10th International Conference on Turbochargers and Turbocharging: 15-16 May 2012, Savoy Place, London 9780857092090, 085709209X


277 77 32MB

English Pages 439 [379] Year 2012

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

10th International Conference on Turbochargers and Turbocharging: 15-16 May 2012, Savoy Place, London
 9780857092090, 085709209X

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

10th International Conference on Turbochargers and Turbocharging

Combustion Engines & Fuels Group Organising Committee Dr Kian Banisoleiman

Lloyd’s Register (Chairman)

Dr Roland Baar

Technische Universität Berlin

Andrew Banks

Ricardo

Steve Birnie

BorgWarner

Dr Chris Brace

University of Bath

Dr Geoff Capon

Ford

Dr Ennio Codan

ABB

Gavin Donkin

Honeywell

Dr-Ing Dietmar Filsinger

IHI Charging Systems Intl.

Pierre French

Cummins Turbo Technologies

Dr Seiichi Ibaraki

Mitsubishi Heavy Industries (MHI)

Per-Inge Larson

Scania

Dr Ricardo Martinez-Botas

Imperial College London

Takashi Otobe

Honda R&D

Alexander Rippl

MAN Diesel & Turbo

Prof Joerg Seume

Hanover University

Dr Les Smith

Jaguar Land Rover

Dr Mahmoud Tarabad

Caterpillar

The Committee would like to thank the following supporters: Gas Turbine Society of Japan (GTSJ) and SAE Japan

10th International Conference on Turbochargers and Turbocharging

15–16 MAY 2012 SAVOY PLACE, LONDON

Conference Proceedings sponsored by:

Oxford Cambridge Philadelphia New Delhi

Published by Woodhead Publishing Limited 80 High Street, Sawston, Cambridge CB22 3HJ, UK www.woodheadpublishing.com Woodhead Publishing, 1518 Walnut Street, Suite 1100, Philadelphia, PA 19102-3406, USA Woodhead Publishing India Private Limited, G-2, Vardaan House, 7/28 Ansari Road, Daryaganj, New Delhi – 110002, India www.woodheadpublishingindia.com

First published 2012, Woodhead Publishing Limited © The author(s) and/or their employer(s) unless otherwise stated, 2012 The authors have asserted their moral rights.

This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. Reasonable efforts have been made to publish reliable data and information, but the authors and the publisher cannot assume responsibility for the validity of all materials. Neither the authors nor the publisher, nor anyone else associated with this publication, shall be liable for any loss, damage or liability directly or indirectly caused or alleged to be caused by this book. Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming and recording, or by any information storage or retrieval system, without permission in writing from Woodhead Publishing Limited. The consent of Woodhead Publishing Limited does not extend to copying for general distribution, for promotion, for creating new works, or for resale. Specific permission must be obtained in writing from Woodhead Publishing Limited for such copying.

Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation, without intent to infringe.

British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library.

Library of Congress Control Number: 2012935885

ISBN 978-0-85709-209-0 (print) ISBN 978-0-85709-613-5 (online)

Produced from electronic copy supplied by authors. Printed in the UK and USA.

CONTENTS

SESSION 1: NOVEL APPLICATIONS C1340/040

HyBoost – An intelligently electrified optimised downsized

3

gasoline engine concept J King, M Heaney, E Bower, N Jackson, N Owen, Ricardo; J Saward, A Fraser, Ford Motor Company; G Morris, P Bloore, Controlled Power Technologies, UK; T Cheng, J Borges-Alejo, M Criddle, Valeo, France C1340/072

Turbo-Discharging for improved engine torque and fuel

15

economy A M Williams, A T Baker, C P Garner, Loughborough University, UK

SESSION 2A: HIGH-BOOST AND TWO-STAGE SYSTEMS C1340/073

Boost system selection for a heavily downsized spark

27

ignition prototype engine C Copeland, University of Bristol; R Martinez-Botas, Imperial College London; J Turner, R Pearson, N Luard, Lotus Engineering; C Carey, S Richardson, Jaguar Cars Limited; P Di Martino, P Chobola, Honeywell Turbo Technologies, UK C1340/069

R2S™ – modelling and consequences for the boost control

43

O Weber, R Christmann, V Gauckler, R Sauerstein, BorgWarner Turbo Systems Engineering GmbH, Germany C1340/031

Application of two stage turbocharging systems on large engines E Codan, T Huber, ABB Turbo Systems Ltd, Switzerland

55

SESSION 2B: TURBINE FATIGUE AND STRUCTURAL C1340/056

A new approach to thermo mechanical fatigue shown on

73

turbocharger housings M Nagode, University of Ljubljana, Slovenia; F Längler, BorgWarner Turbosystems Engineering GmbH; M Hack, LMS Deutschland, Germany C1340/010

On the influence of thermal boundary conditions on the

83

Thermo Mechanical Analysis of turbine housing of a turbocharger C Oberste-Brandenburg, M Gugau, F Kruse, BorgWarner Turbo Systems Engineering GmbH, Germany; K Shoghi, BorgWarner Turbo Systems Ltd, UK C1340/022

Compressor wheel low cycle fatigue calculations for off

97

highway applications – an approach to accurately calculate application duty cycle K Ohri, IPSD Caterpillar; K Shoghi, BorgWarner Turbo Systems Ltd, UK

SESSION 3A: COMPRESSOR AERO-DESIGN OPTIMISATION (CFD) C1340/030

Variable trim compressor – a new approach to variable

111

compressor geometry P Grigoriadis, S Müller, A Benz, M Sens, IAV GmbH, Germany C1340/021

Design optimisation of an impeller with CFD and Meta-

121

Model of optimal Prognosis (MoP) F Frese, Voith Turbo Aufladungssysteme; J Einzinger, ANSYS; J Will, Dynardo, Germany C1340/074

Development of advanced centrifugal compressor for turbocharger, applying control of internal unsteady flow structure M Ebisu, T Shiraishi, I Tomita, Mitsubishi Heavy Industries, Ltd; M Furukawa, Kyushu University, Japan

135

SESSION 3B: ROTORDYNAMICS AND VIBRATIONS C1340/048

Shaft coking resolution using multiple variable bearing

147

system design optimization for commercial vehicle turbochargers A Bhattacharya, M A Hake, T V Barbarie, Honeywell Turbo Technologies, USA; J-M Geoffroy, Honeywell Turbo Technologies, France C1340/075

Advanced rotordynamic simulation of turbochargers using

159

coupled multibody and finite element models M Busch, L Esmaeili, D Lu, B Schweizer, University of Kassel; P Koutsovasilis, U Tomm, BorgWarner Turbo Systems Engineering GmbH, Germany C1340/011

Turbocharger blade vibration: Measurement and validation

173

through laser tip-timing J M Allport, M L Jupp, Cummins Turbo Technologies; A Pezouvanis, G W Janicki, A I Pierończyk, A J Day, P Olley, B Mason, M K Ebrahimi, University of Bradford, UK

SESSION 4A: ADVANCING SIMULATION AND VALIDATION C1340/018

On wide mapping of a mixed flow turbine with regard to

185

compressor heat flows during turbocharger testing B Lüddecke, D Filsinger, IHI Charging Systems International GmbH; M Bargende, Universität Stuttgart, Germany C1340/067

Turbocharger matching methodology for improved exhaust

203

energy recovery A Pesiridis, W S-I W Salim, R F Martinez-Botas, Imperial College London, UK C1340/038

Experimental investigation under unsteady flow conditions on turbocharger compressors for automotive gasoline engines S Marelli, M Capobianco, University of Genoa, Italy

219

SESSION 4B: TURBINE DEVELOPMENT TRENDS C1340/051

Development of a common dual axle VNTTM for single- and

233

two-stage off-highway applications J Wilson, M Avila, P Davies, N Theiss, B Zollinger, Honeywell International, Honeywell Turbo Technologies, USA and France C1340/007

An experimental assessment of the effects of stator vane

243

clearance location on an automotive turbocharger turbine J R Walkingshaw, S W T Spence, D Thornhill, Queen’s University Belfast, UK; J Ehrhard, IHI Charging Systems International GmbH, Germany C1340/055

Testing turbine expanders for high efficiency diesels

257

E Halliwell, Cummins Turbo Technologies, UK

SESSION 5: SYSTEMS AND TRANSIENT RESPONSE C1340/057

The role of turbocompound in the era of emissions reduction

269

R W Kruiswyk, Caterpillar Inc., USA C1340/061

Characterization of a low pressure turbine for turbocompounding

281

applications in a mild-hybrid gasoline engine A M I Bin Mamat, A Romagnoli, R F Martinez-Botas, Imperial College London, UK C1340/019

The transient response of turbocharger turbines

295

H Chen, Honeywell UK Ltd, UK; T Cai, P Li, Honeywell Integrated Technology (China) Ltd, China

SESSION 6: COMPRESSOR DEVELOPMENT TRENDS C1340/063

Performance of a small-size two-stage centrifugal compressor

307

R Numakura, IHI Corporation, Japan C1340/024

Effect of diffuser width and tip clearance on the static pressure distributions in a vaneless diffuser of a high-speed centrifugal compressor A Jaatinen, A Grönman, T Turunen-Saaresti, Lappeenranta University of Technology, Finland

319

C1340/032

Experimental and numerical analysis of a classical bleed slot

325

system for a turbocharger compressor S Sivagnanasundaram, S Spence, J Early, Queen’s University of Belfast; B Nikpour, Cummins Turbo Technologies, UK

SESSION 7: NOVEL PRESSURE CHARGING AND HIGH EFFICIENCY SYSTEMS C1340/052

Dual boost compressor development

345

V-M Lei, M Nejedly, V Houst, V Kares, Honeywell Turbo Technologies, UK C1340/002

Electrically driven supercharger using the TurboClaw®

357

compressor for engine downsizing K R Pullen, City University London; S Etemad, W Thornton, Dynamic Boosting Systems Limited; J Villegas, AVL Powertrain, UK C1340/064

Development of new turbocharger technologies for energy

365

efficiency and low emissions Y Ono, K Shiraishi, K Sakamoto, Y Ito, Mitsubishi Heavy Industries, Ltd, Japan

SESSION 8: TURBINE UNSTEADY FLOW C1340/053

Experimental investigation on the effect of pulsations on

377

exhaust manifold-related flows aiming at improved efficiency A Kalpakli, R Örlü, N Tillmark, P H Alfredsson, Royal Institute of Technology, Sweden C1340/068

A comparison of timescales within a pulsed flow turbocharger

389

turbine C D Copeland, University of Bristol; P Newton, R F Martinez-Botas, Imperial College London, UK; M Seiler, ABB Turbo Systems Ltd, Switzerland C1340/020

Experimental analysis of turbocharger interaction with a pulsatile flow through time-resolved flow measurements upstream and downstream of the turbine F Laurantzon, R Örlü, A Segalini, N Tillmark, P H Alfredsson, Royal Institute of Technology, Sweden

405

ADDITIONAL PAPERS NOT PRESENTED AT THE CONFERENCE C1340/013

Reduced model for the radial turbine based on proper

419

orthogonal decomposition N Winkler, E Alenius, L Fuchs, KTH – The Royal Institute of Technology, Sweden C1340/015

Unsteady flow-based aeroacoustics analysis in centrifugal compressor M Qi, L Cao, Beijing University of Technology; L Hu, J Zhang, National Key Laboratory of Diesel Engine Turbocharging Technology, China

AUTHOR INDEX

429

Your Global Technology Partner Leading turbocharger innovation to meet emissions and power challenges worldwide, Cummins Turbo Technologies offer a complete line of world-class products and technologies. ■ ■ ■ ■ ■ ■

Patented Variable Geometry Technology (Holset VGT™) Two-Stage Turbocharger systems Turbocompound systems Waste Heat Expander Wastegated Turbocharging Robust and durable Fixed Geometry Turbochargers

Cummins Turbo Technologies remain the largest specialist turbocharging engineering centre of excellence in the UK, leading the global drive for better engine performance with improved fuel economy, thermal efficiency, and industry-leading emissions control systems.

AUTHOR INDEX

Alenius, E ......................................................... 419 Alfredsson, P H .......................................... 377, 405 Allport, J M ........................................................ 173 Avila, M ............................................................ 233 Baker, A T .......................................................... 15 Barbarie, T V ..................................................... 147 Bargende, M ...................................................... 185 Benz, A ............................................................. 111 Bhattacharya, A ................................................. 147 Bin Mamat, A M I ............................................... 281 Bloore, P .............................................................. 3 Borges-Alejo, J ..................................................... 3 Bower, E .............................................................. 3 Busch, M ........................................................... 159 Cai, T ............................................................... 295 Cao, L .............................................................. 429 Capobianco, M ................................................... 219 Carey, C ............................................................ 27 Chen, H ............................................................ 295 Cheng, T .............................................................. 3 Chobola, P ......................................................... 27 Christmann, R .................................................... 43 Codan, E ............................................................ 55 Copeland, C D ............................................. 27, 389 Criddle, M ............................................................ 3 Davies, P .......................................................... 233 Day, A J ............................................................ 173 Di Martino, P ...................................................... 27

Early, J ............................................................. 325 Ebisu, M ........................................................... 135 Ebrahimi, M K .................................................... 173 Ehrhard, J ......................................................... 243 Einzinger, J ....................................................... 121 Esmaeili, L ........................................................ 159 Etemad, S ......................................................... 357 Filsinger, D ....................................................... 185 Fraser, A .............................................................. 3 Frese, F ............................................................ 121 Fuchs, L ............................................................ 419 Furukawa, M ..................................................... 135 Garner, C P ........................................................ 15 Gauckler, V ........................................................ 43 Geoffroy, J-M .................................................... 147 Grigoriadis, P .................................................... 111 Grönman, A ...................................................... 319 Gugau, M ........................................................... 83 Hack, M ............................................................. 73 Hake, M A ......................................................... 147 Halliwell, E ........................................................ 257 Heaney, M ........................................................... 3 Houst, V ........................................................... 345 Hu, L ................................................................ 429 Huber, T ............................................................ 55 Ito, Y ................................................................ 365 Jaatinen, A ........................................................ 319 Jackson, N ........................................................... 3 Janicki, G W ...................................................... 173 Jupp, M L .......................................................... 173 Kalpakli, A ........................................................ 377 Kares, V ........................................................... 345 King, J ................................................................. 3

Koutsovasilis, P ................................................. 159 Kruiswyk, R W ................................................... 269 Kruse, F ............................................................. 83 Längler, F .......................................................... 73 Laurantzon, F .................................................... 405 Lei, V-M ............................................................ 345 Li, P ................................................................. 295 Lu, D ................................................................ 159 Luard, N ............................................................ 27 Lüddecke, B ...................................................... 185 Marelli, S .......................................................... 219 Martinez-Botas, R F ........................ 27, 203, 281, 389 Mason, B .......................................................... 173 Morris, G ............................................................. 3 Müller, S ........................................................... 111 Nagode, M ......................................................... 73 Nejedly, M ........................................................ 345 Newton, P ......................................................... 389 Nikpour, B ........................................................ 325 Numakura, R ..................................................... 307 Oberste-Brandenburg, C ...................................... 83 Ohri, K .............................................................. 97 Olley, P ............................................................. 173 Ono, Y .............................................................. 365 Örlü, R ..................................................... 377, 405 Owen, N .............................................................. 3 Pearson, R ......................................................... 27 Pesiridis, A ........................................................ 203 Pezouvanis, A .................................................... 173 Pierończyk, A I .................................................. 173 Pullen, K R ........................................................ 357 Qi, M ................................................................ 429

Richardson, S ..................................................... 27 Romagnoli, A ..................................................... 281 Sakamoto, K ..................................................... 365 Salim, W S-I W .................................................. 203 Sauerstein, R ..................................................... 43 Saward, J ............................................................. 3 Schweizer, B ..................................................... 159 Segalini, A ........................................................ 405 Seiler, M ........................................................... 389 Sens, M ............................................................ 111 Shiraishi, K ....................................................... 365 Shiraishi, T ........................................................ 135 Shoghi, K ...................................................... 83, 97 Sivagnanasundaram, S ....................................... 325 Spence, S W T ........................................... 243, 325 Theiss, N .......................................................... 233 Thornhill, D ....................................................... 243 Thornton, W ...................................................... 357 Tillmark, N ................................................ 377, 405 Tomita, I .......................................................... 135 Tomm, U .......................................................... 159 Turner, J ............................................................ 27 Turunen-Saaresti, T ........................................... 319 Villegas, J ......................................................... 357 Walkingshaw, J R ............................................... 243 Weber, O ........................................................... 43 Will, J ............................................................... 121 Williams, A M ..................................................... 15 Wilson, J ........................................................... 233 Winkler, N ......................................................... 419 Zhang, J ........................................................... 429 Zollinger, B ....................................................... 233

HyBoost – An intelligently electrified optimised downsized gasoline engine concept J King, M Heaney, E Bower, N Jackson, N Owen Ricardo, UK J Saward, A Fraser Ford Motor Company, UK G Morris, P Bloore Controlled Power Technologies, UK T Cheng, J Borges-Alejo, M Criddle Valeo, France

ABSTRACT The UK Technology Strategy Board (TSB) sponsored HyBoost project was a collaborative research programme to develop an ultra efficient optimised gasoline engine concept with “Intelligent Electrification”. The basis of the concept was use of a highly downsized 1.0L boosted engine in conjunction with relatively low cost synergistic ‘12+X’ Volt electrical management system and electrical supercharger technologies to deliver better value CO2 reduction than a full hybrid vehicle. Project targets of 99 g/km CO2 as measured over the European Drive Cycle (EDC) in a standard 2011 Ford Focus whilst maintaining the same performance and driveability attributes as a 2009 production 2.0L version of the car were achieved, and a potential route through to 7% higher than the standard engine configuration (shown in Figure 4). This corresponds to significant reductions of in-cylinder trapped residual mass and therefore will result in reduced cooled EGR demand and potential for secondary fuel economy benefits through, for example, spark advance. At idle (850 rpm, 0 bar BMEP) the average exhaust manifold pressure was measured at 988.7 mbar. This indicated that the exhaust flow was not significantly restricted under idle conditions. Closer inspection of transient cycle pressures both in the cylinder and in the exhaust manifolds (Figure 5) showed the blowdown pulse recovery through the turbine and the associated sub-atmospheric turbine outlet pressure. This translates noticeably into the cylinder with opportunities to further improve pumping work by optimisation of the valve timings. This figure also

Figure 5 – Measured transient preturbine, post turbine and in-cylinder pressures at 3000 rpm, full load

19

shows that a significant amount of the blowdown energy is being lost across the valve. This could be reduced by reducing the high pressure manifold volume. Reducing the turbine size may reduce the losses across the valve, however, this would restrict the flow and reduce the blowdown flow through the turbine and therefore is expected to have only a small effect on the amount of energy recovered. At lighter loads and speeds where there is less blowdown pressure and more time for the blowdown event it may prove beneficial to operate with a smaller turbine. Figure 6 shows the compressor test points from across the engine speed and load range plotted against the compressor steady flow map data. It can be seen that using non-ideal turbo-machines (due to availability) caused the majority of operation to occur at isentropic efficiencies of 60-70%. With improved turbine energy recovery these points would tend towards the optimum efficiency of this turbo-machine. A slightly larger compressor could offer some efficiency advantages by using a more optimum operating point and at a lower speed (therefore, reduced friction).

Figure 6 – Compressor operating points shown on the steady flow compressor map

These first experimental results of a Turbo-Discharging system are very encouraging. Conventional turbocharger hardware has demonstrated the ability to significantly depressurise the exhaust system thereby resulting in reduced pumping work and increased peak engine torque. These results have been used to help validate a Turbo-Discharged engine model that can be optimised for performance prior to further experimental testing. Sections 5 and 6 of this paper discuss where performance improvements will be made and how much improvement can be expected.

5. SYSTEM PERFORMANCE There are a number of developments to the experimental set-up that would improve the performance of a production Turbo-Discharging system resulting in even better depressurisation and, therefore, further fuel economy benefit. These are: 1. Improved sealing between the high and low pressure ports thereby increasing the effectiveness of the pulse energy recovery. The presented hardware set-up utilised port dividers with only a contact seal. Differences in temperature and thermal expansion between the cylinder head and the divider will lead to some leakage between the ports. 2. Reduced heat exchanger pressure drop. The presented system utilised a much longer coolant path which was rated to allow the engine exhaust flow to be cooled to 100 kW litre-1 are possible despite the reduced valve lift. This will be explored further in future studies.

7. CONCLUSIONS Turbo-Discharging is a new approach to maximising the fuel economy and performance benefits of turbine recovered energy from an IC engine exhaust gas flow. It achieves the benefits by discharging (i.e. reducing the pressure of) the engine exhaust system thereby reducing pumping work and potentially extending the engine’s knock limit. This paper presents the first experimental testing of a Turbo-Discharging system with the aim of validating existing modelling methods. These early results are encouraging, showing that: 1. 2.

3.

22

Significant levels of depressurisation can be achieved with a small amount of hardware and negligible changes to the engine combustion system or core architecture. Engine breathing is significantly improved, indicated by >7% torque improvements on a naturally aspirated engine. This effect is expected to be primarily due to reduction of in-cylinder residuals resulting from significantly lower cylinder pressure at the time the exhaust valves closes. When using the chosen small automotive size turbocharger as a TurboDischarger, the operation of the compressor was primarily in a low pressure ratio region. Production systems will recover more energy from the blowdown pulse (for reasons described in Section 5) which will move compressor operation into a more efficient region of the compressor map. Further advantages may be achieved by operating with a larger, slower compressor with proportionately less friction losses.

Overall, these first experimental tests of the system, although not intended to demonstrate a production ready technology, have shown that the concept is feasible and has the potential to significantly reduce in-cylinder pressure during the exhaust stroke and, therefore, enhance both performance and fuel economy of IC engines.

ACKNOWLEDGEMENTS The authors would like to acknowledge the support of the Engineering and Physical Sciences Research Council (EPSRC) and Technology Strategy Board (TSB) for supporting this research as part of the Low Carbon Vehicles Integrated Delivery Programme (Grant EP/H050353/1); the Royal Academy of Engineering; and the dedication and support of Adrian Broster, Steve Horner and Graham Smith.

REFERENCES (1) (2) (3) (4) (5) (6)

M. Alamgir, A.M. Sastry, “Efficient Batteries for Transportation Applications”, Society of Automotive Engineers, SAE Technical Paper 2008-21-0017, 2008. Foresight Vehicle for the Department of Trade and Industry, “Foresight Vehicle Technology Roadmap: Technology and Research Directions for Future Road Vehicles”, Version 3.0, 2008. S.M. Shahed and K-H. Bauer, “Parametric Studies of the Impact of Turbocharging on Gasoline Engine Downsizing”, Society of Automotive Engineering, SAE Technical Paper 2009-01-1472, 2009. A.M. Williams, A.T. Baker, C.P. Garner, “Turbo-Discharging: Predicted Improvements in Engine Fuel Economy and Performance”, Society of Automotive Engineers, SAE Technical Paper 2011-01-0371, 2011. C.E. Möller, P. Johansson, B. Grandin, F. Lindström, “Divided Exhaust Period – A Gas Exchange System for Turbocharged SI Engines”, Society of Automotive Engineers, SAE Technical Paper 2005-01-1150, 2005. R.C. Griffith, S.E. Slaughter, P.E. Mavrosakis, “Applying Ball Bearings to the Series Turbochargers for the Caterpillar Heavy-Duty On-Highway Truck Engines”, Society of Automotive Engineers, SAE Technical Paper 2007-014235, 2007.

23

Application of two stage turbocharging systems on large engines E Codan, T Huber ABB Turbo Systems Ltd, Switzerland

ABSTRACT The first applications of 2-stage turbocharging on large engines are already in operation (1), (2). The main drivers for the introduction of this technology are power density, engine efficiency and low emissions. In this paper some aspects and requirements of engine and turbocharging system design are discussed. Furthermore, some results of detailed studies for the different engine segments are presented. Engine size, operational envelope and performance requirements influence both the design parameters of the system and its development potential. Emissions represent a further challenge for large engines; the contributions of the turbocharging system are discussed.

NOMENCLATURE p pc pmax pme pmi T, t V Vd V298

  C 

1

Pressure (Pa, bar) Compression pressure (bar) Firing, maximum pressure (bar) Brake mean effective pressure (bar) Indicated mean effective pressure (bar) Temperature (K, °C) Volume (m3) Displacement (m3) Reduced volume flow (m3/s) Compression ratio Efficiency Mass ratio of trapped to stoichiometric air Pressure ratio

Subscripts ac Start of compression ex End of expansion HD Closed (high pressure) cycle LWth theoretical gas exchange rec charge air receiver red Reduced TI Turbine inlet Abbreviations BDC Bottom dead centre C Compressor CB Constant boundary CP Constant pressure EVO Exhaust valve open HP High pressure IVC Inlet valve close LP Low pressure T Turbine TDC Top dead centre VOL Volume

THERMODYNAMIC ASPECTS OF THE TURBOCHARGED ENGINE

High pressure turbocharging in combination with the Miller process has recently been intensively investigated. Several studies have been performed by means of advanced simulation tools and confirmed by engine tests (3) (4) (5) (6). These studies show that there is substantial improvement potential for engine

__________________________ © ABB Turbo Systems Ltd, 2012

55

performance by applying 2-stage turbocharging in combination with Miller timing working at a boost pressure in the range of 8 bar and higher. Such a process relies on the reduction of the process temperatures offered by the Miller process. This temperature reduction has a very positive effect on the engine’s efficiency and its raw NOx-emissions. At the same time, the high turbocharging efficiency achievable with 2-stage turbocharging allows the delivery of the required air pressure in an efficient way and additional power due to the improvement in gas exchange work. The above mentioned studies can be seen as a conventional approach to applying 2-stage turbocharging to further developed engines. Another approach is presented here, aimed at analysing and optimising the complete thermodynamic process of a 2-stage turbocharged engine. For this scope, the process is divided into three main parts: closed cycle; exhausting gas to the turbines; and charging air from the compressors. Additionally, some considerations are necessary regarding scavenging the combustion space and early closure of the inlet valve (Miller timing). 1.1 The closed cycle The closed cycle is here defined as a process beginning and ending at BDC at the points start of compression (ac) and end of expansion (ex) (Fig. 1). These points are defined in the p–V diagram as extrapolations to BDC by means of an isentropic expansion of the respective points where the inlet valve closes (IVC) and the exhaust valve opens (EVO). In the case of Miller with IVC before BDC, the point ac is the effective start of compression. Taking the closed cycle as the most important part of the process, one obviously must consider parameters like the compression ratio , firing pressure pmax and its ratio to compression pressure pc. The latter plays a very important role in the trade-off between efficiency and NOx-formation. Experience has shown that increasing pmax/pc leads to better efficiency, albeit with higher NOx emissions.

pp

EVO

TDC

V/V D

V/VD

(IVC)

ex ac BDC

Fig. 1: Pressure – volume diagrams of the closed cycle

In a conventional cycle the points ac and ex are only connecting points to the gas exchange cycle. The point ac defines the available charge air per cycle; it is dictated by the required air/fuel ratio. The point ex depends on the engine energy balance; the pressure difference pex – pac correlates directly with the difference between fuel energy and indicated work. With the introduction of the Miller cycle, the point ac acquires an additional degree of freedom. Theoretically the temperature tac could be reduced considerably, but there are limitations imposed by the increasing energetic losses of extreme Miller timing and by the issues of what we call “cold combustion” - it has been observed that below certain pressure and temperature values, especially at part load, diesel combustion assumes a knocking character with low efficiency and high NOx-emissions. It has already been shown (3) that lowering the temperature at point ac reduces the temperature level of the whole closed cycle and this has a very positive impact on the cycle efficiency: at lower temperatures the specific heat of the charge air is reduced, i.e. less fuel energy is required for producing the same pressure increase in the cylinder. At the same time the lower combustion temperature reduces NOx formation. The

56

temperature at ac also has another consequence: a lower pressure pac is required for preserving the air/fuel ratio and the whole compression curve is lower, giving room for increasing the ratio pmax/pc, even at high  and , without going beyond the design limit for firing pressure. The curves in Fig. 2 show the evolution of the closed cycle efficiency over temperature tac. They have been calculated for a model engine in which IVC is varied under the following boundary conditions: the air/fuel ratio C is maintained constant by varying charge air pressure prec, firing pressure has been kept constant by adjusting the start of injection, while the turbocharging efficiency of a 2-stage turbocharging system is approximately constant. The curves show some variability with different values of C, but the general trend can be expressed as a slope of 1% of closed cycle efficiency improvement for 15 °C tac reduction. This result will be used in section 1.3 for evaluating the induction process.

0.52

HD

Lambda 2.2 Lambda 2.0 Lambda 1.8 Lambda 1.8 - trec -20 Linear (Lambda 2.0)

0.51

0.50

0.49

0.48 250

300

Tac [K]

350

400

Fig. 2: Closed cycle efficiency over tac

1.2 The exhaust process 12 According to different engine pex p cycle simulations for a highly 11 [bar] turbocharged engine with Miller 10 timing and an indicated mean 9 effective pressure pmi = 30 bar, 8 values of pex = 11 bar and Tex = 7 1000 K can be considered 6 representative. For the sake of a 1 5 simplicity no scavenging is pTI considered. The task is now to 4 establish the maximum power 3 4 2 that can be gained from the b 3 2 expansion of this exhaust gas. In 1 this phase it is of secondary TDC 0 importance how this power is BDC 0 2 4 6 8 V/Vd finally used. The work of the exhaust process has been Fig. 3: Pressure – volume diagrams idealised as shown in Fig. 3 (7). of the exhaust process To make the following discussion more understandable, two curves have been plotted in the p-V diagram representing: a) an isentropic expansion of the gas in the cylinder to ambient pressure starting from the point ex at BDC b) a curve derived from a) assuming that the gas is collected in a manifold and expansion work outside the cylinder is dissipated. With the help of the curves a and b, different exhaust processes, the ideal pulse and the ideal constant pressure process can be defined.

57

The ideal pulse process relies on using the blow-down energy. The gas expands to ambient pressure with the piston at BDC. This process would require putting the turbine in the position of the exhaust valve. The available energy is the sum of the areas 1 and 2. The constant pressure process relies on filling a volume at an intermediate pressure pTI. The blow-down energy is not converted into turbine expansion work and increases gas enthalpy to the turbine: the isentropic expansion in the diagram starts from the intersection of the pTI line with curve b. The piston has to provide work during the exhaust stroke. The available energy for the turbine is given by the sum of the areas 2, 3 and 4, whereby area 4 is provided by the piston, i.e. available engine power is reduced by this amount.

C1

CB 1

1

2

3

4

5

CB 4

C2

CB 3

A further idea could be to use both systems in parallel. The availability of two turbine stages opens the theoretical possibility of realising an improved pulse process: a first exhaust valve opens during the blow-down phase and is connected via a pulse manifold to the high pressure (HP) turbine, working with energy corresponding to area 1. A second exhaust valve opens after the first one and is connected to an intermediate constant pressure manifold also receiving the gas from the HP turbine outlet. Gas from this manifold is admitted to the low pressure (LP) turbine which operates with the energy given by the areas 2 and 4. The topology of such a system is represented schematically in fig. 4.

6

T1

VOL 2

LP constant pressure manifold

T2

CB 2

HP pulse manifold

Fig. 4: Two-port pulse system A real pulse process is something between the two ideal cases. A fraction of the blow-down energy (area 1) can be utilised; the remaining fraction is converted into increased enthalpy for the steady flow expansion (area 3). Calling x the conversion rate of the blow-down energy, the available turbine energy is given by the sum of the areas 2 and 4 plus x times area 1 and (1-x) times area 3. The evolution of the available power with the different ideal exhaust processes, pulse and constant pressure, starting with pex = 11 bar and tex = 1000 K, is represented versus the expansion ratio T (fig.5a). The turbine isentropic power with pulse (a) starts from a high value at expansion ratio 1 and increases in linear progression with the expansion ratio. The same line (c) for constant pressure starts from zero and reaches the line for pulse at the highest expansion ratio. The middle lines (b rsp. d) represent the net isentropic power, i.e. turbine power minus the piston power needed to expel the gas from the cylinder. The pulse line shows a slight negative slope; and the constant pressure curve shows again an increase from zero to the same final value. The line e is the net effective power of the exhaust process, applying constant overall turbine efficiency. The curve now shows a very flat maximum at an expansion ratio of 7. The value of the curve is practically constant in the range from 6 to 8, which indicates the optimum range for matching the exhaust process of a very efficient engine.

58

c

a

50% engine output

b d e

1

 T Expansion ratio

3

5

7

Net effective turbine work

Turbine work per cycle

CP Isentropic turbine work Pulse Isentropic turbine work CP Net isentropic work Pulse Net isentropic work CP Net effective work

±100 K Tex ±0.05 turbine efficiency

2

9

a) Turbine power versus T

±1 bar pex

 T Expansion ratio

4

6

8

10

b) Sensitivity of turbine power

Fig. 5: Energy balance of the exhaust process A sensitivity analysis has been performed for the curve of effective power e with a constant pressure exhaust system (fig. 5b). The influences are listed as follows: • ± 1% turbine efficiency ± 1.85% net turbine power ± 1.6% net turbine power, ± 1% mass flow rate • ± 1% pex •

± 0.1% net turbine power, ∓ 1% mass flow rate

± 1% Tex

The curve of the effective power for the pulse system is missing since it is difficult to define an efficiency for the conversion of the blow-down energy. Simulations with different configurations have never produced a point above the line for constant pressure. This confirms that the assertion that (quasi) constant pressure systems are the most efficient full load ones for high pressure turbocharging is also valid for 2-stage turbocharging. On the other hand, the power of the pulse system at full load is only marginally lower and part load has not been studied, which leaves the door open for the use of pulse turbocharging for part load optimisation. Simulations with the system shown in Fig. 4 has shown inferior performance in comparison with a conventional pulse system, thus no further steps have been made for its realisation in practice. 1.3

The charging process 9

Miller loss

p

8

prec

[bar]

3

7

Compressors‘ work

3

6 5

intercooling 4

2

2

pac

3

Piston work

1

1 2 1 0 0

TDC

1

BDC

2

3

V/V D

4

Fig. 6: p –V diagram of the charging process

59

Compression process work

After fixing the maximum power that can be extracted from the exhaust gas, b the next step is to find out which is the Compressor work most efficient way to use this power. a The process is described in fig. 6. Area 1 represents the isentropic Isentropic compressor work compression work for the air entering c the cylinder. The intermediate step in Compressor - piston work the compression curve is the effect of d the intercooler between the Compressor - piston work – closed cycle gain compressor stages. Area 2 is the work that would be gained from piston movement under receiver pressure. 4 6 8 10 12 Area 3 is the loss of piston work due to Pressure ratio C the early IVC typical of Miller. Starting from this idealised process, in fig. 7 Fig. 7: Compressor work versus C the different work contributions have been plotted versus the pressure ratio. The first curve (a) represents isentropic compressor work; the second (b) has been derived by applying suitable compressor maps and adapted intercooler temperatures according to water condensation. It can be noted that due to intercooling, the two curves are approximately parallel, i.e. the efficiency of the equivalent single stage compressor increases with the overall pressure ratio and the additional power required for a pressure increase is rather low in the upper range. In order to calculate the net power required for the charge air, the gain in piston work (c) must still be taken into account (d), in addition to the gain in closed cycle work due to the reduction of tac. It can be noted that the curve d is very flat at its start. The plotted curve, obtained by means of an analytical approach, shows a flat minimum at a pressure ratio below 6. An approach by means of engine cycle simulation shows a monotone course for curve 4, i.e. the minimum occurs at pressure ratio 4, as defined by the C requirement. The conclusion of the results of the simulation, optimising both exhaust and charging processes would lead to an engine with an air pressure ratio of 4 (without Miller) and expansion ratio of 7. This high expansion ratio gives enough power to drive the compressors and, additionally, a turbocompound system producing roughly one third of the engine power. According to simulations, the efficiency of such a turbocompound engine shows 8% better values than the base engine with single stage turbocharging. Some drawbacks of this kind of engine would be higher NOx-emissions, exhaust gas temperatures of around 630 °C and mechanical issues of power transmission. 1.4 The scavenging process So far only the exhaust and charging processes for a non-scavenged engine have been considered. Consequently the exhaust gas temperatures are very high. Some scavenging is always present in reality on large engines. The scavenging process requires a positive pressure difference over the cylinder and leads to some efficiency reduction, but also has several advantages. A closer look into this process is thus necessary. Even without valve overlap there is some additional trapped air in the cylinder thanks to the positive pressure difference. This additional air is due to the fact that the residual gas in the clearance volume is compressed from pTI to prec, leaving some volume free for fresh air, which could be considered as some kind of scavenging air. Starting from this mixture of residual gas and air, the influence of increasing valve overlap is analysed.

60

800

Mass flow compressors Mass flow turbines

-20

0

20

40

60

Power

Scavenged gas

600

Scavenged air

Gas from displacement

Trapped air cl. volume

Mass flow

Trapped air displacement

700 tTI [°C]

500

400

Engine power Turbines' power Compressors' power Exhaust gas temperature 80

100

-20

0

40

60

80

200 100

Valve overlap [°CA]

Valve overlap [°CA]

a) Mass balance

20

300

b) Power balance and tTI Fig. 8: Scavenging process

A set of simulations has been performed with constant pressures before and after the cylinder and with increasing valve overlap. The first point is simulated for reference without pressure difference over the cylinders; the other points have a fixed pressure difference of 2 bar, without considering turbocharging energy balance. A zero overlap is usually connected to high flow losses, due to the reduction in valve flow area before and after TDC. To avoid this, the points of maximum valve lift have been fixed and the overlap has been changed, with corresponding variation of the opening and closing ramps. The scavenging mass balance is shown in fig. 8a. After the step with zero overlap, caused by the reduction of pressure in the exhaust manifold, the mass of air introduced to the clearance volume is progressively increased. Up to about 40°CA the corresponding volume of residual gas is scavenged, but no significant mass of air should be present at the exhaust side. The global scavenging factor, defined as the ratio of the total air flow to the amount of trapped air is always 1, but with the overlap of 40°CA, almost 50% of the residual gas is already scavenged. Further increasing valve overlap produces better scavenging but the optimum range for efficiency seems to be between 50 and 60°CA. This is also shown by the curves in Fig. 8b, where engine and turbine power are slightly increased in parallel with the valve overlap. The compressor power curve has a comparable slope up to about 50°CA, after which compressor power increases more steeply, impairing system efficiency. Under the chosen conditions the exhaust gas temperature reaches 500°C at an overlap of 66°CA. Going to even lower temperatures is not conducive to system efficiency. 1.5 The complete engine process So far the whole exercise has been conducted in such a way that the single processes have been analysed separately. In section 1.3 for example, considerations have been made with the degree of freedom that the turbine power must not necessarily be equal to compressor power. Without turbocompounding the air pressure ratio must be increased and the turbine expansion ratio reduced in comparison to the values given in 1.3. The minimum air pressure ratio of 4 would require a very low expansion ratio in the turbines, which would mean very bad utilisation of the exhaust gas energy. Increasing pressure ratio leads to a higher power requirement for the compressor, but better utilisation of the gas energy. This statement will be discussed further by looking into the complete engine process. Applying the previously presented method of analysing the single processes does not necessarily lead to an understanding of the complete process, due to the various and complex interactions between the single processes.

61

Closed cycle

Theoretical pumping work

p pex

Flow losses

prec pTI

Miller loss

pac 0

TDC

0.2

0.4

0.6

0.8

1

BDC

1.2

V/VD

Fig. 9: Complete process analysis Instead of looking at the discharge, charging and scavenging processes, the overall process as characterised by the indicated mean effective pressure (pmi) has been divided in four part processes (fig. 9): • The closed cycle as defined in section 1.1 (pmi,HD) • Theoretical gas exchange, defined as pressure difference over the cylinders multiplied by displacement (pmiLWth) • The Miller loss, which gives the reduction of piston work during the induction process due to early IVC (pmi,Miller) • Flow losses through the valves; (pmi,flow). The flow losses through the valves are calculated from the difference between the work of the piston during the real gas exchange process and the theoretical work represented by pmiLWth minus pmi,Miller.

0.06

Δ [-] 0.04 0.02

Th. pumping work ΔLW Gain closed cycle ΔHD Engine efficiency ΔEff

0.00

Flow losses ΔFlow These terms, expressed in the form -0.02 of efficiency variations, are plotted in fig. 10 over pressure ratio for the -0.04 same engine model as in the Miller losses ΔMiller preceding sections. The efficiency -0.06 values are obtained by multiplying the indicated mean pressure by the -0.08 factor displacement, divided by heat 4 6 8 10 12 Pressure ratio C* input per cycle. This time the Fig. 10: Efficiency contributions pressure ratio has been varied under following conditions: • IVC has been adapted to for a constant air fuel ratio • The start of injection has been adjusted for constant firing pressure • Valve overlap has been adapted for constant exhaust gas temperature • The temperatures of the intercooler and aftercooler have been adapted for a constant margin against water condensation • Turbocharging efficiency remains approximately constant at a level of 75%.

Under these boundary conditions the curve for the possible gain in engine efficiency shows an optimum at pressure ratio of around 7. This is valid for the full load operation of an engine which is optimised with respect to engine efficiency. Additionally, this configuration gives a substantial reduction in the NOx emissions, while exhaust gas temperature remains moderate. Engine efficiency is about 6%

62

(equal to 3 points efficiency in Fig. 10) better than the basic engine. Thus, the efficiency gain of this engine configuration is only 2% lower than that of the optimised turbocompound engine from section 1.3.

2

REQUIREMENTS ON THE TURBOCHARGING SYSTEM

Looking at the design of the 2-stage turbocharging system for 4-stroke engines under the aspect of fuel efficiency only, the target pressure ratio might be around 8. If we take into account the trade-off between efficiency and emissions, as well as the requirement to improve power density and altitude capability, a single figure compression ratio would not be the right solution. There will be a wide range of pressure ratios in the future; even with values above 10. 2.1 Pressure ratio distribution Even if only total pressure ratio has been mentioned up to now, this still has to be broken down into single pressure ratios for each stage. It is well known that turbocharging efficiency defined according to (11) is a function of the pressure distribution between the low pressure stage C,LP and the high pressure stage C,HP. The ratio C,LP/C,HP which gives the maximum efficiency is a function of the intercooling temperature. It is convenient for reason of size to choose a ratio close to two. In this case the loss in global efficiency is marginal but the LP turbine area is smaller (fig. 11).

T

SeffT

Eta-Turbocharging Seff-High Pressure turbine Seff-Low Pressure turbine 0

1

2

3  C,LP/ C,HP 4

Fig. 11: Pressure repartition

By reducing engine load the pressure distribution is changed. The above mentioned diagram is valid for full load operation but the choice of the pressure distribution at full load also has an influence on its evolution over the engine load profile. It has been observed, that the pressure ratio of the HP turbocharger remains fairly constant in the upper load range. Within this range, the change of overall pressure ratio is determined by the low pressure stage only and the HP stage acts as a constant multiplier. According to simulations, for part load operation it is convenient to extend this constant multiplier range as far as possible. This can be achieved by increasing the ratio C,LP/C,HP. An explanation for this behaviour can be found in Figure 12. For a constant overall pressure ratio this diagram shows the behaviour of the reduced turbine mass flow from both stages, plotted over the stage expansion ratio and the overall expansion ratio. The curves are similar to the flow characteristic of a nozzle. The reduced mass flow increases with the expansion ratio up to a maximum defined by the critical expansion ratio, and then remains approximately constant. In this case the critical expansion ratio is defined with reference to 98% of the maximum flow. The critical expansion ratio for the whole system is about 2.8. When the LP stage has also reached its critical pressure ratio, the expansion ratio of the HP turbine can no longer be changed, even though it is well below its critical value. This condition occurs in the example of Fig. 12 at an expansion ratio of about 2.25 for the low pressure stage, which corresponds to an overall expansion ratio between 3 and 4, depending on pressure distribution at full load. The higher the ratio C,LP/C,HP, the earlier the low pressure stage attains its critical expansion ratio and the longer the C,HP remains constant, leading to higher efficiency at part load.

63

2.5 LP turbine mred=f(T,LP)

LP turbine mred=f(T,overall)

Reduced mass flow

2

1.5

1 HP turbine mred=f(T,HP) C,LP / C,HP = 2.6 2-stage_4.4-1.7 C,LP / C,HP = 2.05 2-stage_3.9-1.9 2-stage_3.4-2.2 C,LP / C,HP = 1.55

0.5

0 1

2

HP turbine mred=f(T,overall)

m& red = m& TI ⋅

Expansion ratio T 3

4

⎡ kg K ⎤ ⎢ ⎥ ⎣ s kPa ⎦

TTI pTI 5

6

Fig. 12: Flow characteristics

2.2 Turbocharging efficiency Turbocharging efficiency has always been important for achieving high pressure ratios at convenient exhaust gas temperatures. Two-stage turbo-charging with intercooling ensures a substantial improvement, but the achievable efficiency is even more important than before. This is illustrated in fig. 13, where the achievable pressure difference over the cylinders is plotted versus the total pressure ratio for different turbocharging efficiencies and exhaust gas temperatures. This pressure difference, which is responsible for the theoretical gas exchange work, represents the most important contribution to engine efficiency (see fig. 10). T == 500 75% tTI °C, eta = 75% tTI = 600 °C The line of the Miller losses is 4.0 tTI T == 500 70% °C, eta = 70% plotted in the same diagram. An pRec - pTI engine optimised for efficiency [bar] only should be matched well 3.0 below the point where the curve of the Miller losses intersects the tTI = 550 °C 2.0 Δp curve. Taking the pressure loss over the cylinder and the Miller losses into account at the same tTI = 500 °C 1.0 time, it is logical to stay on the left side of the Miller loss curve. Δpmi, Miller To achieve high pressure ratios it 0.0 is thus mandatory to have high 3 4 5 6 7 8 9 10 11 12 turbocharger efficiencies and Pressure ratio  C exhaust gas temperatures as high Fig. 13: Pressure difference over cylinder as permissible. 2.3 Economic considerations So far only the pressure ratio and efficiency of 2-stage turbocharging have been discussed. Flow capacity plays an important role in the physical dimensions of the 2-stage turbocharging system. The benefits of 2-stage turbocharging can be accessed in a profitable way by using components with high specific capacity. Fig. 14 shows the evolution of ABB/BBC compressor stages for a constant wheel

64

diameter in terms of pressure ratio and flow capacity. It can be seen that the HP compressor stage is designed for a high capacity at moderate pressure ratio as well as ensuring a wide compressor map (fig. 15). Progress in Compressor Performance at Full Load Aluminium Compressor - Exchange interval 50'000 h 6

C

1-stage compressors

2009

2009

2003

2004

5

2003 1996 1996

4

1992

LP-compressor

1978 1989

1970

3

1983

1954-1964 2010

1946

2

HP-compressor 1924

1

3

V298 [m /s]

Fig. 14: Compressor development 3.0

3.0

C tot/tot

C tot/tot

2.0

2.0

*sV

*sV

1.0

. 3 V298 [m /s]

. 3 V298 [m /s]

1.0

Fig. 15: HP Compressor map vs. standard compressor map (green areas: same efficiency) 9

Size 1

Size 2

Size 3

8

Pressure ratio

2.4 Availability First applications of the new ABB product known as Power2 two stage turbocharging have been documented in (1), (2). Two sizes of the first generation Power2 are available; a third size covering the power range from 2500 to 4000 kW will be launched according to market needs (fig. 16).

