product environmental impact through design optimization for additive manufacturing. J. Clean. Product. 137, 1560–1572 (2016) 2. Duriez, E., Azzaro-Pantel, C., Morlier, J., Charlotte, M.: A fast method of material, design and process eco-selection via topology optimization, for additive manufactured structures. Clean. Environ. Syst. 9, 100114 (2023) 3. Hoschke, K.: Sustainable Design with Topology Optimization for Laser Powder Bed Fusion of Metals. Doctoral thesis, Freiburg (2021) 4. Hauschild, M.Z., Rosenbaum, R.K., Olsen, S.I.: Life Cycle Assessment. Theory and Practice. Springer, Cham (2018) 5. Additive Works GmbH: Amphyon 2021 Documentation (2021) 6. Keller, N.: Verzugsminimierung bei selektiven Laserschmelzverfahren durch Multi-SkalenSimulation. Doctoral thesis, Bremen (2017) 7. Hoschke, K., Kappe, K., Bierdel, M., et al.: A topology optimization based design model for enhanced energy absorption and fail-safe behavior. In: 13th World Congress of Structural and Multidisciplinary Optimization WCSMO-13, Conference Proceedings, Beijing (2019) 8. Hoschke, K., Kappe, K., Riedel, W., Hiermaier, S.: A multimodal approach for automation of mechanical design. In: Abali, B., Giorgio, I. (eds.) Developments and Novel Approaches in Nonlinear Solid Body Mechanics, STRUCTMAT, vol. 130, pp. 301–323. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-50460-1_17 9. Hoschke, K., Kilchert, S., Kim, J., et al.: Sustainability-oriented topology optimization of aircraft components and best practices in LPBF-based metal additive manufacturing. In: Towards Sustainable Aviation Summit TSAS2022, Conference proceedings, Toulouse (2022) 10. Pfaff, A., Bierdel, M., Hoschke, K., Wickert, M., Riedel, W., Hiermaier, S.: Resource analysis model and validation for selective laser melting, constituting the potential of lightweight design for material efficiency. Sustainable Production and Consumption (2020) 11. Bierdel, M., Pfaff, A., Kilchert, S., Köhler, A., et al.: Ecological and economic Assessment of Resource Use: Additive Manufacturing Processes in industrial Production. VDI ZRE Study, (2019)

Process Chain 1: Digital Process Chain

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M: Practical Implementation Within a Digital Thread and Its Advantages Konstantin Poka(B)

, Benjamin Merz , Martin Epperlein , and Kai Hilgenberg

Bundesanstalt für Materialforschung und-prüfung (BAM), Unter den Eichen 87, 12205 Berlin, Germany [email protected]

Abstract. The industrialization of AM is only possible by creating synergies with the tools of Industry 4.0. The system technology of Powder Bed Fusion with Laser beam of Metals (PBF-LB/M) reached a level of high performance in terms of process stability and material spectrum. However, the digital process chain, starting from CAD via CAM and plant-specific compilation of the manufacturing file exhibits media disruptions. The consequence is a loss of metadata. A uniform data scheme for Design for Additive Manufacturing (DfAM), the PBF-LB/M process itself, simulations and quality assurance is currently not realized within industry. There is no entity in the common data flows of the process chains, that enables the integration of these functionalities. As part of test bed for the quality assurance in AM within the initiative Quality Infrastructure (QI)-Digital, an integration of the CAD/CAM chain is being established. The outcome is a file in an advanced commercially available format which includes all simulations and manufacturing instructions. The information depth of this file extends to the level of the scan vectors and allows the automatic optimization and holistic documentation. In addition, the KPIs for the economic analysis are generated by compressing information into a unique file combined with the application of a digital twin (DT). The implementation and advantages are demonstrated in a case study on a multi-laser PBF-LB/M system. A build cycle containing a challenging geometry is thermally simulated, optimized, and manufactured. To verify its suitability for an Additive Manufacturing Service Platform (AMSP), the identical production file is transferred to a PBF-LB/M system of another manufacturer. Finally, the achieved quality level of the build cycle is evaluated via 3D scanning. This evaluation is carried out in the identical entity of the production file to highlight the versatility of this format and to integrate quality assurance data. Keywords: Laser Powder Bed Fusion · Digital Twin · Data integrity · Process chain integration · Computer Aided Manufacturing

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 91–114, 2024. https://doi.org/10.1007/978-3-031-42983-5_7

92

K. Poka et al.

1 Introduction In Product Lifecycle Management (PLM), different representation schemes of the digital model of a part exist for each stage, starting with the CAD file of the model, the tessellation as a preparation for printing, the generation of a production file and as an example the digitization via a point cloud for quality assurance [1, 2]. The requirements for a digital thread, which utilizes all these representations in one software suite with one production file, are manifold due to the high degrees of freedom in process control for AM. The process planning and adjustment based on simulation has to be iterative to come closer to the common goal of “print first time right” [3]. To implement this, media disruptions in the process chain during the pre-process and slicing must be fixed by performing scan path creation within the CAD/CAM environment linked to the model of the part. This approach satisfies the high demand for flexibility and bidirectional adaption of alterations of the digital model and process planning [4]. This work presents a machine-specific compilation of the build cycle in analogy to post-processors of conventional tooling machines. The integration of the socalled build processor for AM into a CAM environment is the key for a holistic process view [5].

2 Digital Framework Currently the AM process chain is implemented by a portfolio of proprietary applications with limited interoperability and data consistency, caused by media disruption [6], see Fig. 1. To dissolve them a software suite, along with a production file entity, must be able store and link all the datasets captured. In addition, an interface for their traceable transfer within the different stages of the process chain is mandatory.

Fig. 1. Media disruptions along the digital process chain according to [6]

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

93

2.1 CAD Modelling for AM The initial step is, to create a geometric model of the part that fulfills both functional and economic demands. There are two different workflows available in the applied CAD/CAM environment for describing the geometric shape. Both workflows are registered within the same World Coordinate System (WCS), and they are discussed below. The combination of these workflows enables the designer to meet the requirements of opportunistic and restrictive design guidelines for functional integration, without switching to different CAD suites. The Design for Additive Manufacturing (DfAM) is supported in addition by algorithms for e.g. topology optimization. Although there are Software Developer Kits (SDK) available for exchanging formats, these can often lead to drawbacks such as information loss and errors [2]. Direct Modeling. Direct modelling follows the concept of an analytical description of the model. All major CAD programs using boundary representations (b-reps) for the elements out of which the solid models are created. B-reps are based on topologies such as vertices, edges, and faces, which are defined by explicit geometry elements such as points, curves, and surfaces [7, 8]. This model structure causes problems during distinguishing between the inside and outside for complex geometries. The direction of the normal vector to the outer surface of the model boundary is defined by the number of crossings between a projected ray and the part. This fact is crucial for slicing during the pre-process, see Fig. 1. The DfAM is restricted in the design freedom and the model is error prone, especially with a high number of internal features [7, 9]. Implicit Modeling. Implicit modelling is a far more efficient way to model complex geometries like Triply Periodic Minimal Surfaces (TPMS) [9]. For this modeling approach no explicit calculation of any edge or vertice is mandatory. The representation is performed through Boolean operations and mathematical functions of x, y and z [10]. As an example, a gyroid structure is shown in Fig. 2 based on Eq. (1): 0 = (cos(ki x)sin(ki y) + cos(ki y)sin(ki z) + cos(ki z)sin(ki x))2 − t 2

(1)

where ki = 2π Lnii , i = x, y, z, ni are the numbers of cell repetitions in x, y and z, Li are the absolute sizes of the structure and t is the cell wall thickness [11, 12].

Fig. 2. Left: Gyroid Cube with a TPMS function periodicities ki of ≈ 1.36 mm and t of 0.8 mm Right: Distance field for 0.1 and 0 as boundary representation of a slice at z 3 mm

94

K. Poka et al.

2.2 CAM Based on a Digital Twin The CAM is a crucial step in the process chain of AM. During this stage the model data gets prepared and enriched specifically for the PBF-LB/M [13]. The parts are optionally manipulated with print marks to ensure clear identification of identical ones within one build cycle. Afterwards, the position and orientation of the parts are defined within the build chamber respectively the Machine Coordinate System (MCS). The connection of the build processor with the CAM software allows a real representation of the specific machine. The user gets support by notifications if a violation of a machine intrinsic constrains of the build envelope is detected. Once the part orientation is completed, the support generation is performed to bridge the gap between aligned parts and the build platform by accessing support libraries [7]. The manufacturing plan with the assigned process parameters can now be evaluated based on restrictive aspects from design guidelines like the VDI 3405 part 3.2 [14]. In addition, it can also be further improved by algorithms for e.g. part reorientation, to comply with the limitations stated in design guidelines. The link between CAM and CAD still allows the manipulation of the as designed part itself, to enable its manufacturing with a reduced risk of process abortions or defects [15]. Changes applied to the part are adopted by the production file. The supports and the scan paths are regenerated automatically. Finally, the production file can be compiled for the specific machine. The mandatory steps are described in detail below. CAD Data Conversion and Modification. To enable the calculation of the scan paths, the hull i.e. surface of the parts and their supports needs to be converted for slicing via tessellation [8]. At the moment, mainly two data formats with different information depth regarding the AM process are deployed [10]. Most common for PBF-LB/M data representations is still the Surface Tessellation Language (STL) [9] with the drawbacks presented in Table 1. The differences between STL and a less error prone representative out of DIN EN ISO 52950 [16] are highlighted based on the example of the gyroid cube of Fig. 2 in Table 1, regarding file size and informativeness. Table 1. Comparison of STL and 3MF STL

3MF

Creation

1987 by 3D Systems 2015 by the 3MF consortium

Human readable

No

Yes via XML

Stored information

Mesh

Units, textures, nesting, materials, colors, mesh, machine

Risk of manifold triangles

Yes

No

Number of stored vertices

16,6848

27,800

Number of facets

55,616

55,616

Size of TPMS gyroid cube Chordal Tolerance: 0.0025 Angular Tolerance: 1.0000

2,716 Kilobyte Stored binary

796 Kilobyte Stored binary

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

95

While the STL only contains the geometry, there is a newer open-source standard the 3D Manufacturing Format (3MF). STL and 3MF are consisting out of vertices V , edges E and faces F. An efficient proof for watertightness via the validation of Euler’s polyhedron in Eq. (2) can be applied to both [17]. V −E+F =2

(2)

3MF uses curved triangular tessellations and a header in the Extensible Markup Language (XML) to store the geometries and the part location within the MCS in a human readable format as well as process related metadata [16]. 3MF allows compressing the declaration of vertices and the use of them of more than in one triangle via indexing. Unlike STL it does not support overlapping or manifold triangles in the tessellation, consequently the file format is less error prone [18]. Even though there is no commercial widespread implementation of implicit model representation, it enables an even more efficient approach for slicing [19]. If the geometry is already described implicitly, there is no need for a mesh conversion anymore. The hull, i.e. surface, can be determined via a distance field of a continuous scalar function of Eq. (3) to derivate the boundaries [10], visualized exemplary for one slice for the gyroid cube, see Fig. 2. f (x, y, z) = 0

(3)

This allows integration of algorithms for support generation within the part modelling. The problem of explicit modelling, that only the surface representation is available, is solved. A holistic support generation is implemented based on heat accumulation, variety in cross section and the distribution of the part volume Vm [20]. Scan Path Calculation. The calculation of the scan paths is performed individually for each slice respectively xy plane of the build cycle beforehand [19]. The start- and endpoints of all vectors of the contour ($$Polyline) and the hatch ($$Hatches) of all layers are stored in the background in a Common Layer Interface (CLI) [21, 22] based format. In this representation of the build cycle, the process parameters like the scan speed vs and the laser power PL are still missing. To underline this, a code snippet of one slice in the MCS of a PBF-LB/M machine for the gyroid cube from Fig. 2 at the coordinates 150 150 and the z level of 3 mm or corresponding layer 50 is provided on the next page. Nevertheless, there are also existing other data schemes or machines which are operating “on the fly”. For this type of operating the scan paths of layer n + 1 are computed synchronous to the exposure of the layer n. For the PBF-LB/M system EOS M300-4, Electro Optical Systems, Krailling, Germany, in the case study, the CLI is the foundation and is discussed further. Based on it, the creation of a proprietary .task file is performed, which can be send to the machine from the CAM suite via an ethernet network connection [23]. Only this entity stores all executable beforehand computed machine specific commands and corrections. It is an extension of the CLI and adds the applied laser source, laser power PL and scan speed vs for the scan paths i.e. vectors. The vectors are further distinguished into contour, hatch, up- and down skin [24].

96

K. Poka et al.

The metadata and linked process parameters are mirrored in a CAM production file which allows the identical creation of this executable file and its visualization [2]. In addition, access to the executable file is possible via the SDK of the Original Equipment Manufacturers (OEM). The instance of the production file can be compiled in an across platform manner via a machine specific build processor. It can be visualized to check the implementation of the desired process control, also for other machines of different OEMs. The sliced build cycle forms the fundamental prerequisite for the scan path calculation [4], along with the process instructions that are stored in a parameter set for a specific material. The current integration depth for assigning and manipulating the process parameters of the used build processor at the multi-laser PBF-LB/M system and the CAM software is summarized in Table 2 [25]. Table 2. Features of 3rd party integration of the post processor of the EOS M300-4 in NX 2212 Feature Assign material set Export task

Category Functionality Build cycle related Exposure View and select the powder material set Save to disk or direct transfer to Export machine Part related

Assign exposure set

Exposure

Laser assignments

Exposure

Laser beam compensation

Exposure

View and select the exposure set Select the application of laser sources Beam offset adjustment

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

97

Build Processor for PBF-LB/M. The production file format serves as a container for various data related to the manufacturing process, such as CAD models of parts, machine configurations at specific points in time, proprietary process parameters with their assignment to parts, and optional optimizations performed via simulations [18]. Figure 3 illustrates the interaction between the EOS M300-4 and the CAM suite of NX 2212 with all the various file types involved. The mandatory inputs are the CAD models in the .prt format and the desired material process parameters. For EOS, the material process parameters are stored through a .eospar file extension and allows the creation and assignment of process parameters that are part specific. An XML file represents the PBF-LB/M machine, which notes the serial number and IP address to establish an inbound connection via an ethernet network. Through this connection, the current machine setup, including the correction matrix of the scan heads, can be downloaded in a file with a .ctb extension. This feature enables creation of a digital twin (DT), which allows the calculation of the coordinates of the scan paths within the MCS of the physical machine and their timestamps beforehand. The build chamber and build platform dimensions are included in another XML file that stores the specific machine type’s dimensions. The .pax file is a proprietary XML clone of NX 2212, enabling visualization. The user can model the build platform and annotate regions on the build surface where part placement is prohibited in a .prt file. Afterwards, this gets linked to the machine via the .pax entity. Once the interaction is initialized, the user only needs to assign the process parameters and send the .task file to the machine or store the .openjz file for manipulations that are not yet available, see Fig. 3.

Fig. 3. CAM integration of EOS build processor with images from [15, 25]

The output is generated by forwarding the production file to the SDK. The configuration and the calculated scan paths are mirrored back to the CAM Suite. This allows visualization and verification of the applied laser power PL , xy- and focus correction of the individual laser source. Afterwards, the CLI is enriched with the information about applied process parameters as well as idealized time stamps which are assuming an ideal process with no additional delays due to deviations. The information is transferred in a

98

K. Poka et al.

JavaScript Object Notation (JSON) file. The last contour vector of the layer 50 out of the CLI snippet is presented below and visualized in Fig. 4.

To decrease file size, different vectors can belong to the same vector data. Their exposures are indexed like in a dictionary with key value pairs. Process insights on the vector level are provided with different vector types and their scanning parameters available within the CAM suite. As a showcase the processed vectors of layer 50 from Fig. 2 are illustrated in Fig. 4. The applied process parameters of the JSON files are additionally available in separate tables in the CAM environment.

Fig. 4. Scan paths of slice 50 at z level 3 mm out of the CLI representation, colors of vectors are blue: down skin; dark green: infill; yellow: CAD contour; light green: contour; purple: up skin; black: jumps; red: last vector of the JSON snippet (Color figure online)

By this functionality, a DT is integrated in the CAM [23]. For machines which are operating “on the fly” it is the only way to receive the scan paths before the build cycle execution. The DT is also available for most of the OEMs of PBF-LB/M systems in an extended version as an additional service. A DT of the machine type is cloud hosted and can be adapted by uploading the configuration of the real owned machine of the customer. The whole interaction of the machine components during the build cycle together with the Human Machine Interface (HMI) is then simulated.

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

99

2.3 Analysis of Process Design and Control Via Multiphysics Simulation The avoidance of common error patterns for AM out of VDI 3405 part 2.8 [26] can be supported via a multiphysics simulation [27]. The process design gets evaluated, and process anomalies occurring with a high probability are detected. Idealization for Mechanical and Thermal Slicing. Two different options are currently available for the idealization, see Fig. 5. The first approach describes the model and the support via a 3D mesh like a tetrahedral STL [22]. The displacement of the nodes is computed. The second option is to voxelize the model, where the displacement of the three-dimensional increments is calculated [28, 29]. The complexity of the voxelation is significantly lower in comparison to the surface meshing, and less error prone due to manifoldness or the creation of no watertight meshes, shown in Eq. (2). Both methods are depending on excluding part regions like print marks and on the compression of the mesh. The compression process leads to a mesh primarily containing reference points. The arrangement of these reference points plays a crucial role in determining the mechanical and thermal characteristics during the build-up of the part. The specific techniques for performing this simplification are still an ongoing research topic [29]. The result of this feature reduction is a lower computational cost with an acceptable loss in model information in irrelevant regions due to the lower granularity of the model description.

Fig. 5. Comparison of two different model representations of the gyroid structure from Fig. 2 Left: Tetraeder mesh idealization with element size of 60 μm with detailed view Right: Voxel idealization with edge length of 60 μm with detailed view

Material Properties. The foundations of the material model are the thermal and mechanical characteristics of the raw material, its powder, and its melt [30, 31]. The effects and the interactions on the level of single scans path i.e. the meso scale are extracted out of a calibration procedure [32]. The aim is to adapt the material and process parameters in the simulation to create a shrinkage deformation behavior that is equal to the measured shrinkage in the printed part. Inherent strain is dependent on the stiffness of the part as it is build up during printing and causes thermal shrinkage [27]. It can be expressed as a product of the dilation coefficient α, here a function of the stiffness C and the difference T between the solidification and the reference temperature in the build chamber [27, 30], see Eq. (4). εth = α(C) · T

(4)

100

K. Poka et al.

Process Parameters. As stated previously, the thermal shrinkage of whole layers instead of single tracks on macro scale is simulated to reduce too high computational cost [30]. Producing a solid cube with 10 mm3 requires already 167 layers of 60 μm with overall 24,000 of exposure vectors (8 mm each) and combined 120 m of scan path. To bring down the runtime further, super-layers, such as 1 mm for the mechanical and 6.4 mm for the thermic slicing, are established in this work and applied instead of process intrinsic z steps of 60 μm. To reduce the approximation error and get effective super-layer parameters the simulation gets composed by the skywriting time. Build Cycle Initialization. To determine the risk of global and local overheating, the process parameters are completed with the optional set delay topt . It is implemented between two consecutive layers and represents together with the recoating time the Inter Layer Time (ILT). The boundary conditions are created for the solver, together with the heat transfer coefficient of the shielding gas of 20 W/(m2 · K). The PBF-LB/M process is described via a complex nonlinear equation system in the Samcef Solver Suite in Siemens NX Simcenter [33]. The tuning takes place iteratively via the coefficients α and C or rather their dependency on the assumed T of Eq. (4). The aim is to reach the global minima for the deviation between the prediction and the actual distortion of the printed calibration models, see Fig. 9.

2.4 KPI Based Cost Calculation Model The generic production cost model contains three different categories such as, design, manufacturing, and post-process. In this work, the focus is only on the manufacturing phase, where a subdivision into machine and material cost is made [34]. The machine hour rate is facility specific due to external conditions like local average energy and operation cost [35]. The Machine type EOS M300-4 with all consumables and auxiliaries is not investigated in detail. Instead, the default of NX 2212 of 92 e/h is applied. The basic rudimentary KPI for the productivity is the Build Up Rate (BUR) or melting rate, it describes how fast the material is printed [36]. It can be calculated out of the process parameters hatch distance hs in m, scan speed vs in m/s, layer thickness tz in m, and number of lasers nL which are working simultaneously, see Eq. (5). BUR = hs · vs · tz · nL

(5)

The layer-wise character of the buildup causes additional idle times for spreading the powder tp and an optional delay topt for controlling the cooling rate, resulting in higher ILTs [37]. The determination of the volume Vm of the CAD models is performed in NX 2212 and allows the forecast [34] of the print time for the build cycle tB , expressed in Eq. (6). The consideration of all the idle times for recoating extended by the time for realigning of the scan heads during the jumps of the exposures in tp is more precise in comparison to the pure BUR [38, 39], see Eq. (5). tB =

  V · tp + topt BUR

(6)

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

101

The second cost component is the material. In this work it is simplified for the use case of new powder and a recycling rate of 0 %. The simplified material cost CM is composed out of the mass density of the powder respectively the tap density ρtap , the volume of not molten powder Vp , the mass density ρsk for solid material, the volume of the CAD models Vm . and the price per kg of powder Pp , see Eq. (7) [36].   CM = ρtap · Vp + ρsk · Vm · Pp (7) Nevertheless, the CAM suite offers to adapt the cost model for the reuse of material. The overall production cost can be expresses by the product of the total print time tB with the machine hour rate of 92 e/h added to the material cost CM afterwards. NX uses the timestamps of the simulated process for every exposure of each vector, shown in the JSON snipped before. 2.5 Evaluation of Geometric Accuracy The geometric accuracy is defined by the designer to fulfill the requirements of the part. The requirements are established due to the engineering constrains to provide the function [18]. The concept of Product Manufacturing Information (PMI) is applied, for linking the dimensions and other requirements directly to the model. A technical drawing itself is not mandatory anymore. The specification of tolerance fields and references can be performed direct on the CAD part in the model environment. PMIs are also offering interoperability with various software packages for processing and evaluating, like in the case study, optically digitized, three-dimensional data of the printed CAD model. A data conversion is not required. The CAD environment supports the import of point clouds of a 3D scanner which is utilized for the digitizing of the manufactured parts. It also keeps track of the defined PMI, the manipulation of point cloud operation and applied filters for the surface reconstruction, illustrated for the gyroid cube of Fig. 2 in Fig. 6

Fig. 6. Conversion of a point cloud (a) of 3D scan with point distance of 60 μm to surface mesh (b) and result of Iterated Closest Point (IPC) algorithm with the reference model (c) for deviation analysis

102

K. Poka et al.

Afterwards, a global tolerance field as a result of the best fit as well as the PMIs for specific features can be evaluated [40]. The global best fit is performed via the mean square objective function minimized by iterative closest point algorithm (ICP) for the alignment between the CAD model {xi } and the point  cloud of the 3D scan {pi }. The → → → → registration vector − q is denoted as − q = − qR |− qT in Eq. (8), where Np is the total number of points in the cloud [41]. Np →−  → 1  2 → pi − − xi − R − qR → qT  f − q = Np

(8)

i=1

The quality of the point cloud can be evaluated within the CAD environment via metrics of point distance and the deviation of points for the position in three-dimensional space. The use of an external software suite would derive features of CAD model. This represents again a media disruption including a loss of meta data [42].

3 Case Study The streamlined pipeline of the model data, the design of the build cycle with the information depth on vector level and the execution of the production file on the EOS M300-4 and the SLM 280 HL are performed and evaluated. The build cycle executed on the EOS M300-4 is additionally simulated and based on the results optimized beforehand. Finally, the quality assurance premised on the point cloud of the 3D scan is carried out for the parts printed on the EOS M300-4 out of AlSi10Mg. 3.1 Design Specification and Parametric Modeling The test specimen is composed out of eight galleries, see Fig. 7, which are having a bounding boxes of 50 mm × 5 mm × 32 mm and are available at the web [43].

Fig. 7. Detailed view of the gallery 01-01 9.00 placed on the build platform with 3 mm volume support highlighted in blue and the PMIs within the production file of the build cycle

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

103

The cutouts are 26 mm in height and 1 mm in width, except the two outer ones which are reduced in width by 0.5 mm. By the amount of cutouts and their resulting thin structures heat accumulation and distortion is provoked [27]. The order of magnitude is evaluated via the control of the PMIs of 50 mm corresponding to the assigned test specification at the top and the root of each gallery, see Fig. 7. The galleries are modelled parametrically to reduce design time and assigning print marks automatically. The print marks are consisting out of the distance between the most outer cutout and the part boundary combined with build cycle ID 01×01 to avoid swapping of common parts from different build cycles. The automatic CAD modeling procedure is explained in Table 3 and creates a portfolio of part entities called part family. Table 3. Parametric modelling and creation of print marks via variables Create a fully parametrized part Specification of expression, attributes, and other properties as a variable in the template of the part family Create the part family spreadsheet Selection of variables for control and initialization of automatic modelling with the list of desired characteristics, implementation for assigning them Add family members as parts Execution of the part creation, storage in network drive, access of different entities during build cycle preparation, versioning via the creation date and print mark

3.2 Process Planning Via CAM The initial build orientation is implemented through an import of the pre-defined coordinates of the part locations in.csv format. The placement of the galleries is carried out in a linear pattern with a gap size of 26.5 mm for a better accessibility via the 3D scan for quality assurance. The whole arrangement is then rotated around 30° around the center of volume around the z axis of the MCS. After completion, 3 mm volume support is added, see Fig. 7 and the process parameters including the exposure strategy of Table 4 are assigned to the parts. Four particularities are deployed. The two first consecutive layers are scanned twice to secure a great binding to the build platform. The scan paths of each gallery are only performed by one laser of the EOS M300-4, no transfer between laser sources in the scanning of one gallery is enabled. The laser power PL gets modulated to reduce the Volume Energy Density (VED) for smaller vectors, see Eq. (9). At last, all vectors of all layers are having the exposure sequence against the shielding gas flow to minimize the deposit of waste products in the powder bed from the interaction of the laser and the powder during melting. VED =

PL hs · vs · tz

(9)

104

K. Poka et al. Table 4. Process parameters AlSi10Mg Parameter

Value Machine

Preheating build platform Shielding gas Layer thickness Recoating

165 °C Argon 5.0 250 m³/h 60 µm Two sided

Scanning pattern Strategy Contour Maximal length Hatch Rotation per layer

Stripes one 8 mm 150 µm 47° Exposure

Infill Laser power : Scan speed :

Contour 370 W Laser power : 1210 mm/s Scan speed :

320 W 500 mm/s

Integration of Manufacturing Equipment. The utilization of another PBF-LB/M machine of a different OEM only requires a varying .pax and XML file to adapt the different constrains of the build chamber, here of the SLM 280 HL. The conversion of the production file into an executable .slm format is performed by the integration of a SLM 280 HL specific Materialise® build processor in analogy to the EOS SDK. Both machines with different print drivers also known as build processors can be accessed via the CAM suite of NX. The only manual tasks are the exchange of process parameters  − →  for Haynes® 282® and the implementation of a translation vector t = 150 150 to equalize different MCS origins of the EOS 300–4 and SLM 280 HL, see Table 5. After the assignment of the process parameters, the file is autonomously sliced for the different layer thickness of 40 μm and includes the scan paths connected to all other commands for controlling the PBF-LB/M machine. Table 5. Applied PBF-LB/M systems in the case study

Construction year Build chamber dimension MCS origin

EOS M300-4

SLM 280 HL

2022

2017

300 mm x 300 mm x 400 mm 280 mm x 280 mm x 365 mm Left bottom corner

Middle

Number of lasers

four

one

Equipped material

AlSi10Mg

Haynes® 282®

Layer thickness utilized

60 µm EOSPrint SDK 2.12

40 µm Materialise SLM Build processor 3.2

Build processor

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

105

The proof of interoperability of the application for the production file across platforms is accomplished by printing the same build cycle with the identical manufacturing plan on the PBF-LB/M systems out of Table 5, see Fig. 8.

Fig. 8. Received part quality for the same geometry in overview and detail Left: EOS M300-4 AlSi10Mg, large deformation and sintered powder particles causing failure Right: SLM 280 HL Haynes® 282® , only deformation at the top region, able to build geometry

The further case study is only focusing on the EOS M300-4 and how the process design is manipulated beforehand with the goal to reduce the otherwise emerging thermomechanical distortion. Nevertheless, the simulation and advanced process design can also be performed with the machine specific file for SLM 280 HL within the identical CAM suite of NX 2212.

3.3 Advanced Process Control Based on Simulation Results The build cycle is thermo-mechanically simulated to detect regions with heat accumulation and therefore the risk of distortion because of the thermal undesired conditions. The galleries are idealized by excluding the print marks visible in Fig. 7, these small geometric features can be neglected [2]. The conversion of the CAD model into the input of the simulation is done via 3D meshing [22]. The element size is adjusted according to the part characteristics, thin-walled or solid [22], due to the ratio of surface to number of tetraeder elements. For the galleries around 95,000 tetraeders of a size of 1.4 mm are created to describe the volume and map the distortion by the displacement of the nodes. The material properties for AlSi10Mg from the Eq. (4) are deduced on beforehand printed reference parts with the identical process parameters from the Table 4 to get thermal shrinkage and stiffness for the super-layers. This calibration is visualized in logarithmic scale in Fig. 9.

106

K. Poka et al.

Fig. 9. Derived material curve out of calibration build cycle with galleries for AlSi10Mg

The derived material calibration is now adapted to the specific build cycle with its manufacturing plan. As an initial warning, overheating in the transition of the thin walls or cutouts to the top part is issued. To minimize the risk of a process abortion and to have more homogenous cooling rates due to varying exposure times for different cross sections, a variable delay topt is established to ensure a constant ILT of 30 s. Even though there is a non-optimal coefficient of determination for the description of the material behavior in Fig. 9, the solver can predict the fundamental distortion systematics for the process with an ILT of 30 s, presented in Fig. 10.

Fig. 10. Predicted dislocation of mesh nodes out of the thermo-mechanical simulation of the production file for the optimized build cycle with an ILT of 30 s on the EOS M300-4

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

107

3.4 Manufacturing Documentation with Integrated Economical Assessment The streamlined economic assessment for the build cycle based on the Eqs. (5) to (7) is summarized in the Table 6 and Table 7. The powder characteristics must be provided for an accurate cost assumption [36]. For the process on the EOS M300-4, AlSi10Mg powder with a particle size distribution of 25 μm up to 70 μm ascertained via laser diffraction is applied. The mass densities of the alloy for this specific powder collective are an apparent density ρtap of 1.31 g/cm3 according to ASTM B417 [44] and a skeletal density ρsk of 2.65 g/cm3 measured via inert gas displacement. With the manually entered densities, the volumes for Vp and Vm of Table 7 and the price per kg Pp , here 50 e/kg for OEM declassified powder [45], the material cost CM can be finally calculated [36]. The print time for the build cycle tB is captured for the deployed process parameters and its intrinsic BUR by accessing the timestamps from the JSON file for the separate exposure of each vector within one layer. By a simple addition of all last time steps of the layers the BUR is determined even more precisely then stated in Eq. (5). Now also jump times for repositioning the optics of the scan heads are considered. The time for spreading the powder tp is defined via the recoater speed of 250 mm/s with a total distance of 474 mm for one layer. Its setting is transferred via the actual machine configuration as well as the optional delay topt for controlling the ILT. This initialization enables the automatic cost calculation via a Visual Basic (VB) journal within the CAM suite for a machine hour rate of 92 e/h for the EOS M 300-4. The machine hour rate consists out of spare parts and auxiliaries, mainly the shielding gas for establishing an inert atmosphere as well as the extraction of arising waste products [34]. The consumable with the highest impact on the machine cost in Table 7 is the electricity for powering the laser sources and the cooler. Table 6. Automatic process documentation via a journal in the CAM suite NX 2212 PBF-LB/M report Build cycle:

2023_01_11_Material_Cal_Across_Platforms

Build density: 1,70°%

Support material: 4692 mm³

Max height: 35 mm

Consumed material: 3150000 mm³

Part material: 50054 mm³

Min height: 3 mm

Isometric view in MCS

Top view in MCS

108

K. Poka et al. Table 7. Automatic part documentation via a journal in the CAM suite

Part

Weight

Mesh quality

Position ( , , )

2.00

13.36 g

Fine

74.47/230.81/3

4.00

14.05 g

Fine

90.22/203.53/3

7.00

14.75 g

Fine

105.97/176.25/3

9.00

15.79 g

Fine

121.72/148.97/3

12.00

16.83 g

Fine

137.47/121.7/3

17.00

18.56 g

Fine

153.22/94.42/3

20.00

19.61 g

Fine

168.97/67.14/3

23.00

19.58 g

Fine

184.72/39.86/3

Material cost: 254.92 € NX Version: v2212.4000 Folder:

Machine cost: 582.36 € Print time: 6:21 h Date: 16.01.2023

Illustration

Total cost: 837.28 € Machine:

Author: Konstantin Poka

In addition, the KPI of machine utilization, here 1.70%, is gathered by the ratio of overall consumed material to molten material of the build cycle. A separate section for each individual part is also provided and allows a specific evaluation, presented in Table 7. The information about chordal and angular tolerances for meshing before the slicing are documented via the mesh quality. The quality fine is equivalent to the setting in Table 1. The position is tracked in accordance with DIN EN ISO ASTM 52921 [46] based on the derived center of volume of each part with their print marks within the MCS. Implementing an extension also allows capturing the eigenvector with the largest eigenvalue to acquire translation and rotation during the manufacturing planning of the build cycle [47]. Finally, the software version of the CAM suite, the file address with the corresponding author, the creation date and machine serial number are recorded to improve data consistency. 3.5 Compliance with Product Manufacturing Information The outer surface of the as-built, de-powdered and still to the build platform connected galleries get digitized by an Optical Coordinate Measurement Machine (OCMM). It is comprising the Hexagon Absolut Arm V2P 8320 7-Axis equipped with an AS1 scan head and has a measurement system uncertainty of 60 μm in accordance with

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

109

DIN EN ISO 10360–8 Annex D [48]. A blue 150 mm width laser line projected with a frequency of 300 Hz is scanning the optical accessible outer surface with a point distance of 30 μm. The point cloud is created by stitching the scan patches together and filtering all the points which are detected under an angle smaller than 20° between the optical axis and the measured surface. Afterwards, a downsampling, with a symmetric tolerance field of 100 μm where all the within detected points are assigned later to the same element, is applied. On the compressed point cloud, a surface reconstruction performed by triangulation of the overall 1,167,490 points and the determination of the outer surface by the assigned detection vectors of the scanned points is carried out. The detection vectors are registered in the MCS of the OCMM. Together with their orientation in combination with the pose of the 3D scanner, a normal vector for each surface element can be computed. The tessellated model is subsequently exported to the CAM suite, where a global best fit is performed by minimizing the Root Mean Square Error (RMSE) [41] for positioning relative to the part location in the build cycle. An RMSE of 0.64 mm is received. Already 100 % of the points are laying inside ±4 times the standard deviation of 0.61 mm with a mean value of −0.2 mm in relation to the ideal CAD model of the eight galleries. Based on the positioned digitized model, the test specification after DIN EN ISO 14405-01 [49] is evaluated. The distances of 50 mm are determined between the midpoints of the Gaussian fitted reference sites on the reference planes A and B for each gallery, shown in Fig. 7. The results of the quality assurance are presented in Table 8. Table 8. PMI evaluation of the galleries with an uncertainty of 60 μm Gallery 23.00 20.00 17.00 12.00 9.00 7.00 4.00 2.00

A2 B2 50.06 mm

Actual dimension

Deviation +0.06 mm

A1 B1 49.98 mm

−0.02 mm

A2 B2 50.02 mm

+0.02 mm

A1 B1 49.98 mm

−0.02 mm

A2 B2 50.03 mm

+0.03 mm

A1 B1 49.95 mm

−0.05 mm

A2 B2 50.02 mm

+0.02 mm

A1 B1 49.95 mm

−0.05 mm

A2 B2 50.08 mm

+0.08 mm

A1 B1 49.92 mm

−0.08 mm

A2 B2 50.14 mm

+0.14 mm

A1 B1 49.85 mm

−0.15 mm

A2 B2 50.13 mm

+0.13 mm

A1 B1 49.84 mm

−0.16 mm

A2 B2 50.12 mm

+0.12 mm

A1 B1 49.73 mm

−0.27 mm

Captured points within IT 14 100.00 % 99.97 % 99.96 % 99.48 % 97.06 % 93.00 % 84.52 % 78.08 %

110

K. Poka et al.

The distance between the midpoints of the reference sites A1 and B1 is the dimension at the top where the other pair represents the root, see Fig. 11. The metrics of the manufactured galleries and their deviation to the CAD model are summarized in Table 8, with a trend to larger deviations in geometric accuracy for a higher number of cutouts, as it is stated in the simulation, see Fig. 10. Additionally, an envelope requirement is defined according to DIN EN ISO 14405-1 [49] with the limits of ±0.3 mm, which are corresponding to the International Tolerance (IT) field 14 of DIN EN ISO 268-1 [50] for the largest basic dimension of 50 mm. An assessment for the overall reached part quality in terms of dimensional accuracy is thereby given [27]. Up to 21.92% of the detected and processed point cloud per gallery are outside of the tolerance class IT 14, see Table 8. The length of the smallest normal vector with the starting point on the CAD model and the intersection with the tessellated model are illustrated in Fig. 11 for evaluation with an applied grid width of 0.5 mm. The lower and upper boundaries are mostly in line with IT 14, see Table 8.

Fig. 11. PMI Evaluation based on 3D digitizing via the OCMM and the deviation between the CAD and the point cloud within the production file of the build cycle with the limits of IT 14

4 Conclusion and Outlook A hierarchical continuous part process relation is created by the concept of a part family and the assignment of the different entities, unique identified via print marks to the build cycles. The combination of direct and implicit modeling enables functional integration of complex TPMS regions to tackle special requirements regarding the part. The requirements stored in PMIs as well as the control of DfAM are embedded already during the modelling and are controlled iteratively during the process planning. Through the registration of the machine configuration, the process parameters and the sliced CAD model in the production file, a holistic digital image of the manufacturing process is

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

111

established. Therefore, the coordinates of every scan path of the scan pattern for each entity of a part with its assigned process parameters in the MCS and the already included correction matrix of the scan heads are provided. The computed data of the execution is accessed via the SKD as an interface. It includes vector-wise optimization like the power reduction for lowering the VED and their timestamps of exposure of an ideal process. In addition to the ILT, the Inter Vector Times (IVT) are determined. The process insights detailed down to the level of vectors facilitating the creation of a viable input for simulations. The conducted workflow with calibration on meso scale presented in Fig. 9 and the simulation on the macro scale is able to identify the distortion in the critical upper regions of the galleries of the case study in Fig. 10 and Table 8. The CAM suite supports the user to achieve “print first time right” with a higher probability by its warnings and suggested changes in the process control. Furthermore, functionalities like prediction of recoater collision or pre-deformation of the part are implemented, but they are beyond the scope of this work. Instead, the quality qualification is addressed as the last step of the case study, by the evaluation of the geometric accuracy. The processing, fitting and the evaluation of the 3D scanned galleries is carried out in the identical production file and allows the verification of the PMIs. The option to register all datasets of the case study in the production file of a final size less than 2 GB at the end makes an interchange between different software suites obsolescent. The presented streamlined approach to unite all the different data representations of the AM part along the PLM has a significantly lower risk of data corruption due to the conversion between different specialized proprietary software suites. For achieving deep process insights and control on a vector level, the adaption of a common concept for subtractive machining in a form of a post processor is inevitable. Only this solution in terms of AM called build processor allows the utilization of systems of different OEMs without the need to create machine specific build cycle files. The core is the production file which can be extended for simulation and quality assurance and is a viable exchange format for an Additive Manufacturing Service Platform (AMSP).

References 1. Pei, E., Ressin, M., Campbell, R., Eynard, B., Xiao, J.: Investigating the impact of additive manufacturing data exchange standards for re-distributed manufacturing. Progress Add. Manufac. 4, 331–344 (2019) 2. Dalpadulo, E., Pini, F., Leali, F.: Integrated CAD platform approach for Design for Additive Manufacturing of high performance automotive components. Int. J. Interact. Design Manufac. 14, 899–909 (2020) 3. Butt, J.: Exploring the interrelationship between additive manufacturing and industry 4.0. Designs 4, 13 (2020) 4. Mustafa, S.S., Lazoglu, I.: A design framework for build process planning in DMLS. Progress Add. Manufac. 5, 125–137 (2020) 5. Rodriguez, E., Alvares, A.: A STEP-NC implementation approach for additive manufacturing. Procedia Manufac. 38, 9–16 (2019) 6. Ingenieure, V.D.: VDI 3405 Blatt 5.1 Entwurf Januar 2020 Additive Fertigungsverfahren Rechtliche Aspekte der Prozesskette. vol. VDI 3405 Entwurf Blatt 5.1, (2020) 7. Zhang, B., Goel, A., Ghalsasi, O., Anand, S.: CAD-based design and pre-processing tools for additive manufacturing. J. Manuf. Syst. 52, 227–241 (2019)

112

K. Poka et al.

8. Habib, M.A., Khoda, B.: Hierarchical scanning data structure for additive manufacturing. Procedia Manufac. 10, 1043–1053 (2017) 9. Ding, J., Zou, Q., Qu, S., Bartolo, P., Song, X., Wang, C.C.: STL-free design and manufacturing paradigm for high-precision powder bed fusion. CIRP Ann. 70, 167–170 (2021) 10. Popov, D., Maltsev, E., Fryazinov, O., Pasko, A., Akhatov, I.: Efficient contouring of functionally represented objects for additive manufacturing. Comput. Aided Des. 129, 102917 (2020) 11. Maskery, I., Aboulkhair, N.T., Aremu, A., Tuck, C., Ashcroft, I.: Compressive failure modes and energy absorption in additively manufactured double gyroid lattices. Addit. Manuf. 16, 24–29 (2017) 12. Simsek, U., Gayir, C., Kavas, B.: Computational and experimental investigation of vibration characteristics of variable unit-cell gyroid structures. In: Sim-AM 2019: II International Conference on Simulation for Additive Manufacturing, pp. 369–380. CIMNE (Year) 13. Koizumi, Y., Okugawa, M.: Digital twin science of metal powder bed fusion additive manufacturing: a selective review of simulations for integrated computational materials engineering and science. ISIJ Int. 62, 2183–2196 (2022) 14. Ingenieure, V.D.: VDI 3405 Blatt 3.2 Entwurf Juli 2019 Additive Fertigungsverfahren Gestaltungsempfehlungen Prüfkörper und Prüfmerkmale für limitierende Geometrielemente. vol. VDI 3405 Entwurf Blatt 3.2 (2019) 15. Systems, S.A.E.G.-E.O.: EOS and Siemens intensify cooperation around industrial 3D printing. Siemens AG and EOS GmbH - Electro Optical Systems, Nuremberg (2018) 16. Standardization, I.O.f.: DIN EN ISO ASTM 52950 Mai 2021 Additive Fertigung Grundlagen Überblick über die Datenverarbeitung, vol. DIN EN ISO ASTM 52950 (2021) 17. Alama, J.: Euler’s Polyhedron Formula. Formal. Math. 16, 7–17 (2009) 18. Qin, Y., Qi, Q., Scott, P.J., Jiang, X.: Status, comparison, and future of the representations of additive manufacturing data. Comput. Aided Des. 111, 44–64 (2019) 19. Adams, D., Turner, C.: An implicit slicing method for additive manufacturing processes. Virtual Phys. Prototyp. 13, 2–7 (2018) 20. Jin, Y.-a., He, Y., Fu, J.-z.: Support generation for additive manufacturing based on sliced data. Int. J. Adv. Manufac. Technol. 80, 2041–2052 (2015) 21. l’Industrie, B.M.W.A.C.R.F.E.O.S.G.M.B.A.C.d.R.S.e.T.d., (BIBA), d.F.M.C.I.f.K.u.K.I.B.I.i.i.B.-u.A.: Development an Integration of Rapid Prototyping Techniques for Autmotive Industry. Community Research and Development Information Service (CORDIS) (1995) 22. Livesu, M., Cabiddu, D., Attene, M.: Slice2mesh: meshing sliced data for the simulation of AM processes. In: STAG, pp. 13–23 (Year) 23. Liu, C., Le Roux, L., Körner, C., Tabaste, O., Lacan, F., Bigot, S.: Digital twin-enabled collaborative data management for metal additive manufacturing systems. J. Manuf. Syst. 62, 857–874 (2022) 24. Denkena, B., Dittrich, M.-A., Henning, S., Lindecke, P.: Investigations on a standardized process chain and support structure related rework procedures of SLM manufactured components. Procedia Manufac. 18, 50–57 (2018) 25. Systems, S.A.E.G.-E.O.: Advancing Additive #11 – The value of EOSNorth America reselling Siemens softwarewith its machines. Siemens AG and EOS GmbH - Electro Optical Systems, Nuremberg (2021) 26. Ingenieure, V.D.: VDI 3405 Blatt 2.8 Dezember 2022 Additive Fertigungsverfahren Pulverbettbasiertes Schmelzen von Metall mittels Laserstrahl(PBF-LBM) Fehlerkatalog Fehlerbilder beim Laser-Strahlschmelzen. vol. VDI 3405 Entwurf Blatt 2.8 (2022) 27. Peter, N., Pitts, Z., Thompson, S., Saharan, A.: Benchmarking build simulation software for laser powder bed fusion of metals. Addit. Manuf. 36, 101531 (2020)

Integration of the Whole Digital Chain in a Unique File for PBF-LB/M

113

28. Bierwisch, C., Butz, A., Dietemann, B., Wessel, A., Najuch, T., Mohseni-Mofidi, S.: PBFLB/M multiphysics process simulation from powder to mechanical properties. Procedia CIRP 111, 37–40 (2022) 29. Zongo, F., Simoneau, C., Timercan, A., Tahan, A., Brailovski, V.: Geometric deviations of laser powder bed–fused AlSi10Mg components: numerical predictions versus experimental measurements. Int. J. Adv. Manufac. Technol. 107, 1411–1436 (2020) 30. Afazov, S., et al.: Metal powder bed fusion process chains: an overview of modelling techniques. Progress Add. Manufac. 1–26 (2022) 31. Gunasegaram, D.R., Murphy, A.B., Matthews, M., DebRoy, T.: The case for digital twins in metal additive manufacturing. J. Phys. Mater. 4, 040401 (2021) 32. AG, S.: Siemens introduces Additive Manufacturing Process Simulation solution to improve 3D printing accuracy. Siemens AG Communications, Frankfurt (2018) 33. AG, S.: Siemens introduces AM Path Optimizer technology integrated in NX for additive manufacturing. Siemens AG Digital Industries, Frankfurt (2019) 34. Bartsch, K., Emmelmann, C.: Enabling cost-based support structure optimization in laser powder bed fusion of metals. Jom 74, 1126–1135 (2022) 35. Uz Zaman, U.K., Rivette, M., Siadat, A., Mousavi, S.M.: Integrated product-process design: material and manufacturing process selection for additive manufacturing using multi-criteria decision making. Robot. Comput. Integrat. Manufac. 51, 169–180 (2018) 36. Barclift, M., Armstrong, A., Simpson, T.W., Joshi, S.B.: CAD-integrated cost estimation and build orientation optimization to support design for metal additive manufacturing. In: International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, pp. V02AT03A035. American Society of Mechanical Engineers (Year) 37. Mohr, G., Altenburg, S.J., Hilgenberg, K.: Effects of inter layer time and build height on resulting properties of 316L stainless steel processed by laser powder bed fusion. Addit. Manuf. 32, 101080 (2020) 38. Baldinger, M., Levy, G., Schönsleben, P., Wandfluh, M.: Additive manufacturing cost estimation for buy scenarios. Rapid Prototyp. J. (2016) 39. Chan, S.L., Lu, Y., Wang, Y.: Data-driven cost estimation for additive manufacturing in cybermanufacturing. J. Manuf. Syst. 46, 115–126 (2018) 40. Ingenieure, V.D.: VDI 3405 Blatt 2 August 2013 Additive Fertigungsverfahren Strahlschmelzen metallischer Bauteile Qualifizierung Qualitätssicherung und Nachbearbeitung. vol. VDI 3405 Blatt 2 (2013) 41. Besl, P.J., McKay, N.D.: Method for registration of 3-D shapes. In: Sensor fusion IV: control paradigms and data structures, pp. 586–606. Spie (Year) 42. Mies, D., Marsden, W., Warde, S.: Overview of additive manufacturing informatics: “a digital thread.” Integrat. Mater. Manufac. Innov. 5, 114–142 (2016) 43. Industries, S.A.D.: Powder Bed Fusion material calibration. In: Industries, S.A.D. (ed.) (2022) 44. Materials, A.S.f.T.A.: ASTM B 417 2022 Standard Test Method for Apparent Density of Non Free Flowing Metal Powder Using the Carney Funnel. vol. ASTM B 417 (2022) 45. Systems, E.G.-E.O.: Quotation metal powder EOS aluminium AlSi10Mg including inspection certificate according to EN 10204. In: Systems, E.G.-E.O. (ed.) (2022) 46. Standardization, I.O.F.: DIN EN ISO ASTM 52921 Entwurf Oktober 2019 Additive Fertigung Grundlagen Standardpraxis der Positionierung Koordinaten und Ausrichtung des Bauteils. vol. DIN EN ISO ASTM 52921 Entwurf (2019) 47. Merz, B., Nilsson, R., Garske, C., Hilgenberg, K.: Camera-based high precision position detection for hybrid Additive Manufacturing with Laser Powder Bed Fusion. Int. J. Adv. Manufac. Technol. 125, 2409–2424 (2023)

114

K. Poka et al.

48. e.V., D.I.f.N.: DIN EN ISO 10360-8 März 2014 Geometrische Produktspezifikation und prüfung (GPS) Annahme und Bestätigungsprüfung für Koordinatenmesssysteme (KMS) Teil 8 KMG mit optischen Abstandssensoren. vol. DIN EN ISO 10360-8 (2014) 49. e.V., D.I.f.N.: DIN EN ISO 14405-1 Juni 2017 Geometrische Produktspezifikation (GPS) Dimensionelle Tolerierung Teil 1: Lineare Größenmaße. vol. DIN EN ISO 14405-1 (2017) 50. e.V., D.I.f.N.: DIN EN ISO 286-1 September 2019 Geometrische Produktspezifikation GPS ISO Toleranzsysteme für Längenmaße Teil 1 Grundlagen für Toleranzen Abmaße und Passungen. vol. DIN EN ISO 286-1, (2019)

Approach to an Automated Method for Load-Optimized Design of Multimaterial Joints for Additive Manufacturing Christoph Leupold(B) and Maren Petersen Institute Technology and Education, University of Bremen, Am Fallturm 1, 28359 Bremen, Germany [email protected]

Abstract. The emerging multimaterial technology is extending the potential applications of additive manufacturing in many areas. However, these possibilities also bring new challenges such as creating a sufficiently strong bond between materials and ensuring their continued recyclability. This paper presents a method for designing structures for joining different materials based on the use of experimentally determined values and artificial intelligence. The objective of the method is to find, for a given design space, an arrangement of the various materials that provides a high strength of the material composite in one loading direction and, at the same time, is designed to result in a significantly reduced strength for a further loading direction. As a result, the material composite should break at this predetermined breaking point when loaded in the second direction, so that the materials involved can be recycled. To be able to perform the calculation as quickly as possible, the geometry is reduced to arrangements of small cubes (voxels). Four steps are provided for the use of this method. In the first step, the maximum resolution that can be achieved with the respective additive process is determined. In a further step, different arrangements of voxels are examined for resulting strength using tensile tests. In the next step, the results of these tests serve as input values for an AI application that finds an arrangement of the voxels that provides the desired strengths in the various load directions, taking into account a given design space. A genetic algorithm is used to geometrically optimize the joint. Finally, these designs are used to automatically build a CAD model that enables additive manufacturing of the components. Initial investigations into the voxel sizes and manufacturability of the multimaterial joints using the material extrusion (MEX) process are presented, but evaluation of the overall method is still pending. Keywords: Multimaterial · voxel-based · MEX · recycling-friendly

1 Motivation Due to the further development of manufacturing systems, the application possibilities for additive manufactured products are continuously increasing. For many processes, the use of different materials in one manufacturing process is possible. This development © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 115–129, 2024. https://doi.org/10.1007/978-3-031-42983-5_8

116

C. Leupold and M. Petersen

makes it feasible to include more functionalities in components and to access new areas of application. For example, electrically conductive tracks can be directly integrated as failure sensors, or joints can be provided on otherwise rigid components that cannot be made from a single material. A challenge with many material combinations, however, is the design of the material transition, since this often represents a weak point with regard to the mechanical load-bearing capacity of the components [1]. Since the conventional design of these joints is time-consuming and calculation-intensive due to the complexity of multi-axial load cases, the advantages of multimaterial additive manufacturing (MMAM) can only be used effectively if simple tools exist for designing customized joints. With the use of different materials in a process, an advantage of additive manufacturing technology also disappears in use: Material extrusion components (MEX) made from a single material, for example, can be recycled back into filament and thus used again as feedstock for the process. However, this form of material recycling can currently only be carried out with unmixed waste [2]. As soon as MMAM uses a combination of different materials, the materials of the components have to be separated again for recycling. To be able to implement this process easily, the design of the MMAM components should already take into account where a subsequent recycling process could start. The method presented here is intended to present a solution approach for these two aspects of MMAM: Joints with increased strength can be designed, which at the same time have predetermined breaking points for recycling in selected load directions.

2 Objective The presented method should automatically create ready-to-use designs for individual load cases and design spaces, which ensures a strong connection of different materials. Also, according to the intended load of the components, it should be possible to provide predetermined breaking points in less heavily stressed load directions, which simplifies material recycling. Figure 1 illustrates these goals: While the loading with a force F1 leads to rupture in the case of a butt joint (1), the tensile strength is to be increased by an optimised arrangement of the materials at the joint (2). To enable recycling of the materials, the arrangement of the materials should be designed in such a way that a predetermined breaking point can be used when the joint is loaded with a differently directed, smaller force F2 (3). The characterisation of the individual material combination required for the design, depending on the process parameters, needs to be carried out with low-complexity tests and should be so low in computational intensity that it can be carried out on CAD-capable computers in order to ensure broad applicability. In addition, the results of the application should be able to be inserted directly into CAD data sets. The method is initially to be designed for MEX and, through extensions, can also be used for other additive manufacturing processes.

Approach to an Automated Method for Load-Optimized Design

117

Fig. 1. Illustration of the objectives of the application: (1): fracture of the non-optimised butt joint under load; (2): increased tensile strength through optimised design of the joint; (3): fracture of the joint under load in the other direction to allow recycling of the sections

3 State of the Art 3.1 Manufacturing of Multimaterial Components Using MEX A number of additive processes, such as MEX, material jetting, direct energy deposition or stereolithography, already offer the ability to produce components from more than one material [3]. In the case of MEX components, however, ensuring the creation of a strong joint between the materials involved represents a challenge for many material combinations and is therefore the subject of research in various studies, as the following overview shows: Freund et al. identify bonding mechanisms and show which properties and process parameters promote the creation of strong joints. They identify the compatibility of the materials described by chemical adhesion as well as the design of the material joint with interlocking elements as particularly important [4]. Dairaba-yeva et al., Ermolai et al., Ribeiro et al. and Lopes et al. investigate different forms of mechanical interlocking by using simple geometric arrangements for different material combinations to optimise the tensile strength of multimaterial components and show that different tensile strengths occur depending on the choice of geometry [5–8]. In similar investigations, Marino et al. show that interlocking elements also make a major contribution to maximising the joint strength under compressive loads [9]. The influence of process parameters on the strength of multimaterial components under bending load and the mechanism influencing the strength are investigated by Rabbi et al. whereby an increased printing temperature in particular leads to higher joint strengths in these tests [10]. How parameter variations in multi-material components affect the tensile strength of the components is shown by Tamburrino et al. for various material combinations. In these experiments, mainly the infill values are investigated, which, if chosen appropriately, lead to a mechanical interlocking pattern [11]. Similar investigations are carried out by Xue et al. on the effects of different nozzle temperatures and layer thicknesses. While an increased nozzle temperature facilitates the wetting of the materials and thus increases the joint strength, this result is also achieved for vertically running joints by smaller layer thicknesses. The reason for this phenomenon is the occurrence of mirco

118

C. Leupold and M. Petersen

interlocking elements, which are created using multiple printheads as a small overlap of the printheads’ paths causes the layers to interlock. As a result, higher strengths are achieved at lower layer thicknesses due to a higher number of these elements [12]. Kuipers et al. develop and evaluate an “interlaced topologically interlocking lattice”, which allows mechanical interlocking in all spatial directions to achieve increased joint strengths under tensile load [13]. Cunha et al. show that the chemical affinity of different materials has a significant influence on the tensile strength of butt joints and that this affinity can be estimated using simple tensile tests on welded filaments without producing 3D printed samples [14]. These studies identify different bonding mechanisms and show how they can be influenced by process parameters and geometry design. In addition to chemical adhesion, which depends on the choice of material and the process parameters, individual parts of a component can be joined together by inserting interlocking elements. The next chapter takes a closer look at these bonding mechanisms. 3.2 Bonding Mechanisms Adhesion Mechanisms: Various chemical properties of the materials used, such as the chemisorption, electrostatic attraction, diffusion between materials, adsorption of components of the joining partner and the polarisation of the material, lead to attraction between the materials, which influences the strength of the joint. Figure 2 shows these mechanisms schematically. While some of these mechanisms depend only on the composition of the materials, others can be influenced by the choice of process parameters such as the printing speed or the printing temperature [4].

Fig. 2. Various adhesion mechanisms; based on [4]

Macro Interlocking Elements: For a component, the shapes of the pieces, which are made of different materials, can be designed in such a way that separation of the pieces is prevented by form locking, since, for example, material would have to be pulled through narrow spaces. In these arrangements, a rupture within at least one of the materials is therefore necessary to release the connection. Figure 3 shows the structure of two widely used geometric arrangements for this mechanical interlocking. Micro Interlocking Elements: In addition to the intended design of interlocking elements, overlaps during the extrusion of the material of the different extruders at vertically aligned joints lead to the creation of micro interlocking elements [12]. Figure 4 shows

Approach to an Automated Method for Load-Optimized Design

119

Fig. 3. Different shapes for mechanical interlocking; based on [13]

schematically how the different materials can overlap at the joint. In the case of horizontally oriented joints, this effect only occurs if cavities in the lower layer are filled with material when the next layer is fabricated. These cavities can be created intentionally, for example by under-extrusion, or they can be the result of defects.

Fig. 4. Mechanism of mirco-interlocking at a joint; based on [12]

3.3 Modelling The calculation of multi-material models includes the respective geometries of the materials and the material behaviour and is limited in resolution by the available computing power [15] Targeted simplifications can expedite the calculation of the model: The description of component volumes via voxels (volume elements) uses geometrically simple bodies to define smallest volume elements [16]. Due to the simple geometry of the voxels (e.g. tetrahedron or cuboid) and the identical structure of each of them, calculations on material behaviour are significantly simplified [15]. On this basis, several studies have already been carried out for the automatic optimisation of component designs: Garcia et al. successfully developed a method to optimise the bending stiffness of multimaterial components using a voxel-based model [17]. Schumacher et al. showed that a voxel-based approach allows the optimisation of the stiffness of components by using different designs of filler structures [18]. Morovic et al. developed a method to improve the colouring and strength of components for the Multi Jet Fusion process using voxel-based optimisation [19]. Gu et al. developed a method to optimise an arrangement of voxels for higher strength using machine learning.

120

C. Leupold and M. Petersen

These studies show that voxel-based optimisation has already been successfully applied. A model that is able to take into account weak bonds between materials and uses the listed mechanisms to optimise the strengths of additively manufactured multimaterial components has not yet been presented.

4 Presentation of the Proposed Method The method presented here aims to distribute the strength-increasing mechanisms of the arrangements of the materials over the joint in a load-adapted manner and thus to design a layout for the joint as a combination of elements of the different materials. Figure 5 shows the process divided into four steps: In step 1, the material combination and the process parameters are specified, and the usable voxel size is determined via tests. This is followed in step 2 by tests to determine the influence of the different bonding mechanisms for the materials and process parameters used. In step 3, the boundary conditions of the specific joint are specified to optimise its design in step 4 on the basis of a model. The individual steps are described in detail in the following chapter.

Fig. 5. Procedure of the method for the automatic design of joints

Approach to an Automated Method for Load-Optimized Design

121

4.1 Step 1: Selection of Process Parameters and Determination of the Voxel Size The listed mechanisms of strength increase depend on the selected material combination and process parameters as well as the voxel size. Therefore, a characterisation of the mechanisms must be preceded by a selection of the material combination and the process parameters. While theoretically the size of voxels can be chosen as small as desired to better represent complex figures with a resulting higher resolution [16], there is a minimum size for printable voxels. In addition to the respective materials and the process parameters, this is also specified by the system used, e.g., by the diameter of the nozzle, and must be determined individually. For this purpose, tensile tests are used to determine the smallest possible interfaces between the materials that do not show a significant loss of strength compared to larger interfaces. The dimensions of the voxels are defined based on the size of these interfaces in the various directions in space. 4.2 Step 2: Characterisation of the Material Combination This step involves the collection of measured values in order to describe the interaction of the various materials used for the model. To keep the model as simple as possible, the material interaction is initially limited to the strengths of the materials and the material combination. The step is divided into three parts, as the different mechanisms of the joint strength are investigated based on the findings of previous studies: Firstly, the adhesion of between the materials is determined by tensile tests of individual elements placed next to each other. However, each set-up for determining the adhesion also includes micro interlocking elements, as these are associated with every possible connection of the materials due to the printing process, as interactions cannot be avoided. In a second step, the influence of micro interlocking on the bond strength is investigated separately by arranging several elements next to each other, when individual elements are enclosed from several sides, which corresponds to a fixed clamping of these. Finally, the effect of macro interlocking of complete elements is investigated. The collected data will be used in the next steps to simulate and optimise the arrangement of materials for specific applications using the model. 4.3 Step 3: Specification of the Boundary Conditions Subsequently, the requirements for individual intended use of the joint are transferred to the model as boundary conditions. Since the required arrangement of materials should optimise both the strength of the joint and the recyclability of the component, both loaded and unloaded directions must be specified. For directions in which high loads are foreseen, the arrangement should accordingly contain many strength-increasing elements, while in non-loaded directions as few as possible of these should be provided to create a predetermined breaking point with which the materials can be separated for material recycling. These requirements can represent a conflict of objectives, which is resolved by prioritising individual requirements. This is done by means of coefficients that describe the extent to which the maximum or minimum strength may deviate from the achievable values as a compromise for the realisation of the other strengths.

122

C. Leupold and M. Petersen

4.4 Step 4: Optimising the Arrangement of the Materials For the model, the available design space is simulated as an arrangement of elements in a matrix. The use of matrices offers the advantage that, on the one hand, mathematical calculations can be carried out directly and, on the other hand, any number of different materials can be integrated. Each element of the matrix corresponds to a voxel. In order to reproduce the three-dimensional construction space, a three-dimensional matrix is initialised, which is extended by a further dimension for storing material identification as well as the strengths to the neighboured elements. Starting from this matrix, an algorithm examines the arrangements for weak points in the form of a collection of elements connected orthogonally to the direction of loading, which in total are least affected by strength-increasing mechanisms. The problem is thus similar to that of a pathfinding task. In these, the absolute minimum of the path length can always be determined with the help of Dijkstra’s algorithm [20]. For this reason, Dijkstra’s algorithm is used as a basis for this method and modified in such a way that instead of the shortest distance, a potential fracture plane of lowest strength is determined. This value has to be maximised for a maximum strength in the corresponding load direction or minimised to create a predetermined breaking point. Therefore, this value for the respective arrangement is in turn used as an aim function for a genetic algorithm that mimics evolutionary processes in order to optimise the arrangement [21]. Based on randomly generated populations of arrangements and mutations of these, an arrangement that meets the requirements is searched for. The results are then processed in such a way that integration into existing CAD models is possible, for which various interfaces have to be defined depending on the CAD software used. The four steps of the method show a procedure with which it is possible to carry out the design of joints automatically. During this process, the strength of the connections is optimised and the recycling of the materials used is supported by introducing predetermined breaking points. Initially, only tensile and shear strengths are used to describe the material behaviour, which leads to a rigid model in which the effects of strain are not taken into account. The extent to which these assumptions represent a good approximation to the real behaviour of the materials must be assessed during the evaluation of the method.