7

6

5 1000

2000

3000

4000

5000

6000

7000

8000

Turbocharged power per Power2 unit

Fig. 16: Power 2 application ranges

65

3

APPLICATION AND POTENTIAL OF POWER2 IN DIFFERENT ENGINE SEGMENTS

3.1 Engine models Most simulations have been performed for a representative engine, and the results have been applied to different engine sizes. In order to achieve a better understanding of the potential of 2-stage turbocharging, an exhaustive series of simulations have been performed for different diesel engine models: • Model A – Large medium-speed engine • Model B – Small medium-speed engine • Model C – High-speed engine The results are summarized in the trade-off diagram in figure 17.

IMO Tier III ECA limits

bsfc in kg/kWh

Brake specific fuel consumption

All engine models have been 0.22 IMO Tier II limits normalised to constant values of mean effective pressure, compression pressure and 0.21 firing pressure. The reference Model C points are valid for single 0.20 stage turbocharging. The most Model B important differences derive from the cylinder dimensions 0.19 (bore and stroke) and rotational speed. As can be Model A expected, fuel consumption 0.18 increases from A to C due to 0 2 4 6 8 10 12 14 NOx emissions NOx in g/kWh size effects (combustion, heat losses, mechanical losses, Fig. 17: Trade-off efficiency emissions turbocharging efficiency). The turbocharging systems and valve timings have been matched for air/fuel ratios and exhaust gas temperatures that are typical for medium-speed engines in HFO applications (A and B) and for high-speed engine running on distillate oil (C). Additional remarks: • Model A. Engine and turbocharging efficiency are high with moderate scavenging. • Model B. Due to the lower efficiency a higher scavenging ratio is required for to attain the same exhaust gas temperature as in A. This engine model has a larger stroke to bore ratio compared to the others. The comparatively lower speed has a positive effect on the efficiency but increases NOx-emissions. • Model C. This model has a very small scavenging ratio and high exhaust gas temperatures. It has the highest specific fuel consumption and the lowest NOx-emissions. The shaded areas represent the results obtained by applying 2-stage turbocharging and Miller timing under the following boundary conditions: • Pressure ratios ranging from 6 to 12.5 • Constant air/fuel ratio and exhaust gas temperatures by adjusting Miller valve timing and valve overlap • Constant compression and firing pressure by adjusting injection timing and compression ratio • Variation in turbocharging efficiency. For further evaluation a slightly higher exhaust gas temperature has also been considered for models A and B.

66

It can be seen that all the engines have a potential for a simultaneous reduction of specific fuel consumption and NOx-emissions. The difference between the 2 NOxemission levels is approximately constant for all models, while model C can profit from the largest relative reduction. Under the given boundary conditions, the pressure ratio related to highest efficiency is about 7 for the medium-speed engines, whereas for the high-speed engine it amounts to values above 8. This is a consequence of the higher exhaust gas temperature in accordance with Fig. 13. The condition of constant compression and firing pressures keep the areas in a relatively narrow range. Increasing the ratio of firing pressure to compression pressure (not shown in the diagram) could lead to lower fuel consumption within the IMO Tier II NOx limits. Reducing the ratio by means of retarded injection (additional lines in fig. 17) could be used for achieving the lowest possible NOxemissions. These lines give an indication of the possibility of reaching the low emission limits announced in IMO tier III and EPA Tier 4 by using only Miller timing and late injection. According to the applied simulation models, it seems that for the medium-speed engines and especially for engine B, the limits could only be approached in connection with a very high penalty in fuel consumption. With engine C the simulations show that very low emissions levels could be achieved. These considerations are limited to the nominal operation points at full engine load. At part load, the effect of Miller on emissions is substantially reduced, calling for additional emission reduction technologies. But the application of the proposed technology allows for substantial reductions in raw engine emissions and, consequently, reductions in the capacity required from the additional technologies. Emission reduction technologies, typically selective catalytic reduction (SCR) and exhaust gas recirculation (EGR) are considered and have interactions with the turbocharging system. These interactions have been studied (8) and first experiments with the application are ongoing (9). 3.2 System flexibility It is well known that engines with extreme Miller timings require some kind of variability in terms of IVC. A solution is offered by ABB under the name VCM (Valve Control Management) (10). Medium-speed engines are typically used in constant speed applications or in a narrow speed range, e.g. baseload power generation and controllable (CPP) or fixed pitch propeller (FPP) operation. In all cases the operating points with maximum power and torque are met at nominal and maximum rated speeds. For these applications IVC variation is necessary for starting and loading the engine as well as for improving part load boosting for FPP applications. High speed engines can cover a wider speed range and the point with maximum torque is not necessarily that with the highest speed. The potential of Power2 for this kind of application has been studied by means of simulations. As the engine operating curve an FPP curve with 10% torque rise from pme = 22.5 bar at 1900 rpm to pme = 25 bar at 1500 rpm has been considered (Fig. 18). With single stage turbocharging (boost pressure 4.7 bar) this operating curve is extremely challenging without additional control possibilities. At part load the maximum exhaust gas temperature is about 720°C with a very low air/fuel ratio (C = 1.3). It can be noted that with Power2 (boost pressure 9.2 bar), Miller timing, and IVC variability, this very demanding curve can be operated with a much lower exhaust gas temperature and with a higher air/fuel ratio. In addition, fuel consumption is considerably reduced and transient behaviour can be expected to be much better.

67

pme [bar] 30 25 20 15 10 5 0

Mean effective pressure

4

700

Turbine inlet gas temperature

3.5

650

3

600

2.5

550

2

500

C 1.5

Air equivalence ratio 450

1

400

140

200 NOx-Emission

130

150 100

[ ]

120 bsfc 110 [%]

NOx [%]

50 Specific fuel consumption

100

0

90

80 800

tTI [°C]

-50

1000

1200 1400 1600 Engine speed [rpm] bmep [bar] bsfc42.7 1-stage no Miller bsfc42.7_2st 2-stage – Miller – Variable valve timing

1800

-100 2000

Fig. 18: Improved torque capacity with 2-stage turbocharging

CONCLUSIONS A theoretical investigation on the turbocharged engine process has indicated that 2stage turbocharging, extreme Miller timing, and variable inlet valve closure represents a very effective solution for engines with high power density. It is a proven measure for satisfying the requirements of high fuel efficiency in connection with low NOx emissions and high operational flexibility. With Power2 and VCM, ABB offers two of the core components for achieving these aims. Power2 is the product name of ABB’s 2-stage turbocharging concept, consisting at its minimum of an LP- and an HP-stage. The LP-stage is specifically designed for operation with high efficiency and high specific flow capacity. The HP-stage is specifically designed for operation at moderate pressure ratios, high volume flow and high absolute pressure with enhanced map width. VCM (Valve Control Management) is offered as ABB’s solution for variable valve timing. It is an electro-hydraulic system offering wide flexibility for control of valve timing as well as valve lift. Additionally, the potential of this engine concept has been studied for different sizes of 4-stroke diesel engines. It has been shown that a range of engines can be positioned in a global trade-off diagram. Large medium-speed engine offer very good values of specific fuel consumption but high NOx-emissions whereas highspeed engines give higher fuel consumption and lower emissions. After a discussion of the specific characteristics of various engines, the feasibility of the concept for

68

application on a high-speed engine featuring high torque requirements at variable speed has been evaluated.

REFERENCES (1)

Raikio, T., B. Hallbäck & A. Hjort, 2010, Design and first application of a 2stage turbocharging system for a medium-speed diesel engine, 26th CIMAC World Congress in Bergen (N) (2) Haidn, M., J. Klausner, J. Lang & Ch. Trapp, 2010, Zweistufige HochdruckTurboaufladung für Gasmotoren mit hohem Wirkungsgrad, 15. Aufladetechnische Konferenz, Dresden (D). (3) Codan, E. & Mathey, Ch., 2007, Emissions – a new challenge for turbocharging, 25th CIMAC World Congress in Vienna, Austria (4) Codan, E., Mathey, Ch. & Vögeli, S., 2009, Applications and Potentials of 2stage Turbocharging, 14. Aufladetechnische Konferenz, Dresden (D). (5) Codan, E., Mathey, Ch. & Rettig, A., 2010, 2-Stage Turbocharging – Flexibility for Engine Optimisation, 26th CIMAC World Congress in Bergen (N) (6) Millo, F., Gianoglio, M. & Delneri, D., 2010, Combining dual stage turbocharging with extreme Miller timings to achieve NOx emissions reductions in marine diesel engines, 26th CIMAC World Congress in Bergen (N) (7) Watson, N. & Janota, M.S., 1982, Turbocharging the internal combustion engine, MacMillan Press Ltd. (8) Codan, E., S. Bernasconi, & H. Born, 2010, IMO III Emission Regulation: Impact on the Turbocharging System, 26th CIMAC World Congress in Bergen (N) (9) Ruschmeyer, K, Rickert, C. & Schlemmer-Kelling, U., 2011, Potential des Caterpillar MaK 6 M32 C mit zweistufiger Abgasturboaufladung, 16. Aufladetechnische Konferenz, Dresden (D). (10) Mathey, Ch., 2010, Variable Valve Timing – A necessity for future large diesel and gas engines, 26th CIMAC World Congress in Bergen (N) (11) CIMAC, 2007, Turbocharging Efficiencies – Definitions and guidelines for measurement and calculation, Recommendation Nr. 27, Conseil International des Machines à Combustion, Frankfurt am Mein (D), (www.cimac.com)

69

Boost system selection for a heavily downsized spark ignition prototype engine C Copeland1, R Martinez-Botas2, J Turner3, R Pearson3, N Luard3, C Carey4, S Richardson4, P Di Martino5, P Chobola5 1 University of Bristol, UK 2 Imperial College London, UK 3 Lotus Engineering, UK 4 Jaguar Cars Limited, UK 5 Honeywell Turbo Technologies, UK

ABSTRACT The Ultraboost project outlined in this paper seeks to develop a highly pressurecharged, downsized, spark ignition engine that is capable of a 35% reduction in tailpipe CO2 emissions over a naturally aspirated 5.0L V8 while still maintaining performance, emissions and transient response. This project is especially ambitious since, in order to achieve this level of fuel economy improvement, a 60% reduction in engine displacement is targeted with a BMEP of greater than 30bar. What is more, achieving these targets in a gasoline engine with stoichiometric fuelling requires careful design and component selection in order to address the many challenges surrounding gasoline combustion under such high boost pressures. This paper describes a critical examination of the competing requirements of a heavily downsized, gasoline engine with a specific focus on the role of the boosting system in delivering these requirements. From this analysis, the optimal base boosting system configuration is investigated. In addition, a number of boosting technologies, ranging from the novel to the more traditional, are discussed in view of their ability to fulfil a role on the Ultraboost engine. A set of assessment criteria is presented in order to facilitate the selection process. Finally, a 1-D GT-Power model of the Ultraboost engine equipped with the different boosting systems was used to generate an informed rating of all boosting options and permit a reliable comparison with respect to the targets of the project.

1 INTRODUCTION It is well recognized that the performance of an internal combustion engine can be maintained in spite of a significant reduction in displacement by increasing the density of intake charge through supercharging, or ‘boosting’. The main aim of this approach for a gasoline engine is to improve efficiency through the reduction of friction and throttling loss under part load conditions. While the concept itself, commonly referred to as downsizing is not new, the challenge is to push the boundaries of size reduction while mitigating the sacrifice in drivability that is often associated with smaller, pressure-charged engines. Conventional pressure charging technology can be seen in mainstream production for engines up to 25 bar BMEP with some research programs looking to push this envelope to beyond 30 bar [1-3].

_______________________________________ © The author(s) and/or their employer(s), 2012

27

Figure 1 demonstrates how the preliminary BMEP target of the Ultraboost project described in this paper compares with other downsize demonstrator programs. This displays the ambitious nature of this project, especially with the aim to deliver 25bar BMEP at 1000rpm and a specific power of 140kW/L at 6500RPM. This extreme downsizing approach has received increasing attention owing to the potential for a significant improvement in fuel economy and CO2 emissions using commercially available technologies. Moreover, the call to reduce CO2 emissions is expected to increase considerably given the increased climate change awareness, rising costs of fuel, and increasing legislative restriction. These pressures are seen as key to shaping the vehicle powertrain demands in the short to medium term and will insure that downsizing must be part of the strategy.

Figure 1: Ultraboost BMEP target This paper focuses on the selection of the best boosting system for the Ultraboost 2.0L, four-cylinder, gasoline engine that aims to imitate the full load torque and transient response of a modern 5.0L, eight-cylinder while delivering a 35% improvement in fuel economy. A gasoline application is especially challenging due to the wider flow bandwidth that the intake system must deliver in comparison to a diesel engine. Also, since the overriding aim of the Ultraboost project is to reduce CO2 emissions, the boost system must consume a minimal amount of useful energy to generate boost. This therefore suggests some level of exhaust gas energy recovery, which in turn, introduces a new set of challenges due to the impact of a turbine on exhaust backpressure and engine breathability. These types of competing requirements are considered carefully in the work presented in this paper.

2 THE ULTRABOOST PROJECT 2.1 Introduction The Ultraboost Project is a collaborative effort between five industrial partners and three Universities funded by the UK Technology Strategy Board as part of the Low Carbon Vehicle Program. The objective of the project is to develop a highly downsized engine concept that is capable of delivering a dramatic reduction in

28

tailpipe CO2. A comprehensive introduction to the Ultraboost project can be found in reference [4]. From the start, the boosting system was perceived as central to the success of the project owing to the heavy reliance on pressure charging to maintain drivability and performance. The analysis of the pressure charging options and the subsequent system selection for the Ultraboost project was led by Imperial College London in close collaboration with Lotus Engineering and Jaguar Land Rover. Although not a project partner, Honeywell Turbo Technologies also provided close support for turbocharger selection and Table 1: Ultraboost Engine matching. Since the Ultraboost Specifications prototype engine development was concurrent with the boost system selection, the performance of the different options was compared using a GT-Power model constructed by Lotus Engineering. The specifications of the Ultraboost prototype engine are provided in Table 1. 2.2 Project Performance Targets Table 2 outlines the primary upper level targets for the Ultraboost project. The most relevant targets for the boosting system are fuel economy and engine performance. These targets are discussed in brief as follows. Other aims for the project that are not expressed in the table are the lambda () and catalyst heating targets. Table 2: Ultraboost Project Targets

NEDC fuel economy benefit: The primary upper level vehicle target is to achieve a 35% CO2 reduction over the baseline 5.0L NA V8 over the New European Drive Cycle (NEDC). Work by Jaguar Land Rover presented in 2009 [5] indicated that an approximate 23% benefit could be achieved by downsizing and boosting alone. The remainder is expected to come from an improvement in thermal and combustion efficiencies (higher compression ratio) as well as base engine friction optimization [4]. To assess the fuel economy benefit of competing boosting options, the drive cycle performance was simulated by running the GT Power model over a number of steady-state speed and load mini-mapping points. Engine fuel flow at these points was predicted using GT Power and these points were weighted over the NEDC to provide a prediction of fuel economy.

29

Torque and power: The steady-state performance target for the Ultraboost engine is to match the full load 5.0L V8 NA torque and power curve shown in Figure 2. The peak power and torque values are therefore 515Nm (@ 3500 rpm) and 283kW (@ 6600 rpm) respectively. The 2.0L Ultraboost concept displacement points to a peak BMEP of 32bar at 3500 rpm. The most challenging requirement for the boosting system to deliver is to reach 25bar BMEP target at 1000rpm. 600

300

500

250

400

200

300

150

200

100

100

50

0 0

1000

2000

3000

4000

5000

6000

0 7000

Speed [rev/min]

Figure 2: Target power and torque curve

Figure 3: Transient response of AJ133, LionV6 and UB100 at 1250rpm Transient response target: Setting a transient response target of a heavily downsized boosted engine is a difficult task as there is little precedent to serve as a direct comparison. Figure 3 shows transient response curves at 1250rpm for a 5.0L V8 NA engine and a 3.0L twin-turbocharged V6 diesel engine. Data for these two engines were collected from engine dynamometer testing. Also included in Figure 3 are two simulated variants of the 2.0L Ultraboost engine equipped with two stage

30

boosting arrangements. The change in slope of the boosted engine curves clearly shows where the ‘naturally-aspirated’ response ceases and the boost system response dominates. It was agreed that the minimum level of transient response acceptability for the Ultraboost engine would be to achieve 90% of max torque in the same time as the 3.0L V6 Twin turbo charged diesel engine whilst using the 5.0L V8 NA gasoline engine as a stretch target for the project. An additional target was to ensure that the profile of the transient response would produce a response that a customer would find desirable. It should be noted that due to the degree of downsizing, the targets in this project are exceptional. For comparison, the time-to-torque of a single turbocharged downsized petrol engine can be expected to be at least double what has been achieved at 1250rpm. Lambda () target: The Ultraboost project has set out an aim to maintain  =1.0 across whole speed and load range, this requirement has particular relevance in considering turbochargers selection. Catalyst heating target: The lead partner, Jaguar Land Rover set out a specific catalyst heating target in keeping with vehicle emissions target indicated in Table 2. This target was especially important when considering a multistage turbocharger arrangement where turbines are present in the exhaust flow prior to the catalytic converter. 2.3 Modelling Assumptions Although the GT-Power model of the Ultraboost engine is to evolve over the life of the project, it is important to outline the initial assumptions that were made when comparing the different boosting systems. First, the combustion was modelled using a simple Wiebe heat release function without a dedicated knock model. Thus, for the purposes of the boost system selection, it was assumed that the air path was the limiting factor in achieving the target performance, not the combustion system. The charge air coolers for all of the boost systems were assumed to be air-to-air with an effectiveness of approximately 85% and pressure loss below 100mbar. Exhaust gas recirculation was accomplished using a cooled, long route circuit. All full load points were modelled with 10% EGR apart from constant speed transient load steps, which were simulated without EGR.

3 BOOSTING ASSESSMENT 3.1 Assessment Criteria A comprehensive set of assessment criteria were required to judge the advantages and disadvantages of each system of boosting components relative to others. These assessment criteria included: • • • • • • • • • • •

Achieving the level of BMEP to enable extreme downsizing Minimize BSFC (part load and full load) Meeting the low end torque target Transient response Pumping loss In-cylinder residual content Performance at altitude Charge cooling load (linked with compression efficiency) Heat to catalyst Impact on vehicle architecture Cost, Package size and Weight

31

• • •

Control complexity Technological readiness Noise, Vibration and Harshness (NVH)

It was interesting to note the tension between some of these criteria. For example, a supercharger typically suffers from a greater BSFC but this is balanced by good transient response, lower pumping loss, in-cylinder residuals and rapid catalyst warm-up. A turbocharger, on the other hand, recovers waste exhaust gas energy but at the expense of higher pumping loss and possible slower catalyst light-off. Thus, the choice between options with different strengths will depend on the importance placed on the different selection criteria. Table 3: Method for attaching a weighting to each of the assessment criteria IMPORTANCE Vehicle Requirements →

Driveability Target

Fuel Economy

Emissions

Delivery & Packaging

Criteria Weight

BOOSTING SYSTEM ASSESSMENT CRITERIA

Downsize Enabler Bsfc (Part Load) Bsfc (Full Load) Low End Torque Transient Response PMEP Residuals Inter-Cooling Altitude Catalyst Warm Up Vehicle Arch. Impact Cost Package Size Weight Control Complexity Technology Readiness NVH

3.2 Criteria Importance It was vital at the start of the selection procedure to agree with the lead partner on the importance attached to each of the assessment criteria listed above. Although this could have been done by simply picking a series of normalized numbers to represent a weighting, a systematic approach was preferable. Therefore, a process was developed with its roots in the Quality Function Deployment (QFD) methodology. In essence, the process links the 17 boosting system criteria listed above to a set of four vehicle requirements: 1) drivability, 2) fuel economy, 3) emissions and 4) delivery and packaging. The importance of these four vehicle requirements were judged and are represented as shaded cells in the top row of Table 3. Note: shades of colour are used here to maintain a level of confidentiality over the absolute numbers. The values in the body of the matrix were selected to represent the ‘relationship level’ between the vehicle requirements (columns) and the boosting system criteria (rows). Finally, the criterion weighting was calculated by multiplying each number in the row by the importance of the vehicle requirement before adding together.

32

Figure 4: Meeting the target torque curve with a single centrifugal compressor stage

4 BOOSTING SYSTEM 4.1 Full Load Torque Target The full load torque target for Ultraboost shown in Figure 1 is an excellent starting point since it determines the maximum boost pressure that must be delivered over the engine operating range. Although the necessary boost level will depend on a variety of factors (EGR, PMEP, intercooling, etc), it was generally found to vary between 3bar and 3.5bar absolute in most of the simulations. This requirement immediately limits the number of boosting options that are capable of delivering this pressure of the entire flow range of the engine.

33

Considered first was an attempt to meet the torque curve with a single boosting stage. There are two main ways to compress the air charge entering the engine: centrifugal and positive displacement compression. Centrifugal Compressor: A high speed centrifugal compressor can deliver significant boost levels from a single stage. One of the main drawbacks, however, is their limited flow range due to the aerodynamic phenomena of surge and choke at low and high flow rates respectively. This limitation is more pronounced in a gasoline engine due to its wider operating range. Figure 4 demonstrates the difficulty of meeting the full load torque curve with a single turbocharger. A small compressor is able to deliver most of the low rpm torque target without surging, but chokes after 3000 rpm. A large compressor can deliver the rated power at maximum speed without choking, but is surge limited under 3000 rpm. Thus, more than one turbocharging stage is necessary. Positive Displacement Supercharger: These are most commonly driven mechanically from the engine crankshaft pulley. There are a variety of designs (roots, screw, hook and claw, etc) that can deliver high pressure ratios. However, there are technical limitations that make it difficult to achieve boost levels above 2.5bar. Even if a single supercharger were able to deliver the necessary pressure ratio of 3.5bar, it still would fail in the most important target: to maximize the fuel economy. Since a supercharger must remove power from the engine to generate boost, in comparison with a well-matched turbocharger, the supercharger will produce higher BSFC. Considering these findings, it was deemed that the Ultraboost engine therefore requires a minimum of two compressor stages to deliver the target torque. In addition, observing that a supercharger brings a penalty in fuel economy, the primary source of boost must be a turbocharger that is able to reclaim exhaust gas energy to generate boost. 4.2 Two-Stage Layouts Two stages of boosting are most commonly arranged in either parallel or series. Typical layouts are shown in Figure 5 below. TWO-STAGE PARALLEL

TWO-STAGE SERIES

Figure 5: Meeting the target torque with two boosting stages arranged in parallel or series

34

TWO-STAGE PARALLEL

TWO-STAGE SERIES

Figure 6: Meeting the target torque with two boosting stages arranged in parallel or series Two-stage parallel systems: In a parallel arrangement, each compressor delivers half of the total air flow but at the full boost pressure. It is also possible to use the stages sequentially, that is, to bypass one of the turbochargers so that the full air delivery is supplied by a single turbocharger at lower engine rpm. Since each of the turbochargers is smaller than a single, larger turbocharger, this allows the system to generate more boost at lower rpm where less energy is available in the exhaust. The top plot in Figure 6 shows that used sequentially, two smaller turbochargers in parallel are able to meet the torque requirements over a wider range compared to a single, larger turbocharger. Unfortunately, two parallel turbochargers still cannot meet the full low end torque requirement of the Ultraboost engine due to limitations of compressor surge. In addition, since the full pressure ratio must be supplied by each turbocharger individually, this requires small build turbochargers with very high pressure ratio capability (tall and thin

35

compressor maps). This proved difficult since smaller turbocharger compressors are typically less efficient at high pressure ratios, thus leading to prohibitively high compressor outlet temperatures for a standard aluminium wheel. Two-stage series systems: In the series arrangement shown in Figure 5, the boost developed from the low pressure (LP) stage is fed into the high pressure (HP) stage so that the final boost is multiplied. This characteristic is especially useful at low engine rpm where the exhaust energy is low. Thus, although the LP stage is sized to cover the top end, it is also able to contribute enough pressure at low rpm to help the HP stage deliver the torque target. The bottom plot in Figure 6 shows the areas where each of the two stages are designed to operate. At 3000rpm, full load, the HP stage is bypassed and the LP stage works alone to supply the entire boost pressure that is required. Since there is ample exhaust gas energy available in this region, the LP stage can be a turbocharger for the BSFC benefit. In addition, since the region which uses the HP stage is reasonably small, a supercharger can be used without a significant impact on BSFC. In fact, only three of the 15 minimap points used to assess drive cycle fuel economy require both the HP and LP stages. 4.3 Three-Stage Layouts With substantial support from Honeywell Turbo Technologies, three stage systems were extensively reviewed. Once a third stage is added, there are a number of ways that the components can be arranged and, indeed, employed over the engine operating range. Figure 7 shows two of a number of possibilities that were investigated for Ultraboost and discussed below briefly.

THREE STAGE SERIES-PARALLEL

THREE STAGE SERIES

Figure 7: Three stage boosting arrangements Series-parallel three stage: The left plot in Figure 6 shows that two turbochargers arranged in parallel fail to reach the low end torque target. To make up this difference, it is possible to add a third stage in the HP position that can make up this shortfall. Along with the added expense, this adds a considerable level of control complexity as well as concerns of packaging space. In addition, the difficulty with high compressor outlet temperatures explained earlier is still present unless the supercharger is active over the entire rpm range.

36

Three stage series: Although it is possible to meet the target torque curve with two turbocharger stages in series, the HP turbine and compressor need to be very small in relation to the LP stage. Since it is generally accepted that a large gap in turbocharger size can be a cause for concern, it is possible to add a third stage to bridge the gap between LP and HP stages. Since this third stage adds further matching options, a number of strategies for control and sizing of the three turbochargers were investigated within the model. None of these three-stage series matches provided a significant enough performance benefit to justify the added cost and complexity of three turbochargers. 4.4 Summary In conclusion, a considerable time was spent in exploring the best way to arrange two and three stages of boosting components to reach the full load and transient targets. Arranged in parallel, two turbochargers cannot reach the target torque at low engine rpm without the addition of a third stage. Three stages were generally disappointing since the expectation of a significant improvement in performance was not realized. In fact, it was found that in some cases, the three stage options actually performed poorly against a well matched two stage setup. Thus, the best boosting arrangement for the Ultraboost project was found to be a two-stage, series sequential system shown on the right of Figure 5.

5 BOOSTING COMPONENTS 5.1 Low Pressure Stage The discussion in the previous section concluded that a turbocharger is necessary to keep BSFC as low as possible. Thus, it was decided early into the boost selection that the low pressure stage must be a high performance turbocharger with the possibility of additional technical enhancements (variable geometry or twin-entry turbine, ball bearings, etc) as the boosting system was developed. During this initial selection procedure, however, the LP stage was generally treated as a fixed geometry, wastegated turbocharger with a standard aluminium compressor wheel. In consultation with Honeywell Turbo Technologies, a Standard GT30 turbocharger was selected to fill the demanding role of the low pressure stage. Despite the excellent map width of the GT30 compressor, the demand for very high boost levels meant that the compressor was often operating near its limit at full load. The runup line was typically as close as possible to surge to maximize low-rpm contribution. The compressor outlet temperatures were also very close to the 230°C limit when operating at maximum boost above 3000rpm. Finally, at maximum power (6500rpm), the compressor was near to the choking limit. The initial test results on the first prototype engine will allow a matching optimization of the Standard GT30 unit to the meet the project targets more effectively. 5.2 High Pressure Stage With the selection of the low pressure stage fixed, most of the exploration surrounded assessing a wide variety of boosting components for the high pressure stage. Most of those considered are listed as follows with a brief outline of their strengths and weaknesses. Turbocharger with a fixed geometry turbine (FGT) with wastegate: With help from the LP GT30 turbocharger, a small build HP GT14 turbocharger generated enough boost to meet the target torque at low rpm. Full load and part load BSFC were marginally better than other options. The restrictive turbine flow area of the HP build does present some difficulties in the exhaust. Up to 3000 rpm, the small turbine restricts exhaust flow, thereby increasing backpressure, pumping loss and in-cylinder residuals. Shorter period exhaust and intake cams can help lower residuals but at the expense of BSFC. Transient response must rely on the inertia of

37

both turbocharger stages and therefore results in an undesirable concave curve shown in blue in Figure 3. Finally, there is a concern that the large wetted area of two turbines in the exhaust could adversely affect cold start emissions due to slower catalyst light-off. This is of considerable concern with new, tighter tailpipe emissions anticipated. Positive aspects are however, a standard fixed geometry turbocharger is inexpensive, compact and reliable. Turbocharger with a variable geometry turbine (VGT): A VGT is able to vary the flow area by adjusting the opening of the nozzle vanes which feed flow into the turbine rotor. Modelling showed that a GT14 VGT turbine helps to reduce the exhaust backpressure, pumping loss and residuals in comparison to the FGT. Unfortunately, the adjustable vane mechanism is more sensitive to high turbine inlet temperatures typical of a gasoline engine. Applying a VGT to gasoline would either require significant enrichment at the expense of BSFC, or a development program to engineer for higher temperatures. Eaton TVS supercharger with clutched, single speed drive: Eaton has developed the TVS supercharger models to generate peak efficiencies that are close to a turbocharger compressor. The main weakness of any crankshaft driven supercharger is the decrease in fuel economy that results from extracting energy from the crankshaft. However, since the HP stage is used sparingly over the NEDC, there is less than a 1% difference in part load fuel economy over the turbocharger option. Critically, this assumes that the supercharger is completely declutched when not in use. Other performance advantages include: low pumping loss and residuals as well as a desirable transient response shape (Figure 3). The transient response can be further improved at the expense of a small rise in BSFC by increasing the fixed drive ratio between the supercharger and the engine. Eaton TVS supercharger with variable speed drive: The Eaton drive ratio determines the level of boost pressure that the supercharger contributes over its operating range. The ability to tailor the drive ratio via an infinitely variable ratio transmission improves the part load BSFC. Unfortunately, the commercial readiness of a variable speed drive is low as of the date of this research. Rotrex centrifugal supercharger: A Rotrex supercharger mates a centrifugal compressor to a high speed epicyclic traction drive with a ~12:1 step up ratio. With a typical crankshaft pulley ratio of 2.5:1, this therefore allows a maximum crank to compressor shaft ratio of 30:1. Thus, at 1000 engine rpm, this drive ratio limits the compressor speed to 30 krpm. This speed limitation is significant since to generate sufficient boost at low rpm, a small centrifugal compressor must typically spin in excess of 150 krpm. Thus, a centrifugal supercharger was simply unable to supply sufficient boost at low engine rpm. Lontra supercharger: The Lontra device is a positive displacement supercharger with very good reported compression efficiencies. It is also able to achieve very high pressure ratios up to 3.0bar absolute without an appreciable loss in efficiency. It is, however, a very large device that would present a challenge to package effectively. CPT Variable Torque Enhancement System (VTES): Control Power Technologies (CPT), have developed an electric boosting solution that mates a standard radial centrifugal compressor to a high speed electric motor. To ensure that the VTES booster can be used on a standard low voltage vehicle electrical system, CPT offer 12V and 24V motor varieties. Unfortunately, this limits the power available and therefore, the maximum boost to approximately 1.6bar absolute. This device is therefore unable to meet the ambitious low-end BMEP targets of the Ultraboost project.

38

AERISTECH: Aeristech are also developing a radial compressor e-booster. To address the lower boost levels from low voltage devices, their electric motor has a high power rating thus requiring a high voltage supply. While this does solve the problem of boost level, it also means that this boosting solution is only applicable in vehicle applications where a high voltage electrical storage system is available. Technological readiness of this option was also low at the time of selection. IPT SUPERGEN: The SuperGen device is an intriguing new supercharger design that is under development at Integral Powertrain. It offers a unique solution to the difficulties observed with CPT and Aeristech e-boosters. The SuperGen device consists of two, low voltage electric motors, an epicyclic traction drive and an input shaft that is driven off the crankshaft pulley. Owing to its unique mechanical/electrical power split arrangement, the device is able to rotate the compressor wheel at high speeds and deliver enough power to generate high levels of boost from a low voltage electrical supply. It can also operate in a ‘sustained mode’ where the first electric motor generates power to supply to the second electric motor such that the battery is not depleted. The main advantage that this device provides over a more traditional supercharger is the excellent transient response.

6 BOOSTING SYSTEM SCORING The final stage of the selection process was to rank all boost systems using the weighted criteria listed in Table 3. For each of the individual boosting systems, all of the assessment criteria were given a score from one to nine that represents the performance of this system relative to others. The matrix shown in Table 4 is shaded to give a sense of the scoring for each of the boosting systems. The final score is then calculated by multiplying each of the numbers by its corresponding criterion weight before adding across all the criteria under consideration. The top three boosting systems that all scored highly for the Ultraboost project are as follows: TURBO-SUPER: HP Eaton TVS Supercharger + LP Honeywell GT30 FGT Turbocharger The two-stage Turbo-Super is ranked best overall since it provided the best combination of low end torque, transient response and fuel economy using well proven technologies. If a variable speed (CVT) drive is added, this has the potential to improve the BSFC and transient response, making this the most attractive system in view of performance and customer requirements. TWIN-TURBO: HP Honeywell GT14 Turbocharger + LP GT30 FGT Turbocharger The two stage turbocharger rates highly in the overall rating despite its poorer performance due to practical considerations of cost, size, and commercial readiness. When scored using the performance based assessment criteria only, however, its rating was relatively poor against other options.

39

ROOTS + FGT

VGT HP + FGT LP

FGT HP+ FGT LP

VTES + FGT

ROTREX + FGT

LONTRA + FGT

THREE STAGE BOOSTING SYSTEMS

EATON + TWO STAGE PARALLEL VGT

EATON + TWO STAGE PARALLEL FGT

THREE STAGE TURBO: PARALLEL-SERIES

THREE STAGE SERIES FGT

AERISTECH +FGT

ELECTRIC SUPERCHARGER + SUPERGEN + FGT LP TURBOCHARGER

CENTRIFUGAL SUPERCHARGER + LP TURBOCHARGER

TWO-SPEED POSITVE ROOTS + FGT DISPLACEMENT SUPERCHARGER + LP TURBOCHARGER VARIABLE SPEED ROOTS + FGT

TWO STAGE TURBOCHARGING

BOOSTING SYSTEM

IMPORTANCE →

DOWNSIZE ENABLER (BMEP)

BSFC (PART LOAD)

BSFC IMPACT ON LOW END TRANSIENT PUMPING INTER- CATALYST (FULL RESIDUALS ALTITUDE VEHICLE TORQUE RESPONSE LOSS COOLING WARM UP LOAD) ARCHITECTURE

Table 4: Boost System Criteria Scoring Matrix

COST

PACKAGE CONTROL TECHNOLOGY WEIGHT SIZE COMPLEXITY READINESS

NVH

SUPERGEN-TURBO: HP IP SuperGen + LP Honeywell GT30 FGT Turbocharger The Integral Powertrain SuperGen device combined with a LP turbocharger rates highly in the performance criteria. This is largely due to the electric boosting capability which produces the best transient response of all of the options. However, this device is currently in a development phase, and consequently its overall rating (including readiness) is less than the others.

7 CONCLUSIONS This paper provides an outline of the process taken to select the best boosting system for the Ultraboost project. This ambitious project aims to decrease the engine displacement by 60% while maintaining the overall performance. Since this can only be achieved by heavily pressure charging the smaller engine, the selection of the best boosting system is critical to the success of the project. It was first necessary to decide on the component layout that could deliver sufficient boost pressure over the entire engine operating range. Also, since the over-riding aim is to reduce vehicle CO2 emissions, it was clear from the outset that the boosting system must be at least partially driven from energy recovered from the exhaust system. Moreover, since a single turbocharger was found to be insufficient to deliver the full target torque due to limitations of surge and choke, at least two stages were deemed necessary. Due to multiplication of boost pressure, the series boosting layout was able to meet the target torque curve using only two stages while the parallel arrangement required a third stage. The final layout of the base boosting system was therefore clear: a two stage, series boosting arrangement with a LP turbocharger and a flexible solution for the HP stage. A number of different boosting components were assessed in GT-Power and rated according to the set of weighted assessment criteria. The Turbo-Super boosting system with a HP Eaton TVS supercharger and a Honeywell GT30 LP turbocharger was consistently rated the highest providing the best combination of performance and practical attributes. It is this solution that has been recommended to the project for future Ultraboost boosted prototype engines.

REFERENCES [1] Bandel, W, G K Fraidl, P E Kapus, and H Sikinger. 2009. “The Turbocharged GDI Engine: Boosted Synergies for High Fuel Economy Plus Ultra-low Emission”, SAE-2006-01-1266 [2] Lumsden, Grant, Dave Oudenijeweme, Neil Fraser, and Hugh Blaxill. 2009. “Development of a Turbocharged Direct Injection Downsizing Demonstrator Engine”, SAE-2009-01-1503 [3] Fraser, Neil, Hugh Blaxill, Grant Lumsden, and Mike Bassett. 2009. “Challenges for Increased Efficiency Through Gasoline Engine Downsizing”, SAE 2009-011053 [4] Carey, C., McAllister, M., Sandford, M., Richardson, S., Pierson, S., Darnton, N., Bredda, S., Akehurst, S., Brace, C., Turner, J., Pearson, R., Luard, N., Martinez-Botas, R., Copeland, C., Lewis, M., Fernandes, J. 2010, “Extreme Engine Downsizing”, IMECHE: Innovations in Fuel Economy and Sustainable Road Transport. Pune, India. [5] McAllister, M., Buckley, D. 2009. “Future Gasoline Engine Downsizing Technologies – CO2 Improvements and engine design considerations”, IMECHE Internal Combustion Engines: Performance, Fuel and Emissions Conference.

41

R2S™ – modelling and consequences for the boost control O Weber, R Christmann, V Gauckler, R Sauerstein BorgWarner Turbo Systems Engineering GmbH, Germany

ABSTRACT Downsized and down-speeded combustion engines as a main trend achieving in the automotive world the lower CO2 emissions targets require as a consequence higher boost pressure levels. This can be accomplished by using a so-called “Regulated 2-Stage” (R2S™) turbocharging system which consists of two turbochargers in series. In addition with the increasing requirements of the turbocharging system the demands on the control system also increase. According to the increased flexibility the number of actuators rises and the manner how to control the boost pressure changes significantly. Understanding the main influencing parameters of the boost control of a R2S is a key to success achieving a good performance of the boost pressure control. Therefore this paper presents a method of a model based approach for a R2S turbocharging system. As a result the requirements for the actuation and overall sensor system can be defined more precisely. BWTS introduced this model for the development of adequate controlling elements. Keywords: Modelling, Turbocharger, Boost Pressure Control

1

INTRODUCTION

Whereas the turbocharger (TC) previously only made use of its advantages over a limited engine speed range, the more recent developments in combustion engine turbocharging must provide optimal support over the entire engine speed range. This necessitates the employment of complex charging systems (e.g. TC with variable turbine geometry or multistage systems). In this paper a so called Regulated 2-Stage” (R2S™) turbocharging system is discussed. The R2S™ consist of two turbochargers in series which allows a sufficient support of boost pressure over the entire engine speed range. The challenge is to achieve and control an optimal transient behaviour. Based on a physical model of the R2S™ system the influence of different controlling elements is demonstrated. The first part of this paper gives a brief overview of the R2S system and the model approach. In the second part the impact of manifold volumes, inertias of the rotating parts and optimised Turbine Bypass Valve (TBV) designs are discussed.

_______________________________________ © The author(s) and/or their employer(s), 2012

43

2

SYSTEM DESCRIPTION

A larger and smaller turbocharger (TC) are connected in series (see Fig. 1, left side) on a R2S™ turbocharger system. The smaller TC on a two-stage system is referred to as the high-pressure stage (HP-stage) and the larger TC as the low-pressure stage (LP-stage). The designation is formulated for the differences in pressure to which both turbochargers are subjected. Depending on the engine operating point, the exhaust gas mass flow can be distributed to both turbines via the Turbine Bypass Valve (TBV). It remains closed at low engine speeds, and the HP turbine has the entire exhaust energy at its disposal. Accordingly, a faster build up of boost pressure producing a higher dynamic response of the overall system is achieved. The TBV opens further with increasing engine speed and directs a share of the exhaust energy directly to the turbine of the LP turbocharger, which undertakes additional compression. In the full-load range, the TBV is opened completely and the HP turbine is completely bypassed from the exhaust gas flow. Control of the boost pressure at this operating point occurs similar to a single-stage system, solely via the waste gate (WG) of the LP turbine. As the LP compressor now assumes the complete compression load, the HP compressor would now represent an obstruction to flow on the fresh-air intake end. For this reason, an additional Compressor Bypass Valve (CBV) is arranged parallel to the HP compressor, which is fully opened in this case. The CBV is used both as a self-regulating valve as well as an independently controllable valve. V2

p2, T2

& E ,in m

m& fuel nE

& E ,out m p3,HPT, T3,HPT

V3,HPT

LLK

& CBV m

CBV

& HPT m

& HPC m

& TBV m

TBV

nHP

p1,HPC, V1,HPC T 1,HPC

V3,LPT

p3,LPT , T3,LPT

& LPT m

& LPC m

WG

nLP

V1,LPC

p1,LPC, T1,LPC

V4

p4, T4

& exh m

V5

p5, T5

Figure 1: Schematic representation of a two-stage system and configuration for modelling

44

& WG m

3

MODELLING

In the first step, a model description of the system must be developed. A theoretical approach is selected here; however, the response of various sub-models is emulated by static maps. In modelling, there is always a fundamental conflict of interest between sufficient accuracy of the prediction of the real system response and the easiest possible mathematical description. From a control system theory point of view, it is sufficient to develop a zero-dimensional mathematical model here. High-frequency mechanisms (e.g. exhaust-gas pulsation) are not considered. In the literature, there are a large number of approaches for modelling of turbochargers or the entire air path of an internal combustion engine. Many of them describe single-stage turbochargers with VTG (e.g. [1]). Modelling of a R2S™ system is shown in [2]. All methods have in common that the modelling method observes the filling and emptying behaviour of the individual volumes in the system. The aim of this model was to develop a flatness-based boost pressure controller. This is the reason why an analytical and physical-based modelling approach is used here. The model itself and the flatness-based approach were presented in a previous paper [3]. As mentioned before the model consists of the filling and emptying behaviour of the several volumes in the system (see Fig. 1 on the right side). Modelling assumes that both the input fresh air as well as the exhaust are ideal gases, and thus the ideal gas equation

p ⋅V = m ⋅ R ⋅T

(1)

with pressure p, volume V, mass m, gas constant R and temperature T applies. Balance equations for the mass and enthalpy are used for the filling and emptying

& can be calculated from the temperature Ti, methods here. The enthalpy flow H i & i [4]: specific heat capacity cp and the mass flow m & = c ⋅T ⋅ m &i H i p i

(2)

Every volume is thus uniquely described by two state variables. Hence, the ideal gas equation (1) can be used to calculate the third variable. For modelling, the masses mi and the pressures pi are selected as state variables. The pressure is particularly useful, because in the control of boost pressure, p2 is a control variable. Thus a conversion to the other state variables is not necessary. The mass flow & i results from the difference between input mass flow and change with time m output mass flow. The first fundamental theorem of thermodynamics and the ideal gas equation (1) are required for calculation of the pressure change &pi [1]:

& &pi =  − 1 ⋅ ∑ H Vi

(3)

Further analysis assumes that heat is not radiated to the surrounding environment. The engine is the central point of the entire air path. For the purpose of boost pressure control, it is sufficient to view and model the engine as a pump and heater from a thermodynamic point of view. The volume flow results from the piston stroke, which is additionally heated by the combustion process. The engine speed demands a volume flow of the system, and the injected fuel quantity determines the exhaust energy. Detailed modelling of the combustion processes in the engine will not be performed at this point. Furthermore, only average values are observed and not resolved by the crank shaft.

45

3.1 Models of the Individual Components Based on the complexity of the overall system and the fact that individual components occur more than once and only differ in their parameters, in the first step the behaviour of the individual components is described, which are subsequently combined to an overall system. Moreover, this results in a modular design, which easily facilitates subsequent extensions and enhancements. The behaviour of the compressor and turbine cannot be described with reasonable time and effort by means of physical equations. For this reason, the manufacturers of turbochargers provide measured characteristic maps for mass or volume flows and efficiency. These maps are read and interpolated or extrapolated with the corresponding algorithms for engine calculation or simulation. However, for the model-based control design used here, an analytical description that is as closed as possible is required. For this reason, an attempt is made to approximate the measured values with suitable mathematical functions. A detailed description can be found in [3]. 3.2 Overall System The basis for the overall model consists of the mass and enthalpy balance equations for the volumes V1;HPC to V3;LPT (Fig. 1, left side). The mass mi and pressure pi are selected as state variables. Furthermore, power balance equations are implemented for both TC shafts, and the speed of the LP-TC nLP and the HP-TC nHP are added as state variables. The system input variables are the orifice areas of the three actuation valves (ACBV, ATBV, AWG) of the R2S™ system, which act as inputs, & fuel which act on and on the other hand the engine speed nE and fuel mass flow m the system as non controllable disturbances. However, both disturbances can be measured at any time. The modelling results are verified by comparing the simulated values with the real system values measured on the test bench. The test bench usually does not utilize the original exhaust system with the corresponding piping. In order to implement the exhaust back-pressure that exists in the vehicle, adjustable throttles are used to set the stationary points compliant to manufacturers’ vehicle installation. Therefore, no differential equations are introduced for the volume V4 of the exhaust system (Fig. 1, right side). The necessary pressure p4 required for the calculation of the mass flows through the LP turbine and the waste gate (WG) is approximated using the measured data dependent on mass flow from the engine. For approximation, a second order polynomial is used that is dependent on the engine mass flow. 3.3 Results The mathematical model is verified with the data measured on the engine test bench. A three litre 6-cylinder Diesel engine delivering a maximum torque of 580 Nm and a maximum power of 200 kW is used as a comparison engine. The verification of the complete model occurs in two steps. In the first step, the static behaviour of the model is compared with the measured data of the engine map grid. Furthermore, the transient behaviour at a load step is compared with the measured results in a second step. On the engine map grid 150 measurement points have been recorded. The boost pressure p2 and both TC speeds nLP and nHP are represented in Fig. 2 on the left side for a reduced number of measurement points. It is evident that the simulated values (particularly the mass flow in the engine) deviate to a greater extent from the measured values at low engine speeds and loads. This is because the exhaust gas recirculation (EGR), which reduces the measured fresh-air share, is not considered in the model. The EGR for this engine is only active in the part-load range at low speeds. No exhaust is recirculated at high speeds, in the full load range and in transient operation.

& E ,in (not shown) and Whereas the measured values during fresh-air mass flow m boost pressure p2 correspond closely with the measured values (excluding the

46

influence of EGR), at both simulated TC speeds, the simulated values diverge more greatly from the measured values. This is due primarily to the modelling of the TBV that distributes the exhaust energy to both turbines. Small changes of the orifice width result in large changes of the mass flow that flows through the valve. Accordingly, a different split in the TC speed is set in the simulation. Nevertheless, the resulting boost pressure corresponds with the measured results, as the entire compression load of both TCs remains the same. In the second step, the transient behaviour of the model is verified with a load step of the engine from Md = 150 Nm to Md = 525 Nm at nE = 3500 rpm (see Fig. 3 on the right side), whereby Md is the engine torque and nE the engine speed. The rise times of the boost pressure and both TC speeds correspond to the measured curve. Divergences with the static values for the TC speeds occur due to a different split distribution of the exhaust energy. Especially the dynamic of the HP-stage differs between the simulated and measured values. Due to the complexity of the overall system, an exact replication with a tolerable level of effort is not possible. For example, it is not possible to exactly model the case where the HP compressor is subjected to flow, i.e. the pressure before the HP compressor p1,HPC is higher than the pressure p2 in the intake manifold. However, the result demonstrates that the simulated model corresponds sufficiently to the real system. Engine speed

Engine speed 4000

[1/min]

[1/min]

6000 4000 2000 0

0

20

40

60

80

100

120

140

3500 3000 10

160

15

20

measuring point

[Nm]

400

200 0

20

40

60

5

[Pa]

3

x 10

80

100

120

140

40

60

5

80

100

120

140

160

180

x 10

2.5 2 15

20

[1/min]

[1/min]

5 0

20

40

60

25

30

35

40

80

100

120

140

160

180

200

30

35

40

30

35

40

mes sim, compl. Mod.

10 5 10

15

20

measuring point x 10

[1/min]

1.5

mes sim, compl. Mod. 0

20

40

60

80

100

120

measuring point

Speed LP-TC

5

10 5

25

time [s]

Speed LP-TC

4

[1/min]

40

15

mes sim, compl. Mod.

10

0

35

Speed HP-TC

4

x 10

15

15

30

time [s]

Speed HP-TC

4

40

mes sim, compl. Mod.

1.5 10

200

measuring point x 10

25

time [s] 3.5 3

20

20

Boost pressure p2

mes sim, compl. Mod. 0

15

Boost pressure p2

2

0

35

200 0 10

160

measuring point

1

30

Engine torque 600

400

[Pa]

[Nm]

Engine torque 600

0

25

time [s]

140

160

180

200

x 10

1 0.5 10

mes sim, compl. Mod. 15

20

25

time [s]

Figure 2: Comparison of measured static engine map grid and measured load step with simulated values

4

POTENTIALS FOR BETTER DYNAMICS AND CONTROLLABILITY

Improving the controllability of the system is a strong request to achieve a satisfying emission control by keeping an excellent transient behaviour and in particular use the potentials of an R2S™ System in an optimal way.

47

With the physical model of the R2S™ system it is now possible to investigate the main influencing parameters improving the controllability of the system. For this investigation three factors have been chosen. In the first part, a modification of the volume of the intake manifold (V2) and exhaust manifold (V3) is examined. The second variation shows the influence of the inertia of the rotating parts. Especially the benefit of reducing the inertia of the turbine wheel material is shown for the LPstage. In the third scenario an improvement of the Turbine Bypass Valve (TBV) is taken into account. 4.1 Improved Volumes In [3] it was already shown that the influence of the volumes between the compressors and turbines can be neglected. Therefore only two volumes need to be considered in this investigation. There are two possible starting points for showing the influence of improved volumes regarding the transient behaviour of the turbocharger: • Reducing the intake manifold volume (V2) or • reducing the exhaust manifold volume (V3). On the left hand side of Fig. 3 the intake manifold volume is reduced to 50% of the standard intake volume (above picture: squared and circle marked line for the speed of the HP-stage resp. diamond and cross symbol for LP-stage and diagram below: squared line for the boost pressure) and compared against the standard one (encircled) during the load cycle. On the right hand side of Fig. 3 the exhaust manifold volume is reduced to 50% of the standard exhaust manifold volume. From a macroscopic point of view no change of behaviour between the different volumes can be observed. In order to make a more specific statement about the influence of the different intake manifold volumes and exhaust manifold volumes Fig. 4 highlights the time of the load step from 12.5s to 15s from Fig. 3 more precisely. Speeds HP & LP-TC

4

16

x 10

14

x 10

14 12

rpm [1/s]

12

rpm [1/s]

Speeds HP & LP-TC

4

16

nHP V2 Std nHP V2 50% nLP V2 Std nLP V2 50%

10 8 6

nHP V3 Std nHP V3 50% nLP V3 Std nLP V3 50%

10 8 6

4

4

10

15

20

25

30

35

40

10

15

20

Time [s]

Boost pressure p2

5

3.5

x 10

25

30

40

Boost pressure p2

5

3.5

x 10

p2 V2 Std p2 V2 50%

p2 V3 Std p2 V3 50% 3

Pressure [Pa]

3

Pressure [Pa]

35

Time [s]

2.5

2

1.5

2.5

2

1.5

10

15

20

25

Time [s]

30

35

40

10

15

20

25

30

35

40

Time [s]

Figure 3: Comparison of turbocharger speeds and boost pressure with improved volumes The reduction of the volume has no big influence regarding the speeds of the HPstage and the LP-stage. A little higher speed of the HP-stage with the reduced V2 is observed (squared line, upper left diagram). Although this influence is not really

48

significant a slightly faster boost pressure generation can be observed (squared line, lower left diagram). The results show even less or no influence of the reduced exhaust manifold volume V3 on the speeds of the turbochargers and the overall boost pressure generation. The right side of Fig. 3, which represents the comparison of the standard V3 and the reduced V3 exhaust manifold shows no different control behaviour even in the more detailed time period of the load step in Fig. 4. Speeds HP & LP-TC

4

16

x 10

14

x 10

14

12

12

nHP V2 Std nHP V2 50% nLP V2 Std nLP V2 50%

10 8

rpm [1/s]

rpm [1/s]

Speeds HP & LP-TC

4

16

6

nHP V3 Std nHP V3 50% nLP V3 Std nLP V3 50%

10 8 6

4

4

12.5

13

13.5

14

14.5

15

12.5

13

13.5

Time [s]

Boost pressure p2

5

3.5

x 10

15

x 10

3

Pressure [Pa]

Pressure [Pa]

14.5

Boost pressure p2

5

3.5

3

2.5

2

p2 V2 Std p2 V2 50%

1.5 12.5

14

Time [s]

13

13.5

14

Time [s]

14.5

2.5

2

p2 V3 Std p2 V3 50%

1.5 15

12.5

13

13.5

14

14.5

15

Time [s]

Figure 4: Detailed comparison of turbocharger speeds and boost pressure with improved volumes 4.2 Improved Inertias Another possibility gaining better controllability is reducing the inertia of the rotating parts of the turbocharger. By reducing the inertia the response behaviour of the rotating wheels and shaft will be improved. To reduce the inertia of the rotating parts, mainly the turbine material is substituted by a material with a reduced specific density. The use of Titanium-Aluminide (TiAl) or ceramic turbine wheels have already been investigated in different papers before for single stage turbochargers [5, 6]. The lower specific density of a TiAl turbine wheel reduces the inertia of the whole rotating shaft and wheels by a factor of two for example. On left hand side of Fig. 5 only the turbine wheel of the HP-stage is replaced by a TiAl material. Due to the lower inertia of the TiAl material the HP-TC accelerates faster and has bigger overshoots until a steady state operation condition is reached. Focusing on the boost pressure generation the usage of TiAl turbine wheel in the HP-stage shows nearly no influence. The minimal influence can be explained with the low absolute inertia of the HP wheel compared to the overall inertia of both TCs. Although the reduction of inertia through the usage of TiAl is significant, the overall influence of the rotating parts of the HP-stage is very small related to the boost pressure increase. This becomes clear on the left hand side of Fig. 5. During the first load step the deceleration of the HP-stage is very steep. The differences between the TiAl wheel and the standard wheel appear and the end of the load step where the LP-stage contributes more and more to the boost pressure generation. Because the difference occurs at the end of the load steps the influence of the LPstage meanwhile increased significantly and overlaps the positive effects of a TiAl wheel in the HP-stage. The left side of Fig. 6 shows this behaviour in more detail.

49

Speeds HP & LP-TC

4

16

x 10

14

x 10

14 12

rpm [1/s]

12

rpm [1/s]

Speeds HP & LP-TC

4

16

nHP Std nHP HP TiAl nLP Std nLP HP TiAl

10 8 6

nHP Std nHP LP TiAl nLP Std nLP LP TiAl

10 8 6

4

4

10

15

20

25

30

35

40

10

15

20

Time [s]

Boost pressure p2

5

3.5

x 10

25

30

40

Boost pressure p2

5

3.5

x 10

p2 Std p2 HP TiAl

p2 Std p2 LP TiAl 3

Pressure [Pa]

3

Pressure [Pa]

35

Time [s]

2.5

2

1.5

2.5

2

1.5

10

15

20

25

30

35

40

10

15

20

Time [s]

25

30

35

40

Time [s]

Figure 5: Comparison of turbocharger speeds and boost pressure with improved inertias On the right hand side of Fig. 5 the LP-stage turbine wheel is simulated with a TiAl turbine wheel. Although the absolute inertia reduction of both TCs in this case is much higher than in the first case the influence on the TC speeds of both stages is astonishingly smaller. The reason is based on the fact that the turbine wheel of the LP-stage contributes a lower amount to the overall inertia of the rotating parts such as shaft and compressor wheel compared to the small HP-stage. Hence the inertia reduction of the TiAL turbine wheel compared to the rest of the rotating parts on the LP TC is smaller in percentage than on the HP-stage. However, as expected the usage of a TiAl turbine wheel in the LP-stage leads to a slightly faster LP-stage acceleration (cross marked line on the upper right side of Fig. 6). This behaviour results also in a slightly faster boost pressure generation by using TiAl material in the LP-stage. Speeds HP & LP-TC

4

16

x 10

14

x 10

14

12

12

nHP Std nHP HP TiAl nLP Std nLP HP TiAl

10 8

rpm [1/s]

rpm [1/s]

Speeds HP & LP-TC

4

16

6

nHP Std nHP LP TiAl nLP Std nLP LP TiAl

10 8 6

4

4

12

12.5

13

13.5

14

14.5

15

12

12.5

13

Time [s]

Boost pressure p2

5

3.5

x 10

14.5

14

14.5

15

x 10

3

Pressure [Pa]

Pressure [Pa]

14

Boost pressure p2

5

3.5

3

2.5

2

p2 Std p2 HP TiAl

1.5 12

13.5

Time [s]

12.5

13

13.5

Time [s]

14

14.5

2.5

2

p2 Std p2 LP TiAl

1.5 15

12

12.5

13

13.5

15

Time [s]

Figure 6: Detailed comparison of turbocharger speeds and boost pressure with improved inertias

50

4.3 Optimized TBV For the controllability it is essential to exactly control the position of the turbine bypass valve (TBV). Because of the geometrical opening behaviour of a flap design the change of mass flow from zero to maximum takes place in the first few degrees of the opening angle of the valve. In order to improve the controllability it is desirable to enhance the usable control range. This can be achieved by influencing the flow coefficient Cq through modifying the design of the flap for example.

& TBV = A ⋅ Cq ⋅ p3,HPT ⋅ m

T3,HPT

 +1 ⎧ − 2  for Π ≤ Π crit ⎪Π − Π 2⋅ ⎪ 1 ⋅⎨ ⋅ R ⋅  − 1 ⎪⎛ 2 ⎞  −1 for Π > Π crit ⎪⎜⎝  + 1 ⎟⎠ ⎩

(4)

& TBV can be calculated using (4) where Π is the ratio of pressure The mass flow m

after the TBV p3,LPT and before the valve p3,HPT. Πcrit is the critical pressure ratio which determines the maximum possible mass flow through the valve with a certain flow area resp. bypass diameter. Other parameters in (4) are the isentropic coefficient , the gas constant R, pressure before the valve p3,HPT and the temperature of the exhaust gas before the valve T3,HPT. Parameters that can be influenced by the design are the opening area A and the flow coefficient Cq. The opening flow area is controlled via the actuator characteristic of the TBV. The flow coefficient first of all is a constant associated to design and depending on the position of the flap. To show the effects of different flow coefficients a simulation is performed at a constant engine operating point (2000 rpm, 240 Nm) in order to eliminate any other influences. Then the TBV is slowly opened to avoid any dynamic side effects. Fig. 7 demonstrates the influence of three different flow coefficients to the mass flow resp. the achieved boost pressure. Massflow / Opening angle @ nE = 2000 rpm / Md = 240 Nm

Boost pressure p2

5

x 10

flow coefficient = 1 flow coefficient = 0,7 flow coefficient = 0,5

3.4

70 3.2

60

3

2.8

Pressure [Pa]

Massflow [%]

50

40

30

2.6

2.4

2.2

2

20 1.8

10

flow coefficient = 1 flow coefficient = 0,7 flow coefficient = 0,5 0

0

1

2

3

Opening angle [deg]

4

1.6

5

6

1.4

0

1

2

3

4

5

6

Opening angle [deg]

Figure 7: Comparison of opening angle function of the valve and the influence on boost pressure for different flap designs The characteristic behaviour of (4) can be seen on the left hand side of Fig. 7. During the first few degrees of flap opening the mass flow changes from zero to

51

levels around 70% of max possible mass flow for this particular operation point. Reducing the flow coefficient means to worsen the flow characteristics. Due to the more linear behaviour of mass flow with flow coefficient Cq=0.5 this design is favourable only because of controllability reasons. On the other side a bad flow coefficient would lead to the need of a larger bypass channel area to compensate the restricted mass flow for the rated power of the engine. So, in the high rpm and torque range of the engine the TBV is opened widely enough to achieve with a low pressure drop a excellent flow behaviour through the TBV. On the right hand side of Fig. 7 the boost pressure p2 over the TBV opening angle is shown. One reason that the boost pressure p2 drops is because the bypassed exhaust gas energy to the LP-stage is not sufficient enough to generate the requested boost pressure. The second reason is based on the fact that the derived basic model cannot describe the situation where the whole compressor side of the HP-stage is bypassed and therefore the pressure before HP compressor is greater than behind the HP compressor. Nevertheless, the results show that a better mass flow through the TBV (higher flow coefficient Cq) leads to a higher boost pressure drop. The chosen engine operating point does not produce enough exhaust gas to accelerate the LP-stage therefore the smaller the flow coefficient is, the more exhaust gas flows through the HP-stage leading to a smaller boost pressure drop. From a controllability point of view the situation where the flow coefficient Cq=0.5 is desirable because small position changes of the valve have a smaller influence on the boost pressure and therefore the control resolution increases. 4.4 Summary The influence of different methods to improve the controllability of a R2S™ turbocharger system has been investigated by dynamic simulations. It was demonstrated that the influence of an intake manifold volume change is slightly bigger than the influence of an exhaust manifold volume change regarding the boost pressure generation. A similar behaviour is observed using a TiAl turbine wheel material in the LP-stage, resp. HP-stage, but it is also obvious that the influence on boost pressure generation of a TiAl turbine wheel in LP-stage is bigger than in the HP-stage. Considering the valve design of the TBV it turned out that the design of the flap and the resulting variation in the flow coefficient has a major influence on the mass flow rate at small opening angles of the TBV. The reduction of the mass flow at small opening angles improves the controllability of the boost pressure due to the better resolution. For the boost pressure generation in transient situations the correct split of the exhaust energy between the HP-stage and the LP-stage is crucial. By flattening the mass flow opening angle function a better controllability can be achieved, because a change in the opening angle leads to a smaller increase of the mass flow and consequently the energy distribution to the LP-stage can be controlled more accurately. To summarise this investigation it is evident that it is more beneficial to rather reduce the intake manifold volume than to reduce the exhaust manifold volume. It also appeared that the usage of an expensive TiAl material in the HP-stage has no big influence on the boost pressure generation. A usage of TiAl might only be reasonable in the LP-stage. The biggest influence on the boost pressure generation was observed by modifying the flap design. Therefore the bottom line is to concentrate on the design of flaps and valves in order to improve controllability, but the influence on the maximum mass flow rate has to be kept in mind, otherwise the controllability may be good, but the maximum power of the whole system can be limited by the valve and flap design.

52

5

CONCLUSION AND OUTLOOK

In the first part of the paper the modelling of the R2S™ is briefly discussed and has shown good results by comparing the simulation data with measurements. Based on the model the influence of different modifications in the air intake and exhaust path has been investigated. Three different modifications of the overall boosting systems have indicated the highest potentials to influence the overall boost pressure behaviour: a) the air intake and the exhaust manifold volumes b) the inertias of the rotating parts of the TCs c) the TBV mass flow characteristic over the opening angle In the simulation it was evident that the change of volumes and inertias did show some benefit, but minor compared to the change of the TBV design. Further progress has to be performed when the simulations will soon be compared to results derived on an engine test bed on a Diesel 2.0-liter engine where the TBV mass flow characteristic was modified. With the results of the measurements the accuracy of the model will be adjusted and simulations will be reiterated. Beside improving existing TBV designs regarding the mass flow characteristic, new TBV designs which do take into account the desired mass flow function over the opening angle need to be considered as well. Crucial for all new designs is that the TBV achieves a perfect sealing at the closed position to maintain the high transient potentials of the R2S™ system. Furthermore to fulfil new emission legislations with Diesel engines the use of a Variable Turbine Geometry (VTG) TC might be advantageous.

6

REFERENCE LIST

[1]

F. Richert: Objektorientierte Modellbildung und Nichtlineare PrädiktiveRegelung von Dieselmotoren, Ph.D. thesis, RWTH Aachen, 2005. D. Schwarzmann: Nonlinear Internal Model Control with Automotive Applications, Ph.D. thesis, Ruhr-Universitaet Bochum, 2007. O. Weber, R. Christmann and S. Liu: Modelling and flatness based control of a R2S™ turbocharger, 13th EAEC European Automotive Congress, Valencia, 2011. H. D. Baehr, S. Kabelac: Thermodynamik - Grundlagen und technische Anwendungen, Springer-Verlag, 2006. B. Engels: Verbesserung des Instationärverhaltens von Abgasturboladern, Technische Akademie Wuppertal, Nürnberg, 1990. B. Engels, R. Lucks, F. Pflüger: Abgasturbolader für zukünftige Ottomotoren, 4. Symposium “Entwicklungstendenzen bei Ottomotoren”, Technische Akademie Esslingen, 1998.

[2] [3] [4] [5] [6]

53

A new approach to thermo mechanical fatigue shown on turbocharger housings M Nagode a, F Längler b, M Hack c a University of Ljubljana, Faculty of Mechanical Engineering, Slovenia b BorgWarner Turbosystems Engineering GmbH, Germany c LMS Deutschland, Germany

ABSTRACT In the last two decades the development time of vehicles has been drastically reduced from eight to three years due to the application of advanced numerical and experimental methods. The specifications including comfort, driving behaviour and durability for every new model are being raised for every vehicle. The development responsibility is passed on to the supplier who, at the start, with limited information, makes a commitment with a fixed price for the production lead time. The aim of the paper is to show the damage operator approach with creep extensions integrated into the LMS Virtual.Lab Durability thermal fatigue module. The subject of the investigation is a turbocharger turbine housing that is exposed to high mechanical and thermal loads leading to considerable creep during its usage. The presented strain-life approach is based on the isothermal cyclically stable stress-strain and Manson-Coffin-Morrow strain-life curves. The finite element model of the turbocharger turbine housing is analysed numerically. A new extension to the Neuber approximation formulas that include viscoplastic correction is used to facilitate the finite element analysis runs. Results are compared with full finite element analysis and with tests. Keywords: thermo-mechanical fatigue; damage operator approach; thermal shock; turbocharger.

ABBREVIATIONS TMF – thermo mechanical fatigue FEA – finite element analysis OEM - original equipment manufacturer DOA – damage operator approach

1

INTRODUCTION

The failure mechanisms associated with TMF are mechanical fatigue, oxidation and creep. The damages due to fatigue, oxidation and creep can be treated separately or together. In the latter case, fatigue damage prediction is often based on the isothermal strain controlled low cycle fatigue (LCF) tests performed at various temperatures, strains and strain rates.

_______________________________________ © The author(s) and/or their employer(s), 2012

73

In TMF evaluations, the thermal and mechanical loadings acting on the specified structure has to be considered. In order to obtain the temperature fields that are later applied together with the mechanical loads in the structural analyses, an assumption has to be made regarding the uncoupling of the thermal and structural analyses. Transient thermal analyses are performed in order to obtain the desired temperature fields for all load cases. The computed temperature fields are then applied in combination with the mechanical loading in the stress-strain finite element analysis (FEA). The second uncoupling regards the separation of the stress-strain response from the damage calculation. The stress-strain response together with the temperature fields are used for the final damage evaluation by the LMS Virtual.Lab Durability module. This non-unified approach has been widely and successfully used for TMF calculations in the automotive industry. For many cases in this latter industry, the durability is no longer mainly determined by mechanical loads only, but by a combined effect of thermal and mechanical loading of the component or system under normal design working conditions. Both kinds of loading result in TMF. In order to address this difficult balance exercise, the development engineer needs a reliable numerical and experimental tool that allows him to tune his task into the increased pace of (virtual) development processes of the OEM’s. Structures that are subject to varying mechanical loads under varying temperatures exhibit different fatigue behaviours at different temperatures, different stress-strain behaviour at different temperatures and creep at high temperatures. This is mainly due to temperature induced stresses. The aim of this paper is to give an overview on the recently developed methodology for stress-life and strain-life finite-element based approaches to TMF. Detailed analysis and description are found in [1-3].

2

DAMAGE OPERATOR APPROACH

The damage operator approach for all cases of stress-life and strain-life under varying temperatures is presented. It is based on the pseudo-elastic stresses that are converted into elastoplastic stress-strain states by using approximate notch analysis formulas for strain-life approach or directly on the elastoplastic stressstrain FEA states. Cyclic hardening and cyclic softening can frequently be assumed to contribute negligibly to the damage. By applying half-life stabilised cyclic stressstrain response the requested FEAs can be shortened drastically. Traditional fatigue damage accumulation uses rainflow counting and linear Miner damage accumulation rule. In the case of TMF one experiences that due to the changes in the environment (like temperature) the fatigue behaviour is changing over time. For variable amplitude loading it is very typical that the largest load cycles - that contribute most to the damage - take a very long time to complete, due to the many nested cycles inside. Rainflow approach which only considers cycles when they are completed can no more be justified in the latter case. Hence one has to take one step back and analyze the behind the rainflow based approaches: the theory of especially the hysteresis operator that describes damage equivalence of the rainflow approach with the damage simple case of constant fatigue behaviour is shown.

74

mathematical theory lying hysteresis operators and accumulation. In [4-6] the operator approach for the

It shall be noted here that the hysteresis operators for elastoplastic stress-strain behaviour can be identified with rheological models [7,8,9]. The operators for damage accumulation have the same structure and therefore can also be identified with rheological models. It should be warned that even though this is helpful for imagination and gives hope for a theoretical fatigue model, the parameters for these models are drawn from empirical tests, more explicitly from SN or εN curves. The oxidation damage is integrated into the SN and εN curves as the tests are carried out under isothermal ambient conditions and not in vacuum. The same holds for the creep master curves. But again from an analogy to the stress-strain behaviour the basic idea of damage operators can be indicated. In kinematic elastoplastic models one can associate the dissipated energy to the total movement of the yield surfaces. Reference [5] details on how to define from this a dissipation operator, for which the total change (Variation) over time gives the dissipated energy. If 0 ≤ t1 ≤ t2 ≤ K ≤ ti gives a sequence of the times the variation is defined as

VarW  ti  =

∑ W t  − W t  i

j =1

j

j −1

(1)

Given the correct operator, the dissipated energy can then be written as W  ti  = VarW  ti 

(2)

For the case of linear damage accumulation one can define a formally similar model and operator, for which the total variation corresponds to the accumulated Miner sum [5]. The methodology and extension to mean stress effects had been further developed in [6]. The thermal influences on stress-strain states, mean stress effect, time dependent material response and damage accumulation have been added recently in [1-3] and are briefly described below.

3

STRESS-LIFE DAMAGE OPERATOR APPROACH

The stress-life approach based on an equivalent temperature calculation [10,11,12] proved to be acceptable for a number of automotive cases. However, there exists a better way to access fatigue damage at constant or variable temperature directly by applying the so called Prandtl operators with the corresponding rheological type of model. In fact, damage D(ti ) at time ti can be expressed as D  ti  = VarD  ti 

(3)

Damage operator D  ti  represents cyclic damage evolution and is connected to the SN curves. The principal idea behind the approach is depicted in Fig. 1. The damage operator can be interpreted as a serially connected spring-slider model and stands for the total of cyclic damage evolutions D j  ti  over nr segments of the model.

This yields

D  ti  =

∑ D t  nr

j =1

j

i

(4)

where the contribution of each spring-slider segment to the total damage is given by

D j  ti  = α j Ti   α j Ti 

(5)

75

The slider shift can be explained as an irreversible dislocation movement. The change in temperature influences the stiffness of the springs that are directly linked to temperature Ti = T (ti ) dependent Prandtl densities α j Ti  and, thus, affects the movement of dislocations. At higher temperatures the movement is relieved, whereas it is hindered at lower temperatures.

Figure 1 – Damage operator in the form of the spring-slider model. We first start from the material data at constant temperatures. Fatigue tests are usually carried out for a certain number of stress levels and constant environmental conditions. Only a limited number of SN curves are available, e.g. nT , for different test temperatures and for a particular material and geometry (see Fig. 2 for an example). To get the SN curve at any temperature, one can interpolate among the experimentally obtained SN curves [1].

Figure 2 – Temperature dependent test and interpolated SN and SD curves. SN curves are transformed into equivalent stress-cycle damage (SD) curves

df =

1 Nf

(6)

to avoid numerical difficulties due to the unlimited number of cycles to failure Nf = ∞ for stress amplitudes  a beneath the endurance limit. The equivalent cycle

damage is always bounded 0 ≤ df ≤ 1 and therefore more convenient than Nf for

further analyses. Therefore, the curves in Fig. 2 can be interpreted as cyclic stressdamage curves. Consequently [1,5,6]

D ti  =

∑ α T   α  T  nr

j =1

j

i

j

i

(7)

In [1] a transformation of the original stress history  (ti ) holding the stress amplitude and mean stress information is transformed into a new one containing the equivalent stress amplitude information solely. In this way mean stress correction is considered. This allows one to use any mean stress correction formula.

76

This is an important difference to the approaches based on pure energy that cannot take the mean stress effects into account. Back stress  α j  ti  depends on equivalent stress  e (ti ) and follows in the original

approach [4-8] the kinematic hardening rule, which can be expressed in the evolution formula as play operator with general initial value

⎪⎧

⎪⎧

 α j  ti  = max ⎨ e  ti  − rj ,min ⎨ e  ti  + rj , ⎩⎪

⎩⎪

⎫⎫ α j Ti −1  ⎪⎪   t  ⎬⎬ α j Ti  α j i −1 ⎭⎪⎭⎪

(8)

In the presented [1] approach additional multiplier α j Ti −1  α j Ti  is included due to

the temperature changes. Presumably, there is no residual stress initially, so  α j  t0  = 0 and D  t0  = 0 . The Prandtl densities α j Ti  as well as fictive yield

stresses r j in the range j = 1K nr are gained from the available SD curves. Eqs. [8]

and [10] are thus a complete substitute for the rainflow counting and Miner damage accumulation rule enabling continuous damage assessment by taking both variable temperatures and mean stress correction into account. For further details refer to [1].

4

STRAIN-LIFE DAMAGE OPERATOR APPROACH

The strain-life approach can be either stress or strain controlled. In both cases the mechanical loading is transformed into the local stress-strain states by carrying out linear or nonlinear FEA. The thermal loading is taken into account by applying temperature to the nodes of the FE model. In the case of strain-life approach also the stress-strain behaviour, which is now temperature dependent, needs to be modelled. Using the above mentioned analogy between stress-strain modelling and damage behaviour, the similar approach [13,14] for temperature dependence is used as already shown in Eq. (8). Special care needs to be taken also for the mean stress correction as temperature changes may lead to independent changes in stresses and strains. A full analysis is given in [2]. The strain-life damage operator approach can be outlined as follows: •

FE model is restrained and loaded mechanically. Thermal loading is applied to the nodes of the FE model. Transient linear or elastoplastic FEA are carried out.



Pseudo-elastic stresses or elastoplastic stresses are converted into elastoplastic stresses and strains by two Prandtl operators based on the single play operator with general initial value.



Mean stress, stress amplitude, mean strain and strain amplitude are calculated continuously.



The stress and strain histories holding the mean stress, stress amplitude, mean strain and strain amplitude information are transformed into the new one containing only the damage parameter P information.



Damage is calculated continuously.

The damage calculation is similar to the one described above. The standardised damage parameter (PN) life curves are transformed into equivalent damage parameter-cycle damage (PD) curves. Consequently, the play operators have to be calculated anew and the equivalent stress in Eq. (8) is replaced by damage parameter P  ti  .

77

5

VISCOPLASTIC APPROXIMATION AND CREEP FATIGUE

The damage operator approach that has initially been developed for independent plasticity and fatigue modelling has further been developed to dependent viscoplasticity [15] and creep [16,17]. The oxidation is taken account indirectly. The lifetime prediction can be based on different structural Further, three cases are listed ranged from the least to the most complex one:

time time into FEA.



linear elastic FEA and approximate notch analysis formula e.g. the Neuber formula [18] with additional viscoplastic approximation,



elastoplastic FEA with kinematic hardening and viscoplastic approximation and



elastoviscoplastic FEA.

It has been shown e.g. by [19,20] that the linear elastic FEA combined with an approximate notch analysis formula is applicable if the plastic regions are small as compared to the supporting elastic material. Otherwise the more time consuming nonlinear elastoplastic FEA is required. The elastoviscoplastic FEA is extremely slow and rarely meets the industry requirements due to the continually shorter development cycles. Therefore a time efficient viscoplastic approximation based on a nonlinear Maxwell model [15] and the corresponding lifetime prediction was developed and tested on several thermal shock experiments on exhaust downpipe [21,22] and turbocharger housing [23] depicted in Fig. 3. If the strain-life and master curves are attained by the tests carried out under isothermal ambient conditions and not in vacuum, oxidation is taken into account indirectly in fatigue Df  t  and creep Dc  t  damages. The total damage is defined

as

D  t  = Df (t ) + Dc (t )

(9)

The low cycle TMF damage calculation is conducted as shown in the latter section, while the creep damage follows the simple Robinson rule [24] Dc (t ) =

∫ t ( (t ), T (t )) t

0

dt

(10)

r

where tr denotes the time to rupture for the given stress and temperature. In [3] the authors explain the application of the nonlinear model to the Neuber type approximation formula in detail.

6

FINITE-ELEMENT BASED DURABILITY SIMULATION

In the past decades fatigue based on finite elements has become an important tool in the design cycle. For a fatigue analysis one needs to calculate the stress-strain trajectories for all the loading cycles. Especially for multitudes of mechanical loads acting on real structures special care needs to be taken on selecting the proper type of FEA that is necessary to get results accurate enough for fatigue estimation. For mechanical loads the main idea for simplification is linearised approaches like quasi-static superposition or modal superposition. The strain-life approach with the Neuber type of approximation uses the concept that often, even though there are plastic zones to be taken into account on a mesoscopic level, one can still work with pseudo-elastic stresses on a macroscopic level.

78

Detailed analyses on how these types of approximations can be used in TMF have been performed in parallel to the development of the methodology as shown in this paper. Especially by further developing the approximation type of approach to viscoplastic approximation [15] an important step has been achieved. For many real life examples simplified FEA have still proved to be adequate.

7

THERMAL SHOCK SIMULATION ON A TURBOCHARGER HOUSING

The tested turbocharger housing (Fig. 3) was exposed to thermal shocks which were applied in cycles.

Figure 3 – Mesh of the full model (a) and mesh of the turbocharger housing (b). One cycle represents heating from ambient temperature to the highest temperature and cooling back to the ambient temperature again in a number of substeps. The time for one cycle is 600 seconds: 255 seconds of heating followed by 345 seconds of cooling. Cycles are repeated until rupture occurs. No additional mechanical load is applied. For the durability calculation of the turbocharger housing, a transient thermal FEA and a structural elastoplastic FEA with kinematic hardening are performed for the stabilised thermal shock cycle. Stress and temperature fields generated by heating and cooling of the turbocharger housing in the stabilised cycle are used as the input data for the damage calculation. In the LMS Virtual.Lab Thermal Fatigue the resulting cycle is multiplied ten times and viscoplastic approximation is determined. The turbocharger housing is made of casting material Ni-resist D-5S. Temperature dependent data are depicted in Fig. 4 and listed in Table 1. Fatigue damage is calculated considering the Smith-Watson-Topper mean stress correction and the open mode (I) critical plane multiaxial criterion. The signed von Mises multiaxial criterion turns out to be inappropriate for this particular application as it tends to drastically overestimate the number of cycles to failure N [21,22]. The critical plane criterion gives more suitable results, but only the Smith-WatsonTopper mean stress correction results in consistently conservative lifetime predictions [21,22,23]. Creep damage calculation is performed using tensilecompressive creep relation [16].

79

Figure 4 – Temperature dependent material data for casting material Ni-resist D-5S; cyclic stress-strain curves (a), creep-rupture curves (b).

Four critical locations occur at the damage simulation. They are denoted as P2, P6, P16 and P20 (Figs. 3b and 5). The damage calculation of the turbocharger housing considering fatigue and creep damage is thoroughly explained in [3]. Table 1 – Material parameters for the law of perfect viscoplasticity with elastic domain for casting material Ni-resist D-5S.

T °C 600 700 800

k MPa 5.340 7.028 0

K MPa s1/N 898 633 304

N 7.200 5.685 6.483

The most critical location is P6 with 41 predicted cycles to failure. P6 also coincides with the area of the highest load temperature. Calculated results are compared to the tested turbocharger housings (Fig. 5).

Figure 5 – Critical locations of turbocharger housings.

Turbocharger housings were tested to failure. Critical locations at the test occurred at the same places as in the simulation (Fig. 5). The comparison between the predicted and observed number of cycles to failure is given in Table 2.

80

Table 2 – Comparison between the predicted and observed number of cycles (N) to failure for the critical locations of the turbocharger housing.

critical location predicted observed

8

P2

P6

P16

P20

N 50 70 ± 15

N 41 25 ± 8

N 108 310 ± 30

N 583 361 ± 60

CONCLUSIONS

At extreme TMF conditions the time dependent viscoplasticity and creep contribution to the total damage become significant. The transient structural FEA with kinematic hardening and additional viscoplastic approximation proves to be appropriate. One stabilised thermal shock cycle has to be analysed by the FEA. The relaxation effects can then be modelled by the presented approach implemented into the LMS Virtual.Lab Thermal Fatigue. The tedious elastoviscoplastic FEA can thus mainly be replaced by the elastoplastic FEA with kinematic hardening. The material parameters can be calibrated on the half-life stabilised cyclic stress-strain curve. The DOA with mean stress correction and viscoplastic approximation implemented into the LMS Virtual.Lab Thermal Fatigue stands as a suitable and time efficient tool for lifetime predictions of TMF loaded components even at extreme conditions. The expected results are conservative. For a turbocharger housing the Smith-WatsonTopper mean stress correction and the critical plane mode (I) approach is the most suitable parameter combination.

9

REFERENCES

1.

Nagode, M., Hack, M. and Fajdiga, M., "High Cycle Thermo-Mechanical Fatigue: Damage Operator Approach," Fatigue Fract. Engng. Mater. Struct. 32:505-514, 2009. Nagode, M., Hack, M. and Fajdiga, M., "Low Cycle Thermo-Mechanical Fatigue: Damage Operator Approach," Fatigue Fract. Engng. Mater. Struct. 33:149160, 2010. Nagode, M., Längler, F. and Hack, M., "A time-dependent damage operator approach to thermo-mechanical fatigue of Ni-resist D-5S," Int. J. Fatigue, 33:692-699, 2011. Brokate, M. and Sprekels, J., "Hysteresis and Phase Transition," Applied Mathematical Sciences 121, Springer, New York, 1996. Brokate, M., Dressler, K. and Krejči, P., "Rainflow Counting and Energy Dissipation in Elastoplasticity," Eur. J. Mech. A/Solids 15:705-737, 1996. Hack, M., "Schädigungsbasierte Hysteresefilter," D386 (Diss Univ. Kaiserslautern), Shaker Verlag, Aachen, 1998. Prandtl, L., "Ein Gedankenmodell zur kinetischen Theorie der festen Körper," Z. Ang. Math. Mech. 8:85-106, 1928. Visintin, A., "Rheological Models and Hysteresis Effects," Rend. Sem. Matem. Univ. Padova 77:213-243, 1987. Krejči, P., "On a System of Nonlinear PDEs with Temperature-Dependent Hysteresis in One-Dimensional Thermoplasticity," J. Math. Anal. Appl. 209:2546, 1997. Nagode, M. and Hack, M., "An Online Algorithm for Temperature Influenced Fatigue-Life Estimation: Stress-life Approach," Int. J. Fatigue 26:163–171, 2004. Kohout, J., "Temperature dependence of stress–lifetime fatigue curves," Fatigue Fract. Eng. Mater. Struct. 23(12):969–77, 2000.

2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

81

12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

23. 24.

82

Kang, H. T., Lee, Y., Chen, J. and Fan, D., "A thermo-mechanical fatigue damage model for variable temperature and loading amplitude conditions", Int. J. Fatigue 29:1797-1802, 2007. Nagode, M. and Zingsheim, F., "An Online Algorithm for Temperature Influenced Fatigue-Life Estimation: Strain-Life Approach," Int. J. Fatigue 26:151-161, 2004. Nagode, M. and Fajdiga, M., "Temperature-Stress-Strain Trajectory Modelling During Thermo-Mechanical Fatigue," Fatigue Fract. Engng. Mater. Struct. 29:175–182, 2006. Nagode, M. and Fajdiga, M., "Coupled Elastoplasticity and Viscoplasticity under Thermomechanical Loading," Fatigue Fract. Engng. Mater. Struc. 30:510–519, 2007. Šeruga, D. and Nagode, M., "Unification of the most commonly used timetemperature creep parameters," Mater. Sci. Eng. A. 528: 2804-2811, 2011. Robinson, E.L., "Effect of Temperature Variation on the Creep Strength of Steels," Trans. ASME 160:253-259, 1938. Neuber, H., "Theory of Stress Concentration for Shear-Strained Prismatical Bodies with Arbitrary Nonlinear Stress-Strain Law," J. Appl. Mech. 12:544-50, 1961. Rosa, U., Nagode, M. and Fajdiga, M., "Strain-Life Approach in ThermoMechanical Fatigue Evaluation of Complex Structures," Fatigue Fract. Engng. Mater. Struct. 30:808-822, 2007. Härkegård, G. and Mann, T., "Neuber prediction of elastic-plastic strain concentration in notched tensile specimens under large-scale yielding," J. Strain Anal. Eng. Des. 38:79-94, 2003. Šeruga, D., Nagode, M., Hack, M. and Hansenne, E., "Thermomechanische Ermüdung - Simulation von Thermoshocks am vorderen Auspuffrohr," NAFEMS Seminar, Wiesbaden, Nov 2010. Nagode, M., Šeruga, D., Hack, M. and Hansenne, E., "Damage Operator-Based Lifetime Calculation Under Thermomechanical Fatigue and Creep for Application on Uginox F12T EN 1.4512 Exhaust Downpipes," Strain, DOI: 10.1111/j.1475-1305.2011.00812.x, 2011. Nagode, M., Längler, F., Hack, M. and Fajdiga, M., "Validation of the Prandtl Damage Operator Approach for Durability Prediction of a Turbocharger Turbine Housing," Proc. Fatigue Design 2009, Senlis, France, Nov. 25-26, 2009. Robinson, E.L., "Effect of temperature variation on the creep strength of steels," Trans. ASME 160:253-259, 1938.

Compressor wheel low cycle fatigue calculations for off highway applications – an approach to accurately calculate application duty cycle K Ohri IPSD Caterpillar, UK K Shoghi BorgWarner Turbo Systems Ltd, UK

ABSTRACT Under certain conditions, the compressor wheel of a turbocharger can be exposed to high stresses due to the mechanical and thermal loading during operation. The resultant stress levels on the compressor wheel give rise to fatigue in the compressor wheel material which can be classified in terms of low cycle fatigue (LCF) and High cycle fatigue (HCF). The low cycle fatigue limits on the compressor wheels can be characterized using occurrence of plastic strain in the wheel material due to high speed in the wheel or high amplitude of the cycle leading to wheel failure. Therefore, it is essential to completely understand the end application (Duty cycle) in order to make a judgement on a suitable compressor wheel for the application. The High cycle fatigue is associated with the excitation of the blades of the wheel matching the blade resonant frequency resulting in significant increase in the amplitude of the vibration. The off highway machine industry incorporates a vast range of duty cycles. Along with the wide range (Over eighty different types within Perkins’s customer base) of applications and operating parameters comes the complication to verify a suitable duty cycle for each application. The exact duty cycle can be influenced by nature of machine operation, the terrain, the climate and skill of the operator. The duty cycle calculation of the compressor wheel have been presented in the papers by R Christmann [1] in 2010 Turbocharger conference. This paper describes a proposed method that allows accurate LCF prediction, tailored to individual duty cycles using a limited effort in terms of data and resource. The paper also highlights the use of an evaluative method, which can be used as a quick tool for initial evaluation as a first step. This tool takes in to account number of average cycles per hour and classifies applications into different categories. This can be made use of during early stages of a ‘New product introduction’ to instate a baseline with some confidence and highlight applications of concern. A detail analysis can then be carried out once the engine calibration is developed for a specific application. The accuracy of the results would highly depend on the accuracy of initial input data. These cycles can be validated under test bed conditions easily before being applied to a range of applications.

_______________________________________ © The author(s) and/or their employer(s), 2012

97

1.

NOMENCLATURE

f = Output signal hc = Constant 44300 (m) h2 = Height above the sea level (m) N1 = Speed at sea level (m/s) N2 = Speed above sea level (m/s) t = Time (s) y = Initial condition

2.

V = Speed of the compressor wheel (m/s) σ = Stress in the compressor wheel (N/m2) ρ1 = Air Density at sea level = 1.225 (kg/m3) ρ2 = Air Density above the sea level (kg/m3) τ = Time constant

INTRODUCTION

Emissions legislation in NOx and particulate matter have been a key driver in increasing the complexity of powertrain systems on off highway applications. On the other hand, increasing customer demands on low purchasing and running cost of machines, high durability targets and increased productivity drive additional pressure on the powertrain to improve performance and yet attain the fuel economy benefits. As a result there is a constantly high demand in boost from the air system on the engine. With this, there comes the challenge of operating within mechanical limitations of various components including the turbocharger. The nature of operations and type of transmission systems seen on off highway machines can make the applications extremely transient. This transient nature coupled with high performance aspirations results in decreasing fatigue life of the turbocharger compressor wheel. Transience can be counted as both the engine’s ability to change speed at constant or variable load and the engine’s ability to accept load at constant speed. The first section of the paper provides an introduction to the standardized approach adopted to define machine duty cycles historically. The second section reflects the revised approach, making use of known parameters to calculate turbo speeds for specific duty cycles. The final section of the paper discusses the affect of altitude, air temperature and other associated factors on the resultant change in speed and therefore, life of the compressor wheel.

3.

APPROACH TO LOW CYCLE FATIGUE

The lifetime calculations of a compressor wheel are a combination of local and nominal stress concept, which is determined from the turbocharger speed profiles. With a large number of off-highway applications with different power ratings, performance requirements and complexity of tasks handled; it becomes a mammoth task to physically measure turbo speeds for each application. Hence, reducing test time becomes essential. Compressor wheels can be subjected to high stresses during operation due to speed, pressure, and temperature. As discussed by R Christmann [1] the influence of temperature on the life is small compared with speed. Hence it will not be considered in this paper. The maximum principal stress due to speed normally occurs in the bore or root of the blade depending on design of the wheel as given by figure 1. The principal stress due to both speed and pressure is shown in figure 2. The pressure has a small effect on the position of maximum principal stress. Since the differential pressure is applied on the pressure

98

side of the blades and the movement of the blades due to speed is towards the suction side there is a very small reduction of 0.7% in the magnitude of the principal stress when it occurs in the root of the blade and an increase of 3.5% near the nose. Figure 2 also indicates the principal stress due to pressure load only and the stress due to pressure is small.