5 Investigations to Date The work already carried out for the evaluation of the presented method comprises step 1 of the method (4.1). Furthermore, the printability of a joint with the found voxel size was tested. 5.1 Determination of the Voxel Size to Be Used For the use of a model, all interactions of the materials with each other in the different spatial directions must be known. Experiments are to be carried out to determine how small the side surfaces of the voxels can be minimally without significant losses in strength compared to larger surfaces. For this purpose, the different arrangements are

Approach to an Automated Method for Load-Optimized Design

123

investigated. Since the anisotropy of components manufactured using the MEX process is limited to the differences in properties from the Z-direction to the X- and Y-direction, not all conceivable combinations of the arrangement of the interface and the loading direction must be used for the characterisation. The differing combinations are outlined in Fig. 6.

Fig. 6. Combinations of intersection arrangement and loading direction

Two sample designs are used to determine the strength values for these arrangements. These are shown in Fig. 7. The specimen shape shown in (1) is used to perform tensile tests when the loading direction is oriented orthogonally to the arrangement of the joints. The specimen geometry shown in (2) is used to perform shear tensile tests in the other cases. To ensure that the edges of the tested voxels are uniform, the smallest voxel size tested is selected in such a way that a surrounding edge can be created with the nozzle. With a nozzle diameter of 0.4 mm, this results in a minimum area with an edge length of 0.8 mm, i.e., an area of 0.64 mm2 . The edges of the next largest surfaces are respectively extended by the nozzle diameter and are listed in Table 1. The largest area was used to establish comparability with larger areas. To facilitate handling of the specimens and to be able to use tensile forces in the same order of magnitude, several of these surfaces were arranged per specimen. Table 1. Dimensions of the samples Nr.

Edge length of the interface area

Size of the interface area

Number of elements per sample

Resulting interface area

1

0,8 mm

0,64 mm2

12

7,68 mm2

2

1,2 mm

1,44 mm2

5

7,2 mm2

3

1,6 mm

2,56 mm2

3

7,68 mm2

4

2,8 mm

7,84 mm2

1

7,84 mm2

For better comparability, these surface sizes are also used in the building directions in which the layer height instead of the line width represents a geometric specification. A step with twice the height of the layer thickness (in this case 0.2 mm) aligned orthogonally to the contact area ensures that these areas are not enlarged or reduced during the slicing process via the staircase effect.

124

C. Leupold and M. Petersen

Fig. 7. Sample geometries: (1) for tensile tests with the interface orthogonal to the direction of loading; (2) for shear tensile tests with the loading parallel to the interface; Exemplary representation of the specimens with 0,64 mm2 contact surfaces

5.2 Tensile Tests Polylactide (PLA) and acrylonitrile butadiene styrene (ABS) are used as example materials. These are among the most commonly used materials for MEX [22] and do not have good chemical adhesion to each other [23]. A full factorial test with the parameters material combination, spatial arrangement and size of the contact area and the values shown in Table 2 and three samples of each combination is carried out. Table 2. Samples used for tensile and shear tests Material combinations

Spatial arrangements (Fig. 6)

Size of contact surfaces (Table 1)

• PLA-PLA • ABS- ABS • ABS-PLA

• • • • •

• • • •

Interface: XZ; load: X Interface: XZ; load: Y Interface: XZ; load: Z Interface: XY; load: X Interface: XY; load: Z

0,64 mm2 1,44 mm2 2,56 mm2 7,84 mm2

In addition, for the combination of ABS and PLA, specimens with ABS as the bottom material and PLA as the top material and vice versa are created for the XY direction interface arrangements, as the wettability of the material with the other material differs. The programme Ideamaker from Raise3D is used to slice the samples. All samples, both

Approach to an Automated Method for Load-Optimized Design

125

single and multimaterial, are produced with the same toolpath. Table 3 shows the process parameters used. Table 3. Manufacturing system and process parameter (selection) Manufacturing system

Raise3D Pro3

Material 1

PolyLite PLA [24]

Material 2

PolyLite ABS [25]

Nozzle size

0,4 mm

Layer thickness

0,1 mm

Temperature nozzle 1 (PLA)

225 °C

Temperature nozzle 2 (ABS)

250 °C

Temperature build sheet

70 °C

Print velocity (contact surfaces)

20 mm/s

Print velocity (standard)

30 mm/s

Infill

100%

Pattern

1 Circular; Inside: 0°;90° Pattern

The samples with a interface in XY orientation and PLA as the bottom material and ABS as the top material could not be produced with the selected set of parameters, as the adhesion of the PLA to the ABS substrate was not sufficient to form a new, solid layer. The successfully produced samples were tested after manufacturing using a tensile testing machine from the manufacturer Hegewald & Peschke with a maximum tensile force of 100 kN and a 500 N load cell, the results of these tests are shown in Fig. 8. These results show that the tensile strength of the PLA-PLA samples clearly exceeds the tensile strength of the other samples. It also shows that especially for the ABS-ABS samples and the ABS-PLA samples there are sometimes large deviations between samples of the same configuration. The ABS-ABS samples have a lower strength than the PLA-PLA samples, but a comparable or higher strength than the ABS-PLA samples. For the ABSPLA specimens, it is noticeable that for the orientation of the joint in the XY direction, when loaded in the X direction, the strength corresponds to 0 for some specimens, as the specimens already broke during clamping before the tensile test began. Looking at the different material combinations, the specimens with the largest contact surface have the lowest strength values for almost all orientations and in the mean value, while the mean value of the other specimens is very similar (Fig. 8; top right). The results show that the strength of the materials and the material combination in the investigated range does not fall with decreasing surface size, so that from this aspect the minimum size of the voxels used is not limited but is based on the production restrictions of the system used. The results, which can be compared with the specifications of the material manufacturer due to the same orientation used (parting plane XY, load direction: Z), are within the range of the scatter of the values, which confirms the measurement results.

126

C. Leupold and M. Petersen

Fig. 8. Results of the tensile tests of the samples made of PLA, ABS and the combination of materials

Since the use of small voxels has advantages when approaching non-rectangular shapes, the edge length of 0.8 mm is used for the size of the voxels in the following. 5.3 Validation of the Printability of the Joints To verify that application-generated joint arrangements are printable, a random joint arrangement of 10x10x5 voxels with an edge length of 0.8 mm was manufactured five times using the process parameters in Table 1 and tested for tensile strength. Although ABS is used as the upper material and PLA as the lower material, which was not printable in the previous study with an obtuse arrangement of the materials, the production of the samples is possible without any problems with the voxel size. Figure 9 shows the tensile strengths (1), as well as the CAD model (2) and one of the samples after the tensile test (3). The tested samples show clear cracks (3). The determined values for tensile strength are within a deviation of + -7% around the average value of all five samples. However, the tensile strength values are reduced by a factor of about 10 compared to the average tensile strength of the specimens with the same contact area size. After the termination criterion of the tensile test (90% force drop), there is still bonding between the individual parts of the specimen.

Approach to an Automated Method for Load-Optimized Design

127

Fig. 9. (1) Tensile strengths of the specimens; (2): CAD model of the tensile sample; (3): a sample after the tensile test

The fact that some joints between neighbouring voxels broke during the tensile test while others remained intact suggests that the different material combinations have different elongations at break. These different strains would explain lower tensile strengths of the sample, as some joints were not loaded to their maximum tensile strength while others already broke. This thesis could be tested by using extensometers in further tensile tests. The lower strengths compared to the previous investigations could also be due to the fact that the expansion of the contact surfaces behaves differently without neighbouring elements than with them. Thus, the heated, viscous material could spread over a larger area during the manufacturing process if no limiting elements exist. An analysis of the fracture surfaces of the fracture surfaces could be helpful in analysing the phenomenon.

6 Conclusion and Outlook The use of different materials during the manufacturing process of a component makes it possible to integrate new functionalities and thus to open up further applications for additive manufactured products. However, for many material combinations, e.g., with MEX, low bonding strengths mean that the arrangement of the materials must be specifically designed. At the same time, the use of multiple materials makes it more difficult to recycle the components. With the presented method, both aspects are to be addressed by an individual, automatic design of the joining zone. The arrangement of the materials in the joint is optimised voxel-based by means of a genetic algorithm in such a way that various strength-increasing mechanisms are favoured or prevented depending on the requirements. The prevention of these mechanisms results in the creation of predetermined breaking points, which enable the material recycling of the individual materials. The investigations to date show that in the MEX process the use of small voxels to describe material arrangements is compatible with the restrictions of the process. However, some aspects of the application still need to be tested or optimised. For example, it can already be deduced from the results so far that the exclusive use of the tensile

128

C. Leupold and M. Petersen

strength to describe the materials reduces the complexity of the necessary tests and the calculation, but at the same time also leads to deficits in the prediction of strengths of the joining zone. For this reason, the model should be extended to include strains of the individual materials to increase the strengths of the optimised joining zones. An analysis of the fracture surfaces should be carried out to investigate the phenomenon of the strongly varying strengths in the tests. After a successful evaluation of the complete method, it is possible to include different materials or constructions such as a support structure in the method and thus expand the design possibilities. Eventually, the use of the method could be applied to other additive processes.

References 1. Yao, X., Moon, S.K., Bi, G., Wei, J.: A multi-material part design framework in additive manufacturing. Int. J. Adv. Manuf. Technol. 99, 2111–2119 (2018). https://doi.org/10.1007/ s00170-018-2025-7 2. Cruz Sanchez, F.A., Boudaoud, H., Camargo, M., Pearce, J.M.: Plastic recycling in additive manufacturing: a systematic literature review and opportunities for the circular economy. J. Clean. Prod. 264, 12160 (2020). https://doi.org/10.1016/j.jclepro.2020.121602 3. Hasanov, S., et al.: Review on additive manufacturing of multi-material parts: progress and challenges. JMMP 6, 4 (2022). https://doi.org/10.3390/jmmp6010004 4. Freund, R., Watschke, H., Heubach, J., Vietor, T.: Determination of influencing factors on interface strength of additively manufactured multi-material parts by material extrusion. Appl. Sci. 9, 1782 (2019). https://doi.org/10.3390/app9091782 5. Dairabayeva, D., Perveen, A., Talamona, D.: Investigation on the mechanical performance of mono-material vs multi-material interface geometries using fused filament fabrication. RPJ 29, 40–52 (2023). https://doi.org/10.1108/RPJ-07-2022-02 6. Ermolai, V., et al.: Mechanical behaviour of macroscopic interfaces for 3D printed multimaterial samples. MATEC Web Conf. 368 (2022). https://doi.org/10.1051/matecconf/202 236801004 7. Ribeiro, M., Sousa Carneiro, O., Da Ferreira Silva, A.: Interface geometries in 3D multimaterial prints by fused filament fabrication. RPJ 25, 38–46 (2019). https://doi.org/10.1108/ RPJ-05-2017-0107 8. Lopes, L.R., Silva, A.F., Carneiro, O.S.: Multi-material 3D printing: the relevance of materials affinity on the boundary interface performance. Addit. Manuf. 23, 45–52 (2018). https://doi. org/10.1016/j.addma.2018.06.027 9. Peralta Marino, G., de La Pierre, S., Salvo, M., Díaz Lantada, A., Ferraris, M.: Modelling, additive layer manufacturing and testing of interlocking structures for joined components. Sci. Rep. 12, 2526 (2022). https://doi.org/10.1038/s41598-022-06521-z 10. Rabbi, M.F., Chalivendra, V.: Interfacial fracture characterization of multi-material additively manufactured polymer composites. Compos. Part C: Open Access 5, 100145 (2021). https:// doi.org/10.1016/j.jcomc.2021.100145 11. Tamburrino, F., Graziosi, S., Bordegoni, M.: The influence of slicing parameters on the multimaterial adhesion mechanisms of FDM printed parts: an exploratory study. Virtual Phys. Prototyp. 14, 316–332 (2019). https://doi.org/10.1080/17452759.2019.1607758 12. Xue, F., Boudaoud, H., Robin, G., Cruz Sanchez, F.A., Daya, E.M.: Influence of layer thickness and nozzle temperature on the interlocking adhesion strength of additive manufactured multimaterial interface (2022)

Approach to an Automated Method for Load-Optimized Design

129

13. Kuipers, T., Su, R., Wu, J., Wang, C.C.: ITIL: interlaced topologically interlocking lattice for continuous dual-material extrusion. Addit. Manuf. 50, 102495 (2022). https://doi.org/10. 1016/j.addma.2021.102495 14. Cunha, P., Teixeira, R., Carneiro, O.S., Silva, A.F.: Multi-material fused filament fabrication: an expedited methodology to assess the affinity between different materials. Prog. Addit. Manuf. 8(2), 195–204 (2023). https://doi.org/10.1007/s40964-022-00322-6 15. Aremu, A.O., et al.: A voxel-based method of constructing and skinning conformal and functionally graded lattice structures suitable for additive manufacturing. Addit. Manuf. 13, 1–13 (2017). https://doi.org/10.1016/j.addma.2016.10.006 16. Bacciaglia, A., Ceruti, A., Liverani, A.: A systematic review of voxelization method in additive manufacturing. Mech. Ind. 20(6), 630 (2019). https://doi.org/10.1051/meca/2019058 17. Garcia, D., Jones, M.E., Zhu, Y., Hang, Z.Y.: Mesoscale design of heterogeneous material systems in multi-material additive manufacturing. J. Mater. Res. 33(1), 58–67 (2018). https:// doi.org/10.1557/jmr.2017.328 18. Schumacher, C., Bickel, B., Rys, J., Marschner, S., Daraio, C., Gross, M.: Microstructures to control elasticity in 3D printing. ACM Trans. Graph. (Tog) 34(4), 1–13 (2015). https://doi. org/10.1145/2766926 19. Moroviˇc, P., Moroviˇc, J., Tastl, I., Gottwals, M., Dispoto, G.: Co-optimization of color and mechanical properties by volumetric voxel control. Struct. Multidisc. Optim. 60, 895–908 (2019). https://doi.org/10.1007/s00158-019-02240-8 20. Dijkstra, E.W.: A note on two problems in connexion with graphs. Numer. Math. (1959). https://doi.org/10.1007/BF01386390 21. Gerdes, I., Klawonn, F., Kruse, R.: Genetische Algorithmen und Optimierung. In: Gerdes, I., Klawonn, F., Kruse, R. (eds.) Evolutionäre Algorithmen, pp. 33–46. Vieweg+Teubner Verlag, Wiesbaden (2004) 22. Khaki, S., Rio, M., Marin, P.: Characterization of emissions in fab labs: an additive manufacturing environment issue. Sustainability 14(5), 2900 (2022). https://doi.org/10.3390/su1405 2900 23. Kumar, S., Singh, R., Singh, T.P., Batish, A., Kumar, A.: Three-dimensional printing of dual thermoplastic materials with different layer combinations: Tensile, flexural, and fractured surface investigations. J. Thermoplast. Compos. Mater. 35(6), 826–845 (2022). https://doi. org/10.1177/0892705720925123 24. polymaker: PolyLite PLA. Technical Data Sheet. V5.0. https://cdn.shopify.com/s/files/ 1/0548/7299/7945/files/PolyLite_PLA_TDS_V5.1.pdf?v=1640828798. Accessed 19 Mar 2023 25. polymaker: PolyLite ABS. Technical Data Sheet. V5.1. https://cdn.shopify.com/s/files/ 1/0548/7299/7945/files/PolyLite_ABS_TDS_V5.2.pdf?v=1640828798. Accessed 19 Mar 2023

Uncoupling Development Time from the Size of a Library of AM Parts Through Complexity Reduction and Modeling of Topology Optimization Results Guilain Lang1(B)

, Gerald Perruchoud1 , David Novo1 , and Stephane Brun2 1 CSEM, Neuchatel, Switzerland

[email protected] 2 Renault Trucks, Lyon, France

Abstract. Additive manufacturing (AM) framework enables the customization of geometries regardless of the number of parts to manufacture. Hence, the right product can be provided to the right customer. However, traditional optimization methods, used to improve performance and mass, are usually non-parametric. Therefore, the process flow shall be repeated even if all parts in a library display similar functions and overall geometry. As such, the relation between development time and the number of parts to design is linear. Hence, one cannot afford to generate large libraries using these approaches. In this paper, we propose an alternative solution for structural design, inspired by Knowledge Based Engineering. First, topology optimization is used to emulate a prior knowledge of optimal geometry. As this result is nonparametric, it is approximated using low complexity elements such as shells and beams. Then, batch optimization is run to model the optimal geometry in function of the load case. Finally, the ready to print part is reconstructed using traditional parametric CAD software. We benchmarked our methodology with the Design for AM (DfAM) of a library of chassis components. Thanks to the selective complexity reduction, the computation time is reduced significantly compared to traditional approaches. Furthermore, the parametrization of the part and the modeling of its behavior rationalize the design of families of parts. Indeed, most activities are only executed once. In this manner, our methodology uncouples the development time from the number of load cases allowing to design for AM entire libraries of similar part. Keywords: DfAM · KBE · Parametric design · FEM · Topology optimization

1 Introduction The development of Additive Manufacturing (AM) has opened the door to a new design freedom: its layered approach enables the manufacturing of intricate and freeform parts without increasing the manufacturing complexity and cost [1]. This revolution has led © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 130–144, 2024. https://doi.org/10.1007/978-3-031-42983-5_9

Uncoupling Development Time from the Size

131

to the emergence of new design tools such as topology optimization (TO) [2]. These tools are quite efficient to generate complex structural parts, impossible to manufacture before [3]. However, the process to transform optimization results into a ready to print geometry is tedious and time-consuming. Furthermore, this final geometry is nonparametric. Hence, it leads to a linear relation between development time and the number of parts to design. In this context, one cannot afford to generate large libraries using these approaches. In this paper, we propose an alternative methodology1 aiming to decouple the development time from the number of designs in a product library2 . We focus on libraries where parts perform the same function, but where the load-cases vary or where each part fosters different performances. Even if we focus on structural applications in this paper, these foundations and concepts can be applied to other domains of engineering. The novelty of this methodology does not lie in the tools used but comes from a selective reduction of complexity. We use the raw results of topology optimization to build a representative reduced model using shell and beam elements. Then a batch optimization is set up to model the optimal design parameters within the product family. Finally, traditional CAD software are used to reconstruct the parametrized part which includes all required manufacturing features. To illustrate our demarch, we apply our methodology to the Design for AM of a library of chassis components for the automotive industry.

2 Context and Use-Case Description MANUELA, Additive Manufacturing using Metal Pilot Line, is an European Union’s Horizon 2020 Research and Innovation Program (grant Agreement no. 820774) [5]. It aims to benchmark and promote AM in Europe. In the context of series production, this is almost impossible to compare the capabilities and cost of AM with traditional manufacturing technologies. Indeed, processes, design methodologies and logistics have been built on traditional manufacturing specificities. Hence, a consortium of AM experts has been gathered to rethink and implement an AM optimized development cycle, from design to manufacturing, including validations. The automotive industry is incredibly competitive. This degree of competition pushes manufacturers to streamline manufacturing processes. As such, small series often share common components to save on the cost of molds. This results in parts being oversized for most. As such, Renault Trucks investigates AM as an alternative to casting. Indeed, even for low volumes, AM allows to customize parts on clients’ demand and thus rationalizes vehicles’ weight, CO2 emissions and resource consumptions. MANUELA’s 1 Note that Önnermalm and al. Have proposed a methodology with a similar philosophy in

2019 [4]. They introduced Altair’s C123 approach which uses multiple levels of abstraction to simplify the conception of steel structures: the abstraction C1 is used to find load path options; the abstraction C2 is used to optimize the corresponding simplified models; and finally, level C3 is an FEM step to compare the different optimized design options. 2 To prevent ambiguity, we will use “product library” or “product family” when referring to similar designs. The “size of a series” will be used only to describe the size of the lot for serial production (e.g., number of parts to manufacture per year).

132

G. Lang et al.

business use-case PARCO aims to design a series of customized truck’s spring anchorage. The design shall not only be conducted for one specific truck, but for an entire range of vehicles. The actual line of trucks has four payload capacities and five structural interfaces (same size and position but different thickness of the chassis). Therefore, a total of twenty load cases shall be considered. The time allocated to this project does not allow us to design these parts separately. Hence, we developed this methodology to tackle this restriction.

3 Design Process Methodology Knowledge-Based Engineering (KBE) is a research topic which focusses on the structure of algorithms/processes to capture design intent and automatizes ill-structured processes using heuristics [6, 7]. These design rules are built on prior experience in solving a particular problem [8]. Our approach, inspired by KBE approach is composed on five main concepts: identification, interpretation, batch optimization, modeling, and reconstruction (see Fig. 1). The following chapters present each of these steps. In parallel, we illustrate our approach with the PARCO use-case.

Fig. 1. Flow chart of the proposed methodology. Highlights of the five main concepts: identification, interpretation, batch optimization, modeling, and reconstruction. Illustration with the PARCO use case.

3.1 Identification: Initial Guess on Geometry The first step of this methodology aims to propose an initial guess on the geometry. Such activity lays into the category of “creative design” based on the taxonomy of

Uncoupling Development Time from the Size

133

Sriram and al [9]. This is one of the most complicated classes of KBE problems which by construction need experience. We propose to use topology optimization to provide the optimal material distribution within a structural part. This result can then be used to emulate a prior knowledge of the optimal structural architecture. Note that this step can be skipped if this knowledge is already available. Topology optimization is a constrained optimization method which distributes material within a design space (Solid Isotropic Material with Penalization method [2]). Its most common formulation minimizes the overall compliance while keeping the mean density under a user-provided value. The density, η ∈ [ηmin , 1] attributed to each FEM elements alters locally the stiffness, through the Young modulus, (SIMP model of degree n: E(η) = E0 · ηn ) and the material density (linear model: ρ(η) = ρ0 · η). ⎧ η < ηobj ⎨   (1) min U T · K(η) · U , s.t. K(η) · U = F η ⎩ 0 < ηmin < ηi < 1 Where, U and K is the generalized displacement and stiffness matrices, respectively. η is the integral of the design density (η), it allows to constrain the amount of material distributed by the algorithm (the user provides this limit through ηobj ). In our methodology, the whole available envelope must be considered as design space. If important interfaces are identified, (e.g., screws wall, important surface of contact) these regions can be defined as non-design space where the density, η, will be kept at 1 by the optimizer. Since topology optimization used a gradient-based approach, the number of elements have a major impact on the computation time. At that stage, we are only seeking a guess on geometry. Hence, to stay coherent with the philosophy of this methodology, a relatively coarse mesh can be used. The selection of the load case has a significant impact on the solution: it drives the topology optimization. Two approaches can be considered: median or extreme load cases. We recommend using the median values which leads, intuitively, to an overall lower mass (in term of 2-norm on the whole library). However, if the stress level is critical, for the highest load cases, choosing extreme values may be more relevant. In PARCO, we used the formulation proposed in Eq. 1. We selected an objective volume fraction, ηobj , of 0.3 and a median load case. Furthermore, PARCO’s requirements impose a minimum stiffness. Therefore, we added a total of 6 constraints on the displacement of the interface with the spring shackle. Note that we do not consider manufacturing constraints during the topology optimization. Indeed, it may complexify the model simplification in the next step. The design space used in PARCO and the TO raw results are presented in Fig. 2. 3.2 Interpretation and Simplification of the Initial Guess Traditional AM design methodologies use smoothing or polyNURBS approximation to interpret the result of TO. Such solutions are usually not parametrizable. Hence, this single solution cannot be adapted to withstand other load cases. The second step of

134

G. Lang et al.

Fig. 2. Design space, and result of the topology optimization for the median load case of PARCO.

our methodology bypasses this limitation. The main objective is to reduce our problem complexity to mainstream and accelerate parametric analysis. In the context of structural design, shells and beams can often approximate the result of TO. Furthermore, such lowdimensional elements highlight possible design parameters through shells and beams thicknesses. In the use-case PARCO, we indeed used only shells and beams (see Fig. 3). In Solidworks, we generate the shells manually from the initial envelope used for the TO of step 1. Then, we import and mesh these surfaces in Altair Hypermesh. A beam element is added to mimic TO results (represented as a blue line in Fig. 3). We group these elements into 5 components whose properties will drive shells’ and beams’ thicknesses during the next step. If we decided to put aside manufacturing constraints before, we deem relevant to consider them during the simplification process. Printing orientation shall be defined, and manufacturing features must be identified. For example, the shape of most surfaces has been adapted to allow a supportless printing of PARCO. However, one shell was perpendicular to the building plate and required the integration of non-sacrificial supports. Since the impact of these structures has been considered negligible on the operational performance, we decided to not implement them yet to keep the model as simple as possible. 3.3 Batch Analysis on Simplified Model The third step generates the data required to model the system. We introduced design parameters in the previous section. If a batch optimization can be built on design parameters presented in the previous section, we think this highly inefficient in the context of product design. Instead, the designs generation can be formulated as a finite number

Uncoupling Development Time from the Size

135

Fig. 3. Illustration of our interpretation of the TO results using shell and beam elements.

of optimization where design parameters are the optimization variables. We introduce “external parameters” to differentiate elements in the batch. The interpretation of these external parameters depends on the objective of the design family. We identified four potential interpretations of external parameters based on the objective of the library (this list is not exhaustive). The Fig. 4 presents a decision flow chart to identify the adequate interpretation. 1. One performance shall be optimized (typ. Mass) for a range of load cases. In that configuration, the load cases are parametrized through the external parameters. The

136

G. Lang et al.

Fig. 4. Flow chart to identify batch construction scheme (interpretation of external parameters) based on the objective of the series.

objective of the batch analysis is then to find an optimal set of design parameters for each load case. 2. Multiple performances shall be optimized for a range of load cases, and their relative importance is known. This problem can be reduced to case 1 using the derived cost function defined as the weighted sum of all performances. 3. Multiple performances shall be optimized for a single load case, but their relative importance (weights) is unknown. In that case, it is not possible to identify an optimal solution as in the first two approaches. Indeed, improving one performance will have a negative effect on the others. In multi-objective optimization, this is the definition of the Pareto front: no solution is dominated by another. The objective of the batch is thus to find the Pareto front and identify its corresponding design parameters. The simplest solution is to reduce the problem to a simple optimization under constraints: if we have m different performances to optimize, we create m − 1 external parameters. These external parameters are then used to set equality constraints on m − 1 parameters while the last one is being optimized. For a model with n design parameters, the multi-objective is optimization is thus simplified into: ⎧ ⎨ lbi < pdesign,i < ubi , i = 1, . . . , n (2) min {perf1 }, s.t. pexternal,j = λj , j = 2, . . . , m pdesign ⎩ ...

Uncoupling Development Time from the Size

137

Where, lbi and ubi are respectively the lower and upper bounds of the i-th design parameter. In this formulation, the set →λ1 , . . . , λm−1  varies for each instance of the series. 4. Multiple performances shall be optimized for a range of load cases, but their relative importance is unknown. In that case, the objective of the batch is to find the Pareto front for each load case. Therefore, we first apply approach three to reduce the multiobjectives optimization problem and use this formulation as input of the first approach. It leads to a total of m+k −1 external parameters, where k is the number of parameters required to control the use cases. In all approaches, the batch is thus a succession of simple optimizations under constraints. Each run represents one specific combination of external parameters. If the behavior of the system’s is a priori known, we recommend reducing the number of combinations using a design of experiment approach. However, if this information is unknown – as it is often the case - a full box evaluation will lead to a batch’s size of: lenbatch =

n 

ki

(3)

i=1

Where, ki is the number evaluation of the i-th external parameter, and n the number of external parameters. Note that additional probes can be defined to constrain the optimization. They can be any output of the FE simulation such as displacements of important nodes, maximum stress, or strain. It does not increase the size of the batch but may influence the optimization time. In PARCO, our design problem lays in the first scenario. Therefore, we define two external parameters: the first to parametrize the front axle load of the truck (loads’ magnitude) and the second to control the thickness of the chassis (which is interfaced in the model). Then, we identify eight outputs: six displacement/rotation of the interface with the spring shackle, the maximum stress, and the total mass. Limits on compliance and safety factor are given by the requirements of the part. Therefore, we minimize the mass while keeping both displacements/rotations and maximum stress below the imposed limits. We implemented this batch using the in-build module of Altair HyperStudy and limited the number of iterations of the optimizer to fifty. We run a full box on four truck weight and five chassis thickness. The result of this batch simulation is thus a list of twenty sets of optimal design parameters and their corresponding masses (one for each external parameters couple). 3.4 Modeling The fourth step aims to model the relation between design and external parameters. Given the external parameters, the model shall provide the corresponding optimal design parameters. This modeling step reduces the number of simulations required. Indeed, for any intermediate load cases, optimal design parameters can then be extrapolated.

138

G. Lang et al.

The selection of the adequate model shall be based on both the evolution of the parameters and their interactions. Therefore, we invite the reader to identify the best identification or approximation method for her/his application. In PARCO, the need for modeling is limited: the sizes of trucks are normalized, and we evaluated all combinations during the batch optimization. Therefore, all optimal parameters are already known. However, we can use any approximation approach to model the optimal shell thicknesses in function of external parameters (see Fig. 5). Hence, if a new truck is being developed or if the chassis thickness varies, optimal designs can be extrapolated without the need of further computation.

Fig. 5. Global (left) and local (right) cubic approximation of the optimal back skin thickness (cyan surface in Fig. 3) in function of truck’s front axle load (FAL) and chassis thickness (interface). All values have been normalized to protect intellectual property.

3.5 Reconstruction The last step of this method is the reconstruction and parametrization of the geometry. As mentioned earlier, usual AM workflows reconstruct the geometry (e.g., solution of the topology optimization) using polyNURBS surfaces. However, free-shape geometries cannot be parametrized in general. Thus, we propose to come back to more traditional CAD approaches where design parameters can be used as input of the construction. The interpretation step already provides the skeleton of the geometry: the surfaces used for the batch processing can easily be extruded. The length of the extrusion is provided by the model generated during step 4. Manufacturing features such as non-sacrificial supports or reference surfaces can be integrated at that stage. In general, such structure will have only limited impact on the stress distribution in the part. However, we propose to validate the final design – at least one in the library– through traditional FE analysis. A total of four modifications/features have been introduced in PARCO (see Fig. 6):

Uncoupling Development Time from the Size

139

1. The beam has been replaced by a shell to facilitate manufacturing and improve symmetry. 2. A thin structure has been used as a non-sacrificial support for the central shell. 3. A total of four reference surfaces have been added to clamp the part during postmachining operations. 4. The back shell has been opened – where stress was minimum – to prevent issues during part’s removal from the build plate.

Fig. 6. Reconstruction of the geometry based on the shell representation of the optimized part. Additional features have been introduced to ensure support-less manufacturing and rationalize post-manufacturing operations.