Figure 1: Principal stress distribution in the compressor wheel due to speed

Figure 2: Principal stress due to both speed and pressure (Left). Principle stress due to pressure only (Right) The finite element analysis (F.E.A) is calculated for the linear elastic range of the material and due to small effect of pressure loading only stress σ due to speed v is considered. The stress due to centrifugal load in the wheel is proportional to its square of speed. The test to determine the number of cycles to failure of the compressor wheel is performed with the compressor wheel accelerating to a high speed well over its operating range and then dual for a few seconds before decelerating to its original low speed. The maximum speed attained during this test is large enough to cause plastic deformation in the compressor wheel material. This plastic deformation if occurred in the bore of the wheel, can be measured. This over speeding during the test, will cause an accumulative damage which will result in crack initiation in the critical locations of the wheel as given by FEA in figure 1 and will result in complete failure of the wheel.

4.

STANDARDISED APPROACH

Low cycle fatigue lifetime of the compressor wheel was historically calculated using standardised cycles. Different applications were categorised in accordance with their transient nature. In-field failure data, compressor wheel stress profiles and

99

application duty cycles were used in categorisation of various applications. An overview of the process followed to categorise the application profiles is shown in Figure 3.

Figure 3: Flow chart showing required steps to calculate number of cycles per hour Mean time to failure obtained from field data can be used in conjunction with the process mentioned above to derive a standardised cycle for each application. This is represented in Figure 4.

Figure 4: Typical generated turbo speed as a function of time

5.

NEW APPROACH

The introduction of the electronic engine on, off-highway applications has enabled a greater control on interaction between engine speed and desired fuelling. This has made it possible for the engine manufacturers to meet the emissions legislation and expected transient response of various applications simultaneously. Therefore a revised methodology to calculate compressor wheel speeds based on the use of available engine data from existing off-highway duty cycles was developed. Since most of the available engine data is not equipped with turbo speeds, engine speed and Fuelling estimates are used to generate turbo speed plots versus time. Subsequent steps shown in figure 5 are followed in order to develop a turbo speed versus time plot.

100

Figure 5: Required procedures to determine turbo speed versus time plot from engine data The first step from the above method is to obtain a turbo speed response across the engine operating range. The most accurate way would be to use a test-bed transient cycle with turbo speed measurements. The data obtained from an engine can then be used to generate a turbo speed response profile across the engine speed and load range. This is given in Figure 6. The application of trilinear interpolation allows the calculation of turbocharger speeds at specific engine speed and load points for the application duty cycle1. Off-highway applications operate over a wide spectrum in terms of engine operating conditions. This depends on various factors including machine type, type of work, kind of control strategies between the engine and application unit etc. In addition to this, complexity of some operations performed makes it very difficult to use a standardised duty cycle for low cycle fatigue evaluation. Composite application profiles constructed using field experience and data collection can be used as a good measure for a major percentile of machine’s operating. These statistically calculated cycles could provide a good average distribution of machine operation. A typical distribution of an offhighway application is shown in Figure 7.

Figure 6: Turbo speed response for corresponding engine speed and torque

The steady state data can be smoothened using rolling averages to eliminate instantaneous changes under transient conditions caused by steady state interpolation.

1

Microsoft excel based ‘add-ins’ including ‘interpolate3d’ and ‘FINT’ can be use to carry out trilinear interpolation in excel.

101

Figure 7: Example of an engine operating profile on an application

6.

EXPERIMENTAL VALIDATION

The method discussed above was validated using a comparison between measured and calculated turbo speeds across the same engine cycle, shown in figure 8. Data from a Non-Road Transient Cycle (NRTC) was used to carry out this verification. The NRTC cycle is a representative of a typical off highway duty cycle. It is an industry standard cycle used to sign off the Tier4 interim emissions compliant engines against transient conditions.

Figure 8: Comparison between measured and calculated turbo speed The work carried out and discussed in this paper is pre-requisition for the final fatigue duty cycle calculation [1].

102

An essential requirement for LCF lifetime calculation is an accurate and representative, time and speed profile of the turbo compressor wheel for the application duty cycle. 6.1 Accuracy assessment Following the initial investigation, good correlation was observed between the measured and calculated data. The graph shown in Figure 6 shows the standard deviation of a difference between the (stead-state) measured and calculated turbo speed on the NRTC cycle. Application of normal distribution principles to the data, suggested that approximately 75% of the values are less than 9Krpm different as per the distribution. The key differences on the test cycle was found in regions well below to limiting LCF speeds for the compressor wheel. Engine system calibration under transient conditions and compressor wheel selection play a key role in determining the operating range of the turbo. This methodology should be built in to the turbo selection process. Due to the simplicity of the approach, there are a number of factors that could affect the accuracy of the results. Some of these factors have been discussed under section 7 of this paper.

Figure 9: Distribution plot of difference between measured and calculated turbo speed 6.2

Key learning obtained and improvements made

6.2.1 First order response Following the initial assessment, the resultant data can be further refined using first order response equation (1).

τ

dy + y t  = f t  dt

(1)

The turbo speed profile created using initial calculations as described under Section 5, negates the effect of system inertia. Therefore, the rainflow counting of the

103

calculated turbo speed profile resulted in higher number of discrete damaging events for a given cycle as compared the measured data. This resulted in higher average stress on the compressor wheel as compared to measured data. First order response was applied to calculated turbo speeds (In section 5) in order to substitute the effect of inertia. A close match was observed in the lifetime calculation using the first order response and measured data. The difference of calculated LCF lifetime between the measured data and calculations is shown in Figure 10. NRTC cycle was used as a basis for comparison.

Figure 10: LCF lifetime calculation using measured, initially calculated and refined data

7.

FACTORS AFFECTING LOW CYCLE FATIGUE

7.1 Measurement frequency Frequency of data measurement is an important constraint in order to accurately capture the compressor wheel speed profile. High frequency can induce undesirable noise inputs, which can add additional variability in to the life expectancy. At the same time, low measurement frequency damping out certain speed fluctuations. The results from high frequency measurements can be averaged in order to produce turbo speed profile for LCF estimation. 10Hz measurement frequency has been typically used to establish a good correlation between calculated LCF life based on a particular duty cycle. 7.2 Altitude Characteristic of the altitude has a significant effect on the speed of the turbocharger. As altitude changes, there will be a change in pressure, temperature and density. In the troposphere (up to 11km height above sea level) the relationship between altitude and density is given by equation 2. The correlation with the values obtained using equation (2) and those given by Thermodynamics and Transport properties of Fluids [5] is very good. According to Robert E. Fromm etel [6] Airflow, speed, and Pressure vary inversely as the density ratio. It also states that pressure varies as the square of the speed ratio. The relationship between speed and density is given by equation 2. The relationship between tip speed and altitude of a compressor wheel is shown in figure 11. It must be noted that different applications may have different slope as shown in figure 11. This figure is based on a typical non-wastegated turbocharger. Most modern off-highway engines use a waste gated turbocharger or a variable geometry turbocharger. The speed increase for a wastegated turbocharger will vary heavily with engine calibration.

104

h ⎞ ρ2 ⎛ = ⎜1 − 2 ⎟ hc ⎠ ρ1 ⎝

4.256

(2)

ρ1 ⎛ N2 ⎞ =⎜ ⎟ ρ2 ⎝ N1 ⎠

2

(3)

Speed of Compressor wheel above sea level ( m/s) 580

570

560

Tip Spees (m/s)

550

540

530

520

510

500

490 0

500

1000

1500

2000

2500

3000

Heigth (m)

Figure 11: Relationship between height above sea level and speed of the wheel 7.3 Effect of speed on fatigue life (LCF) Figure 12 shows the relationship between compressor wheel speeds on the fatigue life. This relationship changes with the design and size of the compressor wheel. To demonstrate this more clearly a typical off road application duty cycle was used.

Figure 12: Compressor wheel life based on wheel tip speed (POF: Probability of failure)

105

The level of tip speed of the duty cycle was increased for all data points in that cycle by the ratio of the new speed/old speed and the new fatigue life was calculated. For example to simulate the effect of fatigue life at 500m/s instead of 480m/s all the compressor wheel speeds in the data for 480m/s were increased by factor of 500/480 and the new fatigue life was calculated for the new duty cycle with a higher speed. 7.4 Operator variability Operator variability is also a factor that could affect the operation of the machinery and hence its duty cycle. Therefore, to remove the affect of operator variability, an average of a number of duty cycles or certain application sign-off cycles can be used for LCF calculation. A field study reflected that transient nature of an application is inversely proportional to the operator experience.

8.

CONCLUSION AND FURTHER WORK

Accurate representation of an application duty cycle is dependent on a large number of factors. Due to the amount of variables and complexity and variety of tasks handled by off-highway machinery, it becomes a very difficult to accurately define the ‘duty cycle’ for a typical application. Based on the wealth of machine development experience within Caterpillar, use of a quick tool was discussed for initial evaluation of LCF. Different categories of machines could be classified under categories of transience and then evaluated for LCF. This method can be further refined against the use of electronic engines on off-highway applications. Limited engine test bed data can be used to easily calculate turbo speeds for some typical duty cycles. The correlation between measured turbo speed and calculated turbo speed was good. This method however, can be further refined by use of algorithms to calculate turbo speed using turbo maps and associated inputs from the engine’s ECM. There are numerous factors that affect the resultant accuracy of the calculations. There are various alternative methods, which could be used to determine turbo speed. Amongst the spectrum of models available, there is a trade off of accuracy, speed of execution, speed of construction, speed of validation, expertise and flexibility. A good judgement of the added value of several other accurate methods is essential. There are also a number of deterioration factors on various engine systems that could affect turbo response over its life. Some of the common issues like fuel injector coking or injector growth, increasing pressure drop in the chargeair lines over the course of engine life are some of the factors. Further work will be undertaken in understanding these models to further accommodate some of the engine system related factors effecting LCF. Accommodating the effect of altitude is a key factor to completely understand the risk of LCF on a particular application. Some of the key parameters affecting the engine performance and response at high altitudes are well understood and documented. Therefore adapting the learning to encompass all the different operating conditions for an engine and its effect on turbo speed can be determined.

106

REFERENCES [1] R Christmann, F Langler, M Habermehl, P-M Fonts, L Fontvieille, P Moulin. Low– cycle fatigue of turbocharger compressor wheels online. 9th International Conference on Turbocharging 2010 London [2] Brouke P: Trilinear interpolation, http://paulbourke.net/miscellaneous/interpolation, 1997 [3] Holeman J.P: Experimental methods for engineers, 7th Ed., McGraw-Hill, New York, 2001: First order systems [4] Baines N.C: Fundamentals of turbocharging, 2005. [5] Thermodynamic and Transport Properties of Fluid. G.F.C. Rogers and Y.R. Mayhew. Fourth Edition. [6] R. E. Fromm, Jr, J. Varon, A. E. Lechin, M. Hirshkowitz cpap Machine Performance and Altitude chest 1995; 108;19578-1580

107

On the influence of thermal boundary conditions on the Thermo Mechanical Analysis of turbine housing of a turbocharger C Oberste-Brandenburg1, K Shoghi2, M Gugau1, F Kruse1 1) BorgWarner Turbo Systems Engineering GmbH, Germany 2) BorgWarner Turbo Systems Ltd, UK

ABSTRACT With the legislation on environmental issues in Automotive Industry downsizing is one solution to reduce emissions. Engine and turbocharger manufacturers have made some significant improvement by selecting the suitable turbochargers for several different applications. Due to the increasing demands, the necessity for higher exhaust gas temperatures arose over the last decade. This has resulted in higher turbine stage temperatures. The material used to manufacture the housing for the turbine stage of the turbocharger has to retain its structural integrity at the elevated temperatures. However due to the design constraints there are some cracks which occur during the engine thermal cycle test in some locations in the turbine housings. If these cracks leak exhaust gases, the criterion for housing qualification has not been fulfilled. Therefore there is a need to validate the design before tooling and manufacturing of the turbine housing. Nowadays, a methodology based on a sequential fluid (CFD) and transient thermomechanical (FEA) analysis constitutes the de facto standard for the determination of the thermo-mechanical behaviour of turbine housings. The approach yields accurate results, however the choice of the boundary conditions is essential and the numerical effort is substantial. This paper addresses questions regarding the interaction of fluid and mechanical calculation dealing with the different time scales of the observed phenomena. It discusses the influence of thermal cycle and compares and examines the effect of time and temperature on the result. Furthermore, a simplified approach is presented which allows omitting the CFD in order to constitute a much faster, but less accurate methodology for an early and rapid assessment. A comparison of the simplified and the complete methodology with experimental results is presented. Keywords: thermo-mechanical fatigue, thermal loads, turbine housing

NOTATION dh = Hydraulic diameter [m] d = pipe diameter [m] L= Length of pipe [m] Nu=Nusselt number Re = Reynolds number p=Pressure [Pa] Pr =Prandtl number

T = Temperature [K] α =Heat Transfer Coefficient (HTC) [W/m2 K] = Thermal Conductivity [W/m K] v = Velocity [m/s] y+ = non dimensional wall distance

_______________________________________ © The author(s) and/or their employer(s), 2012

83

ABBREVIATIONS 3D CFD BC CAD CFD CHT EGR ETHM HTC

1

Three dimensional CFD Boundary conditions Computer aided design Computational fluid dynamics Conjugate heat transfer Exhaust gas recirculation Equivalent turbine housing Model Heat transfer coefficient

k- PC-R2S R2S SST TC TMF TH TW

Type of turbulence model Passenger car regulated two stage turbocharger Regulated two stage turbocharger Type of turbulence model (“Shear stress transport”) Turbo Charger Thermo mechanical fatigue Turbine housing Turbine wheel

INTRODUCTION

The turbine housing of a turbocharger consists of a volute, the shrouded part of the wheel and an area downstream the wheel. Essentially there are two main types of turbine; axial or radial flow wheel. Since the radial flow is the most common type used in the turbocharger industry, the housing for this wheel type is considered for discussion. Ductile cast iron is usually used for the manufacture of turbine housing. In operation the turbine housing is subjected to high thermomechanical loads which can result in crack initiation and propagation as discussed by S. Bist et al. [1]. These cracks do not necessarily result in functional failure of the turbine housing, unless it causes leakage of the exhaust gases in the housing. These cracks occur at different locations of the turbine housing as shown in Figure 1. Traditionally the turbine housing design is validated using experimental methods such as thermal cycle test. Although different sizes have different duty cycles specified, in general these duty cycles can vary according to the requirement. The thermal cycle testing to qualify turbine housing for a particular engine application can be very expensive and time consuming. Finite element methodology can be used in rational design of the turbine housing prior to tooling of the prototype. This method can be employed to reduce the development time and the cost significantly.

Figure 1: Occurrence of cracks in different locations of the turbine housings 2

STANDARD WORKFLOW FOR THERMO-MECHANICAL FATIGUE (TMF)

2.1 General description Nowadays, the de facto standard for the calculation of TMF-phenomena for turbocharger is a sequential approach shown in Figure 2. Starting point of the calculation is the thermodynamic state of the system given by mass flow, pressures

84

and temperatures. Forced by the rapid development process, a compromise between compute time and methodological depth must be chosen. Thus, the CFD is usually done using adiabatic wall and steady-state flow boundary conditions. Based on the extracted near wall temperatures and heat transfer coefficients (HTC), a transient thermal analysis is done (see Figure 2, right approach). An alternative, considering the influence of the wall temperature of the solid on the fluid flow, is depicted in Figure 2, left approach. A conjugate heat transfer analysis (CHT) solves the fluid problem as well as the solid heat transfer problem simultaneously, leading to a more realistic picture regarding the interplay of fluid and the thermal problem as discussed by Bohn et al. [2]. The transient thermal solution delivers information about the temperature distribution of the solid at all time instances of the transient thermal cycle. This information serves as the boundary condition and main driving factor for the elasto-plastic calculation of the mechanical behaviour of the solid. However, the possible opening and closing of contacts in the model may lead to an influence of the deformation of the solid on the thermal problem. Within the standard procedure used in customer projects at BorgWarner Turbo Systems, this interaction is neglected. The resulting stresses and strains from the standard workflow are used to make a, possibly quantitative, lifetime estimation.

Figure 2: Standard sequential workflow The standard workflow is a compromise between the necessity to reflect the complexity of the physical problem and its applicability within the development process of turbocharger for passenger cars and commercial vehicles. Hence two directions of development need to be considered: -

A better reflection of the physical process with its interactions, inclusion of transient phenomena, better material modelling, and improved lifetime calculation needs to be considered. Possible approaches besides CHT, which lead to a methodology closer to physical reality are referred to in section 4. Possible approaches to address the aspect to better estimate the lifetime of the part are explained and its applicability and possible use in a working environment are shown by Längler et al. [3] and Nagode et al. [4].

85

-

Even with the development of computational power available, the standard workflow is a tool which can be used relatively late in the development process due to its complexity. Thus, the need to support the development process in a earlier stage leads to need to have a workflow which delivers results comparable to the standard workflow without its complexity. An approach used today at BorgWarner Turbo Systems is presented in section 3.

2.2 Preparation of Thermal Loads with CFD The standard TMF workflow for a turbocharger turbine sequentially combines the calculation of the turbine flow field by means of CFD and the calculation of the heat flow and the resulting stresses and strains in the material by means of FEA. The CFD as an independent part provides the complete local flow field information, which is used for aerodynamic analysis as well as for the determination of the thermal loads for FEA (wall temperature and HTC). This steady-state CFD process is common practise for all application parts at BorgWarner Turbo Systems, yet not subjected to highly sophisticated CFD methods. Thermodynamic boundary conditions are derived from simulation or engine test. A tetrahedral mesh with prism boundary layers is used for the complete turbine, turbulence is simulated with standard k- or SST-model with standard wall functions and non-dimensional wall distance (y+)>20. This allows for a relatively fast model preparation, short calculation times with a high solution robustness and in particular a comparable and extremely well validated result. The wall is treated as adiabatic, nonetheless a HTC can be provided by the model. Assuming Reynold’s analogy between energy and moment for the wall log law, heat transfer at the wall nearest cell can be treated the same way as wall shear stress, thus giving a value even for adiabatic walls. This approach produces reasonable results for further use in the TMF process. As an example, the wall HTC in the turbine housing exit area of a single stage mono scroll waste gated turbocharger is shown in Figure 3.

a) Rated power (max 5 kW/m2K)

b) Coasting (max 0.5 kW/m2K)

Figure 3: HTC on TH walls of waste gate controlled single stage TC At engine rated power with opened waste-gate (Figure 3 a)), there is locally increased HTC and overall by the order of 10 higher values compared to engine coasted condition (Figure 3 b)) with closed waste gate. 2.3 Transient Thermal Simulation A transient thermal simulation is done based on a cycle as depicted in Figure 4. The stationary points resulting from the CFD calculation are interpolated in time to provide thermal boundary conditions on the exhaust side for the complete cycle. In addition, appropriate thermal boundary conditions are applied on the outer surface of the Manifold and the housing. All relevant material parameters in the analysis are temperature dependent. All cycle parameters may vary according to customer specification.

86

Temp / T max

Cycle Definitions Steps:

3

1,20 1,00

11)

0,80 0,60

2

4

0,40 0,20

5

1

0,00 0

250

500

750

1.000

Time (s)

10 sec. Idle – 1

22)

5 sec. Acceleration Nom Load

33)

500 sec. Nom Load

44)

13 sec. Deceleration Idle – 2

55)

500 sec. Idle – 2

Figure 4: Cycle definition 2.4 Elasto-plastic simulation After the temperature field calculation, a thermo mechanical calculation is performed to obtain stresses and strains during the cycle. The structural simulation considers the same cycle as the transient heat transfer calculation. The values of plastic strain is obtained by evaluation of the complete cycle from the structural analysis. The underlying material data needs to be temperature dependent and elasto-plastic, since plastic deformations usually occur in the turbine housing during the cycle. A lifetime estimation based on the results of the elasto-plastic simulation may be added. As a standard approach for all customer projects, a lifetime estimation based on the resulting plastic strains after the thermal cycle is widely used.

3

SIMPLIFIED TMF PROCESS

3.1 Motivation and goals As described in section 2, the TMF process is a detailed, but also information intensive process. A large number of different information regarding the boundary conditions need to be collected, different large CFD and FEA-models need to be constructed. Therefore, the following aspects need to be taken into account: normal CFD process needs 1-2 weeks, until thermal BC for FEA are prepared FEA takes 4-6 weeks depending on complexity of the turbocharger the complete turbocharger CAD model is required for this process step it is not possible to investigate single parts of the turbocharger in an early project stage, until the complete model is available for example low pressure turbine stage of a regulated two stage turbocharger (R2S) design loops are necessary for almost each application turbine, since each one is a unique part To support the development process in an early stage, the need for a simplified process has been identified. Its realization is described in this section. The main goals of this process are: To construct a simplified model such that the results can be obtained within 1-2 days. The employed methodology delivers results which are comparable to the standard procedure. Thus, the choice of the setup of the simplified process has been again a compromise between physical complexity in order to get precise results and a sufficient simplicity such that it can be handled in a sufficiently short time period.

87

3.2 Description of the simplified process The structure of the simplified process is depicted in Figure 5. Keypoints of this process are: The CFD calculation is substituted by an estimation (cf. section 3.3) The general approach to the thermal and mechanical calculation is identical to the standard procedure, i.e. a transient thermal cycle is considered, the same material laws are used, etc. However, in order to speed up the calculation, only the turbine housing with appropriate thermal and mechanical boundary conditions is considered (section 3.4 and 3.5). Further estimations of the lifetime are identical to those used in for the standard procedure. To speed up the construction of the model, two tools have been developed. The first tool is used to estimate the thermal boundary conditions usually provided by the CFD (“1D-tool”) based on the procedures explained in section 3.3. The second tool automatically meshes the turbine housing based on predefined criteria. This mesh is used in solver templates [8]. A typical model is shown in Figure 6.

Figure 5: Workflow of simplified TMF

Bearing Housing Dummy Bearing Housing Bracket Integrated Manifold Cylinder Head Dummy

Figure 6: Model set up for simplified model (left) and full model (right). The bearing housing dummy is not used for the thermo mechanical calculation

88

3.3 Determination of the thermal boundary conditions There are two main reasons for the idea of simplified thermal loads: The possibility to reduce time and costs of the complete process, if a full 3D CFD could be avoided. The experience that the highly locally resolved field information is not of major importance for the resulting stresses as well as elastic and plastic strains. In accordance with these claims and in order not to lose accuracy of the complete process, a new and easy to use tool (1D-tool) based on an analytical approach was developed that enables the FEA engineer to analytically generate thermal boundary conditions for the complete turbocharger turbine. This approach saves more than a week compared to the conventional full 3D-CFD. For an analytical approach, the HTC (α) for forced convection of flow in a pipe is a function of the thermal conductivity ( ), a characteristic geometrical value, e.g. the hydraulic diameter (dh) and the Nusselt number (Nu), which itself is for a defined geometry a function of the Prandtl (Pr) and Reynolds (Re) number, as given in equation 1.

α = Nu ⋅



dh

Nu = f Pr,Re, Geometry 

(1)

Many models for heat transfer coefficients of a complete turbine stage can be found in the literature, e.g. by Cormerais et al. [5] who used values proposed by Depick and Assanis [6]. Contrary to these approaches, the idea was, to divide the complete turbine housing into separate parts, which fulfil the following conditions: a Nusselt correlation is known for each part’s representing geometry both wall HTC and temperature can be represented by one single value the flow state at the parts’ boundaries should be analytically determinable After partitioning, the complete turbine housing can be reduced to a representing model, called the equivalent turbine housing model (ETHM), see Figure 7. From the experience with many application turbine housings, this partitioning is obviously only an approximate estimation of the real conditions. The number of partitions is minimised and controlled by the number of basic components, e.g. single stage, two stage, waste gated or variable inlet guide vanes, which typify a turbine housing. Figure 7 a) exemplarily shows the partitioning for a single stage mono scroll waste gated turbocharger with integrated exhaust gas manifold. The complete turbine stage is strongly simplified and represented by only 6 different partitions, as shown in Figure 7 c). The model takes account for the variability of the waste gate, which is opened in Figure 7 b). Therefore the turbine housing exit is represented by two different partitions. Consequently, all existing control devices in the turbine housing that have significant effects on the temperature and HTC have to be included in the new tool. Eventually, the required number of partitions depends on the control devices, as shown in table 1. After reducing the turbine housing to an ETHM consisting of a sequence of straight or bended pipes with different lengths and diameters, representing the single partitions as simplified geometries, as shown in Figure 7 c), Nu equations can be used for each of the partitions, e.g. as given in equation 2, used for the manifold [7]. 0.666 ⎛ ⎞ ⎛d ⎞ ⎟ (2) Nu = 0.024 ⋅ Re0.786 ⋅ Pr 0.45 ⋅ ⎜ 1 + ⎜ ⎟ ⎜ ⎟ L ⎝ ⎠ ⎝ ⎠

89

b) T at TH exit area

a) Surface partitioning

5

TW Shroud 1

2

Manifold 1 Manifold 2

3

TH Exit 1

4

TH Volute TH Exit 2 Waste Gate

6

c) Equivalent turbine housing model – ETHM Figure 7: Partitioning and equivalent turbine housing model (ETHM) Table 1: Partitioning for different turbine types Turbine Type

Characteristic

Single Stage Mono Scroll

Waste Gate, integrated Manifold

Partitions 6

Single Stage Twin Scroll (1)

2 Volute Scrolls, Waste Gated

5

Single Stage Twin Scroll (2)

with EGR, Waste Gated

6

R2S for PC

2 Turbines / 2 Flaps

9

VTG

Variable inlet guide vanes

7

R2S for CD

2 Turbines / 1 Flap

8

The Re and Pr numbers are generally functions of aerodynamic (temperature and velocity) and geometric data as given in equation 3.

Re = f T , v, Geometry 

Pr = f T , p 

(3)

Geometry information is known from the turbine housing’s dimensions, thermodynamic data from matching. All this is used to define Nu correlations for each single partition of the ETHM. An iteratively analytical flow solver calculates the necessary values at the partition’s boundaries. Thermal conductivity is assumed to be a function of temperature and pressure [7] and finally an average value for temperature and HTC is calculated for each part of the ETHM. The data generated with this new analytical 1D-Tool are of good quality compared to data from full 3DCFD, averaged within the partition boundaries, as presented in Figure 8 for the single stage waste gated mono scroll TC.

90

2

HTC [W/m K]

Temperature [K] 1150

2500

3D-CFD

3D-CFD 1100

t2 Ex i

t1 TH

Ex i TH

M an

te

ifo ld

t2 Ex i

t1 TH

Ex i TH

Vo lu

M an ifo ld

M an ifo ld

Sh ro ud

900

2

0

Vo lu

950

1D-Tool

1

500

te

1000

Sh ro ud

1000

2

1050

1

1500

ifo ld

1D-Tool

M an

2000

Figure 8: Thermal BC by 3D-CFD vs. 1D-Tool at engine rated power

Similar results are presented in Figure 9 for a passenger car regulated two stage (PC-R2S) application. Due to the high complexity of this system, more partitions are needed to adequately represent the complete turbine housing. Obviously, the model’s accuracy of the models could be improved by using more partitions. However, an increasing complexity of the ETHM is equivalent to a higher effort for preparing the FEA model. Generally, the results for HTC generated by the 1D-tool show a trend to slightly smaller values than full 3D-data, whereas the opposite trend is observed for wall temperatures. This might add up to an almost equivalent heat load. 2

HTC [W/m K] 3000 2500

3D-CFD 1D-Tool

2000 1500 1000 500

M a Vo nif o ld lu te Sh -H ro PT ud -H PT M ix in Pi g pe C ha Vo mb er lu te -L Sh PT ro ud -L TH PT Ex i TH t -1 Ex it -2

0

Figure 9: ETHM for PC-R2S turbine housing and comparison of HTC by 3D-CFD vs. 1D-Tool at engine rated power 3.4 Thermal calculation The thermal calculation is done transiently following the same procedure as the full thermal calculation. It differs regarding the thermal boundary conditions normally provided by the CFD as described in section 2.2 and the fact, that only the turbine housing and a simple replacement of the bearing housing is used. Figure 6 shows the comparison of the simplified model (a) and the full model (b). A thorough choice of the thermal boundary conditions is necessary at the newly introduced boundaries of the model.

The results of the temperature distribution shown in Figure 10 are slightly different at manifold flanges and turbocharger exhaust for full load. The differences as shown in Figure 10 are caused by:

91

• •

Different thermal boundary conditions at the lower side of the cylinder head dummy plate (cooling conditions, i.e. film coefficients and temperatures are different). A full CFD calculation covers rapid local changes of the fluid state, resulting in differing boundary conditions for the thermal calculation of the solid. This can not be achieved by the simplified approach.

Temperatur T / Tmax

a) Simplified model

b) Full 3D-CFD model

Figure 10: Temperature distribution step 3 (Figure 4) nominal load after steady state condition 3.5 Elasto-plastic calculation The simplified elasto-plastic calculation differs from the full calculation – similar to the thermal calculation – only by the fact that only the turbine housing is considered. A thorough choice of the mechanical boundary conditions is crucial, especially at the interface to the cylinder head when a turbocharger with an integrated manifold is investigated.

Figure 11 shows the influence of variations in clamping torque and friction. Increasing friction impacts the stress and strain results much more than increasing clamping torque. This is due to the fact that the movement of the manifold branches on the cylinder head dummy is much more limited by increasing friction than by increasing clamping torque. This directly increases strain. Clamping Torque: 8 Nm Friction : 0.12

Clamping Torque: 8 Nm Friction : 0.3

3

1

Clamping Torque: 35 Nm Friction : 0.12

2

Clamping Torque: 35 Nm Friction : 0.3

4

Figure 11: Influence of clamping torque and friction on stress / strain results

92

Figure 12 shows the comparison of strain results of the simplified model (a,c) and the full CFD model (b,d). Comparing the results of plastification, the different mechanical boundary conditions have to be considered. Firstly, full model contains an additional bracket between turbocharger housing and cylinder head. Secondly, the clamping force is higher. Both are inevitable since customer specifications are not known in an early project stage. Both computational results, simplified and full, are pointing out an obvious critical area of high plastification between cylinder 2 and 3, while lower strain at the same location is observed for the full model, as shown in Figure 12 a) and b). The rear view of the critical manifold branch (Figure 12 c) and d)) shows the effects of different mechanical boundary conditions. The full model indicates higher plastifications as a result of limitation of turbine housing movement by the bracket. The movement of the manifold and turbine housing are converse.

a) Simplified model – front view

b) Full model – front view

c) Simplified model – rear view

d) Full model – rear view

Figure 12: Comparison of plastification for simplified and full model 3.6 Comparison with experiment Figure 13 shows a comparison between a result of the simplified TMF-process and a turbine housing after a thermal shock test. The plastic deformations after the cycle are depicted; areas marked red indicate a high probability of the occurrence of a crack after testing. For this test, it can be concluded that the simplified approach is able to predict the location of the crack. In addition, the level of plastification reaches values which indicate a high probability of crack occurrence. However, one

93

may keep in mind that the simplified approach may lead to less accurate predictions under certain circumstances, e.g. the behaviour of the manifold on the cylinder head is difficult to predict.

Figure 13: Comparison simplified model results – thermal shock test

4

POSSIBLE EXTENSIONS OF THE STANDARD WORKFLOW

4.1 Crank angle resolved simulation The real engine operating condition with its pulsating flow is, for economic reasons, not considered in standard workflow steady-state CFD calculation. Nonetheless, it is well known, that the usage of real unsteady boundary conditions at manifold inlets provides considerably different wall thermal loads. Thus, this more accurate approach is chosen in case of the manifold itself being a critical part, while the evaluation of the turbine housing is not subjected to this. Additionally, there is no commonly accepted opinion about the averaging procedure for the time resolved CFD data with respect to the correct weighting factors. 4.2 Contact interaction As already described in section 2.1, the contact status between parts may change during the transient calculation of the thermomechanical behaviour of the solid. As the heat transfer is contact dependent and dramatically decreases when a gap is present, this effect constitutes a non-linear connection relation between thermal and mechanical problem. This can be addressed by a coupled thermomechanical calculation, including the temperature as the fourth degree of freedom in the FEAcalculation. Unfortunately, this approach leads to larger and non-symmetric matrices in the FEA code, leading to a tremendous increase of computing time. This renders this approach as not applicable within the timescale of the development process. Another approach is a staggered co-simulation. This approach is currently not offered by the commercial FEA codes used for TMF problems.

5

CONCLUSIONS AND OUTLOOK

A fast and effective methodology was established for the thermo-mechanical analysis of the turbine housing. A classical theory was developed to substitute arduous and time consuming CFD calculations, leading to a workflow which enables the user to make an early assessment of the TMF behaviour of a turbine housing. Different studies on the choice of the boundary conditions have been performed. The results of the simplified workflow show potential to support the development process in an early stage. First comparisons with testing results show a good agreement. However, an assessment of the turbocharger employing the complete workflow is inevitable. Possible extensions of the standard workflow, which increase

94

the physical content of the simulation are discussed. Quantitative estimation of the lifetime of the turbine housing is definitely one of the most important development fields in the future.

REFERENCES [1]

[2] [3]

[4] [5] [6] [7] [8]

S. Bist, R Kannusamy, P.Tayal, E Liang: Thermomechanical fatigue crack growth and failure prediction for turbine housings, 9th International Conference on Turbochargers and Turbocharging, Institution of Mechanical Engineers, May 19-20, 2010, London, pp. 207-215 Bohn, D., Heuer, T., Kusterer, K., 2005, “Conjugate Flow and Heat Transfer Investigation of a Turbo Charger”, ASME Journal of Engineering for Gas Turbines and Power Vol.127, pp. 663-669 Längler, F., Aleksanoglu, H., Mao, T. and A. Scholz: Validation of a phenomenological lifetime estimation approach for its application on turbine housing of exhaust turbocharger, 9th International Conference on Turbochargers and Turbocharging, Institution of Mechanical Engineers, May 19-20, 2010, London, pp. 193-205 Nagode, M., Längler, F. and Hack, M: A time-dependent damage operator approach to thermomechanical fatigue of Ni-resist D-5S, Int J Fatigue (33) 5, 2011, pp. 692-699 Cormerais, M., Chesse, P., Hetet, J.-F., 2009, “Turbocharger Heat Transfer Modeling Under Steady and Transient Conditions”, International Journal of Thermodynamics Vol12(No 4), pp. 193-202 Depcik, C., Assanis, D., 2002, “A universal Heat Transfer Correlation for Intake and Exhaust Flows in a Spark-Ignition Internal Combustion Engine”, SAE Paper 2002-01-0372 VDI Wärmeatlas, 8. Auflage, Springer Verlag, ISBN 978-3-540-25504-8 Kruse, F., Dreissig, D., “Using ANSA for automated meshing and model build-up of turbocharger housings for structural analysis”, 4th ANSA & μETA International Conference, 1.-3. June 2011, Thessaloniki

95

Design optimisation of an impeller with CFD and Meta-Model of optimal Prognosis (MoP) F Frese Voith Turbo Aufladungssysteme, Germany J Einzinger ANSYS, Germany J Will Dynardo, Germany

ABSTRACT The CFD method is used to predict the flow and the compressor map. For the optimisation CAE-based parametric optimisation with respect to 15 geometry parameters, based on the primary design is used. The optimisation procedure is divided in two steps: The first one is the sensitivity study combined with the generation of the Metamodel of optimal Prognosis (MoP), where the most relevant input parameters and a quality assurance of the model could be identified. In addition the MoP can be used for optimisation to predict the performance of the compressor in the whole parameter space and to search for optimal designs. The second step, for further design improvement, is an optimisation procedure using additional solver calls, where the ARSM (Adaptive Response Surface Method) algorithm is selected for optimisation. The subdomain of important parameter is defined with respect to the result of the MoP. The whole optimisation strategy is designed to work in large parameter spaces (>10..100) with a minimum number of CFD simulations, “no run too much”; to find an improved design.

NOMENCLATURE b m N

blade thickness mass flow node

L u

blade length circumferential velocity

Greek Letters  blade angle  efficiency

 Π

temperature pressure ratio

Subscripts 1.0 compressor inlet 1.1 rotor outlet 1.2 diffusor outlet 2.0 compressor outlet B blade C compressor CW compressor wheel

ini is max n opt T

initial design isentropic maximum normalized optimised total

_______________________________________ © The author(s) and/or their employer(s), 2012

121

1

INTRODUCTION

At the present time the internal combustion engine is widely used for passenger cars and commercial vehicles applications. Today's newly developed combustion engines must meet five key demands. The engine should cause low costs, have a long life cycle, provide a good response with a low fuel consumption and meet the actual emission targets. To achieve these objectives, increasing of the engine power absolute and specific becomes more and more important. To fulfill the core requirements in engine development, today almost all diesel engines for commercial applications use an exhaust-gas turbocharger. During the design process, the combination of a radial turbine and a centrifugal compressor has a decisive influence on the economic operation of a combustion engine. For matching an exhaust-gas turbocharger with an engine the performance values, such as displacement, number of cylinders, power, mass air flow, fuel consumption, boost pressure, exhaust back pressure, etc. are needed for engine design points for the target engine. Based on the performance values, at first a suitable compressor wheel and compressor housing geometry combination is selected. The compressor map (Figure 1) which is determined based on the selected compressor components is either based on measured data or as a result of numerical data, which are determined with the help of 1D and 3D-CFD programs. The compressor map has three limitations that restrict the map width and hight. These are the surge-line and the choke-line, which exist due to the aerodynamics and the maximum compressor speed ucmax. The maximum compressor speed is limited by the allowable mechanical stresses in the impeller.

compressor total pressure ratio ∏C t [-]

CisT

4.5 4.0

surge-line

extended area

ucmax

0.76 0.74

3.5

0.72 0.70

3.0 2.5

0.78

uc=420 m/s

0.68

OP1

0.66 0.64

2.0

OP2

0.62 0.60

1.5

0.58

choke-line

1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 •

normalised compressor mass flow m

Cn

[-]

Figure 1: Compressor map with extended map range At the surge-line the flow tears by low mass flow and high pressure ratio that the delivering of fresh air flow is interrupted. The air mass flow run backwards through the compressor until a stable pressure ratio is reached with a positive mass flow rate, so the pressure is built up again. By periodic repeating of this procedure, the term "pump" is derived.

122

At the choke-line the flow reaches the speed of sound at the narrowest crosssection at the inlet of the compressor wheel. If this condition is reached, a further increase in flow-rate is not possible even by increasing the compressor speed. All flow-curves run to the maximum flow rate value at a pressure ratio of ΠCT=1. With the help of the identified compressor map the engine design points are investigated whether the map range of the map is sufficient. Depending on the application of the engine, a wide range of the compressor map is needed. If the map range is not adequate a new impeller has to be designed to get an extended compressor map (Figure 1) to reach the requirements. The impeller geometry is generated by special turbo machinery design software. The redesign of the existing impeller could be done manually in which this procedure is enormous time-consuming. Another method which is presented in this paper is the automatic optimisation by numerical methods by using 3D-CFD simulations in combination with an optimiser.

2

PARAMETRICAL IMPELLER GEOMETRY

For the automatic parametric optimisation the initial compressor wheel (Figure 2 a), which has seven main and seven splitter blades, needs to be parameterised. Today the compressor wheels are milled. To guarantee such a production process impeller flank milling is required. Therefore only the blade angle of the hub and the shroud curve for the main and splitter blade are parameterised (Figure 2 b). Figure 2 c) shows the normalised beta-angle (Bn) distribution for the main blade over the normalized blade length from the leading edge to the trailing edge. The angle distribution on hub and shroud side is expressed by a Bezier-spline which is controlled by four control points (locator 1-4). During the optimisation process only the value for Bn is changed, for each control point, to generate a new blade design, while the location of the control point is fixed in blade length. Each control point on the hub curve for the main blade is defined by the following equation: HBPi = HBPis + DXHBi

i = control point 1,2,3,4.

(1)

HBPi represents during the optimisation the new normalized beta-angle at the hub contour at locator i, which is the sum of the normalized beta angle of the start design at locator i (HBPis) and the delta value of Bn with the control point at locator i is moved (DXHBi). The optimiser only changes the value for DXHBi. The blade angle at the leading edge (locator 1) for a random design is described as follows: 0.75 = 0.8 + (-0.05) The nomenclature for the control points at the shroud curve is analogue to the hub curve and defined as follows: SBPi = SBPis + DXHBi with

• •



SBPi SBPis value DXHBi

i = control point 1,2,3,4

(2)

= shroud at main-blade with control-point at locator i = shroud at main-blade with control-point at locator i with start = delta value of Bn at main blade at locator i

123

The locations of the Bezier-control points (1-4) on hub and shroud side are located at the same normalized blade length (0, 0.4, 0.8, and 1.0) (Figure 2 c) and additional the parameter DXHBi is defined also on hub and shroud (equation 1 and 2). This is done to prevent an s-bend in the blade design because the blade is moved simultaneous on hub and shroud at the locator i with one value expressed by DXBi, which is mentioned above the input parameter for the optimiser. The characterisation of the splitter blade is analogue to the definition mentioned above of the main blade. During the optimisation process the blade thickness is unchanged and is exemplary shown for the main blade in Figure 2 d).

hub

shroud a)

b)

locator 1

2

3

4

normalised blade thickness bBn [-]

normalised blade angle

Bn [-]

1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0.0

0.2

0.4

0.6

0.8

1.0

normalised blade length (LE-TE) LBn [-]

c)

1.00

0.75

0.50

0.25

0.00 0.0

0.2

0.4

0.6

0.8

1.0

normalised blade length d) (LE-TE) LBn [-]

HUB main blade bezier spline HUB main blade bezier control points SHROUD main blade bezier spline SHROUD main blade bezier control points Figure 2: a) Initial compressor; b) parametric main blade and splitter blade; c) blade angle distribution on hub and shroud for main blade; d) thickness distribution on hub and shroud for main blade

124

In addition to the blades the meridian flow path is parameterised. Figure 3 shows a sketch of the flow path with the definition of the leading and trailing edge of the main and splitter blade. The locations of the leading edges of the blades are fixed, also the shroud contour because the identical compressor housing of the initial design should be used. inlet domain

inlet

hub HXA1_ANGLE_HUB

shroud

rotor domain leading edge blade leading edge splitter trailing edge diffusor

HXA2_ANGLE_HUB

HYA2_ANGLE_HUB

outlet

HYA1_ANGLE_HUB

HYA3_ANGLE_HUB Figure 3: Parametric meridian flow path Here the compressor wheel is milled and therefore the hub contour could be modified. This contour is parameterised with five angle parameters which are controlling the base points of the spline which describe the hub contour (Figure 3). In summary the compressor model is build up with 13 parameters: •

• 3

eight parameters for the beta-distribution (four @ main blade and four @ splitter blade) five parameters for the hub curve at the meridian flow path

CFD-MODEL

The numerical study was carried out with the CFD program ANSYS CFX 13.0. In the present paper two different models were investigated. One model is a complete compressor stage (Figure 4 b) consists of: • • • • •

inlet domain rotor domain diffusor domain volute domain outlet domain.

A full model is not applicable for an optimisation and therefore a periodic segment of 360/7 degree (Figure 3 and Figure 4 a) is used which includes only the • • •

inlet domain rotor domain diffusor domain.

125

For the inlet, rotor, diffusor and outlet domain a hexahedral mesh was used. Because of the geometric complexity a unstructured tetrahedral and prism mesh was used for the volute. The physical and numerical setup for both models is equal and is defined as follows: •

inlet: total pressure and total temperature (flow direction: normal to boundary) outlet: mass flow rotor: angular velocity wall: adiabatic rotor stator interface: frozen rotor numeric: SST-turbulence model, second order discretisation

• • • • •

The periodic segment uses additional a periodic condition boundary for the periodic faces (Figure 4 a). Before the optimisation starts a grid study for the rotor was done to investigate the influence of the mesh density regarding the results. Therefore the periodic model was used with four different mesh sizes and two different advection schemes (up wind and high resolution). An overview of the grid study is shown in Table 1. Table 1: Overview of rotor grid study Grid 1 N

uC

[-]

[m/s]

674878 High Res. 420

mc [kg/s] 0.275

ΠCWT

CWisT

Grid 2 1366156

Up Wind

High Res.

420

420

Grid 3 2744902

Up Wind

High Res.

420

420

Grid 4 5113334

Up Wind

High Res.

Up Wind

420

420

420

0.275

0.275

0.275

0.275

0.275

0.275

0.275

[-]

0.818

0.782

0.826

0.802

0.826

0.803

0.828

0.814

[-]

2.361

2.339

2.374

2.371

2.375

2.356

2.379

2.375

As an evaluation criterion the total pressure ratio (ΠCWT) and the total isentropic efficiency (CWisT) for the compressor wheel and the diffusor was used. The total compressor ratio

∏CWΤ =

p Τ1.2 p T1.0

(5)

is formed by the total pressure at diffusor outlet pT2.0 divided by the total pressure at compressor inlet pT1.0. The compressor wheel efficiency is calculated by

CWisΤ =

 ∏CWT 

⎡⎛ T ⎢⎜⎜ T1.2 ⎣⎢⎝ T1.0

 −1 

⎤ ⎞ ⎟⎟ − 1⎥ ⎠ ⎦⎥

(6)

with  as the isentropic coefficient for air, the total temperature at diffusor outlet TT1.2 and the total temperature at compressor inlet TT1.0. The grid study has shown that the Grid 2 is a good compromise between numerical accuracy and computational time and will be used for the optimisation study. Following the grid study the full model with a resolution of 12 million nodes was used to estimate the deviation between the test data and numerical results for the initial compressor stage. For the comparison the total pressure ratio and the total isentropic compressor efficiency for the full stage follows equation (5) and (6) was used

126

whereby the pressure and temperature for diffusor outlet was replaced by the quantities at compressor outlet (Figure 4 b). Figure 4 c) and d) show the comparison between the measured and simulated data by a compressor speed uc=420m/s with a satisfying accuracy. inlet

3.0

CisT [-]

CWisT [-]

2.5

2.0

1.5 0.15 0.20 0.25 0.30 compressor mass flow m C [kg/s] e)

full model

b)

outlet

0.80 0.75 0.70 0.65 0.60

0.55 0.15 0.20 0.25 0.30 compressor mass flow mC [kg/s] d) simulation 0.90

efficiency

total compressor

Π CWT [-]

1.5 0.15 0.20 0.25 0.30 compressor mass flow m C [kg/s] c) test 3.0

pressure ratio

0.85

efficiency

2.0

total compressor isentropic

2.5

pressure ratio

total compressor

Π CT [-]

a)

total isentropic compressor

periodic faces

0.85 0.80 0.75 0.70 0.65 0.15 0.20 0.25 0.30 compressor mass flow mC [kg/s] f)

periodic segment model

Figure 4: a) CFD-Model periodic segment; b) CFD-model full model; c) comparison ΠCT @ uc=420m/s: test vs. CFD-simulation full model; d) comparison CisT @ uc=420m/s: test vs. CFD-simulation full model; e) comparison ΠCWT @ uc=420m/s: CFD-simulation full model vs. CFD-simulation periodic segment; f) comparison CWisT @ 420m/s: CFD-simulation full model vs. CFD-simulation periodic segment

127

For assessment the numerical error for the periodic segment the pressure ratio and the compressor efficiency was compared to the full model and is shown in Figure 4 f). Only a difference at the surge line could be seen, because of flow separation. For the optimisation two operating points (OP) are considered (Figure 1): • • 4

OP1: uC = 420m/s; mcn= 0.73; ΠCT = 2.37 OP2: uC = 420m/s; mcn= 0.88; ΠCT = 1.68

OPTIMISATION

The optimisation procedure is carried out in three steps: 1. 2.

3.

In step one a sensitivity analysis is performed in order to determine the most important design variables. This is realized with help of the Metamodel of Optimal Prognosis (MOP, see Most and Will [1]). Using the MOP a response surface-based optimisation is carried out next. For this procedure the search for the optimum requires no direct solver runs. The determined optimum is verified finally with only a single solver call. Considering the results of the sensitivity analysis, only the most important input variables are used as design parameters within this procedure. Using the results of step two as basis, now an optimisation is performed by using direct solver calls.

Some remarks concerning the sensitivity analysis: In order to analyze the influence of the input parameters on a certain response parameter, global variance-based sensitivity measures are determined. As basis, the design space is explored with optimized Latin-Hypercube Sampling. With this stochastic sampling method, design samples are generated which cover the design space optimally by minimizing unwanted correlations between the inputs. After the generation of the samples, for each sample the solver evaluates the response values. Based on these support points in a next step an optimal approximation model is determined. This procedure, called Meta-Model of Optimal Prognosis, determines the optimal variable subspace together with the optimal approximation model, where polynomials and Moving Least Squares approximations are considered. Basis for this procedure is an objective measure to quantify the prognosis quantity of the investigated possible meta-models. For this purpose the so-called Coefficient of Prognosis (Most and Will [2]) is utilized:

CoP = 1 −

SSPrediction Ε SST

(7)

This measure quantifies the sum of squared prediction errors with respect to data, which are not used to build up the approximation model

SSE =



N i=1

 y − ˆy 

2

i

i

(8)

This non-objective error measure is scaled with the total sum of squares of the real response values

128

SST =



 yi − μ Y  i=1

2

N

(9)

The optimal variables subspace is determined by applied advance filter technology as described in detail in [1]. Once the optimal subspace was found, the optimal approximation model in this subspace is used to carry out the sensitivity analysis. Using total effect sensitivity indices, the variance contribution is quantified by the conditional output variance with respect to a single input variable (see Saltelli et al. [3])

ST  Xi  = 1 −

V  Y|Xi  V Y

(9)

The sensitivity indices determined on the approximation model are finally scaled with the CoP in order to obtain the explained variation with respect to each of the considered input variables. Applying the MOP procedure on the compressor (40 design points), see Figure 5, shows that the CoP of the total pressure ratio ΠCWT and the total temperature ratio ΘCWT is pretty high for both operation points. A value over 80% is known as a good and reliable result. Common reasons for a small value are a too small number of design points or “numerical noise” in the simulation. The “numerical noise” is reduced by the best practice study mentioned above and the monitoring of the analysis showed a stable value of the CoP. The isentropic efficiency CWisT, a more sensitive result than the other ones, has significant smaller values; i.e. we can rely to the MOP in terms of ΠCWT and ΘCWT, but we need to be careful about CWisT.

Figure 5: Coefficient of Prognosis (CoP) and important input variables on isentropic efficiency, total pressure and temperature ratio for both operating points OP1 and OP2

129

Please notice, that all results are with respect to the chosen input parameters and their lower and upper limit! The efficiency is the most important output parameter for the optimisation and this is the reason, why the MOP is not used for optimisation (It would be the fastest way!); a direct algorithm is chosen, see below. All further analysis steps are done for the relevant parameters only, which can also be seen in Figure 5. The most important variables here are: x1=DXHB1, x2=DXHB4 and x3=HX_A2_ANGLE_HUB. The parameters x1 and x2 manipulate the inlet and outlet angle of the main blade (Figure 2 c) and the parameter x3 manipulates the hub contour (Figure 3). Figure 6 shows the analysis, based on the Meta-model:

Figure 6: Meta-model for a) ηCis OP1 b,c) ηCis OP2 d) Anthill plot ηCis OP1 vs. OP2 The isentropic efficiency of OP1 depends only on x1 Figure 6a, while at OP2 it is a function of x1, x2 and x3, Figure 6b and 6c. It can be seen that a larger value of x1 would result in a better efficiency at OP1 while it would reduce the value at OP2, i.e. we have a conflict of optimisation goals. Figure 6d shows an Anthill Plot, the efficiency at OP1 vs. OP2, where one can see the assumed Pareto Front. If one chooses a certain point on the Pareto Front, one variable can only increased by decreasing the other one. The Sensitivity Analysis can be summarized: 1. The Meta-model is reliable, due to the CoP values of ΠCWT and ΘCWT 2. A reduced set of parameters was found, 3 out of 15. 3. The Meta-model is plausible, with respect to physics

130

This result is the basis for the optimisation procedure: The fastest way, using the MoP directly (instead of doing numerical simulations), is not recommended, because of the small CoP of the efficiency; i.e. a direct optimisation algorithm is required. We found Pareto conflict for the efficiency, for further resolution of this a Pareto Optimisation is required. Because these algorithms require a high number of design evaluations we did not use it in here. We resolved the conflict in terms of objectives by constraints: • “old” objective: CWisT OP1 = max and CWisT OP1 = max • “new” objective: CWisT OP1 = max and CWisT OP1 > CWisT OP1ini – 1% The new objective means, that the efficiency at OP2 should be as big as possible, while we accept a smaller one at OP1. As Optimisation algorithm we choose the Adaptive Response Surface Algorithm (ARSM). The properties of this are: • Finds the optimum, depending on the start point. From the Sensitivity Analysis we can see, that there is a global optimum • Efficient for a small number of input variables ( 1) as follows:

1

1

1

1

1

(3.6) (3.7)

where b is the constant to control the curve of the low BSRnorm efficiency fit, c is the constant calculated from intercept Z0 given as: Intercept of efficiency at BSRnorm axis,

1



(3.8)

Similarly, a curve is fitted to the MFR against BSRnorm plot using the following equation:

1

(3.9)

where cm is the intercept of the curve at 0.0 BSRnorm and m is an exponent coefficient that controls the curvature of the curve. The values of efficiency and mass flow are extrapolated over the entire range of pressure ratio. The procedure described above is carried out for both the wide and narrow maps prior to the engine simulation. 3.3 Engine Model The engine model that is used in this study is a 4.7 litre direct injection (DI) Diesel engine with some pertinent specifications shown in Table 2. To isolate the effects of turbine maps on the performance of the engine, the engine layout is kept as basic as possible without additional turbine control devices such as wastegates or bypass systems. Table 2 Basic engine specification Parameters Combustion System Capacity Compression Ratio Bore x Stroke Dimension Induction System

4

Specification 4-Stroke, V6, Diesel DI 4.7 litres 16.5 100 x 100 mm Single stage turbocharger

RESULTS AND DISCUSSION

This section discusses the findings of the engine-turbocharger simulation work that was carried out from two aspects: the impact of map data range on the map extension output and the effects of using two different map data range on the engine performance prediction. 4.1 Extension of Turbine Maps The extrapolation for mass flow ratio against velocity ratio is shown in Figure 4 for both wide and narrow maps. Clearly, it is seen that the use of different map ranges has produced a significant difference in the extended region of the map represented by lines in the figure. The use of narrow map data range results in approximately 6% higher mass flow ratio intercept at zero velocity ratios compared to that obtained with wider map data. At high velocity ratios, the effect of using different map ranges is more pronounced; again with the narrow map intercepting higher velocity ratio. This effect is likely due to failure of the narrow map to take into

212

account the nature of the curvature in the actual experimental data, hence the flatter mass flow curve compared to that obtained the wide map. Consequently, the software predicts higher mass flow rates as velocity ratio point is shifted away from the maximum efficiency points along speed-lines. Table 3 Values of curve fits and shape factors used in GT-Power map extrapolation Description Mass flow fit, lowBSR/highPR side: mass flow ratio at zero BSR (cm) Mass flow fit, exponent of the mass flow line (m)

Curve fitting coefficients Wide Map (FR) Narrow Map (NR) 1.1206 1.1801

Efficiency fit, lowBSR/highPR side: shape factor at low BSR (b) Efficiency fit, highBSR/lowPR side: zero intercept at high BSR (z0)

2.6762

1.4451

2.0832

1.9250

1.7726

1.8479

The shapes of the curves depend on the coefficients imposed in the equations that are used for extrapolations. The values of these coefficients which are imposed by GT-Power map processor are shown in Table 3 for the different maps used in this investigation. The designations “FR” and “NR” refers to the different map data ranges used in the investigation with the former and the latter being the wide map (full range) and narrow map (narrow range) respectively. Clearly, with limited data range, the failure of the extrapolation method to account for the mass flow gradient change in the actual data has resulted in extrapolation points to be more spread out over the velocity ratio. It can be seen in Figure 5 that at low speeds, the use of the narrow map results in prediction of higher mass flow compared to the wider map over almost the entire range of pressure ratio. However, at higher speeds, the difference in mass flow prediction is significant at low pressure ratios and less so at high pressure ratios. As higher pressure ratios are present at low velocity ratios, it is anticipated that the map extension at particular area is more likely to affect the outcome of the engine simulation. Therefore the actual performance is expected to be off the mark from the predicted performance if the optimum operating range of the engine is set with the turbine operating within this affected range of velocity ratio.

Mass Flow vs Normalized Velocity Ratio (Fit & Data)

1.00 0.80

MFR

0.10

φ (kg/s)(√K/kPa)

1.20

0.60 0.40

Mass Flow vs Pressure Ratio (Fit & Data)

0.08 0.06 0.04 0.02

0.20 0.00 0.00

1.00

1.50

2.00

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0

NR Fit

FR Data

NR Data

Figure 4 Effect of data width on mass flow extrapolation

3.00

3.50

4.00

PR

Norm. BSR FR Fit

2.50

50% FR Fit 100% FR Fit 50% Data

50% NR Fit 100% NR Fit 100% Data

Figure 5 Predicted mass flow parameter against pressure ratio using different map ranges

213

The extrapolation of efficiency parameter is shown in Figure 6 with experimental data points from several speed-lines. It was mentioned earlier that efficiencyvelocity ratio fit is carried out separately for low and high normalized velocity ratio values. The result shows that at low velocity ratio, the predicted efficiency for narrow maps is slightly lower than that for wide maps. This is due to the presence of data points at low velocity ratio in the wide maps as seen in the figure. At high velocity ratios, due to the presence of more data points, the intercept of efficiency is drawn inwards, thereby predicting lower efficiency. In addition, it is also worth noting that at high velocity ratios (≈BSRnorm > 1.3), the normalized efficiency data points for the speed lines seems to diverge away from each other and no longer lie on a single line. This indicates that the individual speed line efficiency spread is not accurately captured by the GT-power turbocharger model. Rough estimates indicate that this normalized velocity ratio corresponds to the pressure ratio of 1.13 for low speed lines (i.e. 50% equivalent speed) and up to 1.60 for high speed lines (i.e. 100% equivalent speed). What this implies is that the prediction of efficiency will be affected at pressure ratios lower the mentioned values in the simulation as can be observed in Figure 7.

Efficiency vs Normalized Velocity Ratio (Fit & Data)

Efficiency vs Pressure Ratio (Fit & Data) 1.00

1.20 0.80 1.00 0.60

η

ηnorm

0.80

0.40

0.60 0.40

Diverging data points

0.20

0.20 0.00

0.00

1.0

2.0

BSRnorm FR Fit FR Data

NR Fit NR Data

Figure 6 Efficiency fit for wide and narrow map data ranges against normalized BSR

3.0

4.0

PR

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 50% FR Fit 100% FR Fit 50% Data

50% NR Fit 100% NR Fit 100% Data

Figure 7 Comparison of predicted efficiency against pressure ratio using different map ranges

4.2 Engine Performance Prediction The engine simulation was carried out for engine speeds ranging from 1000 to 5000 RPM to capture the behaviour of turbine over a wide operating range. Figure 8 shows the basic predicted performance characteristics of the turbocharged engine in terms of brake power and torque whereas Figure 9 shows the volumetric efficiency obtained using the different map ranges. The volumetric efficiency indicates the difference in the breathing capability of the engine. It has to be stated here that due to the unavailability of a directly matching turbocharger turbine map the results of power, volumetric efficiency, BMEP and BSFC start to deviate out of realistic range at the lower half of the engine speed range. The discussion, however, especially as it is based on comparison of methods, stands for some significant trends to be observed, regardless.

214

Volumetric Efficiency vs Engine Speed

700

400 350 300 250 200 150 100 50 0

500 400 300 200

ηvol

600

Torque (Nm)

Power (bhp)

Power Output and Torque vs Engine Speed

100 0

1000

2000

3000

4000

5000

0 6000

1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Engine Speed (RPM)

0

T01 FR - Power

T01 NR - Power

T01 FR - Torque

T01 NR - Torque

Figure 8 Comparison of engine power output and brake torque for simulations using different map data ranges

1000

2000

3000

4000

5000

6000

Engine Speed (RPM) T01 FR - Vol.Eff

T01 NR - Vol.Eff

Figure 9 Comparison of predicted engine volumetric efficiency from using different map data ranges

The brake power and torque are directly related to the engine brake mean effective pressure (BMEP) as shown in Figure 10. BMEP represents the parameter used to compare the engine power regardless of its capacity. It is interesting to note that although the maximum BMEP predicted using both maps are almost the same, the wide map achieves this value at a lower engine speed at approximately 3000 RPM compared to 3500 RPM for the narrow map thus affecting the rating of the engine, although this could be a case of simulation speed point resolution. The use of narrow map resulted in the prediction of lower BMEP with a maximum prediction being 9.8% lower than that for a wider map at 2600RPM.

BSFC vs Engine Speed

18 16 14 12 10 8 6 4 2 0

600 500

BSFC (g/kWhr)

BMEP (bar)

BMEP vs Engine Speed

400 300 200 100 0

0

1000

2000

3000

4000

5000

6000

Engine Speed (RPM) T01 FR - BMEP

T01 NR - BMEP

Figure 10 Predicted engine BMEP for different turbocharger map data range and baseline engine

0

1000

2000

3000

4000

5000

6000

Engine Speed (RPM) T01 FR - BSFC

T01 NR - BSFC

Figure 11 Comparison of predicted engine BSFC using different map data range

The increase in volumetric efficiency also affects the fuel consumption of the engine as can be seen in Figure 11. As a result of the increase in brake power in a turbocharged engine, the brake specific fuel consumption, which is the ratio of fuel mass to power, is subsequently reduced when using the wider range map.

215

4.3

The Effect of Different Map Ranges on Engine Performance Simulation This investigation was set out to analyze the effect of map data range on the engine simulation output. As can be seen in Figure 8 to Figure 11 above, the use of wide and narrow maps indeed affects the predicted engine performance particularly in the region of 2000 to 3000 RPM engine speed despite the fact that the maps used are of the same turbine. For instance, the maximum differences in performance are 9.8% and 10.8% in BMEP and BSFC respectively at 2600 RPM engine speed.

2.2

50000

2.0

40000

1.8

30000

1.6

20000

1.4

10000

1.2

0 1000

2000

3000

PR

60000

1.0 4000

Engine Speed (RPM) T01 FR - Nact. T01 FR - PR

0.08 Efficiency & BSR vs Engine Speed

0.8

0.06

0.6

0.04

0.4

0.02

0.2

0 1000

1500

2000

2500

Engine Speed (RPM)

T01 NR - Nact. T01 NR - PR

Figure 12 Turbine operating speed and map pressure ratio for different map data range

T01 FR - φ T01 FR - η T01 FR - BSR

ηts (fraction) & BSR

Turbine Mass Flow Parameter and

2.4

φ (kg/s)(√K/kPa)

N-t (RPM)

70000

Turbine Speed and Pressure Ratio vs Engine Speed

0 3000 T01 NR - φ T01 NR - η T01 NR - BSR

Figure 13 Predicted turbine parameters and velocity ratio using wide and narrow maps

Such large difference in predicted engine performance parameters is directly related to the amount of air being delivered to the cylinder at a particular engine speed. At speed and pressure ratio points in the turbine maps where the values of efficiency and mass flow are different, the computed power and consequently compressor mass flow delivery will also be different. The condition for this to happen is when the simulation runs at the points on the maps which are further away from the maximum efficiency points on the speed lines where the values of mass flow are in the extrapolated region. To examine this further, the predicted turbine speed and pressure ratio are compared for both maps in Figure 12. It can be observed that within 2000 to 3000RPM engine speed, the wide map predicts higher turbine speed and slightly higher pressure ratio than the narrow map. This effectively led to an increase in calculated boost pressures and therefore increased mass flows into the engine cylinder. There is a need to explain the source of such a significant difference in prediction using the two maps at hand. To do this, the turbine mass flow parameter and efficiency values in the most affected engine speed range and the corresponding velocity ratio are compared in Figure 13. The specific narrow map used predicts lower turbine efficiency in as much as an average 3.8% over the entire speed range with a peak value of 10.1% difference at 2500RPM. At 2600RPM where the prediction difference is at its maximum, the difference in velocity ratio is also at its maximum with the wide map and the narrow map reading values of 0.62 and 0.49 respectively. These velocity ratio values lie at locations where experimental data is present only in the wide map and the velocity ratio for the narrow map is read from the extrapolated data region (Figure 4 and Figure 6). This is also a testament to the

216

advantage of having a turbocharger facility that is able to provide a wider range of performance data compared to conventional turbocharger facilities. What can be drawn from the above analysis is that using narrow maps, which is usually the case in current practices, may result in under-prediction of the basic performance prediction of an engine. For this particular case, the differences in predicted performance occur in the ‘useful’ range of the engine speed. This would imply that for a given requested BMEP or BSFC curve in an engine operation regime, the use of a narrow turbocharger map in a simulation may result in overspecification of a matching turbine. Inconsistencies in predicted and actual engine performance are often mitigated through calibration and appropriate fine-tuning in the later stages of development. The addition of various turbine and engine control mechanisms such as wastegates may further diminish the impact of these inconsistencies. Nonetheless, the findings from this investigation reveal that these variations can be quite substantial and their impact, more crucial.

5

CONCLUSION

The map extension procedure in commercial one-dimensional-based engine simulation software, namely GT-Power, has been implemented using maps of a turbine with different data range and the predicted basic engine performance characteristics has been compared for the two maps used. It was found that for this specific simulation, the use of the narrow maps (in the mid velocity ratio range) resulted in prediction of lower BMEP with a maximum prediction being 9.8% lower than that for a wider map. In this particular investigation, the associated brake power and torque are significantly under-predicted especially at mid-range engine speeds (2000 – 3000RPM). The use of narrow maps also results in over-prediction of the engine’s BSFC at the said engine speeds to as much as 10.8% (at 2600RPM). Such substantial inconsistencies will result in larger uncertainties in the later stage of engine calibration and drive cycle evaluation. More inconsistencies in the prediction can be expected when load simulations are carried out. The variation in performance prediction can be traced back to how the range of data affects the extension of the map. Here, GT-Power map processor fails to capture the gradient changes in mass flow and efficiency curves as the GT-Power fitting procedure is carried out. At points which are further away from maximum efficiency points along the speed lines, the performance values for each maps digress away from one another. As a consequence, the difference in the values of turbine mass flow and efficiency looked-up from the predicted maps becomes more apparent, in particular, at low pressure ratios. Although the map extension method employed by GT-Power is robust enough for use in one-dimensional software, in this case, it is the range of data entered into the turbocharger component within the software that has the most profound effect on the outcome of the simulation. This calls for an improved modelling method that is able to represent the complete physical behaviour of a turbine as an alternative to the current methods in a gas dynamics code in the future. The ultimate goal would be to have a model that is able to incorporate steady and unsteady performance parameters of a turbocharger turbine and the same time being compatible with the platform of current engine simulation codes.

217

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]

[11] [12] [13]

[14] [15] [16]

[17] [18]

218

S. Szymko, “The Development of an Eddy Current Dynamometer for Evaluation of Steady and Pulsating Turbocharger Turbine Performance,” Imperial College London, 2006. T. Otobe, P. Grigoriadis, M. Sens, and R. Berndt, “Method of Performance Measurement for Low Turbocharger Speeds,” in 9th International Conference on Turbochargers and Turbocharging, pp. 409-419. P. Moraal and I. Kolmanovsky, “Turbocharger Modeling for Automotive Control Applications,” SAE Technical Paper Series, no. 1999-01-0908, 1999. N. Watson and M. S. Janota, Turbocharging the Internal Combustion Engine. London: MacMillan Press, 1982. J. P. Jensen, A. F. Kristensen, S. C. Sorenson, N. Houbak, and E. Hendricks, “Mean Value Modeling of a Small Turbocharged Diesel Engine,” SAE Technical Paper Series, no. 910070, 1991. L. Eriksson, “Modeling and Control of Turbocharged SI and DI Engines,” in IFP International Conference, 2007, vol. 62, no. 4, pp. 523-538. N. C. Baines, “Turbocharger turbine pulse flow performance and modelling – 25 years on,” in 9th International Conference on Turbochargers and Turbocharging, 2010, no. 7, pp. 347-362. G. Martin, V. Talon, P. Higelin, A. Charlet, and C. Caillol, “Implementing Turbomachinery Physics into Data Map-Based Turbocharger Models,” SAE International Journal of Engines, vol. 2, no. 1, pp. 211-229, 2009. L. Jiang, J. Vanier, and H. Yilmaz, “Parameterization and Simulation for a Turbocharged Spark Ignition Direct Injection Engine with Variable Valve Timing,” SAE Technical Paper Series, no. 2009-01-0680, 2009. J. Serrano, F. Arnau, V. Dolz, A. Tiseira, and C. Cervello, “A Model of Turbocharger Radial Turbines Appropriate to be used in Zero- and OneDimensional Gas Dynamics Codes for Internal Combustion Engines Modelling,” Energy Conversion and Management, vol. 49, pp. 3729-3745, Dec. 2008. K. Ghorbanian and M. Gholamrezaei, “An Artificial Neural Network Approach to Compressor Performance Prediction,” Applied Energy, vol. 86, no. 7-8, pp. 1210-1221, Jul. 2009. A. Romagnoli and R. Martinez-Botas, “Performance Prediction of a Nozzled and Nozzleless Mixed-flow Turbine in Steady Conditions,” International Journal of Mechanical Sciences, vol. 53, no. 8, pp. 557-574, Aug. 2011. M. Jung, R. Ford, K. Glover, N. Collings, U. Christen, and M. J. Watts, “Parameterization and Transient Validation of a Variable Geometry Turbocharger for Mean-value Modeling at Low and Medium Speed-load Points,” SAE Technical Paper Series, no. 2002-01-2729, Citeseer, 2002. S. Rajoo, “Steady and Pulsating Performance of a Variable Geometry Mixed Flow Turbocharger Turbine,” 2007. N. Karamanis and R. F. Martinez-Botas, “Mixed-Flow Turbines for Automotive Turbochargers: Steady and Unsteady Performance,” Int. J. Engine Res., vol. 3, no. 3, pp. 127-138, 2002. C. D. Copeland, R. Martinez-Botas, and M. Seiler, “Comparison Between Steady and Unsteady Double-Entry Turbine Performance Using the QuasiSteady Assumption,” Journal of Turbomachinery, vol. 133, no. 3, p. 031001, 2011. S. Rajoo and R. Martinez-Botas, “Mixed Flow Turbine Research: A Review,” Journal of Turbomachinery, vol. 130, no. 4, p. 044001, 2008. Gamma-Technologies, GT-Suite Flow Theory Manual, V 7.1 ed. 2010.

An experimental assessment of the effects of stator vane clearance location on an automotive turbocharger turbine J R Walkingshaw, S W T Spence, D Thornhill, J Ehrhard* School of Mechanical & Aerospace Engineering, Queen’s University Belfast, UK *IHI Charging Systems International GmbH, Germany

ABSTRACT Experimental tests have been conducted on a radial turbine looking at three different stator vane positions equating to a minimum, 25% and maximum MFR condition. Separate tests have been performed investigating performance with either hub side or shroud side stator vane clearance. Through the experimental tests and CFD simulations it was found that better performance was achieved when operating with hub side tip clearance on the stator vanes. CFD revealed that the stator vane tip leakage flow modified the inlet flow angle at rotor inlet. The flow angle moved to more negative values of incidence.

NOMENCLATURE MFR P PR VGT

Mass Flow Rate Static Pressure Pressure Ratio Variable Geometry Turbine

Cθ P0 U/C

Absolute Velocity Tangential Total Pressure Isentropic Velocity Ratio

1 INTRODUCTION Due to emissions and drive cycle legislation a stronger emphasis is being put on automotive manufactures to improve part load performance of the engine system. As a result, turbocharger manufacturers are looking to improve efficiency at offdesign operating points yet meet the maximum load and MFR (Mass Flow Rate) requirements of the engine. Therefore, a turbocharger turbine used for automotive applications requires a wide operating range to deal with the varying inlet flow conditions. One method of better matching the turbine to the engine is the use of a VGT (Variable Geometry Turbine). Typically a swing vane system is employed which varies the throat area, increasing the PR and subsequently the Cθ component at the low MFR conditions (1, 2). The clearance on the stator vanes results in tip leakage and secondary flow structures. The development of these flow structures has an impact on the inlet flow distribution and performance of the turbine. As the trend is moving towards variable geometry systems, works such as that performed by Mulloy & Weber and Fredmonski et al (3, 4) need to be re-examined to assess the effects variations in the inlet flow field, caused by stator vanes, have on turbine performance. Their initial work showed that modification to the inlet

___________________________________________ © The author(s) and/or their employer(s), 2012

243

blade angle of the turbine had an impact on the value of U/C at which peak efficiency occurred. Mulloy & Weber found that at low U/C operating points the rotor with the forward curved blade (blade curved into the approaching flow) was better aligned, increasing the MFR and efficiency. The backward curved blading had the opposite effect, aligning better at high rotational speeds with the round nosed blading being a compromise. Extensive experimental based studies on the effects of stator vanes and incidence angle on a radial turbine were performed by Spence & Artt in the late 90s (5, 6, 7). Although providing an assessment of stator vane to rotor throat area ratio and current incidence loss models, the turbine was never studied at the low U/C values typically found in automotive turbocharger applications. In addition, no tip clearance was present on the stators investigated. O’Neill et al (8) covered the flow development in a stator vane system and highlighted the effects of tip leakage in the stator vane domain but no analysis was carried out on the effects of the flow and the development within the rotor. Studies by Palfreyman et al and Barr et al (9, 10) showed via CFD analysis the flow development within a mixed flow and backswept design turbine respectively. These studies showed the potential benefits of mixed flow and backswept turbines for meeting automotive requirements. More recent computational studies performed by Natkaniec et al and Walkingshaw et al (11, 12, 13) show the complex flow field which is generated as a result of the stator vanes. The study by Natkaniec provided a breakdown of the flow structures found in a modern VGT system. In addition, the study revealed the highly transient nature of the flow but it did not comment on the effect experienced by the rotor. Furthermore, the study was only performed at one stator vane position. Walkingshaw showed the effects of operating with either hub or shroud clearance on the stator vanes. It was established that an efficiency gain was had by operating with hub side stator vane clearance. The study also showed that the stator vane tip leakage produced a vortex which augmented the flow and improved incidence and flow development in the rotor passage from 0% to 10% span at low values of U/C. Due to the lack of detailed flow measurements on a VGT system operating at a range of points typically found in automotive applications a set of experimental tests have been performed in the current study. The tests were carried out on a radial turbine operating at three different stator vane positions equating to a minimum, 25% and maximum MFR condition. Independent tests were performed to ascertain the effect of operating with shroud side and then hub side stator vane tip clearance and the results are presented below.

2 EXPERIMENTAL SETUP A test rig was developed to provide global performance characteristics as well as detailed flow measurements within the turbine. Existing automotive components were scaled and the tests performed at a turbine inlet temperature of 410K. The low inlet temperature combined with rig insulation minimised heat loss and provided accurate efficiency calculations based on the temperature drop across the complete stage. The flow was evenly delivered around the circumference of the VGT system by means of a torus. The flow passed through a set of pre-swirl vanes which imparted the same level of swirl as that offered by the volute on the automotive system. A pinch on the endwalls was used to accelerate the flow and minimise boundary layer effects at stator vane inlet. Following this the flow exited through the turbine and exhaust diffuser. Figure 1 provides a sectioned view of the rig and Table 1 shows the stator and rotor geometry.

244

Figure 1: Test rig assembly Table 1: Stator and rotor geometry Stator Vane Tip Clearance (% Stator Vane Height)

2.3%

Number of Stator Vanes

11 Baseline Rotor

Inlet Diameter

90 mm

Inlet Blade Height

15.1 mm

Blade Number

9

Static Pressure Max MFR SV Position

Static Pressure Min MFR SV Position

Static Pressure 25% MFR SV Position

Figure 2: Static pressure measurements in stator domain

245

Two sets of stator vane nozzles were produced, one with clearance on the hub side the other with clearance on the shroud. Thirty six steady state static pressure measurements were taken in the stator and rotor domains. Due to limitations of space it was only possible to take static pressure measurements for the stator vanes with shroud side clearance. Figure 2 shows the location of the static pressure measurements relative to the three different stator vane positions. Three sets of static pressure, circumferential traverses were taken at locations which equated to the CFD predicted low pressure regions, generated by the tip leakage vortex of the stator vanes (12, 13). In addition, 10 static pressure measurements were taken at the shroud to provide further data for analysis and CFD validation.

3 TEST RESULTS AND DISCUSSION From the test data in Figure 3a and b there is a clear advantage to operating with hub side stator vane tip clearance at the more closed stator vane positions. The "H" in the legend of the graphs represents hub side clearance on the stator vanes and "S" shroud side. At the minimum MFR stator vane position a constant improvement of 4.5% in efficiency can be seen at different PR and rotor speeds. At the 25% stator vane position the improvements in efficiency are 3% and are constant across the speed range and PR. At the maximum MFR stator vane position a drop in efficiency of 1.5% occurs as seen in Figure 3c. The cause of the improved efficiency when operating with hub side clearance was discussed by Walkingshaw et al (12) and is due to the augmentation of the flow angle in the region of the stator vane tip leakage flow. The stator vane tip leakage flow generates two vortices, one at the leading edge and the other just after the spindle of the stator vane. This causes the incidence angle to shift to the more negative direction. When operating with hub side clearance this improved incidence angle can be utilised. However, when operating with shroud side clearance the flow passes through the tip gap present on the rotor. Due to the blade angle distribution of the rotor, the improved incidence angle is not sufficiently turned in the negative direction to result in flow reattachment, on the suction surface of the rotor blade in this shroud region. Furthermore, the CFD study performed by Walkingshaw et al (12) showed increased tip leakage flow in the rotor passage when operating with tip clearance on the hub side of the stator vanes. During the testing some flow visualisation captured the formation of the vortices as seen in Figure 6. Good correlation is seen between Figure 6a and b, the test data and CFD model (12, 13). The vortices were also seen on the hub side at both the minimum and 25% stator vane position. This shift of incidence angle in the more negative direction is beneficial at the off design situations where the rotor typically encounters strong positive incidence. The shift in the incidence angle towards the negative direction however, causes a drop in efficiency at the fully open stator vane position as seen in Figure 3c. The effects of the tip leakage flow diminish as the stator vanes become more open. This is attributed to less blade loading across the stator vane at the more open positions, resulting in less tip leakage flow. The data in Figure 3a, b and c also show that the PR has no impact on the extent at which the stator vane tip leakage flow effects performance. This correlates well with the CFD study performed by Walkingshaw et al (12) where it was found that the ratio of stator vane tip leakage flow to total mass flow was dependant only on stator vane angle and was not affected by PR.

246

Δ+4.0

Δ10%

a)

Δ10%

Δ+3.0

b)

Δ-1.0

Δ10%

c) Figure 3: Efficiency against PR for three different stator vane positions

247

Δ+8.8%

a)

Δ+10.6%

b)

Δ-1.5%

c) Figure 4: MFR against PR for three different stator vane positions

248

Δ10%

Δ0.1

a)

Δ10%

Δ0.1

b)

Δ10%

Δ0.1

c) Figure 5: Efficiency against U/C for three different stator vane positions

249

Static pressure measurements were taken in the stator domain at the minimum stator vane position and show good correlation with the CFD results as seen in Figure 7. This is further proof of the presence of the two stator vane tip leakage vortices. However, the test data would imply that the spindle does not generate as strong a low pressure region as that predicated by the CFD. From the results in Figure 6a it is difficult to assess if the two vortices join before entering the rotor as seen from the CFD in Figure 6b. From the static pressure measurements on the rotor shroud in Figure 8a it is also difficult to conclude that they do join. This plot does show that for the shroud side stator vane tip clearance configuration, holes 1 and 4 have low static pressures. This may be a result of high velocity flow passing from the tip gap of the stator vane through the tip gap of the rotor. It also further supports the fact that operating with shroud side clearance on the stator vanes results in more tip leakage flow in the rotor. Figure 8b shows that the effect of the stator vane tip leakage flow at the maximum stator vane position is well mixed out before reaching the shroud of the rotor. Minimal difference is seen with the two different configurations at the maximum stator vane position. Both plots in Figure 8 show good correlation with the CFD. The CFD static pressure data for the rotor was produced using length averaged turbo lines at the location of the experimental static pressure holes. The mass flow has increased for the minimum and 25% stator vane configuration operating with hub side clearance. This can be seen in Figure 4a and b. The cause can be seen in Figure 9. The tip leakage flow causes a non-uniform flow field but results in improved incidence which has caused the flow to remain attached to the suction surface of the rotor blade near the hub, in Figure 9a. However, in Figure 9b where the clearance is on the shroud side of the stator vane the flow has separated due to the highly tangential flow. In Figure 4c less mass flow is found with the hub side clearance setup at the maximum stator vane position. In this instance the rotor chokes. Most of the tip leakage flow in the rotor is driven by the rotor blade loading at this operating point. However, when operating with hub side clearance the CFD showed that the shift in incidence angle to the more negative direction caused a vortex to form on the pressure side of the rotor. This passed along the pressure surface and then through the tip gap of the rotor resulting in additional tip leakage flow. This rotor tip leakage flow passed through the throat at the exducer of the rotor resulting in additional blockage which reduced the MFR. Figure 5a, b and c show the efficiency against the velocity ratio. Included in these plots are the predicted CFD data. It is worth noting that heat loss effects are present at the lower pressure ratios. It can be seen that at the more open stator vane positions the rotor speed has a larger effect on efficiency. This is due to the fact that the flow velocities as well blade loading has increased. This produces additional loses affecting the efficiency. The CFD model used had the same aerodynamic flow path as the test rig, more details of which can be found in Walkingshaw et al (12, 13). Due to time restrictions three different speed lines were produced at the three different stator vane positions for the hub side clearance configuration. Only one speed line at the 25% stator vane position was run for the shroud side clearance configuration. At the minimum, 25% and maximum stator vane positions, operating with hub side clearance resulted in a difference of 8%, 11% and 5% respectively for the predicted and measured efficiencies as seen in Figure 5. It is suspected that at the 25% stator vane position some heat loss effects have caused the difference to be quite large. The difference in the MFR between the predicted and measured values, for the hub side configuration were, 4%, 3% and 5% for the minimum, 25% and maximum stator vane positions respectively.

250

a)

b) Figure 6: Flow visualisation of stator vane tip leakage flow captured in the stator vane tip clearance gap

Figure 7: Static pressure measurements taken at the minimum stator vane position shroud side

251

a)

b) Figure 8: Static pressure measurements taken at the rotor shroud for a) minimum and b) maximum stator vane position In general it is thought that the more open the stator vanes the more accurate the agreement between the CFD and test data is expected. The CFD model used during this study used stage interfaces between the domains. In Walkingshaw et al (12) a study had been performed on the effects of interface selection and differences had been seen in the time average transient and stage interface results. It is recognised that the mixing losses incurred as a result of the stage interface not only affect the efficiency values but can impact the MFR calculations as well. Even with these differences the CFD has captured the trends of both the MFR and efficiency well. In addition, the CFD has been able to accurately predict the presence and location of the stator vane tip leakage flow as well as the flow passing through the shroud at the rotor.

252

a)

b)

Figure 9: Velocity contour plot at 10% span a) hub side stator clearance b) shroud side stator clearance

a)

Max

Min Velocity

b) Figure 10: Trailing edge wake of a VGT system in the closed position a) experimental b) CFD velocity contour plot When looking at Figure 6 and 10 it can be seen that at the closed stator vane position the flow entering the rotor is highly non uniform. Figure 10a shows the trailing edge wake compared to the entropy losses predicted by the CFD. Good

253

correlation is found between the two. However, both the tip leakage vorticies and the trailing edge wake are mixed out before entering the rotor domain in the CFD model. Therefore, careful assessment needs to be undertaken to decide which CFD modelling approach should be used when analysing a VGT system since it is clear that at the closed stator vane positions the true nature of the flow is highly unsteady.

4 CONCLUSIONS 1)

2) 3)

4)

5) 6)

It has been found through experimental tests that operating with hub side stator vane tip clearance provides a 4.5% and 3% improvement in efficiency at the minimum and 25% stator vane positions. MFR at the minimum and 25% stator vane positions is also increased. At the maximum stator vane position the efficiency and MFR are lower when operating with hub side clearance. Through flow visualisation, static pressure measurements and CFD the presence of two tip leakage vortices have been found. One forms from the leading edge and the other after the spindle of the stator vane. These vortices change the flow angle close to the endwall and shift it to a more negative incidence angle. The more negative flow angle can be utilised at low values of U/C by the rotor when operating with hub side stator vane clearance but results in additional tip leakage flow in the rotor when operating with shroud side stator vane tip clearance. At higher values of U/C the more negative incidence angle results in poorer performance. As the stator vanes become more open the impact of the stator vane tip leakage flow reduces. Due to the presence of the stator vane tip leakage vortices as well as the trailing edge wake, the flow entering the rotor is non-uniform. As a result there is some discrepancy between the steady state CFD model and the test results. Better correlation is found between the tests and CFD at the more open stator vane positions. This is due to the inlet flow being less affected by the presence of the stator vanes and the stator vane tip leakage flows having more time to mix out before entering the rotor.

REFERENCE LIST (1) (2) (3)

(4)

(5)

(6)

254

Japikse, D. & Baines, N. C., 1994, “Introduction to Turbomachinery”, Concepts NREC, Wilder, VT and Oxford University Press. Baines, N. C., 2005, “Fundamentals of Turbocharging”, Concepts NREC, Edward Brother Incorporated. Mulloy, J. M. & Weber, H. G., 1982, “Radial Inflow Turbine Impeller for Improved Off-Design Performance”, ASME Proc. of the 27th International Gas Turbine Conference and Exhibit, New York, USA, 82-GT101 Fredmonski, A. J., Huber, F. W., Roelke, R. J. & Simonyi, S., 1991, “Design and Experimental Evaluation of Compact Radial Inflow Turbines”, NASA Lewis Research Centre, Report AIAA-91-2127. Spence, S. W. T. & Artt, D. W., 1998, “Experimental performance evaluation of a 99.0 mm radial inflow nozzled turbine with different stator throat areas”, Proceedings of the Institution of Mechanical Engineers, Part A: Journal of Power and Energy, February 1, 1998; vol. 212, 1: pp. 27-42. Spence, S. W. T., Doran, W. J. & Artt, D. W., 1999, “Experimental performance evaluation of a 99.0 mm radial inflow nozzled turbine at larger stator-rotor throat area ratios”, Proceedings of the Institution of Mechanical

(7)

(8)

(9)

(10)

(11)

(12)

(13)

Engineers, Part A: Journal of Power and Energy, May 1, 1999; vol. 213, 3: pp. 205-218. Spence, S. W. T. & Artt, D. W., 1998, “An experimental assessment of incidence losses in a radial inflow turbine rotor”, Proceedings of the Institution of Mechanical Engineers, Part A: Journal of Power and Energy, February 1, 1998; vol. 212, 1: pp. 43-53. O’Neill, J. W., Spence, S. W. T. & Cunningham, G., 2005, “An Assessment Of Stator Vane Leakage In A Variable Geometry Radial Turbine”, Proc. of the ETC 6th European Conference on Turbomachinery, Volume II, 065_04/65. Palfreyman, D. & Martinez-Botas, R., 2002, “Numerical Study Of The Internal Flow Field Characteristics In Mixed Flow Turbines”, Proc. of ASME Turbo Expo, Amsterdam, GT-2002-30372. Barr, L., Spence, S. W. T. & Eynon, P., 2008, “Improved Performance Of A Radial Turbine Through The Implementation of Back Swept Blading”, Proc. of ASME Turbo Expo, Power for Land, Sea and Air, GT2008-50064. Natkaniec, C. K., Kammeyer, J. & Seume, J. R., 2011, “Secondary Flow Structures And Losses In A Radial Turbine Nozzle”, Proc. of ASME Turbo Expo, Power for Land, Sea and Air, GT2011-46753. Walkingshaw, J. R., Spence, S. W. T. & Ehrhard, J., 2011, “A numerical study of stator vane tip leakage effects on flow development in a variable geometry turbocharger turbine”, Proc. of the ETC 9th European Conference on Turbomachinery, B269. Walkingshaw, J. R., Spence, S., Ehrhard, J. & Thornhill, D., 2010, “A Numerical Study Of the Flow Fields In A Highly Off-Design Variable Geometry Turbine”, Proc. Of ASME Turbo Expo, GT2010-22669.

255

Development of a common dual axle VNTTM for single- and two-stage off-highway applications J Wilson, M Avila, P Davies, N Theiss, B Zollinger Honeywell International, Honeywell Turbo Technologies, USA and France

ABSTRACT Honeywell Turbo Technologies has long been a leader in both the on- and offhighway commercial vehicle turbocharger segments. Variable geometry turbines have been used in on-highway applications since the 1990's but to date have had limited usage in off-highway. This paper describes how the patented Dual Axle Variable Nozzle Turbine technology has been adapted and applied for use in the modern heavy duty offhighway environment. It details performance targets and mechanical features that were optimized and combined with a new electro-hydraulic actuation system to meet the stringent performance, durability, and emissions targets of a modern TierIV-compliant off-highway diesel engine. The result of the development is the DutyDrive VNTTM which entered production at the end of 2010. The high performance and broad flow range of Honeywell’s Variable Nozzle Turbine allowed John Deere to use one common turbocharger to satisfy over 20 unique power ratings, including both two-stage and single stage applications, greatly reducing complexity and development costs. Additionally, this new generation turbocharger assisted John Deere to increase engine power and torque by 10%, while meeting reliability expectations in the high-temperature, high-vibration under-hood environment of a modern premium off-highway application.

1

INTRODUCTION AND BACKGROUND

Variable geometry turbochargers have been used for a number of years for onhighway applications ranging from passenger vehicles to buses to medium and heavy duty trucks. A key technology which Honeywell has developed for medium and heavy duty truck and bus applications is the patented Dual Axle Variable Nozzle Turbine (DAVNTTM). This technology has been used with both pneumatic and electronic actuation, and has demonstrated field reliability on a wide variety of applications. To support the needs of United States light truck customers, Honeywell Turbo Technologies (HTT) has also developed an electro-hydraulic (EH) actuation system to mate with the turbochargers used by those customers. This is a well established technology, which has been used in millions of light truck turbochargers since 2003.

___________________________________________ © The author(s) and/or their employer(s), 2012

233

Variable geometry turbochargers have only recently been used for off-highway applications with the introduction of more stringent emissions requirements. Offhighway applications can present unique challenges for turbochargers, and specifically for variable geometry turbochargers. Typically off-highway engines spend more time at high engine speed than similar displacement on-highway applications. From the turbo side, this can mean different aerodynamic performance characteristics tailored for the off-highway operating conditions. Also, the variety of applications which an off-highway engine may go into can be significantly more diverse than many on-highway engines. This wide range of applications can represent a challenge to the engine manufacturers to cover all of the end users’ requirements with the minimum number of engine (or turbocharger) part numbers. The wide range of applications also drives a wide range of duty cycles, and some applications may work in especially severe environmental conditions, all of which the turbocharger must be designed to survive. Additionally, the need for clean sight lines on off-highway machines drives tight packaging restrictions that lead to high under-hood temperatures. These considerations can drive specific turbocharger designs and validation requirements for use in the offhighway market. As the development team was evaluating their product portfolio, they found that combining two proven technologies, EH actuation and DAVNTTM, into a single turbo could provide a very strong offering for the off-highway market. The EH actuation system from the US light truck market provides compact packaging size and proven high-temperature actuation capability, both of which were established as critical requirements for this application. The patented variable geometry architecture of the DAVNTTM has demonstrated itself to be robust under the demanding operating conditions and long life requirements of heavy duty trucks. Coupling these two designs seemed like a logical integration of two proven technologies. This paper shows how the resulting system can then be optimized to meet the stringent critical to quality characteristics of off-highway customers for Interim TierIV and Final TierIV emissions levels.

2

PERFORMANCE TARGETS AND AERODYNAMIC DEVELOPMENT

To meet the requirements for Interim TierIV emissions, as well as the range of power ratings required for this engine, the air system architecture included options for both single- and two-stage turbocharging. Figure 1 shows the typical turbine operating regions of single-stage and two-stage turbochargers, as well as typical operating region for regeneration of the engine’s exhaust after-treatment system. Engines with exhaust after-treatment systems will periodically have to go through a phase where the engine is operated under specific conditions to allow regeneration of the after-treatment sub-system. During this phase, it is important to keep the exhaust temperature high enough to facilitate regeneration. On modern engines, using the VNTTM to increase exhaust temperatures during regeneration has added a new critical operating zone to the turbocharger performance requirements. The desire to cover the three distinct operating zones shown in Figure 1 with a common VNTTM turbocharger drove the development of new turbine aerodynamics in order to provide usable efficiency over this required wide range of operating conditions.