4 Results and Discussions 4.1 PARCO Library Performances As mentioned in Sect. 2, the main objective of PARCO was to investigate if AM can be used as an alternative to traditional manufacturing for serial demand. The first interesting consideration is the print-on demand capability of AM. Indeed, we mentioned that trucks often share standard parts which are thus oversized for the smallest vehicles in the range. With no need for expensive molds, AM can thus offer a great alternative to customize such small series. Our methodology enables users to foster the full customization potential of AM. Indeed, it provides a design framework to customize a whole family of parts with almost no additional effort, nor time. In the PARCO use-case, we optimized the mass of a family of spring anchorage while keeping their stiffness and yield safety factor within admissible bounds. We were

140

G. Lang et al.

able to reduce the mass by approximately 40% for the whole library. As an example, the printed demonstrator shown in Fig. 8, (optimized for a low front axial load and an intermediate interface thickness) weights 6.25 kg while its casted counterpart was above 10.5 kg. It is important to notice that the design freedom is not the only leverage we used to reach these performances. Indeed, we tend to forget that the range of material available is dictated by the manufacturing processes. During PARCO, we used the EOS 20MnCr5 powder. This alloy displays a higher yield strength than usual cast steels. Hence, the walls can be thinner – leading to a higher stress in the part – without negative impact on factors of safety. We shared the performances of our final library in Fig. 7. 4.2 Development Time The mechanical performances of our methodology cannot be benchmarked with our usecase PARCO, the material properties of AM and cast alloys being too different. In any case, our methodology is not exclusive to one optimization scheme – which dictates the outcome –, thus, it would be pointless to compare the results. However, our methodology stands out from traditional approaches because it uncouples the development time from the number of – similar – parts to design. This behavior is achieved through two main strategies: the minimization of computation’s impact on the development time and the extrapolation of new solutions. The model simplification step drastically reduces the computation time. Indeed, we estimated that direct implementation of TO would take multiple hours for each load case (in the case of PARCO). Using our simplified model, the optimization takes at most 8.5 min (maximum 50 evaluations and evaluation’s duration 5 the results were omitted, due to the high scanning speed v, resulting in distributions similar to that of geometry RH F.

In Fig. 4 we can observe that for all five printed geometries and the depicted machine parameter configuration, an overall high HPSI score for the inpainted samples can be achieved. A higher HPSI value corresponds to a better match between x ˆ and x. Thus, indicating good image quality of the model prediction. With the increase of available information, the inpainting task should become easier. Our results confirm this assumption, that indeed with increasing τ , which corresponds to the amount of available information, the probability of generating

156

H. A. Zhou et al.

an image with high HPSI increases too. However, our results also show, that images with low HPSI were generated, despite the high amount of available information. This becomes especially relevant, for parts where one layer is printed in a few time steps, like LA and RH F, so that an accurate prediction should be possible after one or two time steps t.

Fig. 5. The top row depicts machine parameter configurations at position p = 3, and the bottom row depicts position p = 5. One square tile consists of xτ (top left), its corresponding mask mτ (top right), the cross-section heat signature x (bottom left) and the predicted image x ˆ (bottom right). Other machine parameter configurations are omitted due to low energy distribution levels and therefore low image contrast.

Besides an overview of the distribution of HPSI values across different geometries and machine parameter configurations, we also show inpainting results with their corresponding HPSI values in Fig. 5. The generated samples display accurate heat signature levels, which can be visually confirmed by comparing the greyscale values of x and x ˆ. But under closer inspection, we observe that the model inpaintings are limited by their degree of detailed features. Especially in images, where bigger regions are missing (see Fig. 5 geometry CA F and p = 5), instead of detailed textures, the missing area was filled with an almost evenly distributed grey value. This indicates that the model learned to average over all possible solutions, instead of distinct detailed solutions. However, features like slits that occur in part geometries like RSA F and RS F are accurately extended, thus confirming that the model is capable of generating images with features of a limited level.

5

Discussion and Conclusion

With our results, we demonstrated the inpainting performance of our model to produce high-quality images with plausible heat signature levels and to some

Fabrication Forecasting of LPBF Processes Through Image Inpainting

157

degree detailed feature continuity. High HPSI values of generated samples compared to corresponding ground truth instances indicate high fabrication forecasting accuracy. However, with the illustrated limited level of detail in the generated samples, small critical features like pores that cover a few pixels will not be predictable. We conclude from our results, that further inpainting performance improvements are required, by either adjusting the presented approach or modifying the model architecture. Although previous work of Gobert et al. [7] and Zhang et al. [8] already applied generative models for LPBF in-situ OT monitoring data, their results are only comparable on an image-based qualitative level. In both works, the authors evaluate their models by manually comparing expected heat signature features between prediction and ground truth. Likewise, we validated the performance of our model through the continuity of slits, which arguably surpasses the level of detail from generated images in previously reported results. Because qualitative evaluations depend on the subjective opinion and domain expertise of the evaluator, we additionally introduced the usage of HPSI as an image quality metric to quantify our model performance. This not only enables better comparability between results, but also provides a model hyperparameter tuning objective. Although HPSI is reported to correlate with human perception, it remains up to debate, how to accurately measure fabrication forecasting performance for LPBF processes. Our contribution and the associated improvement in fabrication forecasting enable more-profound decisions during process control. Although the presented forecasting capabilities are limited to the same layer and regarding the feature level of detail, they leverage the time dependant sequential property of the printing process and monitoring setup, to predict future fabrication outcomes. Given an accurate online forecasting model of the fabrication process, model-based predictive control strategies could use these forecasting predictions to steer the fabrication process towards a defect-free manufacturing result. Acknowledgments. Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy - EXC-2023 Internet of Production - 390621612. Simulations were performed with computing resources granted by RWTH Aachen University under project rwth1228.

References 1. Verein Deutscher Ingenieure: VDI 3405: Additive Fertigungsverfahren: Grundlagen. Begriffe, Verfahrensbeschreibungen (2014) 2. Grasso, M., Remani, A., Dickins, A., Colosimo, B.M., Leach, R.K.: In-situ measurement and monitoring methods for metal powder bed fusion: an updated review. Meas. Sci. Technol. 32(11), 112001 (2021) 3. Mahmoud, D., Magolon, M., Boer, J., Elbestawi, M.A., Mohammadi, M.G.: Applications of machine learning in process monitoring and controls of L-PBF additive manufacturing: a review. Appl. Sci. 11(24), 11910 (2021)

158

H. A. Zhou et al.

4. Yadav, P., Singh, V.K., Joffre, T., Rigo, O., Arvieu, C., Le Guen, E., Lacoste, E.: Inline drift detection using monitoring systems and machine learning in selective laser melting. Adv. Eng. Mater. 22(12), 2000660 (2020) 5. Feng, S., Chen, Z., Bircher, B., Ji, Z., Nyborg, L., Bigot, S.: Predicting laser powder bed fusion defects through in-process monitoring data and machine learning. Mater. Des. 222, 111115 (2022). https://ncedirect.com/science/article/pii/ S0264127522007377 6. Schwerz, C., Nyborg, L.: A neural network for identification and classification of systematic internal flaws in laser powder bed fusion. CIRP J. Manuf. Sci. Technol. 37, 312–318 (2022) 7. Gobert, C., Arrieta, E., Belmontes, A., Wicker, R.B., Medina, F., McWilliams, B.: Conditional generative adversarial networks for in-situ layerwise additive manufacturing data (2019) 8. Zhang, S., Jahn, A., Jauer, L., Schleifenbaum, J.H.: Geometry-based radiation prediction of laser exposure area for laser powder bed fusion using deep learning. Appl. Sci. 12(17), 8854 (2022). https://www.mdpi.com/2076-3417/12/17/8854 9. Zenzinger, G., Bamberg, J., Ladewig, A., Hess, T., Henkel, B., Satzger, W.: Process monitoring of additive manufacturing by using optical tomography. In: AIP Conference Proceedings, pp. 164–170 (2015) 10. Deutsches Institut f¨ ur Normung: DIN EN ISO/ASTM 52902:2020–05: Additive manufacturing - test artifacts - geometric capability assessment of additive manufacturing systems (ISO/ASTM 52902:2019); German version EN ISO/ASTM 52902:2019 (2020–05) 11. Otsu, N.: A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man Cybern. 9(1), 62–66 (1979) 12. Liu, G., Reda, F.A., Shih, K.J., Wang, T.-C., Tao, A., Catanzaro, B.: Image inpainting for irregular holes using partial convolutions. In: Ferrari, V., Hebert, M., Sminchisescu, C., Weiss, Y. (eds.) ECCV 2018. LNCS, vol. 11215, pp. 89–105. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-01252-6 6 13. Zhao, Y., et al.: VCGAN: video colorization with hybrid generative adversarial network. IEEE Trans. Multimedia, 1 (2022). https://arxiv.org/pdf/2104.12357 14. Reisenhofer, R., Bosse, S., Kutyniok, G., Wiegand, T.: A Haar wavelet-based perceptual similarity index for image quality assessment. Sig. Process. Image Commun. 61, 33–43 (2018). https://www.sciencedirect.com/science/article/pii/S0923 596517302187 15. Li, L., et al.: A system for massively parallel hyperparameter tuning. In: Conference on Machine Learning and Systems (2020). https://arxiv.org/pdf/1810.05934 16. Bergstra, J., Yamins, D., Cox, D.D.: Making a science of model search: hyperparameter optimization in hundreds of dimensions for vision architectures. In: Proceedings of the 30th International Conference on International Conference on Machine Learning - Volume 28, ICML 2013, pp. I-115–I-123. JMLR.org (2013) 17. Liaw, R., Liang, E., Nishihara, R., Moritz, P., Gonzalez, J.E., Stoica, I.: Tune: a research platform for distributed model selection and training. arXiv preprint arXiv:1807.05118 (2018)

The Influence of Nozzle Size on the Printing Process and the Mechanical Properties of FFF-Printed Parts Joakim Larsson(B) , Per Lindström, Christer Korin, Jens Ekengren, and Patrik Karlsson Örebro University, 701 82 Örebro, Sweden [email protected]

Abstract. Recent process developments in Fused Filament Fabrication (FFF), such as the possibilities to use high end polymers (for example PEEK) or to manufacture metal parts and parts reinforced with continuous fibers, have increased industrial interest. Previously, this additive manufacturing (AM) technology was mostly popular among hobbyists thanks to its low investment cost. With the increased industrial interest comes higher demands on product strength and production efficiency. The FFF process has many parameters that should be optimized to meet these tougher requirements. One of these parameters is the size of the nozzle through which the filament is extruded. Today a fairly wide range of sizes are available on the market, but most standard-sized printers come equipped with a 0.4 mm nozzle. In this study, a wide range of nozzles of different sizes have been manufactured to investigate how the nozzle size affects both the printing process and the mechanical properties of the printed parts. Tensile bars have been manufactured in polylactic acid (PLA) using 7 different nozzle sizes. The samples were investigated by means of computer tomography (CT) and optical microscopy and subjected to tensile testing. Keywords: FFF · Fused Filament Fabrication · PLA · Nozzle size

1 Introduction With the possibility of manufacturing more advanced materials using Fused Filament Fabrication (FFF) industrial interest has increased. Recent developments have made it possible to manufacture metal parts, parts made in advanced engineering polymers (such as PEEK) and parts reinforced with continuous fibers. Previously, this specific additive manufacturing (AM) technology was mostly used among hobbyists due to its low investment cost and ease of use. In the past, the technology has mainly been used to manufacture parts for low-strength applications, such as design prototypes or toys. However, the new possibilities in terms of material selection may facilitate the manufacture of parts for tougher applications. With increased industrial interest comes higher demands on product strength and production efficiency. In the FFF process, plastic filament is used as a feed stock. This © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 159–170, 2024. https://doi.org/10.1007/978-3-031-42983-5_11

160

J. Larsson et al.

plastic wire is commonly produced with a diameter of 1.75 mm or 2.85 mm, where most of the printers on the market use a 1.75 mm filament. During printing the material is fed through a heated nozzle (nozzle temperature) with a specific opening diameter (nozzle size) forcing the melted plastic through. The nozzle is then moved relative to the build plate (which is normally heated) according to a path predefined using a slicing software. When a layer is completed the distance between the nozzle and the build table is changed (layer height) and the same procedure is followed for the next layer. The FFF process has many parameters that should be optimized to meet the increased requirements. Parameters that are normally controlled and tuned include nozzle temperature, bed temperature, nozzle size, printing speed and layer height. Many investigations have been performed in the field in recent years. Gomes-Gras et al. investigated the fatigue performance of polylactic acid (PLA) specimens manufactured using FFF and found that the infill density and the layer height have the greatest impact on fatigue performance [1]. Zekavat et al. investigated the influence of nozzle temperature on the strength of FFF-made PLA samples. Computed tomography (CT) was used to characterize the printed samples and tensile testing to evaluate the strength depending on nozzle temperature. The authors concluded that higher nozzle temperatures resulted in printed parts with higher strength, mostly due to stronger bonds between the extruded filament lines [2]. Afonso et al. also studied the influence of nozzle temperature and printing speed in a FFF process on the strength and mass of samples printed in PLA. This study showed that the nozzle temperature had the largest impact on the studied parameters; as in Zekavat et al. it was shown that higher temperature equals stronger and heavier parts [3]. Frunzaverde et al. also investigated the influence of nozzle temperature on FFF-printed PLA parts, the study also included the influence of filament color. This study showed that the various colored materials reacted differently to the change in nozzle temperature [4]. No significant change in strength could be seen when changing the nozzle temperature from 200–230 °C for non-colored filaments, while a decrease in strength was observed for colored filaments under the same conditions. These results contradict those of previous studies [4]. Butt et al. studied how nozzle temperature and filament feed rate affect the properties of printed PLA and PLA reinforced with functionalized graphene nanoplatelets. The authors concluded that high feed rates and high temperature lead to heavier printed parts [5]. Chaidas and Kechagias examined the effect of nozzle temperature and layer height on the surface roughness and the dimensional accuracy of FFF-printed PLA/wood parts, concluding that lower temperature and layer height lead to better surface quality and higher dimensional accuracy [6]. Buj-Corral et al. looked at how printing parameters affect the surface roughness of FFF-printed PLA parts, showing that the layer height, followed by the nozzle size, has the largest impact on surface roughness [7]. LuisPérez et al. studied how printing parameters in an FFF process affected the dimensional tolerances of parts printed in PLA. The study showed that the best results were obtained using the smallest nozzle and the lowest layer height [8]. Atakok et al. investigated the strength properties of FFF-printed PLA parts depending on the layer height and infill density and concluded that the highest layer thickness resulted in the strongest parts [9]. Rajpurohit and Dave investigated the flexural strength of FFF-printed PLA samples

The Influence of Nozzle Size on the Printing Process

161

depending on the raster orientation, nozzle size and layer height. It was found that the flexural strength decreased with an increasing raster angle and layer height [10]. Czyzewski et al. studied the influence of nozzle size on functional properties of FFF-printed PLA parts using nozzle sizes from 0.2 mm up to 1.2 mm [11]. Their results showed similar tensile strength for specimens printed using 0.4 and 0.8 mm nozzles and lower strength for the samples printed using 0.2 and 1.2 mm nozzle sizes. However, looking at the microscopy study presented by the authors it is clear that the print made using the 1.2 mm nozzle has been unsuccessful as it is full of air gaps between the extruded filament lines. Syrlybayev et al. have made a review summarizing some of the work done in the field of optimizing the FFF process towards making stronger parts. They found nozzle diameter to be one of the parameters with a possibly considerable impact on strength, but recommended further studies to be done [12]. Thus, the primary topic for the current investigation is how the nozzle size influences the strength of FFF-printed PLA parts. Today a fairly wide range of nozzle sizes are available on the market, but most standard sized printers come equipped with a 0.4 mm nozzle. As previously concluded, the strength of the printed parts is dependent on the bindings between the extruded filament lines [2]. An increased nozzle size means that the number of extruded filament lines and thereby the number of connections between the filament lines would be decreased. In theory, this should result in stronger parts. Another major benefit of using a larger nozzle diameter is that the printing process will be faster since more material per time unit is extruded through the nozzle. In this study, a wide range of nozzles of different sizes have been used to investigate the effect of nozzle size on both the printing process and the mechanical properties of the printed parts. Tensile bars have been manufactured in PLA using 7 different nozzle sizes. The samples were mechanically tested by tensile testing and analyzed by means of CT and optical microscopy.

2 Materials and Methods 2.1 Printing Nozzles To investigate the effect of nozzle size on printing efficiency and printed parts 7 different nozzle sizes were chosen arbitrarily, ranging from the standard 0.4 mm nozzle up to a 2.0 mm diameter nozzle. Nozzles in sizes included in this study were bought when available on the market; the larger diameter nozzles were manufactured using a CNClathe in-house. The nozzles were investigated by means of optical microscopy prior to being used in the FFF process. Nozzle openings were analyzed and the area of the opening was measured in the microscopy images using the image processing software Image J [13]. The micrographs were converted to binary and then imported into Image J where the area was measured. The chosen nozzle sizes and the corresponding actual nozzle opening areas are presented in Table 1. As can be seen in Table 1, there are considerable differences in the nominal and the actual area of the two smallest nozzles; however, as these were

162

J. Larsson et al.

bought from a reputable supplier there is no reason to believe that these nozzles are worse than any other bought nozzles. The 0.6 mm nozzle has the largest deviation: the measured area corresponds to a nozzle diameter of 0.54 mm, which is 10% smaller than the nominal diameter. Table 1. Nozzles used in the study. Nozzle size (mm)

Actual opening area (mm2 )

Corresponding nozzle diameter (mm)

Area difference (%)

Bought/Manufactured

0.4

0.11

0.38

11

Bought

0.6

0.23

0.54

20

Bought

0.8

0.47

0.78

6

Bought

1

0.82

1.02

4

Bought

1.2

1.10

1.18

3

Manufactured

1.8

2.60

1.82

2

Manufactured

2

3.17

2.01

1

Manufactured

2.2 Printing, Design and Material The printer used to manufacture all the samples evaluated in this paper was a slightly modified Creality Ender 3 [14]. The printer was modified by changing the diameter of the plastic filament from the Ender 3 standard 1.75 mm to 2.85 mm, needed for nozzles larger than 1.75 mm. The modification was made by replacing the standard feeder unit with a complete feeder unit from an Ultimaker 2 which utilizes 2.85 mm wire [15]. The feeder was fitted in the same position as the original feeder and the filament movement was calibrated using a handheld caliper with a reported tolerance of ±0.03 mm. The nozzle heater had to be replaced in order to facilitate printing on the Ender 3 using nozzles exceeding 1.2 mm in diameter. The original 40 W heater was therefore replaced with a 60 W heater, increasing the heating capacity by 50%. All the samples were printed using the same nozzle temperature, 250 °C. This temperature was chosen because it has been shown to give the highest density according to a previous study [2]. The bed temperature was kept constant at 60 °C during printing and the printing speed was set to 50 mm/s. One big advantage regarding printing time when utilizing larger nozzles, is that the layer height may be increased as nozzle size is increased. A rule of thumb commonly promoted on the Internet is that layer heights of up to 0.8 x nozzle diameter may be used, meaning that the number of layers can be rapidly reduced when using larger nozzles, leading to an immense increase in time efficiency. However, in this study it was chosen only to change one parameter, namely nozzle size. Accordingly, a constant layer height of 0.15 mm was used for all experiments.

The Influence of Nozzle Size on the Printing Process

163

The tensile test samples were designed according to ISO 527 standard [16]. The dimensions of the middle part of the bars were therefore 10 x 4 mm and the length of the middle part is designed to be used with an 80 mm extensometer. The geometry used for the investigations in this study is shown in Fig. 1a. For each nozzle size, six tensile bars were manufactured, all fitted to one build plate, meaning that all the samples were printed in one build. The samples were placed with a 45° angle on the build plate. With the Ultimaker Cura slicer [17], used to prepare the G-code for the manufacturing process, the chosen orientation led to the plastic being extruded along the x- and y-axis of the samples as shown in Figs. 1b and c. The nominal nozzle diameter was used as input in the slicer software, and not the actual measured value. This was done because in a normal use situation the exact diameter of the nozzle is unknown.

Fig. 1. (a) Dimensions of the tensile bars used in this study. Figures (b) and (c) show the two types of printing patterns used to build the tensile bars.

The material chosen for the study was a polylactic acid (PLA) which is one of the most common materials used in FFF. The specific PLA used for manufacturing all the samples was Kimya PLA-R which is a filament made of minimum 95% recycled material [18]. The material properties of the filament were investigated using tensile tests. The diameter was measured 30 times over 3 different spools of material used to manufacture the samples. The results from the tensile test are compared to the manufacturer’s specification in Table 2.

164

J. Larsson et al. Table 2. Properties of used PLA filament, Kimya PLA-R [18].

Data source

Yield strength (MPa)

Tensile strength (MPa)

Youngs modulus (GPa)

Diameter (mm)

Measured

43

55

3.27

2.84 ± 0.02

Manufacturer

-

55

2.82

2.85 ± 0.1

2.3 Characterization of Printed Samples The different characterization methods used to evaluate how the nozzle size affects the quality of the printed parts are presented in this section. Mechanical Testing As mentioned in Sect. 2.1, samples were printed for tensile testing. An Instron 4458, fitted with a 300 kN load cell and an extensometer for 80 mm samples, was used for tensile testing. All the samples were tested at 5 mm/min. Five samples were prepared and tensile tested. Measurements The dimensions of the waist of the tensile bars were measured using a handheld tactile micrometer with a reported accuracy of ±1.6 µm. The width and the thickness were measured on all samples in randomized spots, repeated three times. The tensile bars were also weighed using a Sartorius BP2215 with an accuracy of ±0.1 mg. All measurements were performed prior to the mechanical testing. Computed Tomography (CT) Prior to tensile testing, specimens were investigated by means of computed tomography (CT) using a Bruker SkyScan 1272 3D micro CT (µ-CT) system [19]. The samples were scanned at acceleration voltage and filament current of 60 kV and 166 µA, respectively. The resulting volumes had an isometric voxel size of 3.75 µm. Reconstruction of CT data was performed using InstaRecon software, correcting for beam hardening, ring artefacts and post alignment correction [20]. The volumes were investigated using VG studio MAX, commercially available software [21]. Microscopy After being subjected to mechanical testing the tensile bars were investigated by means of optical microscopy. The microscopy focused on fracture surfaces. The images were taken using a Zeiss Discovery.V12 [22].

The Influence of Nozzle Size on the Printing Process

165

3 Results and Discussion

Relative printing time (%)

Tensile bars were successfully printed on the modified Ender 3 and were characterized using the methods described in Sect. 2.3. Information regarding printing time is presented in Fig. 2, showing that printing time is radically reduced as the nozzle size is increased, despite keeping the layer height constant throughout the manufacturing process. Utilizing a 2 mm nozzle, printing time is roughly 80% shorter than with the standard 0.4 mm nozzle. The non-linearity at the end of the curve indicates further changes when moving from the 1.8 mm to the 2 mm nozzle; however, no significant change in the printing pattern could be found when examining the planned path in the slicing software.

100 80 60 40 20 0 0

0.4

0.8

1.2 Nozzle size

1.6

2

Fig. 2. Printing time for each nozzle size used.

Figure 3 shows the mean results from the measurements of the samples, namely their geometrical dimensions and weight. As shown in the figure, the weight follows the thickness of the samples, as is to be expected of printed samples with similar porosity. The mean weight varies roughly 0.5 g between the printed sample batches. The thickness is the measurement that deviates the most from the designated value. The samples were designed to be 4 mm. However, all the resulting samples are thinner than 4 mm. Nevertheless, the size of the nozzle does not show a trend in the resulting dimension and weight of the samples. The observed differences seem to depend on other parameters, probably due to inaccurate nozzle diameter parameter settings in the slicer. The weight was also calculated using the weight from the slicing software and the resulting thickness of the samples. There are larger deviations between the actual and calculated weights for the smaller nozzle sizes (see Fig. 3), with calculated values exceeding actual weights. This could indicate that the density of tensile bars printed with the larger nozzle sizes is higher. The difference between calculated and measured weight are very small for the tensile bars printed using 1 mm and larger nozzles, less than 1.5%, the deviation for the smaller nozzles is between 2 and 4%. This measurement could be a good indication of the porosity of the printed samples. Three samples from 3 of the 7 batches were chosen to be investigated using CT. These samples were manufactured using the smallest and the largest nozzles, and one

166

J. Larsson et al.

Measured dimension (mm) / Weight (g)

12 10 8 Width

6

Thickness Weight

4

Calculated weight 2 0 0

0.4

0.8

1.2

1.6

2

Nozzle size Fig. 3. Dimensional measurements and weight (calculated and measured) of the printed tensile bars for each nozzle size used in the manufacturing process. Measurements were taken prior to mechanical testing.

in between, namely 0.4 mm, 1 mm and 2 mm nozzles. The results of the material characterization by means of CT are presented in Fig. 4, where (a) (c) and (e) show a x–y layer of the specimens analyzed in the center of the thickness of the sample. Images (b), (d) and (f) in Fig. 4 show the cross-sections at the middle of the samples. As seen, the material becomes more homogeneous as the nozzle size increases, which supports the theory from the differences in actual and calculated weight. In Fig. 4(a) the printing pattern is quite clearly visible and air gaps can be seen between the extruded lines. In Figs. 4(c) and (e) representing the tensile bars printed using 1 mm and 2 mm nozzles, it is difficult to distinguish the extruding lines because of a more homogeneous material. The bottom layers (approximately the first 2 mm, i.e. half of the thickness of the tensile bar) appear much denser compared to the top halves of the printed samples. This could be due to less heat from the build plate being transported to the printed layer, leading to weaker adhesion between the previous layer and the extruded material. Also, as the previous layer is colder, the extruded material is cooled at a higher rate, which could lead to less spread of the extruded material, resulting in less/no overlap between the extruded lines. Results from the tensile tests are presented in Figs. 5 and 6. The results are mean values from the tensile tests of 5 samples. Figure 5 illustrates the results regarding Young’s modulus, showing similar moduli for all the tested samples with a mean value of approximately 3.6 GPa. The samples printed using the 0.4 mm nozzle have a slightly lower Young’s modulus of 3.5 GPa. Compared to the filaments Young’s modulus of 3.2 GPa the printed samples all have higher modulus.

The Influence of Nozzle Size on the Printing Process

167

Young's modulus (GPa)

Fig. 4. CT-analysis results of investigated samples printed with different nozzle sizes. Arrows indicate the building direction, for nozzle sizes (a) and (b) 0.4 mm, (c) and (d) 1 mm, and (e) and (f) 2 mm.

4 3.5 3 2.5 2 1.5 1 0.5 0

Printed samples Filament

0

0.4

0.8

1.2 Nozzle size

1.6

2

Fig. 5. Young’s modulus for each nozzle size, measured during the tensile tests.

The strength of the printed samples seems to increase with the nozzle diameter. This is in accordance with the hypothesis presented in the introduction. The samples printed using the largest nozzle (2.0 mm) are roughly 50% stronger than the samples printed using the standard 0.4 mm nozzle.

168

J. Larsson et al. 70 Tensile strength Yield stress

Stress (MPa)

65 60 55 50 45 40 35 0

0.4

0.8

1.2 Nozzle size

1.6

2

Fig. 6. Results from tensile tests for each nozzle size used.

Fig. 7. Microscopy of the fracture surfaces from samples pulled in the tensile test. The arrow indicates the building direction, with (a) 0.4 mm nozzle, (b) 1 mm nozzle and (c) 2 mm nozzle.

Regarding the level of the ultimate tensile strength of the samples printed using the four largest nozzles, it should be noted that the results show a higher tensile strength than

The Influence of Nozzle Size on the Printing Process

169

that of the filament used to produce the samples. As all the printed samples have roughly the same cross-section, it seems unlikely that this trend is due to a measurement error. The filament used to manufacture the parts was tested with the same tensile tester used for testing the tensile bars, and the tests were performed after pulling the printed samples. However, other researchers have reported similar trends, where the printed samples are stronger than the used filament [9, 10]. After being subjected to tensile tests the fracture surfaces of the samples were investigated using microscopy. The results for the tensile bars printed with the same nozzle size as the samples investigated with CT are presented in Fig. 7. The microscopy results are very similar to the CT results, indicating that the material becomes more homogeneous as nozzle size is increased. Therefore the printing raster cannot be identified on a larger area of the cross-section on the samples printed with larger nozzle sizes. The brighter areas in the images indicate where the final tensile breaks appeared; these areas probably have lower Young’s modulus due to less homogenous material leading to that these parts held together until the end of the tensile test.

4 Conclusions In this paper, the effect of increasing the nozzle size in a FFF process was investigated. Samples were manufactured using 7 different nozzle sizes ranging from the standard 0.4 mm up to 2.0 mm. The results show that by increasing the nozzle size from 0.4 mm to 2 mm it is possible to increase the production rate by 80% and the strength of the printed parts by 50%. The increase in strength is supported with weight analysis and material characterization by means of CT and optical microscopy of the printed tensile bars, showing increased homogeneity and fewer defects in the printed material with an increasing nozzle size. However, in this study only one material was used, PLA. Other materials might react differently. The increased strength is probably mostly due to the better bonds between the layers and the individual extruded lines, but the temperature cycle might also have influenced the results, and other filament materials could react differently.