234

0.25

Vanes Fully Open

0.2

Turbine Corrected Flow

2-Stage

Single Stage

0.15

A/T Regen

0.1

Vanes Fully Closed

0.05

0 1

1.5

2 2.5 3 Turbine Expansion Ratio

3.5

4

Figure 1: Typical engine operating regions on turbine map The design of the VNTTM turbine stage requires managing multiple trade-offs between performance and durability. This VNTTM turbine stage design process used analytical optimization tools to define the final design based on the key trade-offs. HTT’s aerodynamic design tools as well as the turbine high cycle fatigue predictive methodology described by Kulkarni1 were further refined and calibrated through this development project. In order to make this complex analytical design process most efficient, it was important to break the design optimization problem into logical and manageable sub-processes. The general design and validation process for the turbine stage is shown in Figure 2. This design process was used to define limits for the design in order to meet both performance and durability requirements.

Figure 2: VNTTM turbine stage development process Progenitor designs were created for the turbine wheel, vanes and turbine housing volute. These designs set the starting point for optimization routines. The engineering team selected initial designs based on experience and simple models. It was important to establish sound starting designs to avoid losing time iterating to local optimums in the design space during the next optimization phase. To address the increased complexity of the performance problem statement, an additional design degree of freedom was added to allow for a cambered, instead of a straight, vane shape.

235

Initial optimization of the vane kinematics and wheel aerodynamics sub-systems was completed before starting work to verify acceptable turbine stresses. These two optimizations allowed the team to freeze certain design parameters (vane & blade count, wheel diameter, etc.) before entering the next computationally intensive design phase. It was recognized that optimizing all design parameters at a sub-system level will not likely result in an optimal system-level solution. Definition of the design space at a higher level was made to allow clear understanding of trade-offs between such high-level objectives as efficiency, high cycle fatigue life, and actuation forces. Designed analytical experiments were used to explore the design space for certain wheel and vane design parameters. These analyses quantified the sensitivity of turbine blade stress, turbine stage efficiency, and vane loading to the design parameters. These parameters included location for spline points on the vanes and blade camber as well as thickness distributions. The design parameters were then optimized to arrive at a final wheel and vane design to achieve maximum stage efficiency within known safe limits for wheel high cycle and low cycle fatigue, as well as vane loads to meet actuation response, hysteresis, and kinematic wear requirements. This design was then taken to the next step, which was to produce hardware and conduct validation tests to verify the final product met its targets. These consisted of performance and function tests as well as durability tests consistent with internal and customer product validation requirements. During the development, vane hysteresis due to aerodynamic load reversal was found within the engine operating range. This was not predicted through turbine stage CFD simulation, but was confirmed through empirical measurement on the engine. Subsequently, a CFD model was created of the full engine exhaust system through close collaboration between HTT and John Deere. This full system model showed that the flow field of the exhaust entering the turbine volute affected the loading on the vanes, and the model was able to predict the reversal seen on the engine. This simulation tool was then used to design a new turbine housing volute to eliminate the reversal across engine operating conditions, which was verified empirically on the engine.

Vane Load for New Turbine Housing Linkage Force (N) < -5 -5 – 0 0 – 5 5 – 10 10 – 15 15 – 20 20 – 25 25 – 30 30 – 35 > 35

175 150 125 100 75 50 1.5

2.0

2.5

3.0

Expansion Ratio

3.5

Turbine Corrected Flow (g/s)

Turbine Corrected Flow (g/s)

Vane Load for Original Turbine Housing 200

200

Linkage Force (N) < -5 -5 – 0 0 – 5 5 – 10 10 – 15 15 – 20 20 – 25 25 – 30 30 – 35 > 35

175 150 125 100 75 50 1.5

2.0

2.5

3.0

3.5

Expansion Ratio

Figure 3: Turbine vane aerodynamic loading versus operating conditions Figure 3 shows a summary of vane aerodynamic loads versus turbine conditions for the original turbine housing design as well as the new turbine housing design. The left side diagram of Figure 3 shows the loading with the original turbine housing which exhibited aerodynamic load reversal in the local region around turbine expansion ratio of 1.7 and turbine corrected flow of 200 g/s. The right side diagram shows that the new turbine housing increased vane loading approximately 15N near this critical zone in order to eliminate reversal across in that area, while only increasing the loading between 0 and 5N across other operating conditions.

236

This small increase across other conditions was proven to be safe through analysis and durability testing. The outcome was a design which eliminated the reversal without impacting the performance or durability of the turbocharger, and still staying within the packaging space requirements of the engine compartment. Figure 4 shows the performance benefit achieved from the final production design of the new turbine aerodynamic development versus the previous generation turbine design in production. By making this significant improvement to the turbine efficiency and tailoring the flow range, it allowed use of a common turbocharger to cover all three key operating regions shown previously in Figure 1. This meant a common variable geometry turbine stage could be used to cover the customer’s 20+ power ratings across a wide variety of engine applications. This new turbocharger also assisted the customer to increase peak torque and rated power values by 10% above their previous emissions level engine, through single-stage turbocharger and two-stage turbocharger air systems. Turbine Efficiency at Exp Ratio =2

Turbine Efficiency at Exp Ratio =3

0.65

0.7

5 pts

0.6

Turbine + Mechanical Efficiency

0.75

Turbine + Mechanical Efficiency

0.7

5 pts

0.65

0.55

0.6

0.5

0.55

0.45

New Generation

0.4

Previous Generation

0.35

0.3

0.5

New Generation

0.45

Previous Generation

0.4

0.35

0

100 200 Turbine Corrected Flow (g/s)

300

0

100 200 Turbine Corrected Flow (g/s)

300

Figure 4: Turbine efficiency improvement of new turbine stage design

3

ACTUATION SYSTEM DEVELOPMENT AND OPTIMIZATION

The target to use a common turbocharger design for both single- and two-stage turbocharging options as well as this being an off-highway application drove a challenging set of design targets for the turbocharger actuation sub-system. The actuator would be required to operate within extreme levels of under-hood temperature and vibration. It would be used on applications with high levels of dust and put into some extremely corrosive environments. It was also necessary to minimize the size of the turbocharger in order to allow room for the installation of the two-stage system in applications with limited packaging space. Based on these requirements, HTT recommended the electro-hydraulic actuation system from its actuation portfolio. 3.1 Optimizing actuation for turbo loads The starting point for the actuation system design was the same EH concept currently used in light truck turbochargers, which use a different VNTTM mechanism design specific to those application requirements. The first step was to optimize the actuation system design for the force requirements of the new turbine stage mentioned in Section 2 of the paper. The maximum load values summarized in Figure 3 were used to confirm the robustness of the components of the kinematic mechanism between the actuator and the vanes. The mechanical strength of these components was found to be acceptable as originally designed. However, based on these load measurements as well as other empirical testing together with the customer, it was determined that increased actuation force was required to meet the response time targets against these aerodynamic loads. This led to the design of a larger diameter actuation piston to deliver the required force.

237

3.2 Improving control and reducing hysteresis The tightening emissions and fuel economy regulations and desire for more precise air system control have driven targets for improved precision of turbocharger control and reduced turbocharger system hysteresis. With this in mind, the development team decided to implement an actuator which was an evolution of the mechanical feedback (MFB) system used previously for light truck applications. The EH actuation system’s control valve controls the VNTTM vane position by charging and venting oil on each side of a piston. A rack attached to the piston rotates a gear (a rack and pinion mechanism) to control angular position of the vanes. For the historical MFB system, a cam attached to the rotating gear (pinion) provides displacement feedback to the control valve through the cam follower. This displacement compresses a feedback spring to move the valve’s spool back to the center closed position when the actuator reaches the desired position. As the control signal provided by the engine control unit (ECU) ramps up and down, the vanes go from full open to full closed and back. The mechanical interactions of the cam/ follower as well as the feedback spring within the control valve were seen as areas where friction as well as wear over time could affect the system stability and hysteresis. This led to the development of an electronic feedback (EFB) actuation system which eliminated these components that have friction or could wear over time. Figure 5 shows the schematic diagrams of the MFB and EFB systems.

Figure 5: Schematics for mechanical and electronic feedback systems Similar to MFB, the EFB control valve controls the actuator position by charging and venting oil on both sides of a piston, which moves the VNTTM vanes through a similar rack and pinion mechanism. But in this case, the cam attached to the rotating gear (pinion) provides displacement feedback directly to the ECU via a magneto-resistive (MR) sensor. When the EFB control valve is de-energized, the vanes are in full open position. If the ECU desires to move the vanes to 50% closed, it will provide a current within a certain range to tell the control valve to close the vanes. When the MR sensor confirms the vanes have reached the desired 50% closed position, the ECU will provide the “null” current to keep the control valve in its center closed position, and therefore, maintaining the 50% commanded position. Because of the closed loop system, if the actual position drifts from the commanded, the ECU will provide the necessary current change to bring the position back to where is desired, and then it will move back to null current to maintain it. With the elimination of the mechanical friction components mentioned above, the actuation system hysteresis can be significantly reduced as shown in Figure 6.

238

10

8

% Hysteresis

6

4

MFB EFB

2

0

-2 0

20 40 60 80 Commanded Vane Position (% closed)

100

Figure 6: Actuation system hysteresis for MFB and EFB systems 3.3 Control system development using mathematical models The adoption of the EFB mentioned above to optimize stability and hysteresis also meant that all logic and algorithms for the control of the VNTTM now had to be located in the engine ECU. It was therefore critical that HTT worked closely with its lead customer, John Deere, to define robust and optimized control algorithms. Developing these control algorithms was an important part of the development partnership between HTT and the customer. To minimize the hardware design iterations and speed-up the fine tuning of the control algorithm, HTT developed a Matlab-Simhydraulic analytical model to predict the behavior of the actuation system. This model was correlated to actual parts through bench testing and delivered to the customer to be used in their engine simulation model to predict the effect of PID gains on the actuation system. This model also allowed HTT and the customer to better understand the flow characteristics of the EFB control valve and oil circuit analytically and helped in the understanding of risk areas for stability or control issues. Although the final gain calibration still needed to be defined empirically on the engine, having this analytical tool helped to reduce product development time. The following tasks were completed using this analytical simulation model: 1) 2) 3)

Initial modeling to understand impact of control valve flow gains Effect of control signal on the actuator response Response time predictions of actuation system under a wide range of oil pressures and temperatures

3.4 Application benefits The EH actuation system is designed to fit within the envelope of the compressor housing, thus, not requiring extra valuable space in the engine compartment. This is accomplished by incorporating the kinematic components into the turbocharger’s center housing. Figure 7 shows the packaging space benefit of the EH system integrated into the center housing versus a typical external electric actuator system. The higher temperature and vibration capabilities of the critical components of the EH actuation system are the main enablers allowing this large reduction in package size.

239

Extra space required by a typical external actuation system Figure 7: Application space benefit of electro-hydraulic actuation system Integrating the actuation system within the boundaries of the basic turbocharger not only allowed easier installation of the turbo into a limited space environment, but also prevented the need for different actuator orientations to fit different engine applications. Additionally, due to this EFB control valve utilizing a current driver, it allowed the identical actuation system to be used with both 12V and 24V electrical systems. Together, these benefits allowed turbo part numbers to be reduced 80% from the previous emissions level engine. This was a significant achievement to help manage the complexity of a broad range of off-highway applications. Figure 8 shows examples of the engine designs using a common VNTTM turbocharger for both single-stage and two-stage turbocharging systems.

Figure 8: Example single- and two-stage applications of same VNTTM turbo2

240

4

CONCLUSIONS

By combining two proven technologies for VNTTM architecture and actuation system, then optimizing each of them for the off-highway market, Honeywell was able to make a large step forward in off-highway turbocharger capabilities. This new turbocharger represents a world first use of electro-hydraulic actuation for a VNTTM turbocharger used as a high pressure stage of a two-stage turbocharger system. Through this development, the customer was also able to significantly increase maximum engine torque and power capability while at the same time offering a wide range of applications and power ratings with a common turbo, therefore significantly reducing turbo part number complexity.

REFERENCE LIST (1) A. Kulkarni, et al. “Turbine Wheel High Cycle Fatigue Reliability Prediction.” Institution of Mechanical Engineers 9th Intl Conference on Turbochargers and Turbocharging. May 2011. London, UK. (2) John Deere Corporate Press Release “John Deere Power Systems Receives EPA and EU Interim Tier 4/Stage III B Certifications for Off-highway.” April 2010. Waterloo, USA.

241

Testing turbine expanders for high efficiency diesels E Halliwell Cummins Turbo Technologies, Laboratory Operations, UK

ABSTRACT This paper describes the test equipment built by Cummins Turbo Technologies, which we believe is unique, and which is aimed at optimising small high speed turbines powered using organic fluid as the medium part of a Rankine cycle. The test installation has brought together a number of challenges which include safe handling and the environmental impact of using organic fluids, measurement of power from small, high speed turbines, controls and safety. The installation represents an investment of 1.5 million USD by Cummins Turbo Technologies in a new market direction.

1.0 INTRODUCTION With increasing pressure on the need to reduce CO2 emissions and fuel consumption, engine manufacturers and vehicle integrators are looking for more radical ways to increase engine efficiency. One method is to attach a secondary heat recovery system to the engine, drawing energy from a number of waste heat sources. Using organic fluids, this energy can be turned into useful mechanical or electrical work. A turbine expander is a compact and efficient way of achieving this end and Cummins Turbo Technologies is currently working with a number of customers to develop turbine expander products for specific applications. The technologies used in such devices are allied to those of the turbocharger, and require similar design, testing and optimisation of the product. As organic fluid turbine expander technology is new for Cummins Turbo Technologies, there was no existing test capability within the engineering facility. The Organic working fluids have thermophysical properties that are substantially different from those of our normal test fluids i.e. air and exhaust gas. These differences make it virtually impossible to operate the turbine expander anywhere close to its intended thermodynamic design point if tested on a conventional turbocharger test stand. By not operating at its intended design point, it makes design validation very difficult. To enable comprehensive product test and design validation Cummins designed and built a dedicated test facility. The installation of this facility represents an investment of over $1.5M and provides the capability to evaluate the thermodynamic performance and durability of these high speed micro-turbine devices. Design of the facility has brought a number of challenges which include the

_______________________________________ © The author(s) and/or their employer(s), 2012

257

safe handling and managing the environmental impact of using organic fluids, measurement of power from small high speed turbines, dynamic control and operational safety. The main function of the new test facility is to extract performance data from the turbine expander by controlling the turbine expansion ratio, turbine inlet temperature and shaft speed to predetermined points whilst logging data, just like a conventional turbine mapping cell for a turbocharger. The difference from a conventional turbocharger test cell and an organic fluid expander test cell is that the organic fluid cannot be vented to atmosphere at the turbine outlet as is the case with turbocharger testing. Therefore the cycle has to be closed loop. All the energy that is transferred to the working fluid to power the expander must be rejected from the working fluid before the end of the cycle. It is a Rankine cycle and comprises four main stages: 1. The working fluid is pumped up to high operating pressure in liquid state (1-2 on T-S Diagram). 2. The fluid enters a boiler where it is heated to a super heated vapour state at a constant pressure (2-3 on T-S Diagram). 3. The super heated vapour expands across a turbine, generating power. The pressure and temperature both drop and the fluid may be a saturated vapour (3-4 on T-S Diagram). 4. The fluid is passed through a condenser where it changes back into a liquid before entering the pump for the next cycle (4-1 on T-S Diagram). It was always the aim whilst developing this facility that the on-engine application for the turbine expander would not be simulated, rather that the system would be designed so as to enable stable, repeatable and safe running. The working fluid used in the test cell with prolonged exposure to temperatures above 250 C. Therefore the design of the test cell centred on using electrically heated heat transfer oil which could safely heat the working fluid to the required operating temperature without the risk of overheating.

Figure 1 – T-S diagram

2.0 DESIGN 2.1 Heating system The heating skid uses Transcal N heat transfer fluid (HTF) and includes circulation pump, electric heater, a water-cooled cooling exchanger and control valves. The HTF temperature is controlled by modulating the power to the electric heater using a Eurotherm controller on the heating skid electrical panel. A 3-way diverting valve is also used to pass the HTF through a cooling heat exchanger. The working fluid outlet temperatures from the vaporiser and super heater are independently controlled by separate 3-way control valves diverting enough HTF past the heat exchangers to achieve the temperature set points. These 3-way valves are controlled from the main test cell controller and can be operated either manually

258

via a dial on the control panel or in closed loop control by the test cell control system.

Figure 2 – Heater skid

Figure 3 – Heat exchanger skid

The heat exchanger skid consists of the four main heat exchangers: recuperator, vaporiser, super heater and condenser. The vaporiser and super heater take energy from the HTF to heat and change the state of the working fluid from a liquid through to a super heated vapour; the condenser supplies cooling from the main plant cooling tower water to change the vapour back to a liquid and the recuperator exchanges heat from the turbine exhaust vapour to the liquid fluid entering the vaporiser. The inclusion of the recuperator enabled the heater to be downsized from 300 kW to 250 kW. 2.2 Working fluid loop The packaging of the high pressure liquid line from the pump outlet to the super heater outlet is not critical as regards fluid flow. Care has been taken to allow for expansion of the stainless steel pipe work without over stressing components. The main emphasis was on positioning the recuperator, vaporiser and super heater heat exchangers within the cell to allow uninterrupted pipe runs for both the working fluid and the HTF whilst not restricting personnel access required for operating and maintaining the cell.

Figure 4 – Schematic of the working fluid loop

259

It is important that only super heated dry vapour passes through the turbine expander. At rest, liquid working fluid is present throughout the loop. Before start up, hot HTF is circulated through the vaporiser and super heater. This slowly raises the temperature and pressure of the working fluid in the system. The working fluid is then pumped around the loop against the hot HTF and the fluid states around the loop slowly change to the operating states. Liquid can be present at the outlet from the super heater until the whole loop is at operating conditions. To stop this liquid entering the expander during warm up, the rig includes an expander bypass loop. The working fluid out of the super heater can be diverted through or around the expander. The inlet to the turbine expander rises up to a shut off valve; the other path includes another shut off valve and a fixed orifice expander. This bypass leg feeds back into the turbine outlet pipework before the hot side of the recuperator. There is a third valve on the turbine outlet before the bypass outlet to isolate the turbine expander from the refrigerant loop. The most critical section of the working fluid loop is from the condenser outlet to the pump inlet. The pump has a required net positive suction head, which is achieved by a combination of head of liquid in the receiver above the pump, gravity and the vapour pressure of the working fluid. Also, the outlet of the condenser must always remain clear of liquid working fluid so it has to be at the highest point in the system. Therefore the condenser height determines the maximum receiver height which in turn determines the available net positive suction head. At start up, when the working fluid in the receiver is at ambient temperature, its low vapour pressure is not enough to overcome the practical limitations of the height of the condenser and hence the height of the receiver, so cavitation and vaporising of the working fluid in the pump inlet pipework will occur. The condensing process starts with working fluid entering the condenser as a vapour and ends with liquid draining from the outlet. Depending upon the amount of cooling available, after the state change the condenser may also sub cool the working fluid. The change of state from a vapour to a liquid is at a constant temperature and the sub cooling is at a constant pressure. Therefore the combination of both processes together creates a range of liquid states from either a high pressure saturated liquid down to a low pressure sub cooled liquid.

Figure 5 – Condenser to pump pipe work arrangement

260

The cooling rate of the condenser can be controlled to have minimal sub cooling so that the liquid enters the receiver in a saturated liquid state increasing the liquid vapour pressure, which in turn increases the available net positive suction head. However, when the pump starts, the depression at the inlet causes the saturated liquid to vaporise. To overcome this problem a second cooling heat exchanger is fitted in line between the receiver outlet and pump inlet. As it is only liquid that can be present in this part of the system, the heat exchanger works as a liquid sub cooler dropping the temperature at a constant pressure. The combination of a condenser and liquid sub cooler is what enables the system to run. The condenser is controlled to produce saturated liquid at a high enough pressure to give the required net positive suction head at the pump and the sub cooler lowers liquid temperature and thus the vapour pressure to stop vaporisation at the pump inlet. The outlet pressure of the condenser also affects the turbine expander outlet pressure which in turn affects the expansion ratio of the unit being tested. Therefore a lower condenser outlet pressure allows a lower turbine inlet pressure for a given expansion ratio set point. The condenser outlet conditions can be controlled to maintain a stable pump inlet pressure set point close to the minimum required net positive suction head. 2.3 Dynamometer and oil system The turbine expander is coupled to a Borghi & Saveri FE150-S eddy current dynamometer through a 12:1 reduction gearbox and a Bibby coupling. The dynamometer is operated in open loop mode either manually from a dial on the control panel or via the test cell control system to load the expander and control shaft speed. Lubrication for the turbine expander bearings, gearbox bearings and gear teeth is all from one oil system. This oil system and the gearbox are sealed from the environment. The oil system incorporates an organic fluid / oil separation system which reintroduces separated working fluid back into the main loop before the condenser. The working fluid and the lubricating oil are miscible so under normal operation any working fluid carry over will be mixed in with the oil so as not to affect the lubrication process. Also at start up if the two fluids were not miscible and separated out it may be that pure working fluid could be pumped by the lubrication system. To separate out the working fluid from the lubrication oil a coalescing filter is mounted above the oil tank. There is an oil drain in the bottom of the filter that allows lubricating oil to return to tank under the liquid level line and a vapour line that connects to the main loop with a one way valve to stop fluid flowing back from the main loop. The system

Figure 6 – Schematic of oil system

261

works by heating the oil in the tank until the working fluid vaporizes, pressurizing the tank. When tank pressure is greater than the pressure in the main loop between the recuperator hot side outlet and condenser inlet, fluid vapour will flow back from the tank to the main loop. 2.4 Cell control The main cell control is a real time control and data acquisition system which can control all the rig’s running parameters and log data. Added to this is a safety PLC that monitors safety critical channels such as temperatures, pressures and shaft speed as well as a gas detection system. The PLC is programmed to shut down the cell in the event of a gas leak, over speed of the expander shaft, or any temperature and pressure anomalies that may occur. When the cell is powered down the detection system automatically reverts to a second power source to continue to function and will trigger the ventilation system if a leak is detected. In parallel with the cell control is a lab view program running on a separate windows computer. This displays a real time pressure temperature plot for different parts of the system with the saturation line for the working fluid over plotted. This tool is extremely useful when running a test as it is easy to check fluid state throughout the loop at a glance.

Figure 7 – Lab View screen shot

3.0 CELL BUILD A new building was erected specifically for the waste heat recovery test cell. To speed up the build process and reduce the amount of work within the new cell, before installation all the hardware was built on four separate skids. At the same time the electrical and instrumentation installation started. After the skids were installed and connected together the whole system was leak tested at a pressure of 8barg using oxygen free nitrogen. The high pressure side was then over pressure tested to 1.1 x maximum operating pressure. When the system was proved to be leak free it was vacuumed out to 2 Torr. This process is to remove any contaminants such as oxygen and water. A vacuum decay

262

test is then done. Ideally only 0.25 Torr should be lost over a 24 hr period. If this test is unsuccessful it must be repeated as the decay in vacuum suggests there is still contamination within the system. When the system holds vacuum it is ready to fill. The system holds just over 125kg of organic fluid. When it was filled, approximately 55kg was put in the receiver and pump inlet pipework and the rest was distributed throughout the system.

4.0 COMMISSIONING The heater skid was commissioned and the HTF was taken up to 250C and circulated for several hours to burn off any moisture in the system. All the test cell sub systems were individually checked and a full calibration of the cell transducers was completed. A commissioning expander was manufactured. This was an expander with the shaft and wheel replaced by a fixed orifice so that the working fluid loop can be tested in isolation. The orifice is such that it should replicate the pressure and temperature drops of the real unit. Using this expander, the cell operating envelope of speed, flow and expansion ratio can be explored. Also, procedures for starting and stopping the rig and emergency shut downs could be developed.

5.0 RUNNING SYSTEM IN MANUAL The rig incorporates three sight glasses each with a camera displaying in the control room. There is a turbine inlet sight glass, a bypass sight glass and a pump inlet sight glass. The HTF target temperature is set and the 3-way valves for the vaporiser and super heater are both set via dials on the control panel to 100% oil flow through the heat exchangers. The condenser cooling rate is turned down to 0% and the turbine inlet and outlet valves are closed. All other system valves are opened so that there is an uninterrupted loop through the bypass leg. When the HTF is at temperature the variable speed pump is set to approximately 40% command and the pump is switched on. Liquid will be seen flowing through the bypass sight glass whilst bubbles can be seen at the pump inlet. As the fluid fills the heat exchangers and the system gets up to operating pressure and temperature, the bypass sight glass will show liquid change to wet vapour and then to dry vapour. At the same time the pump inlet sight glass changes from a still coulomb of liquid, to vapour, then refills with turbulent bubbly liquid which slowly calms to a clear bubble free liquid as the cavitation is eliminated. The super heater outlet stabilises at approximately 14 barg with the fluid in a super heated state. The condenser cooling may have to be adjusted to maintain the optimum pump inlet pressure. When the bypass sight glass is showing dry vapour, the pump speed can be reduced slowly to reduce the turbine inlet pressure to approximately 6 barg whilst maintaining super heater vapour conditions. At this point the turbine inlet, turbine outlet and bypass valves are all switched together opening the turbine inlet pipework to test the expander. Now the pump speed can be slowly increased to achieve the required turbine inlet pressure for testing and the condenser control can be adjusted to set the expansion ratio. Also the vaporiser and super heater outlet temperatures can be adjusted to

263

achieve the correct running point. The mass flow at that point can then be measured. If the system is running an expander with a shaft and wheel connected to the dynamometer, the shaft speed can be controlled by loading the dynamometer from a dial on the control panel. All the parameters that are manually controlled via dials on the control panel can also be set to run in closed loop PID control from the test cell controller. However at this stage the system dynamics are still being learned. If the flow is shut off suddenly, the high and low pressures around the loop quickly average which can over pressure the low pressure side of the working fluid loop. To shut down safely and protect the test piece the system pressure has to be reduced to about 6 barg whilst maintaining super heated vapour at the turbine inlet. Then the bypass switch can be triggered closing the turbine inlet and outlet valves and opening the bypass valve. The HTF target temperature can then be set to 50C and the system left to cool down slowly as the temperature reduces. When the bypass sight glass starts to show signs of wet vapour the pump can then be switched off. Finally the receiver isolation valves are shut. The turbine inlet and turbine outlet isolation valves also allow test pieces to be fitted and removed without having to reclaim all the working fluid from the whole system. The operating envelope of the rig is as follows: Operating Envelope Turbine parameter

Min

Max

Speed

0

70000

rpm

Inlet pressure

4

27

barg

Outlet pressure

2

4

barg

130

230

C

0

0.6

kg/s

Inlet temperature Flow

Units

6.0 CONCLUSIONS Although there are similarities in this rig and a conventional turbocharger gas stand, the added complexities from the necessity to manage the working fluid safely drive a whole new set of working practices. When fitting a new expander to the rig for testing, all the pipework between the turbine inlet and outlet shut off valves has to be leak checked at the operating pressure using nitrogen. If leak free, this pipework has to be vacuumed down to 2 torr and the vacuum has be stable. This process can take up to 2 days. Before removing the expander all the turbine inlet pipework has been emptied of working fluid. This involves using a special fluid reclaim unit. This process can take up to 6 hrs. To swap a turbocharger on a conventional gas stand would usually be done within 3 hrs. Calibration of the pressure transducers has to be done via special calibration valves so the transducer diaphragm can be isolated from the working fluid without breaking into the system.

264

The section of pipework that has the biggest impact on running fluid through the Rankine cycle is the condenser outlet to pump inlet. The relative positions of the condenser, receiver and pump inlet are critical for fluid flow. However the cycle would still not work without the liquid sub cooler at the outlet from the receiver. Therefore the liquid sub cooling heat exchanger is the key component within the whole system. Using a large electric heater to control HTF temperature has shown that control of the working fluid temperature can be maintained within +/-2 C. This stability affords similar temperature stability at the turbine inlet during steady state running. During transients, turbine inlet pressure, controlled via pump speed is stable and responsive, however mass flow increases disproportionately at the same time before dropping off again at steady state conditions. Also, the heating and condensing responses lag behind mass flow during transients so a stabilising period is required before any steady state data can be recorded.

7.0 ACKNOWLEDGEMENTS The author would like to thank Cummins Turbo Technologies Ltd for permission to publish the paper. He would also like to thank his colleagues in the Laboratory Operations Facilities Department for their contribution to the design and development of the cell.

8.0 REFERENCES 1. http://www1.eere.energy.gov/vehiclesandfuels/pdfs/deer_2006/session6/2006_ deer_regner.pdf 2. http://www1.eere.energy.gov/vehiclesandfuels/pdfs/…………/session6/2006_deer _nelson.pdf 3. http://www1.eere.energy.gov/vehiclesandfuels/pdfs/deer_2008/session5/deer0 8_nelson.pdf 4. http://www1.eere.energy.gov/vehiclesandfuels/pdfs/deer_2009/session5/deer0 9_nelson_1.pdf 5. http://www.labothap.ulg.ac.be/cmsms/uploads/File/TFE_SD090623.pdf 6. http://www.sciencedirect.com/science/article/pii/S0360544204000179

265

Characterization of a low pressure turbine for turbocompounding applications in a mild-hybrid gasoline engine A M I Bin Mamat, A Romagnoli, R F Martinez-Botas* *Department of Mechanical Engineering, Imperial College of Science, Technology and Medicine, UK

1

ABSTRACT

This paper describes the results obtained from the development of a high performance Low Pressure Turbine (LPT) for turbocompounding applications to be used in a heavily downsized 1.0L turbocharged gasoline engine. The LPT was located at the downstream of the main turbocharger turbine to recover exhaust energy at very low pressure ratio (PR≈1.1) and generate 1 kW of electric power at 50,000 rpm. No commercially available turbines can offer higher efficiency at this limited range. The newly designed turbine is of a mixed-flow nature and it was developed and tested at Imperial College London. The obtained LPT performance maps were then used to assess the impact of the turbocompound unit on the Brake Mean Effective Pressure (BMEP), Brake Specific Fuel Consumption (BSFC) and Brake Torque under full load and part load engine operation. The study found that the turbocompounding unit has increased the BMEP at maximum value of 0.64 bar, raises the Brake Torque by 5.1 N.m at maximum and reduces the BSFC as low as 2.6% from the baseline engine model. Keywords: Low Pressure Turbine, Energy Turbocompounder, Part Load Engine Simulation

2

Recovery,

Engine

Downsizing,

NOMENCLATURE

BSFC CCP cl d FFR GDI GHG ICE LPT k

Brake Specific Fuel Consumption Absolute Flow Velocity Combined Cycle Power Clearance Diameter Fuel Flow Rate Gasoline Direct Injection Greenhouse Gas Internal Combustion Engine Low Pressure Turbine Specific Heat Ratio Mass Flow Rate

MFP

Mass Flow Parameter

ORP

Organic Rankine Cycle

___________________________________________ © The author(s) and/or their employer(s), 2012

[kg/kW/hr] [ m/s ] [m] [kg/hr]

[ kg/s ] kg K . s Pa

281

P PLR PR R SW T U VR WHR ∆ γ 

Pressure Part Load Ratio Pressure ratio Gas Constant Swept Temperature Rotor Velocity Velocity Ratio Waste Heat Recovery Difference Cone Angle Efficiency

[ Pa ] [ kJ/kg.K] [K] [ m/s ]

Subscript 0 Total/Stagnation Condition 1 Volute Inlet 2 Stator Inlet 3 Rotor Inlet 4 Rotor Exit bl Blade is Isentropic t-s Total-to-static

3

INTRODUCTION

Since 1867, Internal Combustion Engine (ICE) is the main mover in the transportation sector and over the years it has become one of the main causes of the world’s CO2 emission [1]. Consequently more and more legislators started to apply more stringent emission regulations by introducing vehicle taxation schemes and green zone areas in order to limit CO2 levels in urban areas. However, despite providing an immediate benefit in terms of air quality, such restrictive solutions need to be supported by an adequate improvement in powertrain systems. It is within this context that automotive manufacturers are currently looking into novel technologies capable to provide reliable and cost-effective solutions to maximize energy recovery and hence mitigate CO2 emissions. Hybrid and full electric vehicles represent the future in the transportation sector, however a full switch over is still long to come and this is one of the main reasons pushing automotive manufacturers to invest more and more resources into exhaust energy recovery systems. Despite the improvement in efficiency of the ICE, 25% to 35% of the overall energy available to the engine is wasted through its exhaust gases. Therefore, it is apparent that there is great scope for all those technologies enabling to recover the excess energy from the exhaust gases and to regenerate either into the engine crankshaft or to power auxiliary powertrain systems. At the present, there are three main exhaust energy recovery techniques which are being studies within automotive makers: (1) Organic Rankine Cycle (ORC), (2) Thermoelectric Generator and (3) Turbocompounding. The key features of each of these techniques are given in Table 1 which also shows the engineering layout associated with each one of them. As one can imagine each technique embeds positive and negative aspects which makes it hard to establish a clear winner. However as a general observation it can be stated that despite the ORC and Thermoelectric generator offer higher improvement than Turbocompounding in terms of fuel consumption, the engineering challenges associated with their implementation are still significant and expensive to be tackled [2-5]. Therefore Turbocompounding seems to offer the best cost-efficiency compromise even though also for this solution there are still

282

several issues like cooling, integration, engine optimization and exhaust-back pressure mitigation which need to be sorted out. Exhaust gas back-pressure mitigation in particular is one of the main challenges which need to be tackled in turbocompounding technology. In fact the addition of an extra element in the tail pipe increases the amount of exhaust gas back-pressure which goes to the detriment of scavenging and therefore engine performance. Hence it is apparent the need of a bespoke turbine design for turbocompounding applications capable to minimize the negative effects of back-pressure whilst maintaining high energy extraction. The development of such a turbine is at the core of the current paper and more details can be found in the following discussion. Table 1: Comparative Study of Exhaust Enthalpy Recovery System Exhaust Enthalpy Recovery Systems

Turbine

Generated Power

Feed Pump

Rankine Bottoming Cycle

Surrounding

Organic Fluid

Condenser

Heat Exchanger

Engine

Exhaust

Working Principle • Thermal Exchange • High exhaust thermal energy transfer in an Evaporator. • High compressed liquid of Organic Fluid receive energy and change to superheated vapour

Advantage

• Not increase pumping work • Higher reduction of BSFC

Disadvantage

• Installation problem • Hazardous liquid • Cost ineffective

Hot Surface n-type semiconductor

p-type semiconductor

Cold Surface

• Peltier-Seebeck • Lightweight effect • Not increase • Temperature pumping difference of work exhaust surface and thermoelectric • Higher reduction of material surface BSFC produce an electric current

• Large exhaust surface area. • Cost ineffective

Thermoelectric Generator

• Exhaust thermal expansion at low pressure. • Recovered energy is used to drive electric or mechanical regenerator.

• Easily Bolt-on • Increase • Lower mass pumping loss flow capacity. • Lower • Newly pressure dedicated • Electric low pressure generator turbine limitation design

Turbocompounder

4

BACKGROUND

As part of the HyBoost project, the current study focuses in the design and the development of a high performance Low Pressure Turbine for electric turbocompounding. The HyBoost project is a TSB1 research programme which aims 1

The Technology Strategy Board is an executive non-departmental public body (NDPB), established by the UK Government in 2007 and sponsored by the Department for Business, Innovation and Skills (BIS).

283

to produce a car running on a 1.0L turbocharged gasoline engine offering the same performance as a 2.0L engine while retaining CO2 emissions below 100g/km. This has to be achieved by the synergistic application of an extremely downsized gasoline engine as well as exhaust gas recovery, electrified boosting, micro-hybrid functionality with stop/start, regenerative braking, a novel energy storage technology, torque assist and electric turbocompounding, as shown in Fig. 1.

Figure 1: HyBoost Engine Architecture The design specifications for the Low Pressure Turbine (given in Table 2) were obtained by mean of a validated 1-D engine model (Ricardo Wave) of the engine under study. The model results showed that the pressure of the exhaust gases leaving the turbocharger turbine is limited to very low values (≈1.05 bar to ≈1.2 bar) and this forced the design pressure ratio to be set at a very low value, PRLPT=1.1. In addition to this, the costs associated with bearings and development of the electric generator also constrained the rotational speed and turbine power output at NLPT=50000rpm and WLPT=1kW respectively. The former has to be considered as a good compromise between bearing life and implementation costs whereas the latter was set thus to achieve a continuous power output throughout a driving cycle. Table 2: Turbine Operating Requirement Turbine Power Turbine expansion ratio Blade rotational speed Total inlet temperature Mass Flow Rate

1 kW 1.1 50000 rpm 1100 K 0.05 kg/s

Based on the turbine operating requirements of Table 2, it is apparent that at such low pressures and speeds, commercially available turbines provide efficiencies which fall below 40%. In fact commercially available turbocompounders use offthe-shelf turbocharger turbines as rotating device even though these turbines are designed to produce high power at high turbine expansion ratios. Consequently very poor performance would be obtained by using a commercial turbine at low speed and pressure. This can be better understood by looking at the Flow Coefficient (Φ) and Blade Loading2 coefficient (ψ) which are two parameters that univocally link turbine operating conditions with its efficiency[7]. From Fig. 2 it is apparent that the use of a commercial turbine does not fulfil the requirements for turbocompounding since an increase in the rotational speed leads to quadratic decrease in the Blade Loading coefficient thus moving turbine efficiency far from its optimum value (dashed blue line). Even by using a small turbocharger turbine at 2

The definition of Stage Loading and Flow Coefficient can be found in turbomachinery book [6] and it won’t therefore be reported here.

284

lower rotational speeds, the efficiency still lies in a region with low Blade Loading and high Flow Coefficient (ψ≈0.23 Φ≈1.5at20000 rpm, ψ ≈0.03 Φ≈0.89 at 50000rpm) which corresponds to an efficiency of about 40%. Such a low efficiency is not acceptable in a radial machine and in order to match with the mass flow rate, some turbocompounder manufacturers “trim” the turbine by reducing the turbine inlet area. However such a practice does not lead to any improvement in turbine performance which actually tends to deteriorate even further. It is therefore clear that a compromise has to be found between optimum rotational speed (Blade Loading) and mass flow rate (Flow Coefficient) in order to accommodate for the turbocompounding requirements.

Figure 2: Superimposed Low Pressure Requirement into Flow Coefficient and Blade Loading Diagram Chen and Baines [7] Table 3: Low pressure turbine geometric specifications and comparison with commercial turbines 3D Geometry

Number of Blades, Z Leading Edge Root Mean Square Radius, r3,rms Trailing Edge Tip Radius, d4 Cone angle, γ Inlet Blade Angle, bl Rotor blade length, l Comparative study on Turbine design characteristics A4/A3 ηt-s,design Speed[rpm] Low pressure turbine (LPT) 0.35 >70% 50000 Turbine A - High capacity 1.1 80% 60000 Turbine B -Medium capacity 0.9 84% 98000 Turbine C- Small capacity 0.8 72% 160000

9 42.2 mm 22.7 mm 20° varied 33.5 mm PR 1.1 2.0 1.6 2.0

285

5

LOW PRESSURE TURBINE CHARACTERIZATION

The development of the Low Pressure turbine presented in this paper was carried out by following the specification given in Table 1. The design was accomplished by mean of successive steps which went from entropy generated meanline analysis, 3D modelling and CFD calculation. The applied design method followed standard turbomachinery correlations and it won’t be described here; however more details can be found in previously published articles by the same authors. In Table 3 it is given a general overview of the LPT configuration together with a comparative analysis between the design features of the LPT and those of commercially available turbines. From the table it is apparent that the design point for the LPT is out of the conventional turbines range and this corresponds to a very low area ratio value (A4/A3) when compared to standard turbocharger turbines. Such a low value for the area ratio is rather unique and such a constraint was one of the main challenges which had to be solved during the design process.

2 x36 KW Heaters

4” Main Valve

3” Control Valve

V-Cone

Inner Pipe

Outer Pipe

Orifice Plate

Eddy Current Dynamometer Measurement Plane

Pulse Generator

Figure 3: Imperial College cold-flow test facility layout In order to fully characterize the turbine, a prototype LPT wheel from Aluminium Al6082 with 1.6μm surface finish was made for experimental evaluation. The turbine volute and the connecting duct instead were made out of polycarbonates (PC) material and manufactured by using a 3-D Fused Deposition Modelling (FDM) machine. The application of polycarbonates for cold-flow testing turbine volute is a novel technique in the turbomachinery research. The main advantages are associated with reduced design lead time and cheaper than a die-cast metal model. Finally the newly designed LPT was tested under steady-state conditions with the cold-flow gas test stand available at Imperial College (Fig. 3). This is a state-of-theart test facility which enables to obtain turbine maps three/four times wider (with respect to pressure ratio) than those generated using a compressor as a loading device. Such a large width of the maps can be obtained thanks to a bespoke dynamic response eddy-current dynamometer which loads the turbine through a magnetic rotor coupled to the turbine wheel thus overcoming the limitation of choking and surge typical of a compressor unit [8]. All the extensive explanations

286

of the steady-state test conditions were previously published by the Imperial 3 College Turbocharger Research Group and the reader is advised to refer to it for further understanding the working principle behind the test-facility. 5.1 Performance Parameters The non-dimensional characteristics which weight the behaviour of a turbocharger turbine can be simplified onto 4 parameters as given in Eq. (1).

,

,

,

(1)

P01/P4 is the total-to-static pressure ratio, PR of the whole turbine stage,

T

is the

pseudo non-dimensional mass flow rate (defined as Mass Flow Parameter, MFP), ηt-s is the total-to-static efficiency and U3/Cis is the velocity ratio, VR. The total-to-static efficiency, ηt-s is defined as the as the ratio between the actual power and the isentropic power of the turbine whereas the velocity ratio, VR is a dimensionless parameter defined as the ratio between the rotor tip speed,U3 and the isentropic velocity, Cis. The Mass Flow Parameter, MFP is usually plotted against the pressure ratio, PR whereas the efficiency, ηt-s is plotted against the velocity ratio, VR. P

5.2 LPT Performance Maps The performance of the LPT was tested for a set of five different constant speed lines spanning from 80% to 120% equivalent speed lines corresponding to 40,000 rpm and 60,000 rpm of the actual rotational speed. The performance parameters as described in the previous section (MFP, PR, ηt-s and VR) were acquired and 4 normalised against the turbine design values . The outcomes of the experimental results are reported in Figs. 4 and 5. PR and Normalised total-to-static Efficiency for LPT and Conventional Low Speed Turbine

Normalised Total-to-static Efficiency, t-s

1.2

Low Pressure Region

1.1

80% LPT 90% LPT

1.0 100% LPT

0.9

110% LPT

0.8 0.7

120% LPT

0.6

50% Conventional

0.5

60% Conventional

0.4

70% Conventional

0.3 0.2

Very low performance for conventional turbines at PR≈1.08

80% Conventional 90% Conventional

0.1 100% Conventional

0.0 1.00 1.05 1.10 1.15 1.20 1.25 1.30 1.35 1.40 1.45 1.50 Pressure Ratio, PR

Figure 4: Comparison of Normalised t-s between LPT and Conventional Turbine

3 4

Imperial College Turbocharger Research Group website: www.imperial.ac.uk/turbochargers For the efficiency the design value of the total-to-static efficiency is greater than 70%.

287

Normalised Mass Flow Parameter, MFP (mdotT012/P01)

Comparison of Normalised Mass Capacity of LPT and Conventional Turbine

2.8

80% 90%

2.4

100% 110%

2.0

120%

1.6

50% Conventional 60% Conventional

1.2

70% Conventional 80% Conventional

0.8

90% Conventional 100% Conventional

0.4 1.00

1.10

1.20

1.30

1.40

1.50

Pressure ratio, PR Figure 5: Comparison of Normalised MFP between LPT and Conventional Turbine Figure 4 shows the efficiency, ηt-s plotted against the pressure ratio, PR obtained for the LPT and compared with that of a commercial low speed turbine which was tested at Imperial College by Szymko [8]. The figure clearly shows that the LPT operates at very low pressure ratios - 1.07 to 1.3 - and that, for the range of speeds tested, all the efficiencies fall between 0.75 and 1.07 of normalised ηt-s. This demonstrates the excellent performance of the newly designed turbine in the specified pressure ratio range. The maximum ηt-s of the LPT was measured at PR ≈ 1.08 which corresponds to the predominant operating condition of the turbocompounding for the particular engine applications. On the contrary the conventional practice of using off-the-shelf turbines for turbocompounding shows that at such low pressure ratio range an extremely low performance, ηt-s would be obtained (ηt-s≈0.28 at PR≈1.08). As for the efficiency, the Mass Flow Parameter, MFP of the LPT was normalised by dividing it with the design value. Thereafter, the normalised MFP was plotted against the PR as it is shown in Fig. 5. The normalised MFP of the LPT was then compared to the conventional low speed turbine that was divided by the LPT design value. From Fig. 5, it can be seen that the flow capacity of the low pressure turbine shows a typical trend common to all centrifugal machines; the MFP decreases as the rotational speed increases. A centrifugal pressure field opposed to the incoming mass flow is generated by the rotation of the turbine thus causing the drop of the flow capacity of the turbine as the speed increases. Meanwhile, the comparison of the swallowing capacity of LPT and the conventional turbine at equivalent PR shows that a higher MFP is obtained for the conventional turbine at higher turbine power output. Thus, the swallowing capacity for the conventional turbine is bigger than the LPT. Therefore, in order to get 1.0 kW power for the conventional turbine, the rotational speed must get smaller and at lower swallowing capacity. However, this is unfavourable because the turbine efficiency will be smaller.