References 1. Gomez-Gras, G., Jerez-Mesa, R., Travieso-Rodriguez, J.A., Lluma-Fuentes, J.: Fatigue performance of fused filament fabrication PLA specimens. Mater. Des. 140, 278–285 (2018) 2. Zekavat, A.R., Jansson, A., Larsson, J., Pejryd, L.: Investigating the effect of fabrication temperature on mechanical properties of fused deposition modeling parts using X-ray computed tomography. Int. J. Adv. Manuf. Technol. 100(1–4) (2019) 3. Afonso, J.A., Alves, J.L., Caldas, G., Gouveia, B.P., Santana, L., Belinha, J.: Influence of 3D printing process parameters on the mechanical properties and mass of PLA parts and predictive models. Rapid Prototyp. J. 27(3), 487–495 (2021) 4. Frunzaverde, D., et al.: The influence of the printing temperature and the filament color on the dimensional accuracy, tensile strength, and friction performance of FFF-printed PLA specimens. Polymers (Basel) 14(10) (2022)

170

J. Larsson et al.

5. Butt, J., Bhaskar, R., Mohaghegh, V.: Investigating the effects of extrusion temperatures and material extrusion rates on FFF-printed thermoplastics. Int. J. Adv. Manuf. Technol. 117(9–10), 2679–2699 (2021) 6. Chaidas, D., Kechagias, J.D.: An investigation of PLA/W parts quality fabricated by FFF. Mater. Manuf. Process. 37(5), 582–590 (2022) 7. Buj-Corral, I., Sánchez-Casas, X., Luis-Pérez, C.J.: Analysis of am parameters on surface roughness obtained in PLA parts printed with FFF technology. Polymers (Basel) 13(14), 1–20 (2021) 8. Luis-Pérez, C.J., Buj-Corral, I., Sánchez-Casas, X.: Modeling of the influence of input am parameters on dimensional error and form errors in PLA parts printed with FFF technology. Polymers (Basel) 13(23) (2021) 9. Atakok, G., Kam, M., Koc, H.B.: Tensile, three-point bending and impact strength of 3D printed parts using PLA and recycled PLA filaments: a statistical investigation. J. Mater. Res. Technol. 18, 1542–1554 (2022) 10. Rajpurohit, S.R., Dave, H.K.: Flexural strength of fused filament fabricated (FFF) PLA parts on an open-source 3D printer. Adv. Manuf. 6(4), 430–441 (2018) 11. Czy˙zewski, P., Marciniak, D., Nowinka, B., Borowiak, M., Bieli´nski, M.: Influence of extruder’s nozzle diameter on the improvement of functional properties of 3D-printed PLA products. Polymers (Basel) 14(2) (2022) 12. Syrlybayev, D., Zharylkassyn, B., Seisekulova, A., Akhmetov, M., Perveen, A., Talamona, D.: Optimisation of strength properties of FDM printed parts—a critical review. Polymers (Basel) 13(10) (2021) 13. Rasband, W.S.: Image J. U. S. National Institutes of Health, Bethesda, Maryland, USA 14. Creality, Ender 3. https://www.creality.com/products/ender-3-3d-printer 15. Ultimaker, Ultimaker 2. https://ultimaker.com/3d-printers/ultimaker-2-plus-connect 16. I. O. for Standardization, ISO 527-2:2012 Plastics – Determination of tensile properties – Part 2: Test conditions for moulding and extrusion plastics (2012) 17. Ultimaker, Ultimaker Cura. https://ultimaker.com/software/ultimaker-cura 18. Kimya, “No Title.” https://www.kimya.fr/en/product/3d-filament-pla-r-kimya/ 19. Bruker, Bruker Skyscan 1272. https://www.bruker.com/en/products-and-solutions/micros copes/3d-x-ray-microscopes/skyscan-1272.html 20. U. of I. InstaRecon, Inc., InstaRecon. http://instarecon.com/ 21. V. Graphics, VG Studio MAX. https://www.volumegraphics.com/en/products/vgsm.html 22. Zeiss, Zeiss Discovery.V12. https://www.zeiss.com/microscopy/en/products/light-micros copes/stereo-and-zoom-microscopes/stereo-discovery-v12.html

Systematical Assessment of Automation Potential in Additive Manufacturing Process Chains Julian Ulrich Weber(B) , Hanna Jörß, and Mirco Jankowiak Fraunhofer Research Institution for Additive Manufacturing Technologies IAPT, Am Schleusengraben 14, 21029 Hamburg, Germany [email protected]

Abstract. Additive Manufacturing (AM) technologies are considered to be essential production processes of the future, as they enable novel products due to the high geometric design freedom at almost constant production costs. In terms of industrialization, the efficient integration of AM systems into lean production lines is essential. Up to 80% of the total lead time and costs of additively manufactured components are currently generated before or after the actual AM printing process, specifically within pre- and post-process steps. A common tool to reduce lead times and costs in production lines is the automation of manual process steps along the entire end-to-end process chain. Since the development of complex automation solutions is usually very costintensive, the application must be analyzed extensively before development. However, due to the high complexity and diversity of end-to-end AM process chains, the assessment of the automation potential for AM technologies is considered to be challenging. The goal of this work is to develop a methodology to systematically analyze the automation potential along the end-to-end AM process chain and identify high potentials of automation. For this, the common end-to-end process chains of Powder Bed Fusion (PBF-LB/M) and Directed Energy Deposition (DED-LB/M) were defined systematically. According to the end-to-end process chains, the mutual literature-based degree of automation was determined for each process step and its linkage. Finally, a value-benefit analysis of automation solutions was carried out for every process step, to highlight high and low automation potentials. Keywords: PBF-LB/M · DED-LB/M · Process chain analysis

1 Introduction Additive manufacturing (AM) technologies are comparably novel manufacturing technologies that can be used to generate complex geometries layer-wise. In contrast to subtractive manufacturing technologies, AM only generates solid material where it is needed, making AM a superior manufacturing technology to generate complex, lightweight, bionic metal geometries. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 171–186, 2024. https://doi.org/10.1007/978-3-031-42983-5_12

172

J. U. Weber et al.

For the generation of metal parts, a highly focused form of energy is combined with an additional material feedstock. Most widely used AM technologies for metals are Powder Bed Fusion (PBF-LB/M) and Directed Energy Deposition (DED-LB/M) processes, therefore this work focusses on the evaluation of these two processes, even if the characteristics the AM parts are different. PBF-LB/M is a type of AM process that utilizes a laser beam to selectively heat and fuse metallic powder particles together in a layer-by-layer fashion, creating threedimensional parts. The laser beam is directed onto the surface of a bed of powder and is programmed to trace a specific pattern to melt the powder and form the desired shape. Further post-processing steps may be necessary for the finished product. Due to its high precision, ability to form complex geometries, resource-efficiency, short lead times and cost-effectiveness, PBF-LB/M is also used in a wide range of production areas and industries [1]. DED-LB/M is a process that utilizes a laser beam and metallic powder feedstock to deposit metallic coatings or near net-shape structures onto substrate materials. This process enables the fabrication of intricate, functional structures with elevated material efficiency and high deposition speeds on existing components. DED-LB/M can be used to generate large metallic components that would otherwise have been conventionally manufactured by subtractive processes, resulting in up to 95% material waste. Structural components generated by DED-LB/M can be used in tooling, machinery, automotive, aerospace, rail and ship applications. Compared to PBF-LB/M, DED-LB/M has a much higher deposition rate and can generate structures on existing part components. The high deposition rates do however result in a rougher surface quality [2]. However, the market share of both manufacturing processes is still comparably low. With respect to the market comparison between metal AM in general and conventional CNC systems it is seen that the market volume of AM (0.99 billion e) is just a fraction compared to CNC machining systems (76.11 billion e) [1]. However, in relation of the growth rate, it shows that based on forecasts at the metal AM, it can reach 26.7% by 2026 [1]. This is 19.2% more than the growth rate of CNC-based machining systems (7.5%). According to recent research, the complexity of the end-to-end process chain of AM is high and diverse resulting in high part costs, high lead-time and low-quality standard [3]. The end-to-end process chain includes all necessary pre- and post-process steps, that are essential for the actual AM process to produce high quality part components. However, the implementation of an AM process chain into an established manufacturing process can be difficult to handle for companies and affects the whole supply chain and their individual participants [4]. Automation of certain pre- and post-process steps can be a suitable tool, to increase productivity, shorten production times, facilitate human work, reduce costs and increase quality [5]. Specific process step automation was shown in [6], where robotic part handling with reduced clamping efforts for AM parts were introduced. Especially for post processing steps, there are several commercially available solutions available, for example by companies like Additive Industries or Solukon. However, these solutions focus on the automation of sub-steps and not the entire process chain. To identify the potential of automation for each process step in PBF-LB/M and DED-LB/M, a methodical approach is developed in this work.

Systematical Assessment of Automation Potential

173

The development of the methodical approach is based on a value-benefit analysis. A value-benefit analysis is commonly used in the evaluation of complex action alternatives and project selection and is also suitable in this context, as it allows for the establishment of both quantitative and qualitative evaluation criteria and the systematic structuring thereof. Characteristic of the value-benefit-analysis is the establishment of a hierarchically structured target system, which divides the target criteria vertically into levels. The objectives of the lowest complexity level or the “leaves” in the target-systemtree represent the evaluation criteria [7]. To evaluate the importance of each evaluation criteria and part target, weights must be defined, typically node gk and level gs weights [7]. The level weights gS represent the weighting factors that are used for the valuebenefit-analysis. In any case, the determination of the weighting is accompanied by a large subjective evaluation margin and must therefore be determined individually for each new project [8]. According to Haberfellner a top-down approach is recommended, therefore the distribution of weights starts with the target subject and ends with the levels of objectives [9]. Value-benefit analyses were performed for general manufacturing processes with the focus on Robotic Process Automation (RPA), to identify the effect of RPA [10]. Khorasani, et al. show which automation technologies can be used to reduce manufacturing times for general manufacturing processes. It was shown that the use of automated robots or machines reduces the error rate [11]. A specific value-benefit analysis for AM processes to identify the automation potential for each process step is not known and therefore scope of this work.

2 Definition of End-to-End AM Process Chains 2.1 PBF-LB/M The end-to-end process chain of PBF-LB/M contains 19 process steps and is depicted in Fig. 1. The process chain starts with the 3D component design by aid of computer-aided design (CAD) software and the subsequent transfer into a second software to prepare the build job for the machine and to use the build space efficiently. In this process, the parts arrangement is determined, the support structures and also a set of material specific parameters are created. Before printing begins, the build platform is prepared by a computer numerical control (CNC) milling machine and a (sand) blasting treatment. The metal powder material is sieved and can be used for the generated build job. The last step of the pre-process is the machine preparation, during which the build platform is mounted and calibrated, so that the first powder layer can be applied. The in-process of PBF-LB/M contains activities of generation of an inert gas atmosphere and optional preheating of the build area before the start of the AM process. In the PBF-LB/M process, every new powder layer is applied and then fused with focused laser beams. After completion of the whole build job, the chamber must cool down and the unmelted powder has to be removed. After removing the platform from the build chamber, the process chamber can be cleaned [12, 13]. The aim of most post-process activities is to achieve the defined component requirements [14, 15]. After removing any remaining powder residues, a stress relieve treatment

174

J. U. Weber et al.

is performed to prevent possible cracks or distortion due to residual stresses. After that, the components are separated from the build platform by technologies such as a band saw or wire EDM [12, 16]. Support structures are then removed to obtain near-net-shape components [16]. Further post-processing steps vary depending on product requirements, including hot isostatic pressing, milling, grinding or blasting [14, 16]. Quality control is performed using optical scanners, coordinate measuring machines or computer tomography to check the surface properties or internal defects and pores. The current trend is to integrate quality assurance into the build process using sensors to monitor each individual layer application [12, 17]. After final quality assurance, the part will be transported to the internal assembly line or to the customer delivery centre. However, the actual assembly process step is not considered in this work. Pre-Process Milling of platform

Post-Process

In-Process Cleaning of process chamber

4

Part cleaning 12

11

Stress-relief annealing

Generation of component data (complete / error-free volume models)

5

Sieving powder

13 6

Unclamping of platform 10

Machine preparation

Removal of support structures

7

Powder removal

- Determining the component layout - Determining the support structures

Obligatory step

16

17

Surface finishing 18 2

1

15

Hot-isostatic pressing 9

Generation of build job data

Preparation of manufacturing data

Cutting of substrate platform

Final part

Blasting of platform

L-PBF process Final quality assurance

Inerting and tempering the process chamber Printing process Cooling down the process chamber 3

19

Transportation 8

20

Optional step

Fig. 1. Process chain of PBF-LB/M (in accordance to [12, 14]).

2.2 DED-LB/M The end-to-end process chain involves 17 process steps and is depicted in Fig. 2. The design of the DED-LB/M part component begins with the definition of all requirements and constraints [2]. After that, a 3D model of the part is generated by the aid of CAD. The powder which is prepared is filled into the powder conveying system, which generates a constant powder flow for the DED-LB/M process.

Systematical Assessment of Automation Potential

175

For the preparation of the machine, it is important to set all relevant process parameters such as laser power, powder feed rate and scanning speed [14]. The activity of machine adjustment is recommended before the production of every part, to ensure the zero points for fixturing and also the component alignment required for optimal manufacturing [18]. The process for the preparation of the basic part component may require pre-processing steps such as milling or grinding to ensure, that the surface has no presence of impurities or geometrical defects [19]. Subsequently, the actual generation of metallic structures starts with the DED-LB/M process, which is defined as in-process step within the end-to-end process chain. After that, activities of removal of excessive powder and unclamping of the manufactured workpiece starts the post processing steps of the process chain. Before beginning of surface finishing activities, the process chamber and also the part is cleaned regularly. To remove all residual stresses within the part, stress-relief annealing can be performed in a separate heating chamber system. To achieve the required part dimensions and also surface quality of the near-net-shape geometry, machining of the surface is performed either by milling or turning [20].

Pre-Process Milling of the basic component

Generation of component data (complete / error-free volume models)

Blasting of the basic component

Post-Process

In-Process

Part cleaning

Cleaning of printing area 4

5

12

11

Sieving powder

Thermal treatment 6

13

Machine preparation

10 7

Generation of build job data

Surface machining 14

Surface finishing 18

Powder removal

- Determining the component geometry - Creating the printing job

9

Final quality assurance 2

1

Preparation of manufacturing data

Obligatory step

Final part

Unclamping of platform

19

Transportation

DED-LB/M process 3

8

20

Optional step

Fig. 2. Process chain of DED-LB/M (in accordance to [2, 22]).

If higher surface quality requirements have to be met, subsequent finishing activities can be performed, such as sand blasting or polishing [14]. Finally, the part quality is assured to meet the required quality standards [2]. Bonding defects or pores are typical defect that occur in DED-LB/M processes. Most defects can be identified by non-destructive testing using x-ray or ultrasonic technologies, however testing coupons are usually generated next to the actual part component for destructive testing processes. These are required for a full quality assurance with high quality standards and may contain physical testing, such as tensile or hardness testing [21].

176

J. U. Weber et al.

After passing several intra-logistical activities (such as transportation to the final assembly lines or the delivery center), the DED-LB/M component will be prepared for customer delivery.

3 Methodical Approach In order to evaluate the automation potential of each step in PBF-LB/M and DEDLB/M processes, a systematized methodology was developed and applied to two process scenarios. The complexity and variety of pre-, in- and post-process steps necessitates the definition and application of a multi-criteria evaluation system. This methodology was formulated based on a value-benefit analysis. The following framework was defined for the evaluation of the automation potential: (1) Definition of process chains and process scenario, (2) Definition of the target system and its evaluation criteria, (3) Weighting of the target system and each evaluation criteria, (4) Data acquisition methodology, (5) Data normalization to consistent point scale and (6) Calculation of partial and total potentials. Each step is described in the following sub-chapters. 3.1 Definition of Process Chains and Process Scenarios Detailed process chains of PBF-LB/M and DED-LB/M were described in chapter 2, based on a literature review and analysis of both processes internally. Process scenarios are necessary to simplify the assessment of each evaluation criteria. The selection of the scenarios is based on typical AM build jobs and parts, that are known for the industry. For both scenarios, a tensile strength specimen is considered to be generated, because it accompanies almost every print job. The selected material for this scenario is Ti-6Al-4V and will be applied on a base plate. For PBF-LB/M, an EOS M290 system with a 400 W laser system was used while a Trumpf TruRobot 5020 with a TruDisk 6001 laser system (max. 6 kW laser power) was used for the DED-LB/M process. 3.2 Definition of the Target System and Evaluation Criteria To evaluate the potential of automation, beneficial factors are compared to effort factors. The beneficial and effort factors consist out of 2 factor groups each. Definition of Potential Benefit Factors Potential of time and cost saving (ZEP). Potential of time and cost saving is directly linked to main objectives of general automation. To evaluate the potential for this objective, three evaluation criteria were defined: the manual process time of each process step, the share of setup times and the frequency of each process step. It is assumed that in processes with a high manual processing time, there is a high probability of optimizing the production flow with the help of an automation solution. At the same time, the manual processing time is also an indicator for the production costs of a process, compare [15]. Thus, the share of manual processing times (ZEP1 )

Systematical Assessment of Automation Potential

177

is the first evaluation criteria for the indication of ZEP. A high ZEP1 indicates a high potential of time and cost saving. Secondly, the ZEP can be approached by identifying the setup times for each process step. Gronenberg et al. analysed the effect of automation to reduce non-value-adding activities in a process chain [23]. While analysing the process times of each process step, the setup time share (ZEP2 ) represents non-value adding share in the lead-time. Here, a high setup time indicates a high potential of time and cost saving. Thirdly, the frequency of each process step (ZEP3 ) is analysed in the AM process chains. Here, a process step that must be completed for every part design contributes less to the potential of time and cost saving compared to a process step, that must be completed for every generated part. Potential quality improvement (QA). Automation of complex processes allows the reduction of potential errors and has the potential to take over risky and for workers unhealthy activities. According to these objectives, two evaluation criteria are defined: the number of manual activities within a process step (QA1 ) and secondly the exposure of workforce to fine metallic powders (QA2 ). A high number of manual activities is considered to indicate a high potential for quality improvement and thus a high potential of automation. Due to work safety reasons, several work safety measures (such as additional safety equipment) have to be fulfilled, if the workforce is exposed to fine metallic powders. The additional safety equipment is assumed to have a negative impact on AM part quality, therefor a high exposure of powder to the workforce indicates a high potential for process automation. The second group of target criteria is the estimation of the effort required for automation (cost factors). By estimating the estimated effort for each automation solution and comparing to beneficial factors, the potential of automation (AP) is estimated. Definition of Criteria to Estimate Required Effort. Technical feasibility (TM). Relevant for the effort estimation of automation is the consideration of the technical feasibility of the implementation of an automation. Heinrich et al. highlight that the effort of automation grows more than the system effectiveness with increasing complexity of the technical structures [5]. Each process step is analysed regarding their process type. The process type describes to which parts a process is executed rule-based or decision-based [10]. For the evaluation of the process type, the number of decisions (TM 1 ) that must be made during the process step is counted. A decision is defined as the need to make a choice between two or more alternative actions. A high number of decisions to be made during the process leads to a lower suitability for automation [10]. This simplification is made as counting the decisions, the complexity of a process step can be determined and helps to evaluate the automation potential for individual situations in the considered process chain. In contrast, automation of process steps that are always executed in the same way usually require a low number of decisions and are therefore assumed to be automatized with less effort.

178

J. U. Weber et al.

State of development (EA). The current development status of automation solutions on the market provides information about the prioritization of the automation of process steps on the market and in AM companies. The number of automated (fully and partly) products (EA1 ) are assumed to indicate the state of development. Here, a high number of EA1 indicates a low effort to be necessary for the process step automation. Secondly, the investment costs (EA2 ) are reviewed for potential automation solutions of each process step. High EA2 is considered to have a negative effect on the necessary effort to automatize a process step and thus results in a lower potential of automation. 3.3 Weighting of the Evaluation Criteria Numbers between 0 and 1 were chosen as factors for weighting, the sum of all factors of the evaluation criteria equal 1. The level weights gS,n,k were distinguished by means of ranking procedure for every level n and evaluation criteria k. To objectify the ranking, the ranking procedure was performed independently by three participants. All weights gS,n,k are summarized in Fig. 3. 3.4 Data Acquisition To acquire the necessary data for every evaluation criterion, three methodologies were used: (1) Practical data acquisition by means of experiments, (2) Theoretical data acquisition by means of literature review and (3) Data acquisition by independent estimation of three technical personnel. All evaluation criteria, that were identified by multiple data acquisition methodologies, an equally weighted mean value was calculated. Table 1 shows the utilization of data acquisition methodologies for each evaluation criterion. Table 1. Overview of data acquisition methods for each evaluation criterion. Evaluation criteria

Practical

Theoretical

Manual processing time ZEP1

x

x

Setup time share ZEP2

x

x

Frequency of each process step ZEP3

x

x

Number of manual activities QA1

x

x

Exposure to powder material QA2 Number of process decisions TM 1

Estimation

x x

x

Number of automated (fully and partly) products EA1

x

Investment cost of EA2

x

Systematical Assessment of Automation Potential

179

3.5 Data Normalization to Consistent Point Scale In order to ensure comparability between the criteria, a uniform scaling (Eq. 1) of the evaluation criteria values zk was implemented using a scale of 0 to 10, where 0 indicates no fulfilment and 10 indicates an extraordinary fulfilment: zk =

yi ymax

· 10

(1)

Evaluation criteria with a low amount of measured data yi had categories defined on the point scale. Table 2 summarizes all evaluation criteria and its corresponding point values zk .

EA

TM

QA

ZEP

Table 2. Overview of partial potentials and its evaluation criteria and scoring values. Evaluation criterion Manual processing time ZEP1 Setup time share ZEP2 Frequency of each process step ZEP3 Number of manual activities QA1 Exposure of workforce to metallic powders QA2 Number of process decisions TM1 Number of automated (fully and partly) products EA1 Investment cost of EA2

Definition and transformation to point value scale (zk) Manual processing time [min], normalized to point scale. Process and setup time shares, normalized to point scale. 0 – Process step as to be finished once for every part design 5 – Process step as to be finished for every build job 10 – Process step as to be finished for every build part Count of manual operations, normalized to point scale. 0 – Powder contact excluded 5 – Powder contact minimized by protective equipment 10 – Powder contact unavoidable 0 – Numerous decisions distributed in process chain 3 – Decisions during the programming of a system 7 – Few decisions necessary to control components 10 – No decisions necessary Research of the number of product solutions Partly automated solutions count as half solution, normalized to point scale. 0 – Price > 500.000 € 5 – 150.000 to 500.000 € 10 – Price < 150.000 €

3.6 Calculation of Partial and Total Potentials The main objective is to evaluate the potential of automation AP for each process step i of PBF-LB/M and DED-LB/M processes, defined in layer 1 of the evaluation methodology. The composition of the potential of automation AP is defined by the two partial potentials in layer 2: Potential of efficiency increase ES and the estimated effort AA. Both partial potentials weigh equally with a level weight of gs, ES = gs, AA = 0.5. ES is calculated by two further part potentials in layer 3: the potential in time and cost saving ZEP and the potential in quality increase QA with the corresponding weights

180

J. U. Weber et al.

gs,ZEP = 0.3 and gs,QA = 0.2. AA is described by the technical feasibility TM and the state of development EA with gs,TM = 0.2 and gs,EA = 0.3. The final layer 4 consists of all evaluation criteria that describe the part potentials in layer 3. ZEP is calculated by point evaluation for ZEP1 (with gs, ZEP1 = 0.15), ZEP2 (with gs, ZEP2 = 0.05) and ZEP3 (with gs, ZEP3 = 0.1), QA of QA1 (with gs, QA1 = 0.1) and QA2 (with gs, QA2 = 0.1), TM of TM1 (with gs, TM1 = 0.2) and finally EA of EA1 (with gs, EA1 = 0.15) and EA2 (with gs, EA2 = 0.15). All point values zk were multiplied by the corresponding weights and summed up to calculate the (partial) potential of the layer above. The definition of all evaluation criteria is described in Table 2 above, the methodology is summarized in Fig. 3. All (partial) potentials were calculated relatively to all process steps within the AM process chain. Therefore, all potentials are calculated in %, delivering the potential between 0 and 100%. A high percentage correlates to high potential for target improvement, also meaning that a high percentage in estimated effort AA describes a low absolute effort, that has to be performed to automate the process step. Layer 1

Layer 2

Layer 3

Layer 4

Potential time

ZEP1

and cost saving

+

ZEP

gs, ZEP2 = 0.05

Potential

Value z1 … zk on point scale [0…10]

+

efficiency increase

ES

ZEP2

Process step 1…i

ZEP3

+ Potential quality

QA1

increase Potential of

QA

automation

AP

+ QA2

+

AP for each

Technical feasibility

process step i

TM AA Estimated effort

gs,TM1 = 0.2

TM1

+ EA1 EA

+

State of development

EA2

Fig. 3. Schematic representation of the systematic evaluation of the automation potential.

4 Results For both AM process chains, the evaluation criteria and scoring values zk were determined, to calculate the partial potentials of layer 2 and 3 and finally the potential of automation AP for each process step i. The number of each process step is to understand in regard to its description, summarized in Figs. 1 and 2.

Systematical Assessment of Automation Potential

181

4.1 Process Step Evaluation of PBF-LB/M For PBF-LB/M, 19 process steps were identified within the process chain. Table 3 presents the findings of the value-benefit analysis conducted on PBF-LB/M, which is classified into partial potentials and total potential. Table 3. Partial potentials and full potential of each process step in PBF-LB/M. Evaluation Process step i in PBF-LB/M criteria 1 2 3 4 5 6 7

8

9

10 11 12 13 14 15 16 17 18 19 20

ZEP [%]

40 52 22 55 43 17 43 17 43 33 43 40 17 -

35 83 17 22 50 38

QA [%]

15 5

5

ES [%]

30 33 15 35 30 30 50 20 42 40 46 40 10 -

23 70 10 13 32 25

TM [%]

0

30 70 99 99 70 99

EA [%]

65 65 60 50 85 85 30 15 90 50 0

AA [%]

39 39 36 42 91 91 46 49 94 70 40 88 70 -

42 52 40 76 76 79

AP [%]

35 36 26 39 60 60 48 35 68 55 43 64 40 -

33 61 25 45 54 52

0

5 0

5

10 50 60 25 40 50 50 40 0

-

30 99 99 70 99 99 99 99 70 99 99 50 -

50 0

50 40 0

0

5

5

60 80 65

Among the process steps, the highest automation potential is observed in the process step 9 (powder removal), with an automation potential AP of 68%. This is followed by process step 12 (cleaning the parts), which has a total potential of 64%. However, both process steps offer only a medium potential for improving the process’s efficiency, with ES = 42% and 40%, respectively. The process step with the highest potential for improving efficiency ES is the removal of support structures, with 70%. This is because the potential for time and cost savings ZEP in this process step is higher than in any other process step (ZEP = 83%). Removing the support structures is a manual, very time-consuming process step in which each component must be handled individually. With the average partial potential of AA = 52%, resulting from a predominantly rule-based process type, but only a moderately developed level of automation on the market, the automation potential AP results of 61%, yielding the third best automation potential. The process steps 3 (preparation of manufacturing data) and 16 (hot isostatic pressing) reach the lowest automation potential AP, with a result of 25.5% and 25%. Due to very low potential in quality improvement and a small amount of manual labor, a low ES is reached for both process steps. 4.2 Process Step Evaluation DED-LB/M The process chain of DED-LB/M consists of 17 process steps. The evaluation of all process steps for DED-LB/M is summarized in Table 4. By applying the developed systematical methodology, process steps 5, 6 and 15 show the highest potential for automation. Process step 5 describes the pre-process blasting of the substrate material and yields a potential of automation AP of 62%. Step 6 the

182

J. U. Weber et al. Table 4. Partial potentials and full potential of each process step in DED-LB/M.

Evaluation Process step i in DED-LB/M criteria 1 2 3 4 5 6 7 8

9

10 11 12 13 14 15 16 17 18 19 20

ZEP [%]

52 8

13 54 49 30 63 41 45 55 45 46 54 56 -

-

-

48 53 46

QA [%]

50 1

25 16 31 54 63 1

19 -

-

-

31 18 8

ES [%]

51 6

18 39 42 39 63 25 50 45 38 30 34 41 -

-

-

42 39 31

TM [%]

0

0

-

-

-

99 30 99

EA [%]

65 60 60 65 70 95 35 50 65 50 25 65 90 65 -

-

-

70 55 60

AA [%]

39 36 36 51 82 97 33 70 67 58 43 67 82 39 -

-

-

82 45 76

AP [%]

45 21 27 45 62 68 48 48 58 51 40 49 58 40 -

-

-

62 42 53

0

58 30 28 8

4

30 99 99 30 99 70 70 70 70 70 0

pre-process sieving of the powder material and yields a final AP of 68% while process step 15 describes the post-process blasting of the generated AM part, to meet specific surface finishing requirements, and yields an automation potential AP of 62%. The highest potential AP is estimated for the powder sieving, mostly because the estimated effort AA is comparably low (AA = 97%) while the potential of increasing the efficiency ES is considered to be moderate (ES = 39%). AA is assumed to be very high due to many fully automated powder processing and treatment solutions that are already available. Focusing on the potential increase of efficiency ES, the automation of machine preparation is expected to have the highest impact on process efficiency, due to many complex and manual process activities within this process step. The lowest potential of automation AP is reached in process steps 2 and 3, describing both data preparation process steps and consist of the generation of build job files and the generation of manufacturing data files. Process step 2 yields an automation potential AP of 21%, process step 3 yields AP = 27%. Step 2 is therefore only necessary for every print job, resulting in a lower process step frequency and therefore lowering the potential of improving efficiency ES. Further, build jobs of DED-LB/M processes often consist of only one build part, concluding in a low lead-time of the process step. The focused DED-LB/M process scenario also consists of only one build part per build job. Focusing on the potential increase of efficiency ES, again process step 2 is expected to have the lowest potential. Step 3 (preparation of manufacturing data) and process step 8 (DED-LB/M process) reach the second and third-least ES, with ES = 18% for step 3 and 25% for step 8. Step 8 yields a comparably low partial potential ES, because the process step consists of mostly automated activities. Additionally, many fully and partly automated solutions are accessible on the market, leading to a low potential of automation AP. 4.3 Comparison of PBF-LB/M and DED-LB/M Process Step Evaluation It is shown that end-to-end process chain of PBF-LB/M consists of 19 process steps, while the process chain of DED-LB/M consists of 17 process steps. This implies a more complex end-to-end process chain for PBF-LB/M, even though the industrial market

Systematical Assessment of Automation Potential

183

share of PBF-LB/M is higher compared to DED-LB/M. Therefore, the difference in the industrial market share of the AM technologies does not only rely on the number of the end-to-end process steps and considerably the process chain complexity. Here, the complexity of the in-process activities (deposition of energy, powder, …) and the AM process stability might be highly influential for the degree of industrialization. To identify high and low potentials for the automation of each process step in PBFLB/M and DED-LB/M, a portfolio matrix was developed, plotting the estimated effort AA over the potential increase of efficiency ES for every process step in PBF-LB/M (Fig. 4, left) and DED-LB/M (Fig. 4, right). very high priority

low

moderate priority

high priority

L-PBF

100%

80%

LMD 6

12 20

19

13

10

8

60%

40%

15

17 3

9 10

16

4

7 12

4

40%

11

2

3

19 11 14

1 7

20%

0%

0% 0%

low

12

60% 18 8

20%

high

13 18 5

80%

20

18

very low priority

low priority

100%

9 56

Estimated effort AA [%]

# Process step L-PBF LMD 1 Generation of component data X X 2 Generation of build job data X X Preparation of manufacturing 3 X X data 4 Milling of substrate material X X 5 Blasting of substrate material X X 6 Sieving of powder X X 7 Machine preparation X X 8 AM process X X 9 Powder removal X X 10 Unclamping of substrate material X X 11 Cleaning of process chamber X X 12 Part cleaning X X 13 Stress-relief annealing X X 14 Surface machining X 15 Cutting of substrate platform X 16 Removal of support structures X 17 Hot-isostatic pressing X 18 Surface finishing X X 19 Final quality assurance X X 20 Transport X X

20%

40%

60%

80%

Potential of efficieny increase ES [%]

100%

0%

high low

20%

40%

60%

80%

Potential of efficieny increase ES [%]

100%

high

Fig. 4. Considered process steps and portfolio matrices for the identification of priorities for each process step of PBF-LB/M (left) and DED-LB/M (right).

For both process chains, specific process steps can be identified with higher priority for automation: for PBF-LB/M process step 16 (support removal) and 9 (powder removal) and for DED-LB/M, process step 9 (machine preparation) show the highest priority for automation. Medium and low priorities can be identified for most process steps, mainly due to a moderate potential in efficiency increase ES and estimated effort AA. The lowest priority in both process chains is identified in data preparation activities, step 2 and respectively step 3. Close to high priority was evaluated for the powder removal in both processes. However, for DED-LB/M, this process step is only required before material change or high powder agglomeration in the process chamber, which was both not considered in evaluation. Generally, post processing steps show a higher priority for automation. Further, it can be emphasized that automation of machine preparation is expected to have a high impact on the efficiency increase ES, but is considered to be a complex process step for automation. Generally, the process chain of PBF-LB/M shows a higher potential for automation, mainly due to a high amount of manual process steps for each part. Here, the removal of support structures shows the highest priority to be automated.

184

J. U. Weber et al.

4.4 Limitations The presented methodology for evaluating the automation potential of two AM process chains is based on a value-benefit analysis and the selection of two explicit process scenarios, including the selection of part geometries. Thus, the results of this methodology is limited due to its subjectivity in weighting of the evaluation criteria and the scoring of the qualitative criteria. Despite the attempt to reduce the subjective influence by using as many quantitative evaluation criteria as possible, complete objectivity of the evaluation remains unattainable. With access to more quantitative data, additional evaluation criteria could be considered that more accurately reflect the fulfilment of the target criteria. Additionally, the evaluation with quantitative values are normalized to max. Values, therefore the calculated potentials are relative values and only considerable within this work. Furthermore, this evaluation reflects the potential of automation for only one distinct process scenario. A general statement of automation potential for every process chain is not possible within this work. In production processes, the automation of process steps is often dependent on certain batch and lot-sizes as well as part costs. These factors could not be considered within this study.

5 Conclusions and Outlook This work investigated high and low potentials for automation in AM processes, particularly within the PBF-LB/M and DED-LB/M process chain. For this, a methodology was developed to evaluate the automation potential for each process step within the process chains. The developed methodology consists of 6 elementary steps that are building the framework for the evaluation of the automation. The automation potential is calculated by the evaluation of potential increase in efficiency compared to the estimated effort for the automation solution. Both partial potentials consist of further partial potentials and corresponding evaluation criteria. It was shown the complexity of both process chains differs: PBF-LB/M consists of 19 process steps while DED-LB/M consists of 17 process steps. The evaluation of the automation potential indicates high potentials especially for post-processing steps, most certainly for the removal of support structures and for the powder removal, especially for the PBF-LB/M process chain. However, both AM process chains indicate a high impact on efficiency increase by automation setup process of the AM systems. The assessment is based on literature review, experimental data acquisition and assumption. To review and improve the accuracy of the methodology, further studies including a practical implementation of automation solutions are necessary, also to validate the determined potentials. Furthermore, the optimization and application to further AM processes represents promising future field of research and is expected to accelerate the development of automation solutions.