288

6

1-D ENGINE GAS DYNAMIC SIMULATION

In order to assess the impact of the LPT on engine performance, a commercial 1-D simulation model (Ricardo Wave) was developed for the 1.0L gasoline engine presented in Fig. 1. Initially the baseline petrol engine model (with no turbocompound unit) was validated against engine test-bed data obtained at full and extreme part-load conditions for a range of engine speeds spanning from 1000rpm to 6000rpm (refer to Fig. 6). Then three reference engine speeds were selected (1500rpm, 2000rpm and 4000rpm) and full engine simulation was run for intermediate engine operating conditions (from full to extreme part-load) which were obtained by setting the fuel flow rate, the intake manifold pressure and throttle opening area. The compressor operating conditions of the turbocharger for the selected speeds are shown in Fig. 65. After having completed the engine simulation for the baseline engine, the turbocompound unit was added to the engine model. An initial assessment on the best location for the turbocompound unit was performed for the full load condition cases. The results showed that the post-catalyst position provides the best compromise in terms of fuel consumption and engine Brake-Power and it was therefore decided to carry out the engine performance analysis for the post-catalyst set-up only. Superimposition of the compressor operating points at different engine conditions

4.0 3.5

Compression Ratio

3.0 2.5

Compressor max. efficiency region

2.0

Engine speed: 6000rpm

Engine speed: 1000rpm

1.5 Compressor Operating Speed Lines

1.0 0.5 0.0 0

0.02

0.04

0.06

0.08

0.1

0.12

Air flow rate [kg/s] Eng. Speed: 1500 rpm Extreme Part Load

Eng. Speed: 2000 rpm Full Load

Eng. Speed: 4000 rpm

Figure 6: Steady-State Full Load and Part Load Boosting Target The turbocompound unit was modelled by inputting the LPT experimental maps into the turbine model. In order to target the 50000rpm and 1kW power output, an external bypass and a shaft torque controller had to be included in the model of the electric turbocompounding. As per the electric generator, no mathematical model was available to replicate its behaviour. Hence in order to evaluate the impact of the turbocompound unit on engine performance, it was decided to consider two 5

It should be noted that in the Figs. 6 to 9, a standard notation data series is used to symbolize different engine speeds. A diamond represents the engine’s speed at 1500 rpm, a square corresponds to 2000 rpm and a circle is stand for 4000 rpm.

289

different scenarios: the first in which the excess power recovered by the turbocompound unit is directly feedback into the engine crankshaft with mechanical efficiency of 100% and the second in which the recovered energy is simply stored and not reused for any operation. In this way it was possible to assess which is the mechanical benefit of the turbocompounding on engine performance and the impact of pumping work as an effect of increased back-pressure. Four parameters were considered to analyse the turbocompounder implication during part load conditions: fuel flow rate, FFR (kg/hr), BMEP (bar) BSFC (kg/kW.hr) and Brake Torque (N.m). The outcomes of the BMEP and Brake Torque are reported as relative to the baseline engine whereas the BSFC is normalised to the baseline engine model. The correlations used for the simulation analysis are given in Equations (2) to (4). For clarity of analysis it should be noted that negative ∆%BSFC and a positive ∆BMEP and ∆BR.TORQUE indicate a benefit to the engine. ∆

∆%



.

.

.

4

It is worth noting that the part load operation changes the engine air-to-fuel ratio (AFR) thus reducing the available blow down exhaust energy. Therefore it is reasonable to present the results obtained for the part load engine operating conditions according to the part load fuel flow rate ratio which is defined as the ratio between the fuel flow rate at part load and that at full load, as given in the equation below.

Figures 7 to 9 illustrate the evaluated parameters such as BMEP, BSFC and Brake Torque during the simulated part load conditions. The standard notation for the turbocompounding unit in Figs. 7 to 9 show the solid and dashed lines which correspond to the condition when the LPT power is either regenerated (refer to Crankshaft legend) or not (refer to No Crank legend) into the ICE. It is worth reminding that the LPT was designed according to the requirements in Table 1 and it operates at best performance for low pressure range and specified swallowing capacity. Thus, in order to maintain the LPT turbine speed and power output, the inlet exhaust gas into the LPT is externally by-passed. The external bypass also facilitates the back pressure lower than the boosting pressure. By optimising the back pressure, the merit of the turbocharger and turbocompounding unit can certainly be drawn. Figure 7 clearly shows the installation of the turbocompounding unit reduces the BMEP of the ICE (refer to no crank legend) because the pumping work is increased [6]. However, as soon as the recovered exhaust energy is recovered into the ICE, a clear benefit in BMEP can be observed. The maximum benefit of the BMEP was found at 1500 rpm (ΔBMEP≈0.64 bar) and it decreases as the engine speed increases. In Fig. 8 is shown the effect of the turbocompounding on BSFC. Despite presence of the LPT increases the ∆BSFC of the engine (refer to no crankshaft), benefit of the LPT is apparent when the recovered energy is regenerated into engine crankshaft. During the part load operations, it can be inferred that

290

the the the the

addition of the turbocompounding unit is beneficial to the reduction of fuel consumption with a maximum reduction in BSFC about ≈-2.6 % at PLR = 0.85 and 1500rpm. It is worth noting that the maximum BSFC for every engine rpm occurs at part load operation rather than full load. This might probably be due to the setting of the LPT external by-pass which at full load conditions is not capable to divert the exhaust gases away from the LPT, thus causing an increased backpressure and hence increased pumping work. The same does not occur at part-load operation where the reduced amount of exhaust gases can more easily be split between the by-pass and the LPT thus reducing the pumping work. 0.75

ΔBMEP

ΔBMEP, (bar)

0.50

0.25

Full Load

0.00 0 PLR

0.2

0.4

0.6

0.8

1

1.2

-0.25 No Crank: Turbocompounder power not regenerated Crankshaft: Turbocompounder power regenerated into engine

-0.50

-0.75 1500 rpm- Crankshaft

1500 rpm - No Crankshaft

2000 rpm - Crankshaft

2000 rpm-No Crank

4000 rpm - Crankshaft

4000 rpm - No Crankshaft

Figure 7: Part Load Implication on BMEP Δ%BSFC

1.0% 0.5%

PLR

0.0% 0

0.2

0.4

0.6

0.8

1

1.2

Δ% BSFC

-0.5% Full Load

-1.0% -1.5% -2.0% -2.5%

No Crank: Turbocompounder power not regenerated Crankshaft: Turbocompounder power regenerated into engine

-3.0% 1500 rpm- Crankshaft

1500 rpm - No Crankshaft

2000 rpm - Crankshaft

2000 rpm-No Crank

4000 rpm - Crankshaft

4000 rpm - No Crankshaft

Figure 8: Steady-State Part Loads Implication on BSFC

291

Similar consideration as the BMEP and BSFC can be drawn for the Brake Torque. At extreme part-load conditions, the exhaust mass flow is small and the LPT rotates at lower rotational speed thus generating less power which is almost negligible at low engine speed. As a result, Fig. 9 shows no changes of the Brake Torque during the extremely part load operation (far left hand of the figure) and the merit of LPT on the Brake Torque can only be observed as the PLR is higher than 0.3. The highest ∆BR.Torque is ≈5.1 N.m at 1,500 rpm and it occurs at similar location as the maximum ΔBMEP and Δ%BSFC. Moreover, when the turbocompound power is not regenerated into the engine crankshaft, the back pressure due to the LPT attachment increases the pumping work and this can be noticed by a reduction in Brake Torque of as much as ≈3 N.m. Δ%BR.TORQUE

6.0

4.5

Δ BR.TORQUE, (N.m)

3.0

1.5

Full Load PLR

0.0 0 -1.5

0.2

0.4

0.6

0.8

1

1.2

No Crank: Turbocompounder power not regenerated

-3.0

-4.5

Crankshaft: Turbocompounder power regenerated into engine 1500 rpm- Crankshaft

1500 rpm - No Crankshaft

2000 rpm - Crankshaft

2000 rpm-No Crank

4000 rpm - Crankshaft

4000 rpm - No Crankshaft

Figure 9: Steady-State Part Load Implication on the Brake Torque

7

CONCLUSIONS

This paper presented the outcomes of the analysis carried out on the characterization of a High Performance Low Pressure Turbine for turbcompounding applications. The study was based around the operating conditions of a 1.0L turbocharged gasoline engine for which the LPT was developed. The analysis was divided in two main parts: the first looking at the design and testing of the LPT whereas the second looked at engine implication of the turbocompounding. The design requirements were set by the operating conditions of the engine which constrained the LPT to operate at very low pressure ratios PR≈1.1 and generate 1kW at rotational speed of 50000 rpm. At such conditions, commercially available turbines provide very poor performance whereas the experimental results showed that the newly designed turbine succeeded in achieving efficiencies greater than 70% at a pressure ratio PR≈1.08. This is an excellent result which cannot be found in any radial machine operating at such low pressure ratio. Then an engine assessment was carried out for the LPT by mean of a validated 1-D engine model developed for the engine under study. The turbocompound unit was modelled by inputting the experimental maps of the LPT into the turbocompound

292

model. The engine model was ran for three reference engine speeds (1500rpm, 2000rpm and 4000rpm) for a set of different loads going from part to full load conditions. For each engine operating conditions two different set-up for the turbocompounding unit were considered in which the recovered energy was either or not regenerated into the crankshaft. The part load engine simulation demonstrated that, when the recovered energy from the turbocompunding is not regenerated into the engine, the BMEP, Brake Torque and BSFC are affected by the presence of the downstream turbocompounding unit installation. However as soon as the recovered excess power recovered is fed back into the engine crankshaft a clear benefit can be observed. The engine BSFC always shows an improvement under any part load condition and engine speed with a maximum variation of Δ%BSFC ≈ -2.6%. The same occurs for the BMEP that it showed an improvement of about ≈0.64 bar at 1500 rpm; similar trend was also observed for the Brake Torque for which an improvement of 5.1 N.m was calculated at the same engine speed.

8

ACKNOWLEDGEMENT

The authors would like to acknowledge CPT, Ford UK, Valeo, and Ealabc. This consortium along with Imperial College London and Ricardo plc are part of the Hyboost project, a funded program by TSB (Technology strategy Board) whose support and encouragement is gratefully acknowledged. The authors would also like to thank Universiti Teknologi MARA (UiTM) for its financial funding for the first author.

9

REFERENCES

[1]

U.S.Energy Information Administration (EIA), "International Energy Outlook 2010," U.S.Energy Information Administration (EIA), DOE/EIA-0484(2010), Washington DC, United States, July 2011. [2] Yamada, N., Minami, T., and Anuar Mohamad, M. N., "Fundamental experiment of pumpless Rankine-type cycle for low-temperature heat recovery," Energy, Vol. 36, No. 2, 2011, pp. 1010-1017. [3] Srinivasan, K. K., Mago, P. J., and Krishnan, S. R., "Analysis of exhaust waste heat recovery from a dual fuel low temperature combustion engine using an Organic Rankine Cycle," Energy, Vol. 35, No. 6, 2010, pp. 2387-2399. [4] Leising, C. J., Purohit, G. P., DeGrey, S. P., and Finegold, J. G., "Waste Heat Recovery In Truck Engines," West Coast Meeting, Society of Automotive Engineers (SAE), Warrendale, Pensylvania, USA, 1978. [5] Fairbanks, J. W., "The 60 Percent Efficient Diesel Engine; Probable, Possible, or Just A Fantasy," 2005 Diesel Engine Emissions Reduction (DEER) Conference Presentations, U.S Department of Energy, Chicago, Illinois, USA, 2005. [6] Watson, N. and Janota, M. S., Turbocharging the Internal Combustion Engine, First ed., The Macmillan Press Ltd, Hong Kong, 1982. [7] Chen, H. and Baines, N. C., "The Aerodynamic Loading of Radial and MixedFlow Turbines," International Jurnal of Mechanical Science, Vol. 36, No. 1, 1993, pp. 63-79. [8] Szymko, S.. The development of an eddy current dynamometer for evaluation of steady and pulsating turbocharger turbine performance. 2006. London, University of London; Imperial College. Ref Type: Thesis/Dissertation

293

The role of turbocompound in the era of emissions reduction R W Kruiswyk Caterpillar Inc., USA

ABSTRACT This paper presents the results of an analytical investigation into the fuel economy benefit of turbocompound on a heavy-duty diesel engine in the presence of emissions control technologies - specifically exhaust aftertreatment and EGR (exhaust gas recirculation). Included in the investigation are the impact of aftertreatment backpressure levels, turbine efficiencies, variable geometry turbine (VGT) technology, and CVT power transmission technology. The fuel economy benefit of turbocompound is shown to be highly dependent on both aftertreatment backpressure and the use or absence of EGR. Application of advanced technologies like VGT and CVT is shown to have a comparatively smaller impact on BSFC benefit.

1 INTRODUCTION Rising fuel prices and heightened environmental concerns are resulting in greater demand for fuel economy improvements from the manufacturers of medium and heavy-duty diesel engines. Higher fuel prices drive fuel economy to the top of the list of critical customer requirements, and greenhouse gas concerns have resulted in the issuing of the first U.S. fuel economy standards for commercial trucks and busses, starting in 2014-2018. As such, the push for more thermally efficient engines is now greater than ever before. Heavy-duty diesel engines used in mobile applications today are typically 37-43% thermally efficient on the lug curve. Of the ~60% of fuel energy that is wasted (not converted to crankshaft power), roughly half is thermal energy in the exhaust stream that is dumped to atmosphere [1]. Recovering some of this ‘lost’ energy from the exhaust stream, known as waste heat recovery, is therefore key to achieving significant fuel economy improvements on future engines. Over the past 50 years, the form of waste heat recovery that has received the most research focus for mobile applications has been turbocompound. Turbocompound involves the placement of an additional ‘power’ turbine in the exhaust stream downstream of the turbocharger turbine, extracting the otherwise ‘wasted’ energy and feeding it to the crankshaft via a mechanical or electrical transmission (Fig. 1a). Fuel economy improvements of 4-6% or more via turbocompound have been demonstrated in many of these investigations [2-6]. Despite these successes and the long history of development, the use of turbocompound on production high-volume mobile applications has been extremely limited. Inexpensive fuel has kept turbocompound from attaining the favorable cost-value tradeoff that would prompt more widespread production implementation.

_______________________________________ © The author(s) and/or their employer(s), 2012

269

Brake Power

Engine

Ambient

Turb

Turb

Comp

Stack

Transmission

Turb

A/C

CVT

Comp

Transmission

Engine

A/C

Ambient Stack

1a. Conventional turbocompound

1b. CVT-based turbocompound

Figure 1. Turbocompound Engine Configurations Now however, the rising demand for improved fuel economy is sparking renewed interest in waste heat recovery for the medium and heavy-duty engine markets. Add in the recent developments in advanced turbocompound systems - using CVT (continuously variable transmission, Fig. 1b) technology to improve performance of the turbocompound system throughout the engine speed and load range [7,8] – and one might think that turbocompound is finally poised for more widespread market penetration. However, an additional trend must be considered when assessing the likely role of turbocompound in future high efficiency diesel engine development, and that is the trend in emission compliance technology. Emissions legislation of the past 10-12 years has driven the widespread use of a number of NOx and particulate control technologies, in particular the use of EGR (exhaust gas recirculation) and various types of exhaust aftertreatment. These technologies have a significant impact not only on engine performance, but also on turbocompound performance and effectiveness. Much of the development and research into turbocompound was done before the addition of EGR and aftertreatment to the engine, and there has been little in the literature in the last decade to update the performance impact of turbocompound in light of these changes to the heavy-duty engine architecture. This paper documents the results of an analytical investigation into the fuel economy impact of turbocompound on a heavy-duty diesel application with EGR and exhaust aftertreatment, and endeavors to assess the viability of turbocompound as a fuel economy building block in the era of emissions legislation and controls.

2 ANALYSIS OVERVIEW Engine simulation was used to evaluate the fuel economy impact of turbocompound on a 15.2L heavy-duty diesel engine over a range of speeds and loads (see Fig. 2). Comparisons were conducted over a range of backpressures to simulate the effects of exhaust aftertreatment (different backpressures generically representing different aftertreatment technologies or the effects of different aftertreatment packaging constraints). A no-EGR comparison was also conducted. The points marked with an asterisk are the four points that were used for hardware optimization/selection, as discussed in section 2.5.

270

BMEP, kPa

2500.0 2000.0

A100*

1500.0

A75

1000.0

A50

500.0

A25

B100*

0.0 1000

1200

C100*

B75

1400

C75

B50*

C50

B25

C25

1600

1800

2000

Engine Speed, rpm

Figure 2. Analysis points – speed and load 2.1 Engine description and limits Basic description of the engine and operating limits used are shown here as Table 1. Table 1: Engine Configuration and Operating Limits Displacement (L) Rated Speed, rpm

15.2 1800

Rated Power, kw

360

Rated BMEP, kPa

1575

Peak Torque Speed, rpm

1200

Peak Torque Power, kw

312

Peak Torque BMEP, kPa

2055

Peak Cylinder Pressure limit, MPa A/F ratio targets

17.0 Rated: 21:1 Peak Torque 19:1

NOx target, ‘with EGR’, g/kw-hr

4

NOx target, ‘no EGR’, g/kw-hr

8

For the cases where turbocompound was used, the lug curve was kept the same, i.e. the total power from reciprocator plus power turbine equals the original lug curve. The A/F ratio targets are set as approximate minimums for acceptable smoke, particulate, and exhaust temperatures. The NOx target for the ‘with EGR’ case is an approximate engine-out NOx limit assuming ~90-93% SCR (selective catalytic reduction) conversion efficiency to meet U.S. Tier4 off-road or 2010 on-highway NOx limits. The NOx target for the ‘no EGR’ case assumes higher SCR conversion efficiency to meet the same limits; it can also be used to assess the turbocompound benefit in markets where emissions regulations are less stringent. 2.2 Air system configurations The base (no turbocompound) and turbocompound configurations are shown schematically in Fig. 3. Fixed geometry turbines are used in all simulations – the impact of variable geometry turbines is explored in the results. There are a few characteristics specific to each system:

271



Base (no turbocompound) engine, configurations 1 and 2 - Both use a divided exhaust manifold and turbine housing. For configuration 1 (with EGR), an asymmetric turbine is used to help drive EGR [9]. For configuration 2 (no EGR) a standard symmetric turbine is used. Turbocompound engine, configurations 3, 4, and 5 - Results are presented for power turbine located upstream and downstream of the aftertreatment. Power turbine is undivided.



EGR

configuration 1

configuration 2

cooler T AT Asymmetric Turbine

T AT Symmetric Turbine

C

configuration 4

EGR

EGR

configuration 3

cooler AT

T

Power Turbine

C

T

cooler

C

Power to crank

configuration 5

Power T Turbine

AT

T

C

Power to crank

AT

T

Power Turbine

T

C

Power to crank

Figure 3. Analyzed configurations: Base (top), Turbocompound (bottom) 2.3 Turbo maps and efficiencies Isolating the effects of a technology like turbocompound and ensuring a clean backto-back is a challenge, so turbo maps for this comparison were handled as follows: •

• •

Turbocharger turbines – a high-efficiency symmetric/divided turbine map was used for all configurations, scaled as needed for flow. For configurations 3 and 4 (undivided turbine) the efficiency was increased 1%. For configuration 1 (asymmetric turbine), the efficiency was decreased 15%, the penalty increasing as the turbine asymmetry ratio increased. Turbocompound turbine – used map from a high efficiency axial power turbine stage, scaled as needed for flow. A 93% efficient transmission of energy to the crankshaft was accounted for in the model (transmission efficiencies as high as 95% have been reported in the literature [3]). Compressor maps – comparable efficiencies used for all configurations.

2.4 Aftertreatment backpressures A calibrated restriction was used to replicate the pressure-drop versus flow characteristic of a measured DPF+SCR configuration. The resulting curve was scaled up or down to represent different aftertreatment configurations. Three cases were investigated: •





272

High BP (Medium + 10kPa backpressure) - a restrictive system, but realistic for the tight installation space of some machine and truck chassis. Medium BP – a mid-level backpressure typical of today’s heavy-duty onhighway truck installations Low BP (Medium – 7kPa) – low backpressure or no-DPF solution.

2.5 Optimization methodology For each backpressure level, the turbomachinery was sized to provide best overall BSFC at the four points noted in Fig. 2, each point weighted equally. This was done to give a balanced view of the fuel economy benefit of turbocompound, without being specific to lower-speed (on-highway truck) or higher-speed (machine) application cycles. Sweeps of turbine size, turbine asymmetry (for configuration 1), and power turbine size (for configurations 3, 4, and 5) were conducted to provide the lowest weighted BSFC for each configuration. For each turbo configuration, injection timing and EGR valve position were swept to determine best BSFC within the constraints of PCP (peak cylinder pressure), A/F, NOx, and exhaust temperature limits.

3 RESULTS 3.1 Turbocompound benefit at low NOx (with EGR) BSFC comparison of the baseline and turbocompound engines over a range of backpressures is shown in Figure 4.

High BP Medium BP Low BP

3.00% 2.00% 1.00% 0.00%

C25

C50

C75

C100

B25

B50

B75

B100

A25

A50

-2.00%

A75

-1.00% A100

BSFC Benefit, %

4.00%

Operating Condition Figure 4. Turbocompound BSFC benefit, configuration 1 vs. configuration 3 For the high backpressure case, the turbocompound configuration produced only a minimal 0-1% BSFC benefit, primarily at lower speeds. The lack of BSFC improvement is due to significant pumping work penalties with turbocompound. The pressure ratio of the power turbine multiplied by the high backpressure yields excessive exhaust manifold pressures, and the high pumping work offsets the fuel economy benefit of the 20kw of power from the power turbine. As backpressure is decreased, the BSFC benefit of turbocompound improves, reaching ~2-3% at higher speed, full load conditions, dropping to ~1-1.5% or less as speed or load is decreased. As backpressure is decreased, the power turbine is downsized, resulting in more compounding power to the crank. Additionally, the downsized power turbine helps limit boost and cylinder pressure as backpressure is decreased, so that injection timings can be maintained or even advanced. This negative impact of aftertreatment backpressure on turbocompound is not unexpected. Past investigations have shown the sensitivity of turbocompound to turbomachinery efficiencies [10, 11], and an increase in backpressure will have a similar impact (as far as the engine is concerned) as a decrease in turbo efficiency.

273

Turbine efficiencies for the medium backpressure simulations are shown in Figure 5. The base engine had lower turbine efficiencies due to the efficiency penalty of asymmetry and the divided turbine housing. Figures 4 and 5 illustrate that turbocompound produced only a 1-2% BSFC advantage at higher loads, despite having significantly higher turbocharger turbine efficiencies. Figure 5 also shows the power turbine efficiency falling dramatically at lower loads due to the low expansion ratio across that stage; the effects of this will be discussed in section 3.3.

Turbine Efficiencies

90

Turbine - Baseline

Turbine - Turbocompound Power Turbine - Turbocompound

80 70 60 50 40 30 20

C25

C50

C75

C100

B25

B50

B75

B100

A25

A50

A75

0

A100

10

Operating Condition Figure 5. Turbine efficiencies, medium backpressure, configurations 1 & 3 Given the negative impact of backpressure on turbocompound, it’s logical to question whether the power turbine should be located downstream of the aftertreatment, so that the power turbine pressure ratio is applied to a lower outlet pressure. Results of such a comparison are shown in Figure 6 for the medium backpressure case. Locating the power turbine downstream of the aftertreatment primarily helps the BSFC at rated conditions. At C100 the power turbine delivered 31kw of power when located downstream of the aftertreatment vs 25kw of power when located upstream. At the A100 condition, the downstream location decreased the BSFC benefit; this was due to a decrease in the EGR driving capability and the need to retard timing in order to maintain NOx emissions below the limit. While the downstream power turbine location helps the rated BSFC improvement, it also limits hardware options. A CVT-based turbocompound (shown in Figure 1b and discussed in section 3.3.3) cannot be used downstream of the aftertreatment. Even a conventional turbocompound with mechanical transmission to the crank may be difficult to package, such than an electrical transmission may be required.

274

Configuration 3 vs. 1 (upstream power turbine) Configuration 4 vs. 1 (downstream power turbine)

3.00% 2.00% 1.00% 0.00% C25

C50

C75

C100

B25

B50

B75

B100

A25

A50

medium backpressure

A75

-1.00%

A100

BSFC Benefit, %

4.00%

Operating Condition Figure 6. Turbocompound BSFC benefit, impact of power turbine location 3.2 Turbocompound benefit at high NOx (no EGR) For the no EGR case, simulations were conducted only at the medium backpressure condition. Results are in Figure 7. medium backpressure

3.00% 2.00% 1.00% 0.00% -1.00%

C25

C50

C75

C100

B25

B50

B75

B100

A25

A50

A100

-2.00%

Configuration 3 vs. 1 (EGR) Configuration 5 vs. 2 (no EGR) A75

BSFC Benefit, %

4.00%

Operating Condition Figure 7. Turbocompound BSFC Benefit, EGR vs. no-EGR comparison The removal of EGR has the following impacts on the BSFC benefit: • •

1-1.5% greater BSFC benefit at higher load conditions. Without EGR, the energy in the exhaust stream increases significantly, resulting in higher power from the power turbine (from 25kw to 40kw at C100) At higher speeds, the BSFC benefit decreases faster with load in the absence of EGR. The turbocompound engine maintains EGR driving capability at lower loads better than the base engine, and can run at more advanced part-load timings. This advantage is lost in the no-EGR scenario.

275

3.3 Improvement potential via application of advanced technologies In the preceding sections, the BSFC benefit of turbocompound was investigated using ‘conventional’ turbocompound technology. The benefits of turbocompound may well be enhanced by the application of advanced technologies, such as higher efficiency power turbine designs, variable geometry turbines, or CVT-based turbocompound (Fig. 1b). The potential incremental benefits of these technologies will be explored here. 3.3.1 Improvement potential via advanced power turbine design In the simulations presented so far, a high efficiency axial turbine map was used for the power turbine, resulting in good power turbine efficiencies of 76-78% at higher load conditions. However, as load and power turbine pressure ratio decreased, the power turbine efficiency fell to 60% or less. Reference [12] describes a turbine designed specifically for high efficiency in the low (50%)

2020 (predicted) 100 73% 2.9 (85%) 1.6 (60%) 1.4 (>85%)

ELECTRIC SUPERCHARGING

Engine downsizing through inlet charging is gaining momentum for reducing engine emissions while maintaining sufficient power for customer acceptance [3]. Downsized engines are readily equipped with turbochargers that boost power at higher engine speed, but fail to provide boost at lower engine speeds. To boost smaller engines of say 1 litre (~1L) conventional radial compressors used in turbochargers become very small and need to operate at very high speeds, see [4]. An axial compressor would need to run faster still. At low engine speeds, an even lower flow rate for charging is required. This can be achieved by scaling down turbomachinery but operating at even higher speeds or cropping down blades to restrict flow but at the expenses of substantial efficiency reduction. Other radial turbomachine designs with low flow characteristics have been reported with wedge type blades [5, 6]. However, efficiency is reduced to ~55% due to greater wall friction losses (windage) and this is for larger flow machine. This is a well-known phenomenon and semi-empirical methods based on the Reynolds number (the dimensionless ratio of inertial forces to viscous forces) can be used to predict the efficiency penalty for scaling down a turbomachine, [7]. However, this does not account for all of the losses present in very small turbocompressors. As the size is reduced, the relative tolerances and imbalance increase, which further reduces the efficiency. Other solutions such as partial admission/emission designs suffer from low efficiency, i.e. 30-40% [6]. The alternative is a positive displacement compressor such as Regenerative, Screw or Roots that are bulky and noisy. Major developments of the latter have seen semi-axial flow as opposed to cross flow as used in VW TSI powertrain [8, 9]. One other important point relates to engine charging methodology. Engine charging through exhaust, mechanical, or hydraulic driven systems are engine dependent and hence linear with engine output. However, the solution required for boosting engine transient power is an engine independent rapid response boosting system. The emerging solution is a twin charge system, an exhaust-driven turbocharger for high engine speeds and an electrically driven supercharger for charging at low engine speed.

358

The electric supercharger using conventional turbomachine designs still require high speed operation with significant bearing, motor, and drive problems to overcome. Due to these considerations, and the lack of availability of very high speed motors, such compressors are not technically or economically feasible. Even at a size where turbomachinery is feasible (for 2L engine 60 g/s is required with radial turbomachine at 100 krpm), there is still a significant problem providing shaft power at high speed and at a reasonable cost. Low speed motors with gearboxes are large, and until recently, a powerful high-speed electric motor of sufficient power has been cost prohibitive.

3

INNOVATION

To overcome these limitations a novel solution has been developed, which can extend the specific speed range of radial compressors by radically changing the geometry of a low specific speed compressor rotor with highly forward swept blades [10-13], TurboClaw®. The high forward sweep (see Figure 1) allows a very low flow coefficient, and also allows a higher head than would normally be possible at the low blade speeds required. Mixing losses are increased due to the very high blade loading, but they are not the dominant loss mechanism in this regime. The design is sensitive to the Reynolds number, increasing in efficiency at larger sizes. This is because skin and wall friction is proportionally more significant at small sizes. Increasing the tangential component of rotor exit velocity means that the ratio of tangential to radial velocity at the exit of rotor can be much greater than four. It is noted that Watson’s [14] warnings of unstable operation are not valid for the present innovative design with almost tangential outlet angle and the resulting unique blade geometry (see Figure 1). This is promoted with the thick blades (see Figure 1) which reduce the flow. It was found that the radial velocity can be extremely low relative to the tangential velocity without causing instability problems. Since the radial velocity is very low, the increase in tangential velocity arising as flow rate is increased is also very low. This means that the Euler work input does not increase markedly with flow hence the theoretical slope of a constant speed line on the map is only marginally positive. In reality, the increase in losses with increased flow tends to mask this effect leading to a similar compressor map constant speed line shape as a conventional backswept machine (see Figure 2). It is noted that the design includes a diffuser to convert kinetic energy to pressure gain.

Conventional backswept

Cm2

C2

W2 U2

TurboClaw® forwardswept

C2

Cr2

U2

Figure 1: Turbocompressors and velocity triangles: left: conventional backswept; right: TurboClaw® forwardswept

359

This low flow-rate technology is uniquely suited for inlet charging of small engines at low flow rates. TurboClaw® may be readily employed at low engine speed and in conjunction with a conventional, relatively large turbocharger to provide boost at higher engine speeds. The electric motor spools up the low inertia TurboClaw® quickly, and the compressor is only used when required. Significantly the TurboClaw® geometry lends itself to low cost mass production techniques. The low tip speeds of TurboClaw® enables lower operational speeds and subsequent ease of bearings and electrical drive. A typical TurboClaw® compressor map from an 85 mm impeller used as an electric supercharger is depicted in Figure 2.

1.3

Av Pressure Ratio

Shaft speed, rpm

85 mm Impeller Compressor Experimental Map

1.25 1.2

14985 20002

1.15

25053 1.1

29990

1.05

35054 39993

1 0

0.005

0.01

0.015

0.02

0.025

0.03

Av Mass Flow [kg/s] Figure 2: The TurboClaw® compressor map used in twin charge HIL tests It is noted that a comparably diameter centrifugal turbocompressor would typically need to operate a 48,000 rpm in order to develop a pressure ratio of 1.29:1 achieved by TurboClaw®. The operating mass flow for this condition would be 0.065 kg/s for an efficient design of reasonable blade height.

4

EXPERIMENTAL VALIDATION

A proof of concept project sponsored by the UK Technology Strategy Board was successfully completed in 2010. In this project, scoping studies were carried out using AVL BOOST and AVL CRUISE [15] for various duty cycles in order to size the turbocharger and TurboClaw® supercharger. Once this was done, the TurboClaw® supercharger was evaluated in a Hardware-in-Loop (HIL) environment that simulates steady state and transient engine performance. The simulation involved constructing a mean value engine model capable of running in real time for the 1.0L engine. This model was then combined with the actual TurboClaw® supercharger hardware to form the HIL simulation. The layout schematic of the test hardware in the simulated twin charge HIL test is detailed in Figure 3. The supercharger test hardware consisted of a brushless direct current (BLDC) motor driving TurboClaw®. This was designed for the engine speed range between idle and 2800 rpm.

360

Figure 3: Twin charge hardware-in-the-loop (HIL) test schematic Engine torque versus speed data for a naturally aspirated 1.0L (dotted line) and 1.4L (solid line) engine are shown in Figure 4. This shows the 1.0L naturally aspirated engine to be substantially inferior to the 1.4L engine in terms of engine torque. The 1.0L engine was being simulated to show possible performance improvements with charging. The aim was to achieve performance equivalent to the 1.4L engine but at significantly lower fuel consumption for most driving conditions. The 1.0L engine was simulated with turbocharging to compare to 1.4L engine characteristics. This same turbocharged engine was further charged by the electrically driven TurboClaw® as Hardware-in-the Loop in the simulation. Data shows the turbocharged engine with marked torque performance improvements, and equivalent performance to the 1.4L engine, at higher engine speeds. At lower engine speeds, however, the 1.0L engine was short of boost with persistent turbo-lag. In this lower speed range, twin-charging by EDS TurboClaw®, restricted to a pressure ratio of 1.25 (electric drive < 1kW) shows engine torque performance better than the 1.4L engine. Peak torque is shown to be 100 Nm at engine speed of 2000 rpm, an improvement of 31% in engine torque. The system however fails to deliver the required boost in the engine speed range of 2000 to 4000 rpm due to limited motor power and compressor trim. More recent data using a variant of EDS TurboClaw® restricted to a pressure ratio of 1.4 (electric motor to 2.2 kW), not shown in Figure 4, peak at 2250 rpm at 130 Nm. Clearly the 1.0L

Figure 4: Hardware-in-the-loop test results

361

naturally aspirated engine can perform even better than the 1.4L engine with better EDS engine matching. The transient response of the present TurboClaw® geometry has also been investigated. Results are depicted in Figure 5, from top: speed (s), Pressure (kPa), and flow (g/s). Full speed operation (from rest to 40,000 rpm) was achieved in less than 2 seconds. Idle to full speed (15,000 to 40,000 rpm) in 1 second and 10,000 rpm accelerations in 500 milliseconds. These results were optioned using control software with limitations for acceleration to safeguard safety and durability of the prototype. This does not represent a case optimised for maximum acceleration. For transient performance optimisation of electrically driven superchargers, mass (πr2Lρ), moment of inertia (mr2/2 and hence πr4Lρ/2), as well as stored energy Iω2/2 need to be considered since it determines the rate of acceleration. In these formulations m is mass, r is impeller radius, ρ is material density, I is moment of inertia, and ω is rotational speed. On first consideration the fifth power of length scale in the formulation of moment of inertia is in favour of lower diameter conventionally backswept designed impellers operating at higher speeds. For TurboClaw® this is off set against the following: 1. at least a four fold advantage in operating speed, rotational inertial energy being proportional to the square of speed. 2. a 20% lower diameter for the same pressure and mass flow delivered due to lower tip speed, see (13) reducing inertia. 3. a much shorter impeller axial length (10% diameter as opposed to axial length similar to radius for standard backswept designs). 4. Higher torque motors are possible at the lower speeds without incurring punitive motor windage losses.

Figure 5: Hardware-in-the-loop results (Transient Response) The TurboClaw® prototype is presently manufactured from aluminium, but due to low stresses (tip speed 175 m/s at 40,000 rpm) the design has scope for optimisation for lower moment of inertia. In addition, and manufacture from plastics (lower density than aluminium by 2/3) is feasible. For the case discussed

362

here, diameter increased from 18 mm to 85, TurboClaw® is predicted to have a comparable or shorter acceleration time than conventional backswept turbomachine in low specific speed applications.

5

ACKNOWLEDGEMENTS

Financial investment by the Technology Strategy Board is gratefully acknowledged for “EDS TurboClaw® – project BS088J”. Also acknowledged is contribution from project partners AVL Powertrain and Turbocam Europe.

CONCLUSION The TurboClaw® compressor is a new form of radial turbocompressor. The innovation is uniquely placed as it is amenable to being electrically driven at speeds substantially lower than conventional turbocompressors. One application, electrically driven supercharging in conjunction with a turbocharger for small engine downsizing has been described. Results from a hardware-in-the-loop test are described showing substantial improvements in torque at lower engine speeds.

REFERENCES (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)

Deloitte ‘A new era: Accelerating towards 2020- An automotive industry transformed’ http://ec.europa.eu/clima/policies/transport/vehicles/cars_en.htm An Independent Report on the Future of the Automotive Industry in the UK ‘New Automotive Innovation and Growth Team (NAIGT)’ Ford reveals new turbocharger supplier (Oct 2011). http://www.enginetechnologyinternational.com/news.php?NewsID=33842 Rogers C ‘Efficiency of centrifugal Compressor Impellers’ paper 22, AGARD proceedings 282, May 1980. M V Casey et al ‘Radial compressor stages for low flow coefficients’, C403/004 IMechE 1990. Casey M V. The effects of Reynolds Number on the Efficiency of Centrifugal Compressor Stages Journal of Engineering for Gas Turbines and Power APRIL 1985 Vol 107 541-548 Feel the Force, “Engine Technology International”, March 2012, pp36-40. www.volkswagen.co.uk/technology/petrol/tsi Patent GB2.366.333A, dated 06-03-2002. PCT/GB04/003752, dated 9-03-2005. GB916901.2, dated September 2009 Pullen et al ‘TurboClaw- A low cost turbocompressor solution for fuel cells and other application’. International Conference on Compressors and their Systems, 7-9 September 2009, IMechE. Watson, N and Janota, M. S. Turbocharging the internal combustion engine Macmillan, 1982 ISBN 0333242904 www.avl.com/boost1 and www.avl.com/cruise1

363

A comparison of timescales within a pulsed flow turbocharger turbine C D Copeland University of Bristol, Mechanical Engineering Department, UK P Newton, R F Martinez-Botas Imperial College London, Mechanical Engineering Department, UK M Seiler ABB Turbo Systems Ltd, Switzerland

ABSTRACT Most modern turbocharger turbines are driven by a highly pulsating flow generated at the exhaust valve of an internal combustion engine. The amplitude and frequency of the exhaust pulses can influence the performance of the turbine when compared to steady state operation. It is useful to seek to simplify the problem of unsteadiness such that greater understanding may result. This paper uses a combination of experimental and computational results to study the various timescales associated with pulsed turbine operation. The effect of pulse amplitude and frequency on the unsteady flow in the volute and rotor is discussed.

NOMENCLATURE Area (m2) Length (m) Pressure (Pa) Strouhal number Particle residence time (s) Time (s) Fluid velocity component in a Cartesian coordinate system (m/s) Speed (m/s) Volume (m3) Component of distance in a Cartesian spatial coordinate system (m) Ratio of specific heats Pressure weighted unsteady parameter Pressure amplitude weighting factor Density Superscripts * Non-dimensional value Subscripts reference value Component in the ith direction of a Cartesian coordinate system

___________________________________________ © The author(s) and/or their employer(s), 2012

389

1

BACKGROUND

Turbochargers are unique among other turbomachinery applications since the turbine must be able to harness the energy contained in a highly dynamic, pulsating flow generated at the exhaust valve. In fact, most modern applications of turbocharging aim to maintain a separate flow path for each of the cylinders right up to the turbocharger turbine in order to preserve as much of the pulse energy as possible. Although this is well known, steady-state design rules are still applied to most turbochargers turbines since it is assumed that the turbine will respond to the varying flow in a steady-state manner. This can be described as the ‘quasi-steady assumption’.. Most commercial 1-D engine simulations work under this assumption. That is, the dynamic behaviour of the engine and manifold is modelled with wave action, yet the response of the turbocharger is determined from a steady-state lookup table extrapolated from test-stand map data. Indeed, even the most comprehensive studies of engine manifold dynamics may resort to the use of steady state map based modelling of the turbine [1]. What does not seem to be clear from the published literature, however, is how valid this assumption is under the range of pulsating conditions that a turbocharger turbine is subjected to. Early researchers into the unsteady performance of turbochargers such as Benson et. al [2, 3], Wallace et. al. [4, 5, 6], Kosuge, et al [7] and later Capobianco et. al [8, 9, 10] all sought to quantify the degree to which the turbocharger could be assumed quasi-steady when exposed to a pulsating flow. The work published by Benson and Wallace in the late 1960s [2-6] suggested that there could be a significant performance discrepancy if quasi-steady behaviour was used to predict the averaged unsteady performance. In the mid 1970s, Kosuge et.al [7] published results suggesting a tendency of the quasi-steady prediction to under-predict averaged unsteady mass flow and power. The quasi-steady mass flow predictions approached the unsteady value as the frequency increased, whereas the quasisteady output power prediction diverged from the unsteady with increasing frequency. In addition to the influence of pulse frequency, there also seemed to be a correlation between the pulse amplitude and the degree to which the turbine was quasi-steady. The work of Capobianco et. al [8, 9, 10] also echoed the work of Kosuge et. al. [7] in suggesting that pulse amplitude was a dominant factor in determining the validity of the quasi-steady assumption. Thus, the earlier research [2-10] was not unanimous but most authors suggested that the quasi-steady assumption under-predicts the true unsteady performance averaged over a pulse cycle. Since the approach of comparing time-mean performance data was not able to provide a definitive consensus, research that sought to measure time-resolved unsteady data began to appear. Dale and Watson [11] were among the first, followed by Baines et. al [12, 13] Arcoumanis et.al [14, 15], Karamanis et.al [16, 17], Szymko et.al [18], Szymko [19], Rajoo et. al [20, 21, 22] and Copeland et.al [23, 24]. Measuring all the time-resolved flow and torque data meant the instantaneous performance of the turbine could be directly compared to the steady-state operation. Immediately, the instantaneous relationship between the mass flow parameter (MFP) and pressure ratio (PR) was noted to vary from the standard nozzle behaviour seen in steady flow [11]. Szymko et. al. [18] showed that an increase in the Strouhal number (equation 3) appeared linked to an increasing difference between the trend of the unsteady MFP versus PR and the trend of the steady line. Costall et.al [25, 26] took this approach further by identifying a ‘transition’ to a fully unsteady regime using the slope of the unsteady orbit calculated from a 1-D numerical code. These results seem to show that an increase in pulse frequency (and Strouhal number) will lead to a departure from the steady-state relationship between mass parameter and pressure ratio. In this work the volute was treated as unsteady whilst the rotor wheel was treated in a quasisteady manner: an assumption deemed admissible due to the small passage length and high flow velocities. Also, this assumption was not without precedent since the

390

same conclusion was reached by Yeo and Baines [12] after measuring the instantaneous flow angles into the rotor under pulsating flow. While the body of research work presented has provided an improved understanding of the influence of pulsating flow, a definitive view of the relative importance of unsteady effects in turbochargers is yet to be proposed. One reason for this lack of clarity may be a difficulty in finding a common thread among the diverse range of turbine geometries that have been tested. For example, pulsating flow studies on a variable nozzle turbine by Rajoo [21, 22] suggested that the presence of a nozzle can have a marked influence on the quasi-steady versus unsteady agreement whilst Capobianco and Marelli [27] showed the effect the a waste-gate can have on turbocharger performance under pulsed operating conditions. Work by Copeland [23, 24, 28] showed that the way a multiple entry turbine is divided can also have a significant impact on both steady and unsteady performance. Thus, this paper seeks to take a broader view of the question by analysing the time associated with different events within a turbocharger turbine using a combination of both experimental and computational results. By comparing the wide spectrum of timescales that occur as well as the magnitude of changes associated with these unsteady events, this can be helpful in providing a greater insight into the effects of pulsating flow.

2

UNSTEADY CRITERIA

Figure 1 shows the frequency spectrum of a range of unsteady flow phenomenon that a turbocharger encounters. Also shown in this graph is an estimation of the maximum amplitude of the unsteady changes that are associated with each event. It is helpful to begin by considering the extremes of this plot. On the far left, there is a timescale that can be associated with the transient engine load response. The change of flow conditions due to engine transients is typically on the order of a second (1 Hz) for most automotive applications. At the extreme right of the spectrum, the highest frequency flow oscillations are a result of turbulence. Turbulent frequency can cover quite a broad range but typically take the form of very high frequency, low amplitude fluctuations. What is interesting about these two extremes is that they are both dealt with in a steady-state manner. That is, at very low frequencies (1 kHz) the flow variations can clearly be treated as quasisteady since the time associated with an engine load change is much longer than the time for the flow to pass through the turbine. At very high frequency (1 MHz), the turbulence eddies are small enough in magnitude that the flow can be treated as a steady, mean flow. Thus, if the two extremes of Figure 1 exhibit predominately steady-state characteristics, why are exhaust pulsations different? First, the frequency range of exhaust pulses (~10-100Hz) means that the changes of flow conditions are generally too fast to be dismissed as quasi-steady. Secondly, unlike turbulent events, the amplitude of the pulsations is significant, thereby suggesting that the unsteady effects cannot be simply averaged. In fact, as suggested in Figure 1, the amplitude of pulsating exhaust flow can, in some cases, cause the greatest change of turbine flow conditions compared to any other transient event.

Figure 1: Spectrum of unsteady events seen by a turbocharger turbine

391

Several researchers [19, 25, 26] have used the Strouhal number as a parameter to define the onset of unsteady effects in a flow. In order to understand the source of this non-dimensional parameter, consider the dimensionless form of the continuity equation. This can be non-dimensionalised with reference values of density (ρ0), time (t0), length (L0) and velocity (v0) as follows: ;

;

;

(1)

It is important to choose each of these reference values such that the dimensionless variable is on the order of unity. From here it is straightforward to show that the dimensionless form of the continuity equation for compressible flow becomes: (2) Where the Strouhal number (sometimes also referred to as Reduced Frequency) is defined as: (3) Equation 2 shows that the Strouhal number is a parameter whose size indicates the significance of the time derivative term. As pointed out by Greitzer et.al. [29], there is a physical interpretation of the Strouhal number that is helpful for understanding its role. If a fluid is flowing at a speed through a domain of length that is also pulsating at a frequency , the Strouhal number represents the ratio ⁄ of time needed for a particle to travel the length of the domain to the 1⁄ . If the travel time time associated with the disturbance being considered is comparable to the time for a local change of condition due to some unsteady event, the value of the Strouhal Number will be close to unity, thus indicating that both the unsteady, time dependent term and the convective term may be important in describing the flow. This physical interpretation of the Strouhal Number is shown visually in Figure 2. The left plot in Figure 2 shows the influence of two different pulse frequencies but the same mean velocity (and therefore travel time T0). The low frequency case should be close to quasi-steady behaviour since the flow only experiences a small unsteady change Δp2 over the distance. However, the higher frequency pulse will cause the flow to ‘feel’ a much larger change in flow conditions from the mean (Δp1) as it traverses the distance L0. The value of the Strouhal number reflects this effect since its value will be much higher in the second, higher frequency pulse. Yet as explained with the help of Figure 1, there are very high frequency unsteady events (turbulence, etc) that can be treated as steady due to their much lower amplitude. Consider the right plot in figure 2 where the two pulsating flows have very different frequencies and amplitudes. Although the Strouhal number will indicate that the higher frequency pulse is much more ‘unsteady’ (the value will be higher than the lower frequency pulse), the actual change of flow conditions (Δp) in time T0 resulting from the high frequency pulse is on par with the low frequency pulse because of the lower amplitude. Thus, it would seem that while the Strouhal number is useful in showing when the unsteady effect can or cannot be ignored, it does not necessarily provide an indication of the dominance of unsteadiness without consideration of the amplitude of the disturbance. Or, put another way, while a small Strouhal number does give a clear indication of a quasi-steady flow, once the Strouhal number approaches unity, the amplitude of the flow pulsations must also be considered.