Systematical Assessment of Automation Potential

185

References 1. Munsch, M., Schmidt-Lehr, M., Wycisk, E., Führer, T.: Additive Manufacturing Market Report, Hamburg, März (2022) 2. Möller, M.: Prozessmanagement für das Laser-Pulver-Auftragschweißen. Springer Vieweg, Berlin (2021) 3. Vafadar, A., Guzzomi, F., Rassau, A., Hayward, K.: Advances in metal additive manufacturing: a review of common processes, industrial applications, and current challenges. Appl. Sci. 11(3), 1213 (2021). https://doi.org/10.3390/app11031213 4. Holmström, J., Partanen, J.: Digital manufacturing-driven transformations of service supply chains for complex products. Supply Chain Manag. Int. J. 19(4), 421–430 (2014). https://doi. org/10.1108/SCM-10-2013-0387 5. Heinrich, B.: Grundlagen Automatisierung: Sensorik, Regelung, Steuerung, 2nd edn. Springer Vieweg, Wiesbaden (2017) 6. Ferchow, J., Kälin, D., Englberger, G., Schlüssel, M., Klahn, C., Meboldt, M.: Design and validation of integrated clamping interfaces for post-processing and robotic handling in additive manufacturing. Int. J. Adv. Manuf. Technol. 118(11–12), 3761–3787 (2022). https://doi. org/10.1007/s00170-021-08065-4 7. Wartzack, S.: Auswahl- und Bewertungsmethoden. In: Bender, B., Gericke, K. (eds.) Pahl/Beitz Konstruktionslehre. Springer Vieweg, Berlin, Heidelberg (2021). https://doi.org/ 10.1007/978-3-662-57303-7_11 8. Haag, C., Schuh, G., Kreysa, J., Schmelter, K.: Technologiebewertung. In: Schuh, G., Klappert, S. (eds) Technologiemanagement. VDI-Buch(). Springer, Berlin, Heidelberg (2010). https://doi.org/10.1007/978-3-642-12530-0_11 9. Haberfellner, R.: Systems Engineering: Grundlagen und Anwendung, 14. Aufl. Zürich: Orell Füssli Verlag (2018). Verfügbar unter. https://ebookcentral.proquest.com/lib/kxp/detail.act ion?docID=6795525 10. Plattfaut, R., Koch, J.F., Trampler, M., Coners, A.: PEPA: Entwicklung eines Scoring-Modells zur Priorisierung von Prozessen für eine Automatisierung. HMD 57(6), 1111–1129 (2020). https://doi.org/10.1365/s40702-020-00670-3 11. Khorasani, A., Gibson, I., Veetil, J.K., Ghasemi, A.H.: A review of technological improvements in laser-based powder bed fusion of metal printers. Int. J. Adv. Manuf. Technol. 108(1–2), 191–209 (2020). https://doi.org/10.1007/s00170-020-05361-3 12. Möhrle, M.: Gestaltung Von Fabrikstrukturen Für Die Additive Fertigung. Springer, Heidelberg (2018). Verfügbar unter. https://ebookcentral.proquest.com/lib/kxp/detail.action? docID=5445999 13. Yang, L., Wiener, S., Medina, F., Menon, M., Baughman, B.: Additive Manufacturing of Metals: The Technology, Materials, Design and Production. Springer, Cham (2017) 14. Lachmayer, R., Lippert, R.B.: Entwicklungsmethodik für die Additive Fertigung, 1. Aufl. Springer, Heidelberg (2020). Verfügbar unter. http://nbn-resolving.org/urn:nbn:de:bsz:31-epf licht-1728383 15. Gebhardt, A.: Additive Fertigungsverfahren: Additive Manufacturing und 3D-Drucken für Prototyping - Tooling - Produktion, 5. Aufl. Hanser; Ciando, München (2016). Verfügbar unter. http://ebooks.ciando.com/book/index.cfm/bok_id/2203967 16. Wohlers, T., Campbell, R.I., Diegel, O., Kowen, J., Mostow, N., Fidan, I.: Wohlers report 2022: 3D printing and additive manufacturing: global state of the industry. Wohlers Associates, Fort Collins, Colo., Washington, DC, ASTM International (2022) 17. Colosimo, B.M., Huang, Q., Dasgupta, T., Tsung, F.: Opportunities and challenges of quality engineering for additive manufacturing. J. Q. Technol. 50(3), 233–252 (2018). https://doi. org/10.1080/00224065.2018.1487726

186

J. U. Weber et al.

18. Elser, A., Königs, M., Verl, A., Servos, M.: On achieving accuracy and efficiency in additive manufacturing: requirements on a hybrid CAM system. Proc. CIRP 72, 1512–1517 (2018). https://doi.org/10.1016/j.procir.2018.03.265 19. Nellian, A.S., Pang, J.H.L.: Laser metal deposition characterization study of metal additive manufacturing repair of rail steel specimens. Virtual Phys. Prototyping, 18(1), e2134042 (2023). https://doi.org/10.1080/17452759.2022.2134042 20. Lachmayer, R., Rettschlag, K., Kaierle, S.Hg.: Konstruktion für die Additive Fertigung 2019, 1. Aufl. Springer, Heidelberg (2020). Verfügbar unter. http://nbn-resolving.org/urn:nbn:de: bsz:31-epflicht-1675877 21. Chen, Z., Han, C., Gao, M., Kandukuri, S.Y., Zhou, K.: A review on qualification and certification for metal additive manufacturing. Virtual Phys. Prototyping 17(2), 382–405 (2022). https://doi.org/10.1080/17452759.2021.2018938 22. Ehmsen, S., Yi, L., Aurich, J.C.: Process chain analysis of directed energy deposition: energy flows and their influencing factors. Proc. CIRP 98, 607–612 (2021). https://doi.org/10.1016/ j.procir.2021.01.162 23. Groneberg, H., Horstkotte, R., Pruemmer, M., Bergs, T., Döpper, F.: Concept for the reduction of non-value-adding operations in Laser Powder Bed Fusion (L-PBF). Proc. CIRP 107, 344– 349 (2022). https://doi.org/10.1016/j.procir.2022.04.056

Emerging AM Technologies

Numerical Modeling of Part Formation in Volumetric Additive Manufacturing Roozbeh Salajeghe1(B) , Daniel Helmuth Meile1 , Carl Sander Kruse1 , Deepak Marla2 , and Jon Spangenberg1 1

2

Department of Civil and Mechanical Engineering, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark [email protected] Department of Mechanical Engineering, India Institute of Technology Bombay, Mumbai, India

Abstract. Volumetric additive manufacturing (VAM) is a promising manufacturing method that enables the printing of high-accuracy structures in a short time. However, it faces difficulty with printing fine structures, highlighting the need for better process understanding. In this study, we propose a numerical method based on the energy threshold model to predict the shape of printed parts in VAM. We have combined a solver developed in OpenFOAM with an in-house code to interpolate images on the computational domain, which provides accurate VAM predictions. The framework has been used to study the development of the energy within the resin.

Keywords: Volumetric additive manufacturing model · Photopolymerization modeling

1

· Energy threshold

Introduction

Volumetric additive manufacturing (VAM) is a new additive manufacturing (AM) method capable of printing the whole part simultaneously, instead of the layer by layer approach adopted by the conventional methods. VAM method, shown in Fig. 1, presents faster print time, better surface quality, no requirement for overhang structures, higher isotropy of mechanical properties in different directions, and less demanding post-processing [1,2]. This method utilizes computed tomography (CT) concepts to produce intensity-modulated images from an input geometry file. These images are then projected onto a photocurable resin from various angles, and through accumulation, they create an optical dose distribution that selectively cures specific sections of the resin to achieve the desired shape of the input geometry. The photo-curable resin is composed of monomers, oligomers, and photo-initiators. When exposed to UV light with a specific wavelength, the monomers and oligomers attach to each other, forming crosslinked polymers, which causes the liquid photopolymer to solidify. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2024  C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 189–197, 2024. https://doi.org/10.1007/978-3-031-42983-5_13

190

R. Salajeghe et al.

Fig. 1. Schematic of VAM technique

The set of reactions that describes the photopolymerization process has been well described in the literature [3–6]. VAM was first conceptualized and applied by Kelly et al. [7]. They further improved the efficacy of the method by carefully characterizing the properties of the photo-curable resins and looking into some parameters that can affect the speed and fidelity of the model [1]. By enhancing the optics of the setup, Loterie et al. [2,8] improved the precision of the model and demonstrated its potential for clinical applications. Furthermore, they utilized an open-loop feedback method to showcase the potential of the VAM technique in improving fidelity and accuracy [2]. Rackson et al. [9] introduced a uniform illumination step at the end of the projection period and just before the part’s solidification to shorten the print period, remove the non-desired surface effects, and improve the print quality. In an effort to create structures with a spatial gradient in their mechanical properties, a method was proposed by Wang et al. [10], in which they utilized two different resins each of which was sensitive to light with different wavelengths. Cook et al. [11] applied VAM on thiol-ene resins to make the method capable of producing materials with diverse mechanical properties. The applicability of VAM has been expanded by Kollep et al. [12] and Toombs et al. [13], who utilized this method to produce ceramics and glass, respectively. Madrid-Wolff et al. [14] compensated for the light distortions in their intensity-generating algorithm to further broaden the applicability of the method to scattering resins. Further improvements to the intensity-generation algorithm have been discussed by Bhattacharya et al. [15], and Rackson et al. [16]. VAM has been employed for various applications, notable among them is the work by Bernal et al. [17]. They utilized VAM to create cell-laden structures from hydrogel resin, and reported a high cell viability.

Numerical Modeling of Part Formation

191

Many researchers have strived to model the photopolymerization process to provide a means of predicting and optimizing the results of the vat additive manufacturing processes. The first model was introduced by Jakobs [18], in which he evaluated the actinic energy development in a resin that is exposed to a Gaussian beam following a linear path. Further, he defined a threshold for actinic energy above which the resin is cured and below which the resin remains in its liquid form. The portion of the resin that has reached this value is at the so-called gelation point. This model, also known as the energy threshold (ET) model, is widely used in the industry due to its simplicity and valid predictions. Other models, such as those proposed by Goodner et al. [3,5], Jariwala et al. [4], and Tang et al. [6], involve solving the reactions and are generally more complex, but do not necessarily result in significantly improved accuracy of predictions. For instance, although Jariwala’s model [4] provides more information about the printed part than ET predictions, it was less accurate than the ET model in predicting the cured height for some exposure values. Although there have been studies investigating the sedimentation of the printed part in VAM [19,20], no attempts have been made to model the printing process itself. In this study, we have modified the ET model and implemented it in OpenFOAM to make it capable of simulating the VAM process. The methods of this numerical model and the results are presented in the following sections.

2

Methodology

The ET method, developed by Jakobs [18], represents the laser beam as a Gaussian beam and integrates the actinic irradiance field with respect to time to obtain the actinic energy within the domain as,  E = I dt (1) in which E is the energy, I is the irradiance, also called intensity field herein, and t stands for time. In VAM application, however, fluid is not stationary and the energy that is generated by the irradiance is transported by the flow. Accordingly, Eq. 1 can be modified to take into account the transport of the energy by the flow, ∂E  E) = I + ∇ · (U ∂t

(2)

 is the velocity vector field within the domain. The velocity field in which U can be obtained by solving the Navier-Stokes equations subject to the rotating wall boundary condition. For this simple case, however, using the cylindrical coordinates and ignoring the part’s sedimentation, it is straightforward to show that only the angular velocity is non-zero which rises linearly from zero at the center line to the maximum value at the wall according to Uθ = rω

(3)

192

R. Salajeghe et al.

where θ and r are the angular and radial coordinates, respectively, and ω is the rotational velocity of the cylinder. It is important to note that when the printed part starts to sediment, the flow cannot be considered fully rotational anymore. While the effect of part sedimentation on the flow behavior within the VAM printing cylinder has been explored previously by Salajeghe et al. [19,20], the current study focuses on the printing process without the sedimentation. The above equations are implemented in OpenFOAM, which is an opensource package that uses the finite volume approach. The computational domain is considered to be a cylinder with a diameter of 3 cm and a height of 1.2 cm, part of which is irradiated by the projector. A hex-dominant mesh is used to discretize the domain, and the mesh is refined in the region of light transfer. The discretized computational domain is shown in Fig. 2a. Simulating the VAM process presents a challenge in defining the irradiance or intensity field, as the shape of the intensity profile changes over time unlike SLA, which uses a constant beam. Several algorithms have been proposed to generate intensity-modulated images for use in VAM, including those put forward by Rackson [16] and Kelly et al. [1]. In this study, the algorithm proposed by Kelly et al. [1] was utilized and a C++ code was developed to read and scale the generated images to a maximum intensity value of 2 mW/cm2 . The selection of this maximum intensity value falls within the range used by Kelly et al. [1] (0.1–2 mW/cm2 ). Using the Beer-Lambert law, the intensity field is calculated based on the scaled intensity on the surface of the cylinder and the position of each point within the domain, as shown in −δx

I = I0 e Dp

(4)

where I is the intensity field, I0 is the irradiance on the cylinder surface, Dp is the penetration depth, and δx is the distance that the beam has traveled within the resin in the x-direction. Afterwards, the intensity field is interpolated on the mesh using a trilinear interpolation scheme. The projection frame rate is then calculated based on the rotation rate of the cylindrical container, which in turn determines the number of images that should be interpolated on the mesh within each simulation second. Figure 2b shows the interpolated image on the surface of the cylinder at the initial time, while Fig. 2c illustrates a cut within the domain to provide a better representation of light propagation.

3

Results

To model the energy distribution in the VAM process and the resulting shape of the printed part, a 24-second simulation was conducted for a cylinder rotating at 15 ◦ /s. Figure 3 illustrates the energy field in a y-plane that passes through the center of the cylinder, showing how energy accumulates within the domain to create the input geometry. Sub-figure (a) displays the energy field at t = 2 s, where the energy within the in-part region is not distinct from other parts of the domain, and the desired geometry is not evident. At t = 4 s, sub-figure (b) shows the initial traces of the geometry, with more energy accumulated within the part.

Numerical Modeling of Part Formation

193

Fig. 2. (a) discretized computational domain, (b) intensity-modulated image interpolated on the surface of the cylinder, (c) clipped domain showing the intensity field interpolated within the domain obeying the Beer-Lambert law

Sub-figure (c) presents the energy field at t = 6 s, displaying a higher contrast between the in-part and out-of-part regions, but with non-uniform energy distribution in the bear; the legs are pale compared to the body, indicating lower energy accumulation within the legs. Sub-figure (d) shows the energy field at t = 8 s, where the bear shape has a better contrast and uniformity. The quality of the resultant shape improves with time as the energy from different images accumulates. In Fig. 4, the simulation results are compared to the input STL file. Sub-figure mJ (b) displays a contour of E = 5.4 cm 2 from the simulation, which effectively captures the main features of the input geometry in sub-figure (a). While most positive features are accurately represented in the simulation results, smaller negative features, such as the holes in the ears, eyes, and nose, are not well captured due to the limitations of the VAM method used to generate the image set. This issue is also reported by Loterie et al. [2] who obtained smaller positive fea-

194

R. Salajeghe et al.

Fig. 3. Energy field within the domain at (a) t = 2 s, (b) t = 4 s, (c) t = 6 s, (d) t = 8 s.

tures (80 µm) compared to negative features (500 µm). Additionally, the details of the claws are not well preserved in the results, which may be due to either the image-generation algorithm or insufficient mesh resolution in that area. A more refined and geometry-tailored mesh could potentially improve the simulation’s capability to capture smaller features. However, due to the large number of pixels in the input images (approximately 1 million pixels), it is not feasible to conduct a grid independence study as it would require an enormous number of cells to match the resolution of the images. Therefore, a mesh with 6.2 million cells is used in the current study, as discussed in the Methods section. mJ It should be mentioned that the contour value of 5.4 cm 2 is set to better correspond to the desired geometry. As no specific resin material has been considered for the current study, the simulation has been continued for an arbitrary value of one single rotation. For a specified material, the solidified shape is the portion of the resin that has reached the energy threshold of conversion, and the simulation may at least continue up to the time when the energy in a portion of the domain has reached the energy threshold. As a result, if the energy threshold mJ of a given resin is higher or lower than the value of 5.4 cm 2 used in the current study, it will take more or less time for the energy to reach the desired level. However, this variation does not affect the validity of the model. The presented model can be used not only to understand the VAM technique and optimize its parameters, but also to assess the effectiveness of the imagemodulation algorithm. However, it should be noted that this model does not account for some factors such as part sedimentation, light refraction, and thermal

Numerical Modeling of Part Formation

195

Fig. 4. Comparison between the input geometry (a) and the contour of E = 5.4 mJ/cm2 of the simulation (b).

196

R. Salajeghe et al.

effects, which can adversely affect the quality of the print. Therefore, this model can predict the best possible outcome of the VAM technique.

4

Conclusion

A numerical framework is proposed in this study, which predicts the outcome of the volumetric additive manufacturing setup based on the concepts of the energy threshold model. After generating the intensity-modulated images using an algorithm presented in the literature, an in-house code is developed to calculate the intensity field within the domain and interpolate it on a mesh at different times. Additionally, the energy threshold theory has been modified to tailor it for the VAM application. The results suggest that a good estimation of the printed part can be provided by the numerical model. The slight non-uniformity of the energy field within the part for the current algorithm suggests that some parts of the geometry appear sooner than the others, which might adversely affect its resolution by changing the refractive index or inducing sedimentation. In the future, the framework developed here will be employed to study the different parameters that affect the print quality.

References 1. Kelly, B.E., Bhattacharya, I., Heidari, H., Shusteff, M., Spadaccini, C.M., Taylor, H.K.: Volumetric additive manufacturing via tomographic reconstruction. Science 363(6431), 1075–1079 (2019) 2. Loterie, D., Delrot, P., Moser, C.: High-resolution tomographic volumetric additive manufacturing. Nat. Commun. 11(1), 852 (2020) 3. Goodner, M.D., Bowman, C.N.: Modeling and experimental investigation of light intensity and initiator effects on solvent-free photopolymerizations. In: Solvent-Free Polymerizations and Processes. ACS Publications, vol. 713, pp. 220–231 (1998) 4. Jariwala, A.S., et al.: Modeling effects of oxygen inhibition in mask-based stereolithography. Rapid Prototyping J. 17(3), 168–175 (2011) 5. Goodner, M.D., Bowman, C.N.: Development of a comprehensive free radical photopolymerization model incorporating heat and mass transfer effects in thick films. Chem. Eng. Sci. 57(5), 887–900 (2002) 6. Tang, Y.: Stereolithography cure process modeling. Ph.D. dissertation, Georgia Institute of Technology (2005) 7. Kelly, B., Bhattacharya, I., Shusteff, M., Panas, R.M., Taylor, H.K., Spadaccini, C.M.: Computed axial lithography (CAL): toward single step 3D printing of arbitrary geometries. arXiv preprint arXiv:1705.05893 (2017) 8. Loterie, D., Delrot, P., Moser, C.: Volumetric 3D printing of elastomers by tomographic back-projection, vol. 2(20027), p. 46889 (2018). Preprint at https://doi. org/10.13140/RG.2.2.20027.46889 9. Rackson, C.M., et al.: Latent image volumetric additive manufacturing. Opt. Lett. 47(5), 1279–1282 (2022) 10. Wang, B., et al.: Stiffness control in dual color tomographic volumetric 3D printing. Nat. Commun. 13(1), 367 (2022)

Numerical Modeling of Part Formation

197

11. Cook, C.C., et al.: Highly tunable Thiol-Ene photoresins for volumetric additive manufacturing. Adv. Mater. 32(47), 2003376 (2020) 12. Kollep, M., et al.: Tomographic volumetric additive manufacturing of silicon oxycarbide ceramics. Adv. Eng. Mater. 24(7), 2101345 (2022) 13. Toombs, J.T., et al.: Volumetric additive manufacturing of silica glass with microscale computed axial lithography. Science 376(6590), 308–312 (2022) 14. Madrid-Wolff, J., Boniface, A., Loterie, D., Delrot, P., Moser, C.: Lightbased volumetric additive manufacturing in scattering resins. arXiv preprint arXiv:2105.14952 (2021) 15. Bhattacharya, I., Toombs, J., Taylor, H.: High fidelity volumetric additive manufacturing. Addit. Manuf. 47, 102299 (2021) 16. Rackson, C.M., et al.: Object-space optimization of tomographic reconstructions for additive manufacturing. Addit. Manuf. 48, 102367 (2021) 17. Bernal, P.N., et al.: Volumetric bioprinting of complex living-tissue constructs within seconds. Adv. Mater. 31(42), 1904209 (2019) 18. Jacobs, P.F.: Fundamentals of stereolithography. In: International Solid Freeform Fabrication Symposium, pp. 196–211 (1992) 19. Salajeghe, R., Kruse, C.S., Meile, D.H., Marla, D., Spangenberg, J.: Investigating the influence of thermal and mechanical properties of resin on the sedimentation rate of the printed geometry in the volumetric additive manufacturing. In: Solid Freeform Fabrication Symposium: 33rd Annual Meeting. The University of Texas at Austin 2022, pp. 882–889 (2022) 20. Salajeghe, R., Meile, D.H., Kruse, C.S., Marla, D., Spangenberg, J.: Numerical modeling of part sedimentation during volumetric additive manufacturing. Addit. Manuf. 66, 103459 (2023). https://www.sciencedirect.com/science/article/ pii/S2214860423000726

Cost Saving Potential of a Shell-Core Strategy of Combined Powder Bed Fusion of Metals with Laser Beam and Hot Isostatic Pressing Lukas Bauch(B) , Leonie Pauline Pletzer-Zelgert , and Johannes Henrich Schleifenbaum RWTH Aachen University, Campus Boulevard 73, 52074 Aachen, Germany [email protected]

Abstract. Powder Bed Fusion of Metals with Laser Beam (PBF-LB/M) offers the possibility to manufacture various complex geometries with integrated functions in one build job independent of tools. However, due to the long process duration and high machine investment, hourly machine cost rates are an obstacle to positive business cases. One idea to reduce machine cost per part is to additively generate a shell geometry with a loose powder core to decrease PBF-LB/M process time in a first step and achieve high density in another step by Hot Isostatic Pressing (HIP). This idea to use hybrid manufacturing leads to a trade-off between reduced manufacturing costs for PBF-LB/M and additional manufacturing costs for HIP. In this work, the cost saving potential of a shell-core strategy is quantified for sample parts. This provides information whether investigating the technological challenges of PBF-LB/M manufactured shell-core geometries and subsequent HIP makes sense in further research from an economic perspective. Keywords: Additive Manufacturing · Laser Powder Bed Fusion of Metals with Laser Beam · Hot Isostatic Pressing · Cost

1 Introduction Additive Manufacturing (AM) is a process in which small elements of material are joined to generate parts from 3D model data [1]. Different AM processes exist, which have a layer- or unit wise work principle to create a workpiece [2]. Due to technological advance, the AM process PBF-LB/M is widely used in the industry [3]. In PBF-LB/M a layer of metal powder is spread on a powder bed. The powder bed is then selectively scanned with a laser beam to melt and solidify the metal powder according to a digital plan. Afterwards, another metal powder layer is applied and scanned. This cycle, consisting of applying a powder layer and scanning, is iterated until the workpiece is generated [1, 2]. PBF-LB/M is referred to as a tool-less manufacturing process. No tool-dependent economies of scale and less manufacturability restrictions for the design exist [4, 5]. Because of the advantage of high design freedom, PBF-LB/M is used to manufacture components with lightweight [6], functionally integrated [7], individualized [8], © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 198–212, 2024. https://doi.org/10.1007/978-3-031-42983-5_14

Cost Saving Potential of a Shell-Core Strategy

199

or monolithic design [9]. Due to the absence of tool-dependent economies of scale, PBF-LB/M is used in particular to produce components with small quantities. However, machine investment-dependent economies of scale are present. Low productivity in combination with high machine prices is an obstacle for using the advantages coming from design freedom for series production or to produce components with large volume [4, 5]. To overcome this obstacle machine hourly cost rates must be reduced, productivity increased, or both. In HIP, high temperature and high pressure are applied to one or more components in a closed process chamber under an argon atmosphere [10]. This leads to deformation of the components. Internal pores and voids are closed, resulting in parts with high density and improved mechanical properties. Due to the higher fatigue properties of densified parts, HIP after PBF-LB/M is used in the aerospace industry [11, 12]. One idea to increase productivity is to additively build a shell geometry with relative density close to 100% that contains a loose powder core to reduce PBF-LB/M manufacturing time in a first step and achieve high density by subsequent HIP [13, 14]. This idea is referred to as shell-core strategy in this work. The shell-core strategy leads to a trade-off between reduced cost for PBF-LB/M and additional cost for HIP. However, to the best authors’ knowledge, an analysis of the cost saving potential of this idea, meaning the cost difference of manufacturing a part with a shell-core strategy compared to direct PBF-LB/M, has not been published. Thus, the aim of this work is to quantify the cost saving potential per part cspp of a shell-core strategy for sample parts. This provides information whether investigating the technological challenges of manufacturing shell-core geometries with PBF-LB/M and subsequent HIP, like compensating shrinkage during the HIP process, or mechanical properties, [14] makes sense in further research from an economic perspective.

2 State of the Art in Reducing Costs of PBF-LB/M by Combination with HIP The idea of combining PBF-LB/M and HIP to increase productivity was published for the first time in [15] and is also addressed in other publications [13, 14, 16]. In [15] manufacturing of cylindrical parts made from Ti-6Al-4V by PBF-LB/M and subsequent HIP for densification are demonstrated. A gas-impermeable shell with at least 92% relative density that encloses an inner geometry with lower relative density is manufactured additively. It is suggested to sinter the internal geometry to a relative density greater than 80% or leave it in the powder state with a relative density of 65% before further densification by HIP [15]. Herzog, Bartsch et al. propose selecting PBF-LB/M process parameters for a high build rate by increasing the scan speed, accepting a low relative density of at least 95% in the as-built state. With a demonstrator build job composed of 3 fuel connectors, made from Ti-6Al-4V on an SLM500L machine, a build time reduction of 24.5% is realized, compared to the parameter set that would achieve relative density close to 100%. In subsequent HIP, 99.8% relative density is achieved [16]. In [14], the idea of manufacturing shrinkage-compensated shell-core geometries with PBF-LB/M is investigated. Different shelled and solid geometries are designed

200

L. Bauch et al.

and manufactured using PBF-LB/M, whereas the shelled geometries are densified in subsequent HIP. The shell-core geometries in [14] are not designed for minimizing PBF-LB/M process duration and have thicker shells than possibly needed. Additionally, more support structures than possibly required have been used for bracket sample parts, resulting in reduced time savings. However, the total recorded build time of shelled geometries of 8:48 h is approximately 11% shorter compared to the solid geometries of 9:53 h. Moreover, as an extreme and hypothetical example representing the maximum build time saving potential, the build times of a solid and a shelled block geometry with edge lengths of 240 mm, 240 mm, and 290 mm, are estimated for an EOS M290 machine. The estimated build time of the shelled geometry is 61 h and orders of magnitude shorter, than the build time of the solid block of 836 h [14].

3 Model to Calculate the Cost Saving Potential In previous publications a combination of PBF-LB/M with HIP to decrease costs is proposed [13– 16]. However, cspp is not quantified. This research gap is addressed in this work. Therefore, cspp of a shell-core strategy is assessed in a case study. In the following section, the process steps and technologies considered in the cost model, the definition of sample parts and PBF-LB/M build jobs, and the cost calculation are described. 3.1 Considered Process Steps and Technologies It is necessary to consider the process steps where a change of cspp is expected due to a shell-core strategy. Because shrinkage during HIP needs to be compensated in a shellcore strategy, less parts fit into one PBF-LB/M build job. That is why fixed costs per job must be allocated to fewer parts and cspp is decreased. After PBF-LB/M, parts and the build plate are separated, typically by using a bandsaw or electric discharge machining. After that, the build plate is prepared for another build job by removing remaining supports structures and creating a flat surface by milling. Additionally, the build plate is treated by sandblasting to increase the surface roughness for good connection with the first layer of a build job [17, 18]. The costs for build plate separation (BPS), milling (BPM), and sandblasting (BPB) need to be allocated to the number of parts a job consists of. Because the number of parts per job is determined in PBF-LB/M nesting for those process steps, an increasing effect of BPS, BPM, and BPB on cost per part needs to be considered in the cost calculation model, to not overestimate cspp . 3.2 Definition of Sample Parts and PBF-LB/M Build Jobs To calculate cspp , different sample parts are considered. Therefore, six solid cylindric geometries are defined. For each solid cylindric geometry, one shell-core geometry is derived, that corresponds to the as-print state after PBF-LB/M and before HIP. For each

Cost Saving Potential of a Shell-Core Strategy

201

sample part, a PBF-LB/M build job is defined, that consists of multiple nested instances of that part. Definition of Solid and Corresponding Shell-Core Sample Parts An overview of the corresponding sample part geometry pairs is given in Table 1, where p is the identification number of the part, g = 1 indicates a solid geometry, and g = 2 indicates a shell-core geometry. The radius of the solid geometry with a volume of Vp,1 is expressed by rp,1 and the height by hp,1 . The outer radius of the shell of the shell-core geometry is rp,2,s and the height is hp,2,s . The volume of the shell, that is fused during PBF-LB/M to achieve the as-built shell-core geometry is Vp,2,s . The wall thickness of the shell is wp,2 . The inner radius of the shell is equal to the radius of the powder core rp,2,c . The height of the powder core is hp,2,c . Table 1. Defined corresponding shell-core and solid geometries p

g

rp,1 [mm]

hp,1 [mm]

V p,1 [mm3 ]

g

rp,2,s [mm]

hp,2,s V p,2,s wp,2 [mm] [mm3 ] [mm]

rp,2,c [mm]

hp,2,c [mm]

1

1

25.00

50.00 98,175

2

29.09

58.17 29,751 2.00

27.09

54.17

2

1

35.00

30.00 115,454

2

39.00

37.99 35,343 2.00

37.00

33.99

3

1

15.00

70.00 49,480

2

17.74

75.48 18,992 2.00

15.74

71.48

4

1

12.50

25.00 12,272

2

13.97

27.95

6,359 2.00

11.97

23.95

5

1

17.50

15.00 14,432

2

18.92

17.84

7,613 2.00

16.92

13.84

6

1

7.50

35.00

2

8.36

36.72

3,905 2.00

6.36

32.72

6,185

cspp is expected to depend strongly on the influence of the chosen solid and derived shell-core geometries on the decrease of PBF-LB/M scanning duration, because of a different volume to be fused. Therefore, the following influences on cspp regarding the PBF-LB/M scanning duration are considered in the definition of sample geometries: – – – –

Change of PBF-LB/M part scanning time per part t p,sp Change of PBF-LB/M support scanning time per part t p,ss Change of PBF-LB/M part layer time per part t p,lp Change of PBF-LB/M support layer time per part t p,ls

The difference of volume to be fused of corresponding solid and shell-core geometries is assumed to decrease with higher ratio of surface area to volume of the solid geometry SVRp,1 for a given wp,2 . Therefore, a negative correlation of SVRp,1 and t p,sp and in consequence SVRp,1 and cspp is expected. Furthermore, SVRp,1 has an influence on PBF-LB/M scanning duration. For high SVRp,1 , more slower contour vectors compared to faster hatch vectors and scanning delays occur, resulting in a lower build rate [19]. Thus, sample parts with a SVRp,1 range of 0.1200–0.3238 are considered in this case study. A value of 2 mm is defined for wp,2 to achieve a gas tight shell [20]. It is assumed that support structures are built solid, not as a shell-core geometry, and removed before HIP. Shell-core geometries are expected to have a larger area that

202

L. Bauch et al.

needs to be supported due to consideration of shrinkage compensation during HIP in the design. Therefore, shell-core geometries require more support structure volume. This leads to an increase of t p,ss for a given support scanning build rate. To account for this influence in this case study, the geometries are oriented in the build space, so the differently sized circular surface areas of the cylindric geometries are parallel to the build plate. Support structures with height of 3 mm are generated using Autodesk Netfabb that connect the downward facing circular surface with the build plate. Furthermore, the z-height hz p,2 of shell-core geometries is higher, than of corresponding solid geometries hz p,1 with equal orientation in the build space and support height, because shrinkage during HIP must be considered in the design. Because the support height is set to 3 mm for both solid and shell-core geometries, there is no change of t p,ls due to a change of support z-height. However, t p,ls and t p,lp depend the quantity of parts per job. If the build job consists of less parts if a shell-core strategy is applied, layer dependent time per job is distributed to less parts. This leads to an increase of t p,ls and t p,lp and in consequence higher cost per part. Corresponding solid and shell-core geometries must have the same mass. Otherwise, densification of a shell-core geometry to the corresponding solid geometry during HIP with equal volume would not be possible. Therefore, Vp,1 must be equal to the sum of Vp,2,s and Vp,2,c multiplied with the relative density of the powder core d , as expressed in Eq. (1). According to [21], the relative apparent density of IN718 powder is 54.8%. Therefore, 54.8% is assumed for d . Vp,1 = Vp,2,s + Vp,2,c ∗ d

(1)

hp,2,s = hp,1 + 2xp

(2)

rp,2,s = rp,1 + xp

(3)

hp,2,c = hp,1 + 2xp − 2wp,2

(4)

rp,2,c = rp,1 + xp − wp,2

(5)

Shrinkage of cylindric shell-core geometry during HIP is not expected to result in an exactly cylindric solid shape, analog to experiments performed with shell-core geometries in [20]. However, it is assumed that the PBF-LB/M build time difference of a solid cylindric geometry and the exact geometry that would result after HIP is neglectable. Therefore, with isostatic pressure during HIP, meaning equally distributed force per outer surface area of the shell geometry, movement of the walls with an equal shrinking distance of xp towards the part center with the result of an exact cylinder geometry is assumed. To derive a shell-geometry in the as-built state according to Eqs. (2)–(5), the value of xp is required. xp is calculated based on the solid geometries, for a given relative density of the powder core. For the calculation of xp , Eq. (6) is transformed to Eq. (8). Equations (2)–(5) are used to supplement unknown variables with known variables of Eq. (7). This leads to Eq. (8),

Cost Saving Potential of a Shell-Core Strategy

203

where xp is the only unknown variable with a given solid geometry of a cylindric part. Thus, Eq. (8) is parametrized for each solid geometry and solved for xp .

2 rp,1

rp,1 2 ∗ hp,1 = rp,2,s 2 ∗ hp,2,s − rp,2,c 2 ∗ hp,2,c + rp,2,c 2 ∗ hp,2,c ∗ d

(6)

rp,1 2 ∗ hp,1 = rp,2,s 2 ∗ hp,2,s + (d − 1) ∗ rp,2,c 2 ∗ hp,2,c

(7)

∗ hp,1 = (rp,1 + xp )2 ∗ (hp,1 + 2xp ) + (d − 1) ∗ (rp,1 + xp − wp,2 )2 ∗ (hp,1 + 2xp − 2wp,2 ) (8)

Definition of PBF-L/M Build Jobs In Table 2, the characteristics of the defined build jobs are shown. Less shell-core parts fit into one build job compared to corresponding solid parts due to shrinkage compensation. Thus, an influence on PBF-LB/M as well as BPS, BPM, and BPB cspp of due to the quantity of parts per job qp,g,ps is expected. The volume per job the outer hull of the solid respectively the shell geometry is indicated by Vjob,p . Table 2. Characterization of the defined build jobs p

g

hz p,g [mm]

Vjob,p [mm3 ]

qp,g,ps

1

1

53.00

1472,622

15

1

2

61.17

1079,922

11

2

1

33.00

923,628

8

2

2

40.99

692,720

6

3

1

73.00

2177,124

44

3

2

78.48

1533,882

31

4

1

28.00

773,126

63

4

2

30.95

613,593

50

5

1

18.00

461,814

32

5

2

20.84

389,656

27

6

1

38.00

1088,562

176

6

2

39.72

878,272

142

The influence of qp,g,ps on cspp is considered by setting a maximum nesting density. An integer number of equal instances of each solid respectively shell-core part per build job is calculated for a defined maximum nesting density. Therefore, Vjob,p , the EOS M290 build space with an edge length of 250 mm [22] and hz p,g are considered. Because this case study is oriented towards a scenario of economical manufacturing of a high production volume, a maximum nesting density of 50% is assumed. 3.3 Cost Calculation In this section, the equations and required input data for the cost calculation are explained. Calculation of the Cost Saving Potential The cost saving potential is calculated per part cspp , to assess the absolute cost saving

204

L. Bauch et al.

potential per part. cp,ps is the change of cost per part p and process step ps. cspp =



cp,ps where ps ∈ {PBF − LB/M , BPS, BPM , BPB, HIP}

(9)

ps

Additionally, an assessment of the cost saving potential relative to the part weight and the cost per solid part is used to evaluate the effectiveness of a shell-core strategy in general. Therefore, the cost saving potential relative to the part mass cspm,p , and relative to the cost for the solid geometry csprel,p is calculated. The mass of a solid geometry is indicated by mp,1 . cspm,p = csprel,p =

cspp mp,1 cspp Cp,1

(10) (11)

cp,ps is calculated based on the cost per part and process step Cp,g,ps of the shell-core geometry Cp,2,ps and solid geometry Cp,1,ps . cp,ps = Cp,1,ps − Cp,2,ps

(12)

To calculate Cp,g,ps , the cost per job and process step Cjob,p,g,ps is divided by the quantity of parts of that job and process step qp,g,ps . Cp,g,t =

Cjob,p,g,ps qp,g,ps

(13)

The equations for calculating Cjob,p,g,ps for all considered process steps are derived from [23–25]. Cjob,p,g,ps is the sum of direct materials cost Cmat , direct labor cost Cl , and manufacturing overhead Cmach . Cjob,p,g,ps = Cmat + Cl + Cmach

(14)

To calculate Cmat , the cost rate of a material cr mat , is multiplied with the required quantity of material per job qmat . Cmat = cr mat ∗ qmat

(15)

According to [26], 95% of the unfused IN718 metal powder can be recovered by recycling for multiple cycles. However, the amount of lost powder is higher, for instance due to shield gas filter residue, aerosol emissions, and cleaning losses [27, 28]. Therefore, the lost amount of powder depends to on the specific use case. To consider powder losses, qmat is defined to depend on the material efficiency mp,job , which is the ratio of all parts in a build job and the part mass to overall used powder emat [28]. qmat =

mp,job emat

(16)

Cost Saving Potential of a Shell-Core Strategy

205

Cl is calculated based on the labor cost rate cr l and the required duration of labor per job. Cl = cr l ∗ dl

(17)

For Cmach the cost for machine depreciation Cdep , interest Ci , room Cr , electricity Cel , auxiliary materials Cam , and wear parts and maintenance Cwm per job are considered. Cmach = Cdep + Ci + Cr + Cel + Cam + Cwm

(18)

Cdep is calculated based on the acquisition cost of a machine Cac , its depreciation period in years Cac , and the quantity of jobs per year qjobs . Cdep =

Cac pdep ∗ qjobs

(19)

qjobs is calculated by dividing the operating hours per year oh by the duration per job djob . qjobs =

oh djob

(20)

Ci is calculated based on the interest rate i, Cac , pdep , and qjobs . Ci =

i ∗ 0, 5 ∗ Cac pdep ∗ qjobs

(21)

Cr is calculated based on the cost rate per area unit and year cr r , the quantity of blocked shopfloor area qr , and qjobs . Cr =

cr r ∗ qr qjobs

(22)

Cel is the product of the electricity price cr e , the average power consumption of the machine pe , and the duration power is consumed de . Cel = cr e ∗ pe ∗ de

(23)

Cam is the product of the cost rate for auxiliary material cr am and the average required quantity of auxiliary material qam . Cam = cr am ∗ qam

(24)

Input Data The used input data for the cost calculation is shown in Table 3. The abbreviation pgd. is used to indicate that data depends on the sample part and its’ geometry, that a job consists of. The used assumptions regarding input data of the considered process steps are explained in the following sections.

206

L. Bauch et al. Table 3. Input data of the cost calculation model Unit

PBF-LB/M

BPS

BPM

BPB

HIP

qp,g,ps

°

pgd

pgd

pgd

Pgd

pgd

cmat

e/kg

90

-

-

-

-

emat

°

0.62 [28]

-

-

-

-

crl

e/h

37,30 [29]

37,30 [29]

37,30 [29]

37,30 [29]

37,30 [29]

C wm

e/job

pqd

0.16

2.86

0.02

42.99 [30]

dl

min/job

47.8

22.83

20 [18]

12

1.6 [25]

C ac

103 e

480 [24]

51.76 [31]

30 [18]

10 [18]

1.000–5.000

pdep

a

6 [24]

6 [32]

7 [32]

5 [32]

5 [32]

oh

h/a

7884 [24]

2000

2000

2000

5054.4 [25]

d job

min/job

pgd

22.83

80

12

468 [25]

crr

e/m2 /h

130 [24]

130 [24]

130 [24]

130 [24]

130 [24]

qr

m2

17 [24]

12.915

9.435

3.45

94.4

cre

e/kWh

0.1694 [33]

0.1694 [33]

0.1694 [33]

0.1694 [33]

0.1694 [33]

qe

kW

8.6 [22]

3 [34]

6 [35]

0.87 [36]

66.67 [25]

de

min/job

pgd

20.83

60

20

7.800 [25]

cram

e/h

0.228 [24]

-

-

-

-

qam

h/job

pgd

-

-

-

-

cram

e/m3

-

-

-

-

0.65 [25]

qam

Nm3

-

-

-

-

13 [25]

Assumptions for PBF-LB/M The PBF-LB/M machine EOS M290 is selected for the use case in this work because it is an established PBF-LB/M system that is used for industrial production. qp,g,PBF−LB/M depends on the specific build job with a defined maximum nesting density of 50%, as explained in 3.2. For cmat of IN718, 90e/kg is assumed. Besides the duration of PBF-LB/M for additive generation, djob consists of the time for loading the build job of 34 min, mounting the build plate and coater blade of 19.4 min, calibrating and checking of 10.5 min, heating up and flooding the build chamber with inert gas of 78 min, cooldown of 50 min, and removal of powder and build job of 17.9 min [18]. The duration of additive generation depends on the specific job and is estimated using EOSPrint. For dl of PBF-LB/M, only the manual steps of mounting the build plate and coater blade, calibrating and checking, and removal of powder and build job of are considered. For de , heating up and flooding with inert gas and additive generation are considered, because the electricity consumption of the other process elements are assumed to be neglectable.

Cost Saving Potential of a Shell-Core Strategy

207

Because of insolubility of argon in metallic materials, HIP of a shell-core geometry with Argon in the core does not result in a fully dense part [37]. As nitrogen can be used for IN718 in PBF-LB/M [38] and to not introduce Argon in the powder core, nitrogen is assumed as inert gas for PBF-LB/M. To consider Cwm of PBF-LB/M, Cac is multiplied with 0.1 and divided by qjobs , following the approach of [24]. Assumptions for BPS, BPM, and BPB For BPS, BPM, and BPB, qp,g,ps is equal to PBF-LB/M. At the beginning of BPS, the build plate and parts are still connected. In BPM and BPB, the build plate is already separated from parts. However, the build plate must be processed by BPM and BPB due to the quantity of part the PBF-LB/M build job consists of. The duration of BPS, BPM, and BPB consists of the main processing and setup before and afterwards. For the total BPS setup time, a duration of 2 min is assumed. The sawing duration for BPS is calculated by dividing the feed distance by the defined feed rate of the band saw. Based on calculations of the internal software of a Klaeger 3D Cut band saw, a feed rate of 12 mm/min is suitable for IN718 and a width of 250 mm to be cut through, as a conservative assumption for the width of a densely nested PBF-LB/M build job. For milling in BPM, a duration of 60 min and for blasting in BPB of 10 min are assumed for one EOSM290 build plate with IN718, based on information from lab personnel experienced with EOSM290 machines and IN718. It is assumed for dl of BPS and BPB that a person is present during the whole process. For BPM, occupation of personnel is only considered for manual setup operations, because other tasks can be carried out during automated milling. The value of oh is set to 2000 h/for BPS, BPB, and BPB, which corresponds to the assumption of two-shift operation on 250 d/a, with the machines being productive for 50% of the time personnel is available. Moreover, it is assumed, that qr equals three times the occupied area by the machine of 4.305 m2 for BPS [34], 3.148 for BPM [35], and 1.15 [36] for BPB, to account for additional space for intralogistics and operators. To consider Cwm BPS, BPM, and BPB, Cdep is multiplied with 0.1, following the approach of [18]. Assumptions for HIP In recent scientific sources, not all required input data for HIP could be found. However, relevant data is available from a whitepaper [25] by the Swedish HIP machine supplier Quintus Technologies. In [25] a calculation of operational costs for processing of additively manufactured turbine blades from IN718 is presented, where the two HIP machines QIH48 and QIH15L with different size and productivity are considered. For the case study in this work, IN718 and the more productive HIP machine QIH48 are chosen, because of the motivation of efficient AM of high quantities or parts with high volume. The acquisition cost of the HIP machines, the machine depreciation period, and the blocked shopfloor area are not provided [25]. On request, different HIP machine suppliers stated to only disclose HIP machine prices in confidential quotes to potential customers. To still achieve the aim of this work, a best-case and worst-case scenario are considered for Cac for HIP. According, to [39], published in the year 1985, the price for a mid-size HIP machine with a process chamber

208

L. Bauch et al.

width of 0.6 m and height of 1.5 m has been 3 million German mark. In [18], acquisition cost of 1 million e is assumed for one HIP machine, but no information regarding the HIP machine model is provided. A span of 1 million e for a best-case to 5 million e for a worst-case scenario for Cac are considered in this work. For HIP, qp,g,HIP is calculated by dividing the maximum payload per job of 175 kg of a QIH48 machine [25] by the part mass. Moreover, it is assumed, that qr equals three times the blocked area by the machine dimensions of 31.5 m2 [30], to account for additional space for intralogistics and operators.

4 Assessment of the Cost Saving Potential of a Shell-Core Strategy In this section, the cost saving potential per part, per part mass, and relative to the cost per part without shell-core strategy are assessed. Moreover, the change of cost per process step and the change of duration of additive generation in PBF-LB/M are assessed. 4.1 Cost Saving Potential Per Part As explained in Sect. 3.2, a negative influence of SVRp,1 on cspp , is expected. SVRp,1 is known for solid parts, without definition of corresponding shell-core geometries. Thus, SVRp,1 is suitable as indicator to assess cspp without knowledge and effort to derive of the exact corresponding shell geometry.

Fig. 1. Cost saving potential per part

In Fig. 1 cspp and in Fig. 2, cspm,p , and csprel,p are depicted depending on SVRp,1 . For all considered sample parts, cspp is positive. This indicates that a shell-core strategy has the potential to achieve lower cost compared to direct PBF-LB/M with the assumptions made in this case study. Also, cspp,m and cspp,rel are positive and show negative correlation with SVRp,1 . Thus, especially parts with low SVRp,1 should be considered for the application of a shell-core strategy to successfully reduce cost. 4.2 Change in Cost Per Part and Process Step To identify the influence of the considered process steps, cp,ps is assessed. The values are displayed in Table 4. The result indicates that the reduced cost for PBF-LB/M clearly

Cost Saving Potential of a Shell-Core Strategy

209

Fig. 2. Cost saving potential per part mass (left) and cost saving potential relative to the cost per part without shell-core strategy (right)

exceeds the increased cost due to less parts per job in BPS, BPM, and BPB and the additional process step HIP with a shell-core strategy, especially for low SVRp,1 . This is the case for both the best- and worst-case scenario. The process step with the highest cost increasing effect is HIP, whereas the influence of BPS, BPM, and BPB is only minor. Table 4. Change in cost per process step per sample part due to a shell-core strategy p

SVRp,1 [1/mm]

cp,PBF [e/part]

cp,BPS [e/part]

cp,BPM [e/part]

cp,BPB [e/part]

cp,HIP best case [e/part]

cp,HIP worst case [e/part]

1

0.1200

85.62

–0.40

–1.14

–0.19

–2.48

–9.89

2 3

0.1238

100.11

–0.69

–1.96

–0.32

–2.92

–11.63

0.1619

36.02

–0.16

–0.45

–0.07

–1.25

–4.98

4

0.2400

6.03

–0.07

–0.19

–0.03

–0.31

–1.24

5

0.2476

7.23

–0.10

–0.27

–0.04

–0.36

–1.45

6

0.3238

1.53

–0.02

–0.06

–0.01

–0.16

–0.62

4.3 Change of Duration of Additive Generation in PBF-LB/M The assessment in Sect. 4.2 indicates that a decreased PBF-LB/M build time is crucial to achieve positive cspp with a shell-core strategy. Thus, the change of duration for additive generation per part t p , which consists of scanning and application of new powder layers, and its components t s,sp , t s,ss , t s,pl and t s,sl are evaluated. The change of duration for additive generation of PBF-LB/M per part due to a shell-core strategy is shown in Table 5. For all considered sample parts, the value of t p is positive, because the increasing effect of t p,sp is bigger than the decreasing influence of t p,ss , t p,lp and t p,ls . These results indicate, that due to a shorter PBF-LB/M build time, machine cost for additive generation during PBF-LB/M can be decreased by a shell-core strategy for the chosen sample parts.

210

L. Bauch et al.

Table 5. Change of duration for scanning and application of new layers of PBF-LB/M per part due to a shell-core strategy p

SVRp,1 [1/mm]

t p,sp [h]

t p,ss [h]

t p,lp [h]

t p,ls [h]

t p [h]

1

0.1200

4.283

–0.024

–0.122

–0.011

4.1275

2

0.1238

5.083

–0.032

–0.161

–0.018

4.8729

3

0.1619

1.800

–0.010

–0.053

–0.004

1.7337

4

0.2400

0.317

–0.004

–0.010

–0.002

0.3006

5

0.2476

0.383

–0.006

–0.012

–0.003

0.3634

6

0.3238

0.083

–0.001

–0.004

–0.001

0.0776

5 Conclusion In this work, the cost saving potential of a shell-core strategy was assessed. The shell-core strategy is characterized by a hybrid manufacturing process, where a shelled geometry with loose powder core is generated using PBF-LB/M and densified to a relative density close to 100% by subsequent HIP. This leads to a higher productivity and thus less cost per part for PBF-LB/M, but higher costs for BPS, BPM, BPB, and the additional process step HIP. Depending on the geometry of a part, the shell-core strategy can result in a cost reduction compared do direct manufacturing of a solid part geometry with PBF-LB/M. The results indicate a high cost saving potential for parts with a low surface area to volume ratio. A limitation of the presented work is missing data for the acquisition cost of a HIP machine. However, a positive cost saving potential is possible, even for parts with high surface area to volume ration in the worst-case scenario with assumption of HIP acquisition of 5,000,000 e. Furthermore, the influence of the wall thickness of the shell-geometry was not investigated in this work. Thus, future research should assess the influence of the wall thickness on the cost saving potential. Moreover, future research should investigate the estimation of cost saving potential with low effort based on the surface area to volume ratio in consideration of the wall thickness for real world applications including all post-processing steps. This should be combined with research of methods to determine shell-core geometries for parts with complex target geometry. Furthermore, possible anisotropic shrinkage behavior of parts during HIP due to their orientation in PBF-LB/M should be considered. Acknowledgement. The authors would like to thank the Federal Ministry of Economic Affairs and Climate Action for funding the research in the project “Ressourcenschonende Prozessroute für hochintegrierte Hydrauliksysteme am Beispiel einer elektrifizierten mobilen Arbeitsmaschine” (HyRess, grant number 03LB3030G).

Cost Saving Potential of a Shell-Core Strategy

211

References 1. DIN Deutsches Institut für Normung e.V., 2022: Additive Manufacturing - General principles - Fundamentals and vocabulary. Beuth, Berlin 01.040.25; 25.030 2. Verein Deutscher Ingenieure: Additive Manufacturing Processes, Rapid Manufacturing: Basics, Definitions, Processes. Beuth, Berlin 25.020 (2014) 3. Khorasani, A., Gibson, I., Veetil, J.K., Ghasemi, A.H.: A review of technological improvements in laser-based powder bed fusion of metal printers. Int. J. Adv. Manufact. Technol. 108(1–2), 191–209 (2020) 4. Baumers, M., Dickens, P., Tuck, C., Hague, R.: The cost of additive manufacturing: machine productivity, economies of scale and technology-push. Technol. Forecast. Soc. Chang. 102, 193–201 (2016) 5. Hedenstierna, C.P.T., Disney, S.M., Eyers, D.R., Holmström, J., Syntetos, A.A., Wang, X.: Economies of collaboration in build-to-model operations. J. Ops. Manag. 65(8), 753–773 (2019) 6. Orme, M.E., Gschweitl, M., Ferrari, M., Madera, I., Mouriaux, F.: Designing for additive manufacturing: lightweighting through topology optimization enables lunar spacecraft. J. Mech. Des. Trans. ASME 139(10) (2017) 7. Willkomm, J., et al.: Design and manufacturing of a cylinder head by laser powder bed fusion. IOP Conf. Ser.: Mater. Sci. Eng. 1097(1), 12021 (2021) 8. Willkomm, J., Jauer, L., Ziegler, S., Schleifenbaum, J.H.: Development of individual medical implants with specific mechanical properties manufactured by Laser Powder Bed Fusion. EPiC Ser. Health Sci. 4, 297–300 (2020) 9. Gradl, P.R., Protz, C., Greene, S.E., Ellis, D., Lerch, B., Locci, I.: Development and hot-fire testing of additively manufactured copper combustion chambers for liquid rocket engine applications. In: 53rd AIAA/SAE/ASEE Joint Propulsion Conference. 53rd AIAA/SAE/ASEE Joint Propulsion Conference, Atlanta, GA. American Institute of Aeronautics and Astronautics, Reston, Virginia (2017) 10. Atkinson, H.V., Davies, S.: Fundamental aspects of hot isostatic pressing: an overview. Metall. Mat. Trans. A 31(12), 2981–3000 (2000) 11. Ahlfors, M., Bahbou, F., Eklund, A., Ackelid, U.: HIP for AM-optimized material properties by HIP. Mater. Res. Proc. 10, 1–10 (2019) 12. Seifi, M., et al.: Progress towards metal additive manufacturing standardization to support qualification and certification. JOM 69(3), 439–455 (2017) 13. Du Plessis, A., Macdonald, E.: Hot isostatic pressing in metal additive manufacturing: X-ray tomography reveals details of pore closure. Addit. Manuf. 34, 101191 (2020) 14. Du Plessis, A., et al.: Productivity enhancement of laser powder bed fusion using compensated shelled geometries and hot isostatic pressing. Adv. Indust. Manufact. Eng. 2, 100031 (2021) 15. Das, S., Wohlert, M., Beaman, J.J., Bourell, D.L.: Processing of titanium net shapes by SLS/HIP. Mater. Des. 20(2–3), 115–121 (1999) 16. Herzog, D., Bartsch, K., Bossen, B.: Productivity optimization of laser powder bed fusion by hot isostatic pressing. Addit. Manuf. 36, 101494 (2020) 17. Kratzer, M.J., Mayer, J., Höfler, F., Urban, N.: Decision support system for a metal additive manufacturing process chain design for the automotive industry. In: Meboldt, M., Klahn, C. (eds.) Industrializing Additive Manufacturing, pp. 469–482. Springer International Publishing, Cham (2021) 18. Möhrle, M.: Gestaltung von Fabrikstrukturen für die additive Fertigung. Springer, Heidelberg (2018) 19. Bauch, L., Winklbauer, F., Stittgen, T., Collet, A., Schleifenbaum, J.: Estimation of printing time for laser-based powder bed fusion of metals. In: ASTM International Conference on Additive Manufacturing (ICAM 2021), pp. 14–28 (2022)

212

L. Bauch et al.

20. Riehm, S., et al.: Tailor-made functional composite components using additive manufacturing and hot isostatic pressing. Powder Metall. 64(4), 295–307 (2021) 21. Georgilas, K., Khan, R.H., Kartal, M.E.: The influence of pulsed laser powder bed fusion process parameters on Inconel 718 material properties. Mater. Sci. Eng., A 769, 138527 (2020) 22. EOS GmbH - Electro Optical Systems: EOS M 290 Betriebsanleitung (2014) 23. Ostwald, P.F., McLaren, T.S.: Cost Analysis and Estimating for Engineering and Management. Prentice Hall (2004) 24. Yi, L., Gläßner, C., Aurich, J.C.: How to integrate additive manufacturing technologies into manufacturing systems successfully: a perspective from the commercial vehicle industry. J. Manuf. Syst. 53, 195–211 (2019) 25. Ahlfors, M., Hjärne, J., Shipley, J., et al.: Cost effective hot isostatic pressing—a cost calculation study for AM parts. Quintus Technology (2018) 26. Ardila, L.C., et al.: Effect of IN718 recycled powder reuse on properties of parts manufactured by means of selective laser melting. Phys. Procedia 56, 99–107 (2014) 27. Lutter-Günther, M., Gebbe, C., Kamps, T., Seidel, C., Reinhart, G.: Powder recycling in laser beam melting: strategies, consumption modeling and influence on resource efficiency. Prod. Eng. Res. Devel. 12(3–4), 377–389 (2018) 28. Lutter-Günther, M., Hofmann, A., Hauck, C., Seidel, C., Reinhart, G.: Quantifying powder losses and analyzing powder conditions in order to determine material efficiency in laser beam melting. AMM 856, 231–237 (2016) 29. Statistisches Bundesamt: One hour worked cost an average of 37.30 euros in 2021: Labour costs in Germany in the upper third of the EU. Statistisches Bundesamt (2022). https://www. destatis.de/EN/Press/2022/05/PE22_190_624.html 30. Quintus Technologies: Quintus Compact HIP Systems: Highest performance and easy to use (2016) 31. Schachermayer Deutschland GmbH, 2023. KASTO Hochleistungs-Bandsägeautomat KASTOwin A 4.6. https://webshop.schachermayer.com/cat/de-DE/product/kasto-hochleistungsbandsaegeautomat-kastowin-a-4-6/109964671. Accessed 21 May 2023 32. Bundesministerium der Finanzen: AfA-Tabellen (2001). https://www.bundesfinanzminist erium.de/Content/DE/Standardartikel/Themen/Steuern/Weitere_Steuerthemen/Betriebsprue fung/AfA-Tabellen/AfA-Tabelle_Maschinenbau.html. Accessed 22 May 2023 33. Statista, 2022. Strompreise für Gewerbe - und Industriekunden in Deutschland in den Jahren 2012 bis 2022. https://de.statista.com/statistik/daten/studie/154902/umfrage/stromp reise-fuer-industrie-und-gewerbe-seit-2006/ 34. METZLER GmbH & Co KG, 2018. Gesamptprogramm: Hermann Klaeger GmbH Maschinenfabrik. https://www.metzler.at/media/downloads/pdf/maschinenprofis_2/saegen/ Klaeger_Gesamtprogramm.pdf 35. DMG MORI. Vertival High Speed Machining Centers: DMP 35 DMP 70 36. ABS Strahltechnik. DRUCK-STRAHLKABINE ECO-PF. https://www.strahltechnik-exp ress.de/sandstrahlkabinen/druckstrahlkabinen/druck-strahlkabine-eco-pf 37. Kaletsch, A., Qin, S., Herzog, S., Broeckmann, C.: Influence of high initial porosity introduced by laser powder bed fusion on the fatigue strength of Inconel 718 after post-processing with hot isostatic pressing. Addit. Manuf. 47, 102331 (2021) 38. Pauzon, C., Markström, A., Dubiez-Le Goff, S., Hryha, E.: Effect of the process atmosphere composition on alloy 718 produced by laser powder bed fusion. Metals 11(8), 1254 (2021) 39. Bousack, H.: Das heißisostatische Pressen (HIP): Technik, p. 2000. Publikationen vor, Anwendungsmöglichkeiten und Wirtschaftlichkeit (1985)

Automated Design of 3D-Printed Silicone Parts: A Case Study on Hand Rehabilitation Gloves for Stroke Patients Felix Weigand1(B)

, Julia Föllmer1 , and Arthur Seibel1,2

1 Fraunhofer Research Institution for Additive Manufacturing Technologies IAPT,

21029 Hamburg, Germany [email protected] 2 Hamburg University of Technology, 21073 Hamburg, Germany

Abstract. This paper presents an innovative way to customize silicone products by utilizing a design automation process and extrusion-based silicone 3D printing as a manufacturing technique. A customizable rehabilitation glove for stroke patients is used as an example to demonstrate the process chain. Soft robotics offers unique possibilities for the use of rehabilitation devices in a homecare setting and has the potential to support the healing process throughout the different steps of neurorehabilitation. Furthermore, customization of such devices can make physical therapy more comfortable and effective for the patient. The presented process chain combines pose detection with parametric design in order to adapt a glove consisting of soft bending actuators to the hand of a patient, which can be manufactured in a single day. Keywords: Design Automation · Soft Robotics · Silicone 3D Printing · Hand Rehabilitation

1 Introduction The term “design automation” refers to the automated creation of computer-aided design (CAD) models based on existing designs. It ranges from the automated adaptation of simple parameters to the complete configuration of an entire product. Modern manufacturing methods such as 3D printing enable the production of increasingly complex components and assemblies. To make optimal use of these methods, companies and engineers require specific knowledge, training and experience. Design automation has the potential to not only reduce these requirements but also to reduce the error rate and time required for repetitive modeling tasks. It enables low-cost, customizable designs and reduces development costs for products with many variants or versions. A design automation workflow only needs to be set up once. Afterwards, no further manual work is required to create new designs. One field of applications relevant to design automation is medical technology. In particular, applications where the physiology of the respective patient needs to be considered and the medical product must be individually adapted can profit from an automated © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 213–224, 2024. https://doi.org/10.1007/978-3-031-42983-5_15

214

F. Weigand et al.

solution. Possible examples of such products are shoe inserts, dentures, certain implants, surgical guides or liners for prosthetic devices. In this paper, we take a look at the process chain for our design automation approach using an individualized rehabilitation glove for stroke patients as an example. Every year, around 12.2 million people worldwide are affected by a stroke [1]. More than two thirds of those affected subsequently suffer from hand impairment of varying severity [2]. Hands are our most important tools in performing daily tasks and are essential to living an independent life. Therefore, the recovery of the original hand function or at least the improvement of the current hand function is of considerable importance for the individual as well as employers, insurance companies and the welfare system as a whole. The often complex movement disorders of stroke patients are usually treated with neurorehabilitation – a medical process, that focuses on targeted therapy of the areas with limited mobility. The success of the rehabilitation measures depends on the coordination of the treatment, the structures of clinical care as well as the follow-up treatment. In Germany, a phase model of the Federal Working Group for Rehabilitation for the staged care after a stroke is used (Fig. 1). Each phase of this model must always be tailored to each patient. But in general, targeted, functional exercises and many repetitions are crucial for restoring or improving the functionality of the hand [3].

Fig. 1. German phase model for the staged care of stroke patients (according to [3])

Physical therapy is thus an important component of the phase model and is used to varying degrees in the different phases. However, current treatment methods are laborintensive, costly and require ongoing support from physical therapists. Robotic systems have been used for some time to support rehabilitation facilities and physiotherapists in this task. Typically, rigid exoskeletons are used to ensure that training sequences are repeatable and scalable [4–8]. However, recent developments have also made it possible to use soft robotic systems in this context [9–13]. Using soft and flexible components such as fluidic actuators for the training has several advantages. They are more compact, lighter, easier to put on and take off and, due to their flexibility, can be used safely for homecare rehabilitation. For many patients, it is a challenge to maintain a regular training

Automated Design of 3D-Printed Silicone Parts

215

regimen and the correct movement sequences unsupervised. A soft robotic system can support the patient at home and accelerate the healing process. The complexity of a soft robotic rehabilitation glove makes its production a key challenge for becoming an efficient and economic application. In this paper, we introduce 3D printing as a viable option for the manufacturing of a soft robotic rehabilitation glove. It is manufactured using an extrusion-based additive manufacturing technology for silicones, which enables the production of complex parts and hollow structures that are required for fluidic actuators, without the need for additional joining processes. The technology has the added benefit of enabling an individualization of the glove without requiring customized tools, which are essential in any silicone casting process.

2 Design Automation Process Chain The process chain for the automated design of a rehabilitation glove consists of four main steps (Fig. 2). The process is based on a parametric CAD model, which needs to be created once and includes design variables that can be changed and determine the final size and layout. The input values for the design variables are specific to each patient and need to be identified using pose detection. Once the parameters are set and the CAD model is updated to the specifications of the respective patient, it can be exported as an.stl file. The preparation of the build job happens in the slicer software. Here, the layer information is created and all process parameters are configured. Now, the model can be printed. Once the print job is finished, the rehabilitation system can be assembled and tried on by the patient. The individual steps are described in detail in the following sections. The proposed process chain could be partially or fully implemented at hospitals or physiotherapy facilities. It is designed to enable the adaptation of a CAD model to the patient without any specialized design or production knowledge. Furthermore, it should be easy to implement and produce good results, without the requirement for specialized equipment. A physician or physical therapist would take the image and perform image calibration and pose detection. In the future, additional input options will be available for the medical professionals to fine-tune the results of pose detection. The parametric CAD design would automatically be updated (step 3) and then fabricated (step 4). This last step could either be conducted locally, if the required 3D-printer is available, or by a specialized orthopedic workshop. The design automation workflow is adaptable to other medical applications (e.g. patient-specific implants for craniomaxillofacial reconstruction [14]) and can also utilize 3D imaging or CT scans. In both cases, different types of image recognition would be used to identify and locate the relevant anatomical features. 2.1 Image Calibration The first step of the design automation process chain is to take a picture of the patient’s hand. To use this picture for determining the required values for each design variable, it needs to be calibrated. For this purpose, a fiducial marker is included in the picture.

216

F. Weigand et al.

Fig. 2. Process chain for the automated design of a rehabilitation glove

Such markers are objects or shapes that are placed within the field of view of the camera and can then be used as a point of reference for measurements. We use the ArUco open source library and the corresponding markers (Fig. 3) as a reference system. ArUco markers consist of two-dimensional black and white codes. The codes have a black border and the inner region is characterized by a binary pattern. The pattern is unique and therefore identifies each marker [15].

Fig. 3. Examples of ArUco markers

The program for the hand key point detection utilizes the ArUco marker as a reference object for measuring distances. The ArUco module and its marker detection function can be integrated in OpenCV, which will be further outlined below. The module is used to detect the marker and its distinct features. ArUco markers have a fixed edge length of 50 by 50 mm. By counting the number of pixels of the marker’s perimeter from corner to

Automated Design of 3D-Printed Silicone Parts

217

corner, a ratio of pixels per millimeter can be established. This ratio can then be used to determine the dimensions of and the distances between other features inside the image. 2.2 Pose Detection Once the image is captured and calibrated (Fig. 4 A), the second step is to conduct a pose detection procedure and create a digital skeleton from the results. Pose detection belongs to the field of computer vision and is used to detect and track movements of the human body. Using machine learning, the algorithm analyzes an image and locates key points on the body, for example elbows, knees or wrists. Considering these points in relation to each other, the general body position and orientation can be derived. In the context of this work, the method is used to locate the wrist and finger joints in the image of the patient’s hand. Two open source tools are combined to process the image and to detect key points. The first tool is OpenCV1 . It provides an optimized, real-time computer vision library and supports model execution for machine learning. The second tool is OpenPose2 , which is based on a Convolutional Architecture for Fast Feature Embedding (CAFFE). This is a deep learning framework and provides machine learning models for the classification and cluster analysis of image data. A CAFFE model requires two inputs. The first one is an architecture for the desired deep neural network, which is provided in form of a PROTOTXT file; we use pose_iter_102000.caffemodel3 as our architecture. The second input is a 4D blob, which is a preprocessed version of the original image. The first step of the pose detection algorithm is to preprocess the image using the blobFromImage function of OpenCV. The preprocessed image is then passed to the input layer of the neural network of the CAFFE model. The output of the network is a prediction of the location of the key points. Each location is indicated by a probability map (Fig. 4 B). In total, 22 probability maps are created, one for each joint location. The last step is to scale the probability maps to the size of the original image. The maxima of the maps now indicate the locations of the joint in the original image. To create a digital skeleton that can be used for the parametric design, each key point is marked in the image of the hand (Fig. 4 C). They are numbered and connected by lines, following a list of predetermined point pairs (Fig. 4 D). By determining the length of each line and using the AruCo marker as a reference, the true distances between the hand joints of the patient are determined. Between the left and the right hand, only the numbering of the auricular finger and the thumb need to be changed. The robustness of the key point detection is dependent especially on the hand position and lighting. To receive the best results for the distances required for the parametric model, the fingers should be placed next to each other and not spread widely. The thumb should be roughly at an angle of 45°. While testing the detection, bad lighting and shadows could cause the detection of key points to fail.

1 Downloadable from https://github.com/opencv/opencv. 2 Downloadable from https://github.com/jarunraj/pose. 3 Downloadable from https://www.kaggle.com/datasets/changethetuneman/openpose-model.

218

F. Weigand et al.

Fig. 4. Pose detection steps, (A) input image with ArUco marker, (B) combined probability maps for all joints, overlaid on original image, (C) key points on the hand, (D) digital skeleton

2.3 Parametric Design The third step in the process chain is the adaptation of the parametric design to the patientspecific measurement values determined in the pose detection step. In this section we first describe how the parametric design was created and then how to update the design with new input values for the specified design parameters. The CAD model of the rehabilitation glove was created in SolidWorks 2019 (SP 5.0). However, the workflow is not restricted to this software solution and can easily be transferred to other CAD tools that offer similar parametric design functionalities. For the model to be adjustable to the respective patient, it needs to be built around the data provided by the pose detection from the previous step. In order to utilize this data, the skeleton formed by interconnecting the key points (i.e. the finger and wrist joints) is recreated in a CAD sketch. The location of the metacarpophalangeal joint of the auricular finger is used as the starting point. Using this location as the origin, all other joint positions are referenced accordingly (Fig. 5 A). The base is made up of four triangles, formed between the wrist and the five metacarpophalangeal joints. From here, four vertical lines are drawn for the index, middle, ring and auricular finger. The skeleton of the thumb is constructed with an additional triangle between the index and thumb metacarpophalangeal joints and the thumb interphalangeal

Automated Design of 3D-Printed Silicone Parts

219

joint (Fig. 5 B). To account for the opposability of the thumb, the two representing lines of the skeleton were shifted parallel by 17 mm (Fig. 5 C). After the construction of the skeleton, dimensions are added to each line connecting two of its key points. SolidWorks offers the possibility to define these dimensions as design parameters. The parameters are saved in a separate.txt file, where they can easily be accessed and modified. Since the parameters in SolidWorks are linked to the text file, any changes to the design parameters can easily be adopted by simply updating the model. The pose detection algorithm described above automatically writes its results into a.txt file with the corresponding design parameters for the CAD model, thus updating the design to the specific measurements of the patient’s hand.

Fig. 5. Implementation of the digital skeleton into a parametric design, (A) digital skeleton from pose detection, (B) skeleton recreated as a sketch in CAD, (C) CAD design using skeleton as reference

Using the skeleton from the pose detection step as a basis, the actual model of the rehabilitation glove is now created. In the proposed design, each finger contains a separately actuated pneumatic bending actuator. The actuators used consist of separate cells, which expand under pressure and thus lead to a change in shape. The cells of each actuator are divided into three groups and placed on the finger joints. The points in the sketch representing the positions of the respective joints of the skeleton described above serve as the midpoints for the groups of cells. Five inflatable chambers are located on the metacarpophalangeal joints, four on the proximal interphalangeal joints and two on the distal interphalangeal joints. The auricular finger is an exception since here, only two chambers are placed on the proximal interphalangeal joint. All chambers on a finger are connected by a central channel. To ensure printability and air-tightness, the chambers are designed according to the design guidelines for extrusion-based silicone 3D printing of pneumatic soft grippers [16]. 2.4 Manufacturing The rehabilitation glove is manufactured using silicone 3D printing. Same as any other additive manufacturing technology, silicone 3D printing requires the geometry to be

220

F. Weigand et al.

translated into layer information. This is done using a slicing software. Since extrusionbased silicone 3D printing is based on the same principles as fused deposition modeling, the same slicing software can be used for this purpose. In our case, Cura (version 5.0) was used to prepare the build job (Fig. 6 A). The most important printing parameters are listed in Table 1. Table 1. Cura printing parameters used to manufacture the rehabilitation glove prototypes Material diameter

2.85 mm

Line width (nozzle diameter)

0.4 mm (0.41 mm)

Layer height

0.4 mm

Flow

300%

Print speed

10.0 mm/s

Retraction distance

7.5 mm

Coasting volume

0.24 mm3

Cura offers the possibility to review the results of the slicing process. This is an important step, especially for 3D printing of silicone, to ensure printability and quality of the end product. During the creation of the parametric design, the slicing process was performed several times to identify problem areas for manufacturability and to adjust the design accordingly. The most critical issue was to ensure a continuous extrusion of the outer wall and to avoid very short extrusion paths, which can lead to defects and a lack of air tightness of the final part (Fig. 6 B) [17].

Fig. 6. Slicing and layer-wise manufacturability check, (A) sliced CAD model, (B) problematic short extrusion paths, (C) printing process on the LiQ320

Two complete prototypes of the hand rehabilitation glove were manufactured using the silicone 3D printer LiQ320 3D (InnovatiQ GmbH & Co KG) (Fig. 6 C). The Material used was Silastic 3D 3335 (Dow Chemical). Its material properties are summarized in

Automated Design of 3D-Printed Silicone Parts

221

Table 2. The first prototype took about 14 h to print. It was used to prove the printability and functionality of the final design for the pneumatic bending actuators. For this purpose, it was connected to a pressure source after its completion and all actuators were pressurized multiple times to detect any leakage. The test showed that all actuators were airtight and could be actuated repeatedly. Manufacturing-related defects did not occur on the prototype, and the 3D printing process requires no post-processing. The prototypes are ready to use right off the build plate. Table 2. Properties of Silastic 3D 3335 liquid silicone rubber Shore hardness

A 50

Tensile strength

9.5 MPa

Elongation at break

480%

Relative density

1.12

Temperature resistance

200 °C

After the functional test of the first prototype, only two changes were made to the design. First, the inlets for the compressed air supply of the actuators were improved to ensure a tighter fit with the connectors for the compressed air hoses (Fig. 7 A). Second, eyelets were added to all fingers as attachment points for Velcro straps. The straps are put through the eyelets securing the actuators to the hand (Fig. 7 B). For comparison, a state-of-the-art hard robotic system is shown in Fig. 7 C. It took approximately 20 h to print the second prototype. The manufacturing costs for the second prototype are estimated at 95 e, control box and pressure source not included. Optimization of the process parameters and falling material costs can reduce these costs further. A reduction in costs to below 50 e could be conceivable in the future. At this price point, silicone 3D printing and individualized medical products can become a viable option for industrial manufacturing chains.

3 Conclusion and Future Work In conclusion, the paper shows the successful implementation of the automated design process chain suggested by the authors to create a customizable rehabilitation glove to support the physical therapy of stroke patients. Image calibration and pose detection were successfully used to create a digital skeleton of the hand based on the locations of the finger and wrist joints, including all relevant design parameters. By reproducing the skeleton in CAD and creating a glove CAD model around it, a parametric design is created. The design parameters are linked to a.txt file that can be updated using the pose detection algorithm. To ensure printability, design guidelines for extrusion-based silicone 3D printing were implemented in the parametric design. In this way, no prior knowledge of the printing process is required. Two fully functional prototypes were successfully manufactured using silicone 3D printing. The final prototype fit well to the hand of the test subject and could be actuated

222

F. Weigand et al.

Fig. 7. Finished prototype, (A) evaluation of the fit to the hand, (B) assembled first prototype (hoses for compressed air are not connected), (C) hard robotic rehabilitation system [18]

repeatedly without any air leakage. To correctly design the actuator for the thumb remains a challenge because the images used for pose detection show each finger from above, except for the thumb, which is shown in a side view (Fig. 5). The following steps are suggested by the authors to further develop the design automation process chain as well as the rehabilitation glove. • The pose detection algorithm can be developed further in order to utilize as much data provided by a 2D image as possible. For example, the width of the patient’s fingers could be included as an additional design parameter. In a next step, the design process should be adapted to also include 3D information, such as provided by a 3D scan. This would open up further customization potentials. The rehabilitation glove could basically incorporate a negative of the patient’s fingers to make it a perfect fit. An alternative example where 3D information is needed for a successful customization are silicone liners for prosthetic devices. Here, a 3D adaptation to the stump of a patient’s leg is required. • The next step with regard to the glove’s parametric CAD model is the optimization of the design according to the requirements of physical therapy. The main goals in are to achieve the required forces and to enable the necessary motion sequences by optimizing the presented design. This could include the evaluation of the effects of different gap sizes between the air-filled chambers of the actuators, different chamber geometries and local stiffness changes. Additionally, inputs that are relevant to the user will be identified in order to enable healthcare professionals to manipulate and fine tune the design parameters determined by the pose detection algorithm. • Future work regarding the manufacturing process includes the optimization of the process parameters to reduce production time and costs and the implementation of suitable quality assurance processes. Furthermore, the selection or development of suitable medical grade silicones as well as the certification of the 3D printing process are required in order to become a viable manufacturing solution for the medical sector.

Automated Design of 3D-Printed Silicone Parts

223

Another possible area of research could be non-planar printing for silicones. This would improve the surface quality of certain models and enable optimized designs of the glove with a more secure fit to the hand. • Further development of the rehabilitation glove includes the development of a suitable, compact pneumatic control box that is able to power the rehabilitation glove for the duration of the training and enable the defined motion sequences. Furthermore, clinical studies are required to evaluate the added value of such a system for the patient. These studies should be performed directly on the patient to monitor the progression of the recovery process. The results should be compared with the recovery progression of classical rehabilitation methods to evaluate the benefit for the patient.

References 1. Feigin, V.L., et al.: World stroke organization (WSO): global stroke fact sheet. Int. J. Stroke 17(1), 18–29 (2022). https://doi.org/10.1177/17474930211065917 2. Duncan, P.W., Bode, R.K., Lai, S.M., Perera, S.: Rasch analysis of a new stroke-specific outcome scale: the stroke impact. Arch. Phys. Med. Rehabil. 84(7), 950–963 (2003). https:// doi.org/10.1016/S0003-9993(03)00035-2 3. Kelle-Herfurth, K., Siegmar, N.: Rehabilitation nach einem Schlaganfall. https://schlaganfall begleitung.de/nachsorge/rehabilitation. Accessed 7 Feb 2023 4. Nef, T., Riener, R.: ARMin—design of a novel arm rehabilitation robot. In: 9th International Conference on Rehabilitation Robotics, pp. 57–60 (2005) 5. Loureiro, R., Amirabdollahian, F., Topping, M., Driessen, B., Harwin, W.: Upper limb robot mediated stroke therapy—GENTLE/s approach. Auton. Robots 15(1), 35–51 (2003). https:// doi.org/10.1023/A:1024436732030 6. Heo, P., Gu, G.M., Lee, S., Rhee, K., Kim, J.: Current hand exoskeleton technologies for rehabilitation and assistive engineering. Int. J. Precis. Eng. Manuf. 13(5), 807–824 (2012). https://doi.org/10.1007/s12541-012-0107-2 7. Jones, C.L., Wang, F., Morrison, R., Sarkar, N., Kamper, D.G.: Design and development of the cable actuated finger exoskeleton for hand rehabilitation following stroke. IEEE/ASME Trans. Mechatron. 19(1), 131–140 (2014). https://doi.org/10.1109/tmech.2012.2224359 8. Lee, S.W., Landers, K.A., Park, H.S.: Development of a biomimetic hand exotendon device (BiomHED) for restoration of functional hand movement post-stroke. IEEE Trans. Neural Syst. Rehabil. Eng. 22(4), 886–898 (2014). https://doi.org/10.1109/TNSRE.2014.2298362 9. Wang, L., et al.: Soft robotics for hand rehabilitation. In: Intelligent Biomechatronics in Neurorehabilitation, X. Hu, Ed.: Elsevier, pp. 167–176 (2020 10. Polygerinos,P., Galloway, K.C., Savage, E., Herman, M., Donnell, K.O., Walsh, C.J.: Soft robotic glove for hand rehabilitation and task specific training. In: 2015 IEEE International Conference on Robotics and Automation, pp. 2913–2919 (2015) 11. Connelly, L., Jia, Y., Toro, M.L., Stoykov, M.E., Kenyon, R.V., Kamper, D.G.: A pneumatic glove and immersive virtual reality environment for hand rehabilitative training after stroke. IEEE Trans. Neural Syst. Rehabil. Eng. 18(5), 551–559 (2010). https://doi.org/10.1109/tnsre. 2010.2047588 12. Reymundo, A.A., Muñoz, E.M., Navarro, M., Vela, E., Krebs, H.I.: Hand rehabilitation using soft-robotics. In: 6th IEEE International Conference on Biomedical Robotics and Biomechatronics, pp. 698–703 (2016) 13. Toya, K., Miyagawa, T., Kubota, Y.: Power-assist glove operated by predicting the grasping mode. J. Syst. Des. Dyn. 5(1), 94–108 (2011). https://doi.org/10.1299/jsdd.5.94

224

F. Weigand et al.

14. Imgrund, P., et al.: Digital work flow and process for additive manufacturing of patientspecific-implants for craniomaxillofacial reconstruction. Trans. Addit. Manufact. Meets Med. 4(1) (2022). https://doi.org/10.18416/AMMM.2022.2209680 15. OpenCV, Detection of ArUco Markers. https://docs.opencv.org/3.4/d5/dae/tutorial_aruco_ detection.html. Accessed 23 Feb 2023 16. Weigand, F., Nguyen, A.M., Wolff, J., Seibel, A.: Toward industrial silicone 3D printing of soft robots. In: 2021 IEEE 4th International Conference on Soft Robotics, pp. 523–526 (2021) 17. Weigand, F., Seibel, A.: Additive manufacturing of soft robots. Innov. Product Dev. Addit. Manufact. 2023, 101–112 (2021) 18. memorialhermann, Intelligent Hand Rehabilitation Using an Exoskeleton Driven by Intent and Muscle Activities. https://memorialhermann.org/services/specialties/tirr/healthcare-pro fessionals/journal/2018/summer-2018/intelligent-hand-rehabilitation. Accessed 18 Aug 2023

Investigation of the Feasibility to Process NiTi Alloys with Powder Bed Fusion for Potential Applications Rico Weber1,2(B) , Adriaan B. Spierings1 , and Konrad Wegener2 1 2

Innovation Center for Additive Manufacturing Switzerland, inspire AG, 9014 St. Gallen, Switzerland Institute of Machine Tools and Manufacturing, ETH Zurich, 8005 Zurich, Switzerland [email protected]

Abstract. NiTi is a versatile material with a broad range of functional properties such as shape memory effect and superelasticity in combination with high internal damping capabilities and biocompatibility. Processing NiTi with powder bed fusion for metals (PBF-LB/M) enables new potential to solve engineering problems. Due to the layerwise manufacturing technique associated with additive manufacturing, complex shaped geometries can be realized, which are not possible with conventional manufacturing methods. Hence, the combination of functional material characteristics with powder bed fusion has high potential for novel applications and complex structures. In this work, potential and performance of manufacturing complex NiTi structures is demonstrated. Keywords: NiTi · shape memory alloy powder bed fusion

1

· functional structures ·

Introduction

Powder bed fusion for metals (PBF-LB/M) is an additive manufacturing technology capable of producing complex shaped parts which are not possible to manufacture with conventional methods. The process is characterized by a high energy laser which melts powder in a layerwise technique. Due to the layerwise manufacturing technique complex shapes and geometries can be realized without the need of cost intensive additional equipment. 3D geometries can be manufactured directly after design in a CAD software which enables decentralized and short supply chain manufacturing. Powder bed fusion is one of the most relevant additive manufacturing technologies and is undergoing constantly development to increase maturity of the manufacturing technology in the industry. In order to enable broad application of the powder bed fusion process in industry it is crucial to process diverse materials. Nickel-titanium alloys (NiTi) are gaining huge interests due to their inherent functional properties which present high c The Author(s), under exclusive license to Springer Nature Switzerland AG 2024  C. Klahn et al. (Eds.): AMPA 2023, STAM, pp. 225–236, 2024. https://doi.org/10.1007/978-3-031-42983-5_16

226

R. Weber et al.

potential for aerospace and biomedical applications [1]. The most well-known property of NiTi is the shape memory effect, which is caused by a diffusionless transformation from martensite to austenite and vice versa. Due to the shape memory effect nearly equiatomic NiTi alloy can recover its shape after deformation and heating to a certain temperature. Shape recovery rates from 3–5% are currently the benchmark for NiTi alloys processed with powder bed fusion [2]. The shape memory effect can be exploited for engineering novel actuation systems replacing heavy hydraulic, electric, or pneumatic actuators [3]. Especially in aerospace NiTi actuators can benefit from a low cost per tons to orbit which is a crucial factor for space missions. NiTi actuation of a jet engine nozzle with a high technology readiness level was demonstrated by Hartl et al. [4]. Recently, additively manufactured and tailored NiTi rods were used to pre-stress sensors and actuators in aerospace components [5]. The two way shape memory effect (TWSME) offers the possibility to achieve shape deformation in both directions during heating and cooling, which can be used for cycling actuators. The TWSME can be programmed into the material by a training procedure. The training consists of mechanical constrained heating and cooling >10 cycles. Self-fitting NiTi parts produced with powder bed fusion was demonstrated by Nespoli et al. exploiting the two way shape memory effect [6]. However, degradation of the transformation strain is still an issue regarding the usage of the two way shape memory effect. The effective transformation strain can drop 0.5–1% and is dependent on the training cycles limiting applications so far [7,8]. Besides the shape memory effect other functional properties are superelasticity and biocompatibility [9]. Superelasticity has great potential for seismic applications to dissipate large amount of mechanical energy [10]. The superior biocompatibility of NiTi is exploited extensively in medical implants such as stents and bone rupture implants [11]. Stents are medical products in which shape memory and biocompatibility properties are combined. The activation temperature, which is the energetic threshold for shape recovery, is tuned to body temperature. Subsequently the predeformed stent is inserted into the artery and recovers the initial shape due to heat of the body providing the thriving force to build up pressure against the artery wall. Powder bed fusion has significant potential to combine the advantages of both the process and unique material properties to develop and engineer novel applications. The possibility to design freeform and complex shaped structures combined with shape memory properties can lead to novel actuation mechanism in a wide range of scaling only limited by the spot diameter of the laser beam. In this work manufacturing and actuation of a broad range of complex NiTi structures shall be demonstrated. These structures can be manufactured in lot size one without the need of additional molding or casting dies at decentralized locations. Aerospace with its low volume market, lightweight and cost driven applications has huge potential for additive manufactured NiTi actuators [12]. Lattice structure are a lightweight technology to reduce weight and simultaneously maintain stiffness and mechanical integrity of a structure. With powder

Potential Applications for NiTi Manufactured with PBF-LB/M

227

bed fusion NiTi lattices can be manufactured to achieve enhancement with functional properties such as shape recovery or superelasticity. NiTi lattices can be used for crash bumpers or energy absorbers. In this work manufacturing and testing of two different lattices is demonstrated. Summarized, this work shall demonstrate potential applications for NiTi parts manufactured with powder bed fusion.

2

Materials and Methods

Pre-alloyed NiTi powders from Eckart (former TLS Technik) were used. The powders had an Ni content of 55.2 ± 0.7 and 53.1 ± 0.7 wt% Ni measured with EDX (FEI NanoSEM 230, Oxford X-Max SDD EDX system). Processing was performed with an industrial Concept Laser M2 machine (Concept Laser GmbH) and with a research machine Aconity Midi+ (Aconity GmbH). Process parameters were set to laser power 80 W, scanning speed 30 mm/s, hatch distance 0.12 mm, layer thickness 0.03 mm and spot size 0.08 mm. The spider like structure in this study consists of eight feets of which each feet has a diameter of 1 mm. The total height is 8 mm. The body is represented with a cylindrical shape of 8 mm diameter and 2 mm height. With a tensile test machine (Galdabini Quasar 10) the spider structure is compressed. Two types of lattices were selected which exhibit a low and high stiffness. First type is a body-centered cubic sandwich lattice with strut thickness 0.2 mm with 10 × 10 × 11 mm. The sandwich plates have a thickness of 0.6 mm. Second type is a octahedral lattice with strut thickness 0.5 mm and 10 × 12 × 12 mm. Cross sectional areas for stress calculation are 100 and 120 mm2 for the BCC and octahedral lattice respectively. Each lattice was loaded with 1 mm/min and unloaded with 5 mm/min for four different loads. The BCC lattice was loaded with 200, 400, 600 and 800 N and the octaherdral lattice with 400, 800, 2600 and 3600 N.

3 3.1

Results and Discussion Demonstrators Built with Powder Bed Fusion

In Fig. 1 various demonstrator geometries are manufactured to investigate influences of cross section variations on the powder bed fusion processing of NiTi and to show the capability of potential applications. All geometries could be manufactured successfully. The use of a titanium plate resulted in delamination of some parts, which can be seen in Fig. 1. Especially geometries with a high length to width ratio are vulnerable for delamination. Delamination of NiTi parts on titanium buildplates is addressed by several researchers [14–16]. The main factors for delamination can be identified as difference in coefficient of thermal expansion and coefficient of thermal conductivity of NiTi and titanium. The coefficient of thermal expansion of NiTi is 11.0 * 10−6 1/K compared to 8.6 * 10−6 1/K of titanium. Coefficient of thermal conductivity is 8.6 W/mK for NiTi and 17 W/mK for titanium. As a result, heat and stresses

228

R. Weber et al.

Fig. 1. Titanium baseplate was used for this built. a) BCC sandwich lattice with strut thickness of 0.2 mm b) Octahedral lattice with strut thickness of 0.5 mm c) spring structure d) Lissajous curve with 0.5 mm strut thickness e) spider like structure (CAD adapted from [13]).

are accumulated in the NiTi parts during manufacturing leading to delamination. Besides, Scheitler et al. observed a brittle Ti2 Ni between the NiTi part and buildplate which leads to a weak breaking point. An effective countermeasure is proposed by using a lattice structure underneath the parts. When designing a proper lattice structure there needs to be a tradeoff in minimum contact cross section of lattice and buildplate as well as stiffness of the overall lattice structure. The lattice contact points to the buildplate need to be large enough to ensure proper bonding and force transmission into the buildplate from the bending of the parts above. Using standard block support with single scanline wall thickness resulted also in delamination. An octahedral lattice with strut thickness 0.5 mm proposed from McCue et al. is used [14]. The contact cross section to the buildplate is circular with diameter of 0.5 mm. However, in this study the maximum height of 4 mm suggested from McCue et al. is reduced to 1.6 mm to increase productivity and reduce material waste. As shown in Fig. 1 delamination could not be eliminated completely with the shortened support. Reason could be the increased stiffness of the support lattice vertical to the buildplate. Stress accumulation therefore could not be fully compensated. Finally, a trade off in the usage of NiTi and titanium plates need to be considered. Titanium plates are more economic, however delamination occurs which could be eliminated by NiTi plates.

Potential Applications for NiTi Manufactured with PBF-LB/M

3.2

229

Shape Memory Effect Demonstration

NiTi material has an unique ability to recover its initial shape after deformation and subsequently heating to a specific activation temperature. The underlying principle for the shape recovery is a diffusionless transformation from martensite (B19‘) to austenite (B2) and back [17]. The microstructure at a certain temperature defines the final functional property. Martensite leads to shape memory properties whereas austenite results in superelasticity. The initial microstructure is twinned martensite. Mechanical deformation leads to detwinning of the microstructure. This detwinned martensite is transformed into austenite when heated and shape recovery is initiated. Large strains can be achieved which is not the case in the cool down phase. Cooling is characterized by transformation from austenite to detwinned martensite without the release of significant strain compared to the heating transformation. In Fig. 2 the shape recovery is demonstrated for diverse structures. Apparently, the shape recovery is not lost after the processing with powder bed fusion. The experiments are conducted at room temperature and for heating a heat gun or gas burner is used. After each shape recovery cycle there are two types of strains which need to be distinguished: recoverable and irrecoverable strain [18]. Recoverable strain consists of an elastic and shape recovery strain. Leftover is irreversible strain which cannot be recovered. In Fig. 2 a) irreversible strain can be visualized by comparing the first and last sequence. The initial shape could not be fully recovered leading

Fig. 2. Demonstration and visualization of shape memory effect with a) rod and b) spider structure (CAD adapted from [13]) and c) thin lattice (CAD adapted from [22]) additively manufactured.

230

R. Weber et al.

to a slight deflection in bending direction. Shape recovery strain can be seen from second last to last sequence. Elastic strain is not shown here but occurs after release of the bending force resulting in rod shape of the second sequence. Depending on the design of the structure, geometrical influences can be exploited to gain high deformation ratios. Powder bed fusion with its capability to produce complex formed structures has great potential to realize designs which cannot be manufactured with conventional methods. For NiTi processing this results in a broadening of actuator applications which were not considered yet. For instance, small and compact actuators with high strains can be implemented in microdevices. But also medium sized actuators to replace hydraulic and electric actuators in aerospace are subject of interest. For heating of the NiTi material joule heating is mostly used [19–21]. Main issue remains the low heat conductivity of NiTi resulting in slow cooling rates. From an engineering perspective for actuation, elastic and irreversible strain shall be minimized whereas shape recovery strain shall be maximized. Researcher should focus in optimizing these parameters through powder bed fusion where as design engineers need to consider these strains to create NiTi actuators. 3.3

Two Way Shape Memory Effect (TWSME)

The two way shape memory effect (TWSME) is a less known functional property of NiTi alloys. In contrast to the one way shape memory effect (OWSME) actuation strain is realized in both heating and cooling phases enabling the possibility for cycling actuation. Prior to heating and cooling the TWSME first needs to be programmed into the material. This is done by proper mechanical and thermal training [8]. Therefore, the NiTi structure is mechanical restrained e.g., with a fixture. This shape will be the cold state later. In our study the additively manufactured NiTi rod is inserted into a metallic loop which can be tightened with a screwdriver enabling increase of mechanical strain onto the rod resulting in shape deformation (see Fig. 3). Afterwards the fixed NiTi rod is exposed to several heating and cooling cycles. This was implemented by heating in a water bath and cooling with pressurized air. This procedure is repeated 18 times. The resulting shape can be seen in Fig. 3 in the starting sequence. The tip of the rod undergoes a total displacement of 4 mm after heating, which can be read from the scale in the background. During cooldown under natural convection to not influence the experimental set up with high pressure air flow, a total displacement of 0.5 mm in respect to the hot state is achieved. The extreme loss of the original shape after training and cooling could not be fully explained. Degradation of the TWSME could be an indicator especially in the first actuation cycles [7]. The degradation of the TWSME is a common issue and is highly dependent on the training cycles and training stress induced. Although a high enough training cycle number is chosen near 20 cycles the training stress was not quantified. During the second cycle heating is resulting in 0.25 mm tip movement and cooling in 0.25 mm back to the initial position of cycle two. No severe degradation could be visually observed. In the third cycle the tip moves

Potential Applications for NiTi Manufactured with PBF-LB/M

231

again by 0.25 mm while heating but only