392

Figure 2: Comparison of a change of flow condition (pressure) under different frequencies and LEFT: Similar amplitudes, RIGHT: different amplitudes In order to include the effect of fluctuation amplitude in the prediction of the degree of unsteadiness a further analysis is undertaken using an order of magnitude approach in a similar manner to that of Greitzer et al. [29]. They point out that one essential characteristic of fully steady operation is the absence of any mass storage within the flow domain. Considering this, it is sensible to start with the fundamental form of the mass continuity equation in a 1D internal flow: (4) For an incompressible fluid flowing through a domain of fixed volume V0, it is clear that the term on the right hand side of this equation will reduce to zero since the density of the fluid will be time invariant. Consequently at a given moment in time, the mass flow entering the domain will exactly match the mass flow at exit thereby resulting in a purely quasi-steady flow even if flow conditions at the boundaries are unsteady. As soon as a time dependant flow is introduced to a compressible fluid, it is clear that the right hand side of the equation will give a finite value and the mass flow entering and leaving will not be balanced at all times. However, if the mass flow through the domain is very large in comparison to the time dependant change of mass in the domain ( ), then the mass flow entering and leaving at any time will remain closely matched and the flow may be treated as quasi-steady. Clearly, when the opposite is true and the change of mass in the domain is a much larger proportion of the through-flow, there will be a significant filling and emptying effect that is no longer quasi-steady. In order to determine the significance of mass storage, it is useful to compare the magnitudes of the terms on each side of Equation 4. To estimate the order of magnitude of the unsteady mass flow term, the two quantities at the inlet and exit can be approximated as a nominal time averaged mass flow: ~

~

(5)

Considering the term on the right hand side of Equation 4, we assume that the domain volume is time invariant and that the mass of the fluid in the domain will be approximately proportional to the incoming fluid density. If a sinusoidal variation of inlet density is then assumed with a frequency, 1⁄ , and with ∆ being the peak to peak amplitude such that: ∆

sin

π

t t

(6)

393

We can then develop a term for the cycle averaged magnitude of the rate of mass change in the domain: ∆

~

(7)

As discussed, the ratio of these two terms will provide an indication of the importance of the mass accumulation effects within the flow domain: ∆





(8)

It is interesting that the final term above can be written as the product of the Strouhal Number (Equation 3) and a weighting factor that is based on the density fluctuation. If an adiabatic system is assumed, it is then possible to interchange the density with pressure, which is measured directly in the lab. Thus, the final unsteady criterion can be expressed as a product of the Strouhal number St and a pressure amplitude weighting factor given the symbol of capital Pi ( ): ∆





(9)

In the interest of brevity, it is useful to assign this parameter the symbol of capital Lambda ( ). If this parameter is near to 1 it suggests that the average rate of change of mass within the flow domain is of a similar magnitude to the nominal time averaged mass flow rate traversing the domain. Thus, the filling and emptying of this volume will cause a significant discrepancy between the mass flowing in and out of the domain and it will not be acceptable to treat the flow as quasi-steady. For a sinusoidal pressure waveform, the weighting factor is bounded by a maximum value of 4⁄ when the base of the curve sits on zero (i.e. if ∆ ⁄ ). 3

EXPERIMENTAL DATA

In order to consider the different timescales involved within a pulsed flow turbocharger both experimental and 3-D computational fluid dynamic (CFD) analyses were drawn upon. The experimental conditions that will be considered here are shown in Table 1 below. The data were obtained from the Imperial College turbocharger turbine test facility and have been presented previously by Copeland et.al. [23, 24, 28]. This facility has been described in detail by several researchers [18-24], accordingly only a brief overview is given here. The Imperial College turbocharger test facility is a cold flow test facility, allowing much greater instrumentation compared to a hot flow test stand. The main advantage of the Imperial College facility over other similar test stands is the use of a high speed electromagnetic, eddy current dynamometer which allows a broad range of test conditions without the aerodynamic limitations of a compressor. The facility also has the ability to operate under either a steady flow or a dynamic pulsating flow using a set of ‘chopper plates’ that periodically cut the air at a given frequency in order to simulate the exhaust pulsations that the turbine may experience under real operation. Under pulsed operation the instantaneous mass flow and pressure are measured at the inlet to the turbine along with the instantaneous power generated by the turbine wheel. Szymko [19] showed that the temperature fluctuation follows an adiabatic relationship with the pressure, thus allowing the instantaneous temperature to be inferred using the average temperature and instantaneous pressure. Table 1 provides the data for the two unsteady operating points that will be considered in this paper. The turbine wheel speed was held at approximately 830rps across two different pulse frequencies: 56Hz and 84Hz. This meant that the turbine wheel rotated 15, and 10 times respectively over a single pulse cycle. The

394

load on the turbine was selected to achieve an average velocity ratio of approximately 0.65 corresponding to the peak efficiency in steady-state. Table 1: Two unsteady test cases Speed

P.R. (avg)

Freq. (Hz)

835

in

1.72

56.3

0.656

15

828

in

1.74

84.3

0.66

10

(rps)

4

U/Cis

Phase

(avg)

N/f

COMPUTATIONAL MODEL

Although the Imperial College test facility allows detailed measurements of instantaneous mass flow, pressure and turbine power there are some measurements which are close to impossible to obtain experimentally. One of the most important time scales to consider is the time taken for the flow to pass through the different components that make up the turbine stage. Therefore to evaluate flow residence times, a CFD model of the turbine was used. It was also used to generate flow data such as pressure amplitude in locations that were not measured experimentally. A thorough validation of the model against experimental data was not considered necessary in the scope of this paper since the detail of the flow inside the turbine is not part of the analysis here.

Figure 3: Particle paths through the turbine stage used to generate average resident times Table 2: Mesh properties Component

No. Mesh Elements

Mesh type

Rotor wheel

2,014,704

Structured hexahedral

Nozzle ring

1,056,240

Structured hexahedral

Volute

1,333,907

Swept and unstructured

Exit Duct

121,267

Swept and unstructured

TOTAL

4,526,118

395

In order to replicate the unsteady effects in the test facility, the whole turbine stage was modelled from the location of the inlet ‘measuring plane’ to the exit duct. The model included the full volute, the nozzle ring with 24 passages, and the rotor wheel consisting of 12 passages. Table 2 gives details of the mesh used in this simulation. Fully transient computational analyses were run to model each of the unsteady experimental conditions given in Table 1. For each condition, the turbine speed was set to the average turbine speed measured in the experimental case. The measured atmospheric pressure was used as the exit boundary condition and the measured static pressure profile was applied at the inlet to the volute. A turbulence model with scalable wall functions was used to model the turbulence effects. Explicit rotor rotation was applied to give 1˚ of rotation per timestep; this meant that for a single pulse in the 56Hz case (15 rotor rotations per pulse) 5,400 timesteps were needed. A method was developed to track an imaginary fluid particle being convected with the bulk flow from inlet to exit. By applying this technique, estimates of the particle travel times through each component of the turbine system (volute, nozzle passage, nozzle-rotor interspace and nozzle passage) could be calculated. The particles were “released” at several different times throughout a pressure pulse in order to get an idea of how the residence times in each component changed over this time. In total, six imaginary particles were traced through the turbine flow domain at each given start time; three at each volute entry so that the average transit time for each volute passage could be calculated. The particle paths through the stage are shown visually in Figure 3. This figure also shows one of the unique features of a double entry design, namely, the difference in flow lengths between the inlets. However, since the two pulse shapes were all in-phase at each inlet, the two residence times in the double entry volute can be thought of as the upper and lower bound of a standard single entry turbine.

5

TIMESCALES

5.1 Flow Residence Time Table 3 shows the average residence time of the particles tracked through the stage from the inlet of the volute to the exit of the rotor in the CFD analyses. The residence times have been normalised by the rotor rotation time (1.205 ms) to give a sense of scale. Note that, to some degree, the residence times will change depending when in the pulse the particle is released. Here, the “start time” provides the time in the pulse cycle (shown in Figures 4-7) when the particle is released at the volute inlet and allowed to flow through the different stages. Two start times are considered in this table, one near the beginning of the pulse event (0.005 sec), and one nearer to the middle (0.0083 sec). Table 3 demonstrates that the time taken for the fluid to pass through the volute will be between 1.2 and 4.3 times the full rotor rotation time. The time to flow through the nozzle and interspace (the gap between nozzle and rotor) is typically under 20% of rotor rotation time whereas the residence time through the rotor can be up to 40%. The comparison between the rotor rotation and the flow transit time through the rotor is of particular interest in the double entry turbine design. Copeland et. al. [27] showed that these times can influence how the flow in the wheel is established when there are unequal flow conditions feeding each inlet. Under unequal admission, the flow in each rotor passage only has a ½ rotor rotation to develop before being abruptly exposed to a different flow, creating a highly dynamic flow into the turbine wheel, even under steady inlet conditions. Nonetheless, the effect of unequal admission will not be considered at length in this paper due to the constraints of space and the desire to make the conclusions as general as possible.

396

Table 3: Flow residence time in different stages of the turbine domain Normalized by the time for one complete rotor rotation Normalised Time (Normalised by 1 rotor revolution) Frequency (Hz) 56

84

Start time (s) (See Figs. 4-7)

volute (short inlet)

volute (long inlet)

nozzle + interspace

rotor

0.0050

1.436

3.465

0.143

0.292

0.0083

1.297

3.796

0.159

0.313

0.0050

1.823

4.329

0.202

0.397

0.0083

1.253

3.271

0.151

0.302

Figures 4-7 demonstrate how the times associated with the flow through each component of the rotor system compare to the overall time of the pressure pulse. Figures 4 and 5 show the 56Hz static pressure pulse measured at the inlet of the shorter and longer turbine entries respectively. Similarly, Figures 6 and 7 show the 84Hz static pressure pulses. Superimposed on these plots are blocks of colour to represent the residence times of the particles in the volute, nozzle, interspace and rotor. The start of the particle track occurs near the beginning of the pulse (0.005 sec) and, as time progresses, the particle travels with the bulk flow while the upstream pressure is changing due to the unsteady pulse. For example in Figure 4, the particle is released at 0.005s into the inner (shorter) volute. It then leaves the volute at approximately 0.0067s to enter the nozzle. During this 0.0017s that this particle is traversing the volute, the upstream inlet pressure rises from just below 160kPa to approximately 200kPa. These plots show, in a very visual way, the essence of the Strouhal number. As discussed in Section 2, the Strouhal number simply provides a comparison between the time taken for the fluid to pass through a domain, to the time of the unsteady event. One can immediately see from the plot that the time required for the flow to pass through the volute is a significant proportion of the pulse event. It also clearly shows that the flow requires much more time to traverse the longer, outer inlet and therefore sees a greater change of inlet conditions during this time (Figures 5 & 7).

Rotor exit

Volute inlet

Figure 4: 56Hz unsteady static pressure pulse measured at the inlet to the inner (shorter) volute with fluid residence times through the turbine system superimposed on top

397

Figure 5: 56Hz unsteady static pressure pulse measured at the inlet to the outer (longer) volute with fluid residence times through the turbine system superimposed on top

Figure 6: 84Hz unsteady static pressure pulse measured at the inlet to the inner (shorter) volute with fluid residence times through the turbine system superimposed on top

398

Figure 7: 84Hz unsteady static pressure pulse measured at the inlet to the outer (longer) volute with fluid residence times through the turbine system superimposed on top

5.2 Strouhal Number To formally calculate the Strouhal number for each of the stages, the appropriate time to represent the pulse ‘event’ must be selected. The time for the full cycle (1/frequency) could be used but since the actual pulse (rise and fall in pressure) occurs over a fraction of the cycle, using the full cycle time does not seem appropriate. Using the fraction of time that the exhaust valve is open, Φ, to only account for the pulse ‘event’ itself seems more in keeping with the intent of the Strouhal number. Table 4, therefore, calculates the Strouhal number using Φ = 1/3 of the full pulse cycle time; this corresponds to the chopper plate geometry used in the Imperial College unsteady test facility. This has the same intention as the Modified Strouhal Number used by Szymko et al. [18] and Costall et al. [25, 26]. The Strouhal numbers corresponding the outer volute passage show values that are close to unity when exposed to frequencies of 56Hz and 84Hz. Figures 5 and 7 show this more visually by demonstrating that the time taken for a particle to flow through the volute represents a significant proportion of the unsteady event. Thus, it seems clear that the outer volute cannot be considered quasi-steady for turbochargers exposed to most pulsating exhaust flows. The Strouhal Numbers calculated for the shorter, inner volute passage are accordingly smaller than those in the longer outer volute passage. Even so, the values are not small enough to treat as quasi-steady - especially in the 84Hz case. On examination of Figures 4 and 6 it seems that the particle residence time within the inner volute is still a significant portion of the whole pulse event in both the 56 and 84Hz cases. It is also intriguing to take note of the magnitude of the local change of conditions which occur as the fluid flows through the volute. On the rising portion of the pulse, the local change of pressure in the shorter inlet (Figures 4 and 6) is approximately 40kPa in the time it takes a fluid particle to travel to the nozzle inlet, under steady operation this would cause a significant change in operating condition. Figure 5 shows the larger residence time in the longer inlet can lead to an even greater change of pressure and can encompass most of the amplitude of the waveform.

399

Table 4: Strouhal numbers for each stage of the turbine Using 1/3 of the pulse cycle time Strouhal number Frequency (Hz)

Pulse 'event' (normalized by rotor rotation time)

Volute (inner)

Volute (outer)

nozzle + interspace

rotor

56

5.0

0.277

0.735

0.030

0.061

84

3.3

0.467

1.154

0.054

0.106

The Strouhal numbers for the nozzle (+ interspace) and the rotor are, as expected, significantly smaller than the volute due to the short path length and the larger bulk flow speed. The nozzle Strouhal number is almost two orders of magnitude smaller than unity thus suggesting that, if considered on its own, it can be treated as quasi-steady. However, the rotor passage is more interesting to consider. Although the Strouhal Number in the rotor is much smaller than the volute, it is not so small as to dismiss immediately. In an attempt to define a threshold which defines fully unsteady flow, other authors [18, 25] have considered the difference between the unsteady and quasi-steady mass parameter trend as explained in Section 1. Costall et al. [25] suggested that a Strouhal Number (if calculated in the same manner as in this work) between 0.13 and 0.26 will lead to a deviation of 5% between the slope of the steady and unsteady mass flow parameter. However, specifying such a threshold will always be somewhat arbitrary unless it can be quantitatively linked with the error between the true turbine performance and that predicted by assuming quasi-steady behaviour. Figure 4 is interesting here as it demonstrates that, even over this short time period, a 10 kPa change of volute inlet pressure is possible while the flow traverses the rotor. This pressure change, however, does not necessarily correspond to the pressure change seen at the rotor inlet due to the damping effect of the nozzle which tends to reduce the unsteadiness seen by the rotor. Figure 8 demonstrates this effect, showing a pressure pulse profile before and after the nozzle row. The pressure profile downstream of the nozzle here was obtained from the computational model.

Static Pressure (Pa)

250000 200000 150000 100000 Pressure post nozzle 50000 Pressure before nozzle 0 0

0.005

0.01 Time (s)

0.015

Figure 8: Pressure pulse profile upstream (solid line) and downstream (dashed line) of the inlet nozzle row, corresponding to the inner (shorter) entry at a pulse frequency of 56Hz 5.3 Lambda Unsteady Criteria A second unsteady criterion is proposed in this paper in Equation 9 that accounts for both the amplitude of the pulse shape as well as the frequency. This can be represented as the product of the Strouhal number and a weighting factor based upon the amplitude of the pressure disturbance. Table 5 shows the value of the

400

pressure amplitude weighting factor, , the Strouhal number, St and the unsteadiness criterion for the two volute passages and the rotor wheel. In the calculation of the pressure amplitude weighting factor, , it is most appropriate to use the amplitude of the pulse that is entering the domain under consideration rather than the amplitude of the disturbance measured at the inlet to the volute. Thus, the static pressure amplitude at the rotor inlet was obtained from the CFD model in order to calculate this factor for the rotor wheel. The value of at the rotor inlet is, as suspected, less than in the volute. This illustrates how the unsteady effect of the pressure pulse into the rotor wheel is reduced due to the damping influence of the nozzle. The calculation of the Strouhal Number here was again based on a disturbance time scale of 1/3 of the total cycle time as in Section 5.2. Table 5: Lambda (Equation 9) for the volute and rotor Volute Inner

Volute outer

Rotor passage

Pulsation Frequency (Hz)

Π

St

Λ

Π

St

Λ

Π

St

Λ

56

0.904

0.277

0.250

0.777

0.735

0.571

0.526

0.061

0.032

84

0.696

0.467

0.325

0.499

1.154

0.575

0.490

0.106

0.052

As outlined earlier, represents a ratio of the cycle averaged mass flow through a domain compared to the rate of change with time of the fluid mass within the domain. It is evident that the values of shown for both the inner and outer volute entries here suggest that the volute cannot be considered as quasi-steady. Even the smallest value of for the inner volute, under a 56Hz pulsation frequency, suggests that the rate of filling and emptying of mass into the volute will be about 25% of the nominal mass flow which would cause a significant discrepancy between the mass flow in and out of the domain. It is also interesting to note the effect of the amplitude weighting factor in table 5. As pulse frequency increases, this factor decreases due to a rise in the average pressure P0 and the decrease in amplitude ΔP as shown by Copeland et.al. [24]. Thus, for the outer volute, both pulse frequencies produce a similar value of , suggesting that the higher pulsation amplitude at 56Hz contributes to the unsteady effect in a similar way to increasing the pulsation frequency to 84Hz. The value of for the rotor wheel is an order of magnitude lower than for the volute showing that the rate of change of mass within a rotor passage may account for up to 5% of the mass flow. Although it is evident that this will cause the rotor wheel to deviate slightly from a purely quasi-steady operation, the values of are small enough to suggest that treating the rotor as quasi-steady will only lead to a small error. This is in keeping with the assumption of other authors [12, 13] who also felt the rotor must be predominantly quasi-steady without such a rigorous examination as presented here.

6

CONCLUSIONS

Acknowledging that it is common practice to assume that the turbine is quasisteady in commercial wave action codes, it is important to present a thorough review of the validity of this assumption. If any of the turbine components (volute, nozzle or rotor) are not quasi-steady, modelling the entire stage as if it behaved in a steady-state manner will produce inaccuracies not only in modelling the turbine, but also the engine performance predictions.

401

Two unsteady criteria have been developed from the continuity equation in order to assess whether the different turbine stages can be considered quasi-steady. The Strouhal number (or reduced frequency) has been used by a number of authors as a test of the quasi-steady assumption. It compares the time for a fluid particle to flow through the domain, to the time associated with an unsteady event. A fully unsteady CFD model of the turbine with a pulsating flow inlet boundary condition was used to obtain fluid residence times at various stages along the fluid path from inlet to exit. This allowed a more accurate calculation of the Strouhal number for the volute, nozzle and rotor than previously available. The values of Strouhal number for both volute passages (shorter and longer) were sufficiently close to unity to clearly indicate that the volute cannot be considered quasi-steady. In addition, a visual demonstration of the rise in static pressure that occurs at the volute inlet during the residence time of the fluid confirmed that a fluid particle would feel a significant change in pressure as it travels from inlet to nozzle. After the volute, the flow quickly passes through the nozzle due to the short path length. The associated Strouhal number in this case is much smaller than unity, indicating quasi-steady behaviour across the nozzle stage. The radial turbine wheel is more interesting to consider, for a pulse frequency of 84Hz, a Strouhal number of 0.1 was calculated, suggesting that the operation of the turbine wheel could be bordering on unsteady. Therefore, a second unsteady criterion was developed to include the effect of both pulsation frequency and amplitude. This unsteady criterion, , compares the order of magnitude of the mass flow through a domain to the periodic change of fluid mass in time within the flow domain. It is shown that this can be represented as the product of the Strouhal Number and a pressure amplitude weighting factor, . On calculation of this parameter for each volute entry, it was apparent that the unsteady mass accumulation effect cannot be ignored - thus confirming the Strouhal Number analysis. For the rotor wheel, the CFD model was used to obtain the static pressure fluctuation at the inlet that is needed to calculate for the blade passage. This showed that the pressure pulse entering the rotor is significantly damped by the nozzle thereby resulting in a value of that was almost two orders of magnitude less than unity. Thus, although the rotor is not wholly quasi-steady, the mass accumulation is insignificant enough that employing the quasi-steady assumption rotor stage is deemed appropriate for most cases. In summary, this work has demonstrated that the turbine stage as a whole will not behave in a quasi-steady manner because of the mass storage occurring in the volute during unsteady operation. However, it has been shown that if a methodology can be developed to account for the unsteadiness of the volute, it is generally valid to assume the turbine wheel will act in a manner very close to quasi-steady. This conclusion could be used to develop a more sophisticated method of modelling the turbine in 1-D wave action software that includes a mixture of unsteady modelling (volute) and steady-state map-generated behaviour (nozzle and rotor).

7

ACKNOWLEDGEMENTS

The authors would like to thank ABB Turbosystems for their continued support and insight throughout this work.

402

8 [1]

[2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]

REFERENCES D’Errico, G., Montenegro, G., Onorati, A., Piscaglia F., “Integrated 1D-3D Fluid Dynamic Simulation of a Turbocharged Diesel Engine with Complete Intake and Exhaust Systems”, Proc. of SAE 2010 World Congress & Exhibition, 201001-1194, April 2010. Benson, R.S. and Scrimshaw, K.H., “An Experimental Investigation of NonSteady Flow In a Radial Turbine”, Proc. IMechE, 1965, 180, Part 3J, Paper 23. Benson, R., “Nonsteady Flow in a Turbocharger Nozzleless Radial Gas Turbine”, SAE National Combined Farm, Construction and Industrial Machinery and Powerplant Meeting, 9 - 12 September, 1974. Wallace, F.J and Blair, G.P., “The Pulsating-Flow Performance of Inward Radial-Flow Turbines”, 65-gtp-21, ASME, 1965. Wallace, F.J., Adgey, J.M. and Blair, G.P., “Performance of Inward Flow Turbines under Unsteady Flow Conditions”, IMechE Proc., 184(1), 1969-1970. Wallace, F. and Miles, J. “Performance of Inward Radial Flow Turbines under Unsteady Flow Conditions with Full and Partial Admission”, IMechE Proc.,Vol.185, Paper C77/71, 1970-1971. Kosuge, H., Yamanaka, N., Ariga, I. and Watanabe, I., “Performance of Radial Flow Turbines under Pulsating Flow Conditions”, Journal of Engineering for Power, 1976, pp53-59. Capobianco, M., and Gambaraotta, A., “Influence of the pulsating flow operation on the turbine characteristics of a small internal combustion engine turbocharger”, Proc. of IMechE, C372/019, 1989. Capobianco, M., and Gambaraotta, A., “Unsteady flow performance of turbocharger radial turbines”, Proc. of IMechE, C405/017, 1990. Capobianco, M., and Gambaraotta, A., “Unsteady flow performance of turbocharger radial turbines”, Journal of Engineering for Gas Turbines and Power, Vol. 114, 1992, pp553-560. Dale, A. and Watson, N., “Vaneless Radial Turbine Performance”, Proc. IMechE, Paper C110/86, 1986. pp. 65-76. Yeo, J. and Baines, N. “Pulsating Flow Behaviour in a Twin-Entry Vaneless Radial-Flow Turbine”, Turbocharging and turbochargers, IMechE, 4th, 1990. Baines, N. C., Harjilouy-Benisi, A., and Yeo, J. H., “The Pulse Flow Performance and Modelling of Radial Inflow Turbines”, Proc. IMechE, 180, Part 3J, 1994, Paper 23. Arcoumanis, C., Hakeem, I., Khezzar, L. and Martinez-Botas, R.F., “Performance of a Mixed Flow Turbocharger Turbine Under Pulsating Flow Conditions”, Transc ASME 95-GT-210, 1995. Arcoumanis, C., Karamanis, N., Martinez-Botas, R. F., and Su, C.C., “Unsteady characteristics of a mixed-flow turbocharger turbine”, IMechE. C557/030, 1999. Karamanis, N., Martinez-Botas, R.F., and Su, C.C., “Mixed Flow Turbines: Inlet and Exit Flow Under Steady and Pulsating Conditions”, ASME Journal of Turbomachinery, Vol. 123, 2001, pp.359-371. Karamanis, N., Martinez-Botas, R.F., “Mixed-Flow Turbines for Automotive Turbochargers: Steady and Unsteady Performance” IMechE Int. J. Engine Research, 2002 Vol. 3 No.3, 2002. Szymko, S., Martinez-Botas, R. F. and Pullen, K. R., “Experimental Evaluation of Turbocharger Turbine Performance under Pulsating Flow Conditions”, Proc. of ASME Turbo Expo, GT 2005-68878, 2005. Szymko, S. “The development of an eddy current dynamometer for evaluation of steady and pulsating turbocharger turbine performance ”, PhD Thesis, Imperial College London, 2006. Rajoo, S. and Martinez-Botas, R. F., “Experimental Study on the Performance of a Variable Geometry Mixed Flow Turbine for Automotive Turbocharger”, Proc. of IMechE. C647-09, 2006.

403

[21] Rajoo, S. and Martinez-Botas, R. F., “Unsteady Effect in a Nozzled Turbocharger Turbine”, Proc. of ASME Turbo Expo, GT2007-28323, 2007. [22] Rajoo, S., “Steady and Pulsating Performance of a Variable Geometry Mixed Flow Turbocharger Turbine”, PhD Thesis, Imperial College London, 2007. [23] Copeland, C., Martinez-Botas, R. M. and Seiler, M. “Unsteady Performance of a Double-Entry Turbocharger Turbine with Comparison to Steady-Flow Conditions”, ASME Journal of Turbomachinery, TURBO-06-1021, 2008. [24] Copeland, C., Martinez-Botas, R. M. and Seiler, M. “Comparison between Steady and Unsteady Double-entry Turbine Performance using the Quasisteady Assumption”, ASME Journal of Turbomachinery, TURBO-09-1109, 2009. [25] Costall, A., Szymko, S., Martinez-Botas, R. F., Filsinger, D., Ninkovic, D., “Assessment of Unsteady Behaviour in Turbocharger Turbines”, Proc. of ASME Turbo Expo, GT 2006-90348, 2006. [26] Costall, A., Rajoo, S. and Martinez-Botas, R. F., “Modelling and Experimental Study of the Unsteady Effects and their Significance for Nozzleless and Nozzled Turbine Performance”, THIESEL Conference on Thermo and Fluid Dynamic Processes in Diesel Engines, 2006. [27] Capobianco, M., Marelli, S., “Experimental analysis of unsteady flow performance in an automotive turbocharger turbine fitted with a waste-gate valve”, Proc. of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering 2011 225: 1087, 2011. [28] Copeland, C., Newton, P., Martinez-Botas, R. M. and Seiler, M. 2010 “The Effect of Unequal Admission on the Performance and Loss Generation in a Double-entry Turbocharger Turbine”, Proc. of ASME Turbo Expo, GT201022212, Glasgow, Scotland [29] Greitzer, E. M., Tan, C. S., Graf, M. B., “Internal Flow, Concepts and Applications” Cambridge: Cambridge University Press, 2004, ISBN 0521343933.

404

Experimental analysis of turbocharger interaction with a pulsatile flow through time-resolved flow measurements upstream and downstream of the turbine F Laurantzon, R Örlü, A Segalini, N Tillmark, P H Alfredsson KTH CCGEx, Department of Mechanics, Royal Institute of Technology, Sweden

ABSTRACT The inflow to and outflow from turbochargers are highly complex and, in particular, pulsating. Nevertheless, most studies of turbocharger performance are conducted under steady conditions. Hence, there is a great interest in determining and understanding turbocharger performance maps under pulsatile conditions. The highly complex flow field constitutes a challenge for time-resolved flow measurements by means of conventional measurement techniques. In a recent paper by Laurantzon et al [Meas. Sci. Technol. 20 123001 (2010)], time-resolved bulk flow measurements under pulsatile conditions have been obtained via wavelet analysis of the signal from a vortex flow meter. Here, this method has been used in order to obtain time-resolved performance maps based on the mass flow both upstream and downstream of the turbine. The results show that the turbine has a large damping effect on the mass flow pulsations, but that the pulse shape is to a high degree preserved while passing through the turbine, and that the time-dependent filling and emptying of the turbine case make the quasi-steady assumption invalid, if the whole turbine stage is considered.

1 INTRODUCTION At KTH CCGEx, Competence Centre for Gas Exchange, KTH works together with vehicle industry in Sweden within the area of gas management of internal combustion engines, where turbo charging is a focal point. Essential in this respect are turbocharger performance maps, which in most cases are obtained under steady conditions. The inflow to and outflow from turbochargers, on the other hand, are highly complex and, in particular, pulsating, see e.g. Ref. (1,2). Hence, there is a great interest to understand how turbocharger performance maps behave under pulsatile conditions. The highly complex flow field, including high speeds, back flow, strong secondary flows, non-isothermal conditions, strong and rapid pulsations etc. constitutes a challenge for time-resolved flow measurements. While commonly employed measurement techniques such as thermocouples and Pitot-tubes do not even guarantee a correct time-averaged reading under pulsating conditions, hot-wires provide time-resolved mass-flux readings (see review in Ref. (3)). Thermal anemometry techniques also need time-resolved temperature measurements, due to the necessity of temperature compensation. Time-resolved temperature measurements in the context of typical engine pulsating frequencies are not possible with standard thermocouples, but cold-wire anemometry may give the frequency response needed. However, cold-wires are impractical due to their fragil-

_______________________________________ © The author(s) and/or their employer(s), 2012

405

ity, see e.g. Ref. (4). The highly inhomogeneous flow field also necessitates several measurement points across the flow in order to obtain a good estimate of the averaged mass flow rate. In a recent paper by Laurantzon et al. (5), time-resolved bulk flow measurements under pulsatile conditions were obtained via timefrequency analysis using wavelet analysis of the signal from a vortex flow meter. The advantage of this method is that neither velocity calibration, nor temperature compensation are needed, thereby constituting an optimal technique for the purpose of the present work. Furthermore, as shown in Ref. (5) the utilization of this technique directly provides an estimate of the time-resolved bulk velocity from a single point measurement thereby qualifying the technique in particular for the inhomogeneous and non-isothermal flow upstream and downstream of a turbine. For a long time many authors have debated the use of steady state turbocharger maps, i.e. maps obtained without inflow pulsations, since this would imply that the flow through the turbine is quasi-steady. However, the unsteady flow through the turbine due to pulsations is close to quasi-steady for low pulse frequencies, e.g. Ref. (6), and the rotation rate of the impeller of the turbine is fairly constant during the pulsating cycle due to high angular momentum of the turbine-compressor combination. However the filling and emptying of the volute is highly unsteady giving rise to a failure of the quasi-steady model, see for instance Ref. (7). Hence, the turbine overall characteristics will be dominated by the transient behavior of the gas in the volute. This complex problem remains a challenge, especially for engine simulations, but also for turbine design. Within the CICERO laboratory at KTH CCGEx, a pulsatile flow rig has been developed and has recently been used to investigate the effect of pulsations on the flow field upstream and downstream a turbine and thereby assess the turbine performance under pulsatile conditions, Ref. (8). In the present paper, the vortex flow meter has been utilized in order to obtain turbocharger performance maps under pulsatile conditions. In particular the effects of mass flow rate and pulsation frequency have been investigated. Results indicate that the pulse shape of the mass flow rate signal to a high degree is preserved, albeit damped, while passing through the turbine. Furthermore, the expected time-dependent mass storage in the turbine housing is confirmed and analyzed. The paper is organized as follows. In Section 2 a description of the flow rig design including the utilized instrumentation is given, while Section 3 outlines the employed time-resolved mass flow rate measurement technique by means of wavelet analysis. Section 4, presents the results in terms of pulsation frequency and flow rate, whereas conclusions and plans for future work are given in Section 5.

2 EXPERIMENTAL SET-UP The flow rig is supplied with dried compressed air through a pipe system from compressors and tanks stored away from the laboratory. The mass flow rate is accurately measured by a hot film flow meter connected to the laboratory inlet line. In the laboratory, upstream the flow rig, the air passes an electric heater (2 in Figure 1) increasing the gas temperature to avoid it from dropping below the dew point downstream the turbine (5 in Figure 1). To obtain pulsating flow a valve that can be rotated to give a time varying open area (3 in Figure 1) is connected to the line downstream the heater. The valve consists of a ball, planed on two opposite sides along the axis of rotation, and tightly fitted into a circular pipe. A frequency controlled AC motor rotates the ball causing the valve to open twice per revolution (see detail in Figure 1). To be able to vary the pulse modulation degree of the turbine inlet flow, the ball valve is connected in parallel with a single pipe with a con-

406

trol valve (4 in Figure 1). The turbine is connected through a 1 m long straight pipe to the pulse generator.

Figure 1. Schematic representation of the CICERO flow rig system and its instrumentation. 1. Inlet regulator valve, 2. Electric heater, 3. Pulse generator, 4. By-pass line, 5. Turbine, 6. Compressor. To resolve the large temporal flow variations the sensors mounted upstream and downstream of the turbine have a high frequency response, a feature not necessary in the compressor line where the flow is almost steady. The pressures in the turbine line were measured with piezo-resistive pressure transducers, whereas the pressure transducers in the compressor line were of differential membrane type. The time varying recovery temperatures were measured with cold-wires (made of tungsten with a diameter of 5 micron) connected to a Dantec Streamline anemometer. For the flow speeds used here the recovery temperature is close to the stagnation temperature. Two sensors were used, one positioned upstream and one downstream the turbine. The calibration of the cold-wires was done in-situ at low speed by means of thermocouples. On the compressor side thermocouples were used to measure the temperature. The unsteady inlet and outlet turbine flow velocities were obtained by vortex shedding flow meters, which will be described in some detail in Section 3. The flow upstream of the turbine is fairly uniform and without swirl, whereas the swirl on the downstream side is substantial and has to be suppressed through a honeycomb section, in order to obtain a reliable output from the vortex meter. The sampling of the various transducer signals was done phase-locked with respect to the angular position of the ball valve. Typical sampling time was 20 seconds for all measured quantities with a sampling frequency of 15.6 kHz.

407

3 TIME RESOLVED MEASUREMENTS WITH A VORTEX SHEDDING FLOW METER Vortex shedding flow meters are based on the detection of vortices shed periodically from a body. Here a circular cylinder located along a diameter of the pipe was used as the shedding element. The circular cylinder generates a flow pattern known as the von Kármán vortex street and experimental evidences (9,10) show that the shedding frequency scales with the flow velocity and the cylinder diameter. The non-dimensional frequency St=fd/U (also called Strouhal number) is approximately constant (≈0.20), although there is a small Reynolds number dependence. The shedding frequency f has been shown to be proportional to a velocity close to the bulk velocity U, for a given cylinder with diameter d, see Ref. (5). This measurement technique has been developed in our laboratory in order to obtain time resolved flow measurements in pulsating flows. One condition that has to be fulfilled is that the pulsating frequency fp is sufficiently low relative to the vortex shedding frequency f, so that there is a clear distinction between them. A general rule of thumb to avoid signal ambiguity is to let f/fp>4.4 (Ref. 11,12), see also the discussion in Ref. (5). By choosing a suitable diameter of the cylinder it is usually possible to obtain such a scale separation. The vortex shedding frequency is detected by means of hot-wire anemometry and wavelet analysis described below is used to obtain the instantaneous frequency. There are two major advantages with this technique as compared to direct measurement of velocity with hot-wire anemometry; namely that here the hot-wire does not need to be calibrated since the hot-wire is only used to detect the shedding frequency. Moreover, the measured quantity has been found to be close to the bulk velocity see Ref. (5). In order to get the instantaneous mass flux also the density needs to be determined, which is obtained by using the gas law with the measured time resolved pressure and temperature. A schematic design of the meter is depicted in Figure 2. For uni-directional flow the shedding occurs downstream the cylinder and only hot-wire 2 is used. However, if reversed flow occurs, the upstream hot-wire will detect the flow velocity variation during the flow reversal. This design is thus suitable for pulsating flow, where quite large back flow rates may occur.

Vortex street

Hot-wire 1 Normal flow direction

Hot-wire 2 Figure 2. Schematic of a pipe section with the vortex shedding flow meter. If the bulk velocity U(t), is time dependent, the vortex shedding frequency is frequency modulated by the variation in flow rate. The instantaneous frequency cannot be obtained with classical Fourier analysis and instead the wavelet technique has been used as a straightforward way to find the time varying frequency. A continuous wavelet transform of a function u(t) is defined through a convolution with a wavelet function ψ(t) ∞ ⎡ 1 ∗ ⎛ t − τ ⎞⎤ % τ , a) = ∫ ⎢u(t) u( ψ ⎜ ⎟⎥dt ⎣ a ⎝ a ⎠⎦ −∞

408

(1)

In a sense, the wavelet transform can be viewed as the correlation of the function u(t) with the complex conjugate of the wavelet function ψ(t) shifted and dilated in time through the parameters a and τ. Thus, the stronger the resemblance of the function u(t) with ψ[(t-τ)/a], the larger the magnitude of the wavelet coefficient. The time shift τ defines the window middle-point, while the scale a defines the width of the observed signal and is related to the instantaneous frequency f(t) through the relationship f(t)a=K where K is a proportionality constant which depends on the chosen wavelet. More details about the wavelet transform properties can be found in specialized textbooks such as Ref. (13). The analysis can be summarized by means of Figure 3, see also Ref. (5). This case shows pulsating flow at a pulse frequency of fp=80 Hz. In Figure 3(a) the time signal for the anemometer output voltage for two periods is shown. The pre-multiplied power spectral density (PSD) can be seen in Figure 3(b); showing the large energy content at the pulse frequency fp, and some at its harmonics. Another aggregation of energy starts at about 50fp, which is related to the vortex shedding of the cylinder. Since the PSD does not give more information than just the frequency range where most of the energy is concentrated, other evaluation methods have to be employed. In Figure 3(c) the so-called wavelet spectrogram for the signal is illustrated. From this the time varying frequency of the vortex shedding can be extracted to give f(t) and hence U(t) as shown in Figure 3(d). The example shown below is just for two periods, but if the entire time series is considered, a phase average can be obtained from the individual periods, as in Figure 3(d). Noteworthy is that for this test case there was no backflow and the lowest velocity corresponded to a shedding frequency at approximately 25fp, hence there was no ambiguity between the pulse and shedding frequencies.

Figure 3. Evaluation of the signal from the vortex shedding flow meter for two pulse periods. (a) Time signal. (b) PSD of the time signal. (c) Wavelet spectrogram. (d) The extracted time resolved frequency and corresponding velocity.

409

4 RESULTS In this section, results from both steady and pulsating turbine measurements will be demonstrated. Specifically the effect of phase shifting the signals will be shown. Typically the data is phase averaged for 400 pulse cycles. Furthermore, all phase averaged data will be presented with two pulses i.e. corresponding to a full revolution of the pulse generator. There is a certain distance between the locations of the upstream and downstream sensors, and this gives rise to a time lag between the phases for the respective quantities under pulsating flow. This time lag should equal the time it takes for the disturbance pulse to travel between the two stations i.e. (2) where Δx is the path length between the stations, u is the average velocity along 1/2 the path and a the speed of sound which has been estimated as a≈20T (where T is the mean temperature in Kelvin). However, there exist several problems involved in this method to find the time lag τ0. Firstly, the distance Δx cannot be unequivocally defined through the turbine without more detailed measurements of the path lines in the space between the two sensors, secondly both u and a will vary in both space and time. Hence, instead we determined τ0 from the peak in the cross correlation between the same quantities, e.g. between the static pressure upstream and downstream the turbine

φ (τ ) = lim

1 T→∞ T



T /2

−T /2

p1 (t)p2 (τ + t)dt

(3)

It should however be emphasized that it is important to do the correlation between the same quantities, since for instance velocity and pressure in one point in a flow field are in general out of phase in unsteady flow. Despite the uncertainty related to employing Eq. (2), it can be used as a check for the order of magnitude of the cross correlation. For the tested flow cases, the local velocity was small compared to the speed of sound (roughly 0.15a at most for this study), moreover the path travelled in the turbine was a small fraction compared to the rest of the piping system, between the sensors. An approximate delay can be obtained by using the time average of u and a, and by approximating the axial distance travelled through the turbine with some characteristic length scale of the turbine e.g. the distance from the inlet to the outlet. The time lag obtained from the cross correlation, Eq. (3) is approximately 4.5 ms for all cases reported, which corresponds fairly well to the time lag estimated with Eq. (2). Note that this time lag will vary depending on the flow case, i.e. the pulsating frequency and the mean flow rate, and is to a high degree governed by the speed of sound a. But since the temperature variations were moderate, the time lag did not change considerably from case to case. In addition to this phase shift an adjustment for the different sensor locations upstream the turbine was done. The time lag between two instruments located downstream the turbine was negligible and all the upstream measured quantities were related to the same reference plane, namely the location for the upstream stagnation pressure sensor. 4.1 Effects of pulsation frequency



In Figure 4, the mass flow downstream the turbine m 2 , is shown vs. the mass flow •

upstream m 1 . Both the time dependent variation during the pulse cycle, which shows a closed loop, and the mean values (shown as a filled circle in Figure 4) are

410

shown. In Figure 4(a), the directly measured values are shown, i.e. without any phase shifting. The values of the mean mass flow rates at the upstream and downstream sides are similar and also close to the value of the system mass flow meter (105 g/s) which gives confidence in the flow measuring technique. One can also observe that the pulsation amplitude is heavily damped downstream the turbine. The phase shifted flow rates are plotted in Figure 4(b); and for this case the relation between the flow rates is close to a straight line. To further estimate the similarity between any two upstream and downstream quantities, the correlation coefficient can be used. The correlation coefficient between two fluctuating signals u1 and u2, is defined as

ru1 u2 =

Σu1i' u '2i

Σ(u1i' ) 2 Σ(u '2i ) 2

(4)

where prime denotes the fluctuating part of the signal. The correlation coefficient •



between m 1 and m 2 , was for all cases greater than 0.95, where a coefficient of 1 in Eq. (4) means perfect correlation. Hence, although damped with more than a factor of two, the pulse shape was to a large degree preserved. 0.25

0.25

f p = 40 Hz f p = 60 Hz f p = 80 Hz

(a)

0.2

m ˙ 2(t − τ ) [kg/s]

m ˙ 2(t) [kg/s]

0.2

0.15

0.1

0.05

0 0

f p = 40 Hz f p = 60 Hz f p = 80 Hz

(b)

0.15

0.1

0.05

0.05

0.1

0.15

m ˙ 1(t) [kg/s]

0.2

0.25

0 0

0.05

0.1

0.15

0.2

0.25

m ˙ 1(t) [kg/s]

Figure 4. Relation between outlet and inlet mass flow. (a) Directly measured values. (b) Outlet mass flow phase shifted to correlate with inlet mass flow. Mean mass flow rate 105 g/s, at fp 40 Hz (solid), 60 Hz (dashed) and 80 Hz (dotted) pulse frequency. The circles show mean values and arrows indicate loop path. Figure 5 shows the turbine map for the three different pulse frequencies, where the upper row is the unaltered values and the lower row shows the phase shifted ones. As a comparison four points from steady flow measurements are plotted together and a polynomial of second degree is applied for visual aid. Here we see that the steady points are close to the mean as obtained from the pulsating flow case for all frequencies. As pointed out in (2), the quasi-steady assumption is usually made in the zero and simple one-dimensional models. However as shown in Figure 5, the area spanned during the pulse cycle, sometimes called the hysteresis area, is just slightly decreased when phase shifting is applied, which indicates that the quasisteady assumption is invalid for the flow through the turbine. Further, as can be seen in Figure 5, the loop path upstream the turbine is clockwise, whereas it is counter clockwise downstream the turbine. The reason for this is illustrated by means of Figure 6, which shows the flow case in Figure 5(a). Here, the acceleration and retardation phases are marked for a given expansion ratio. If one considers the upstream flow rate, it is clear that it is higher at the acceleration phase as com-

411

pared to the retardation phase. Thus, the hysteresis path must be clockwise for the upstream mass flow and vice versa applies for the downstream mass flow. 0.25

f p = 40 Hz

m ˙ [kg/s]

(a)

f p = 60 Hz

(b)

0.25

(c)

0.2

0.2

0.2

0.15

0.15

0.15

0.1

0.1

0.1

0.05

0.05

0.05

0

1.4

1.6

0.25

1.8

0

2

f p = 40 Hz

(d) m ˙ [kg/s]

0.25

1.4

1.6

0.25

1.8

f p = 60 Hz

(e)

0

2

0.2

0.15

0.15

0.15

0.1

0.1

0.1

0.05

0.05

0.05

1.6 p 01/p 2

1.8

0

2

1.4

1.6 p 01/p 2

1.8

1.8

2

(f )

0.2

1.4

1.6

0.25

0.2

0

f p = 80 Hz 1.4

0

2

f p = 80 Hz 1.4

1.6 p 01/p 2

1.8

2

Figure 5. Turbine map at a mean mass flow rate 105 g/s, at different pulse frequencies. The upper row plots are instantaneous, the lower row plots are phase shifted. Upstream flow (solid line) and downstream flow (dashed line). Black dashed line is obtained from steady flow measurements. Full and open circles show mean value for upstream and downstream flow respectively. Arrows indicate loop path.

m ˙ 1 [kg/s]

0.3 0.2 0.1 0

m 2 [kg/s]

0.15 0.1 0.05 0

p 01/p 2

2 1.8 1.6 1.4 0

0.1

0.2

0.3

0.4

0.5 t/T

0.6

0.7

0.8

0.9

1

Figure 6. Phase averaged data for one period (flow case from Figure 5a.). Mass flow rate upstream turbine (upper), mass flow rate downstream turbine (middle) and expansion ratio (lower). The two vertical lines are for visual aid and shows acceleration and retardation phase respectively at the same expansion ratio.

412

4.2 Effects of flow rate Figures 7 and 8 shows the same quantities as in Figures 4 and 5 respectively, although here the pulse frequency is instead held constant whereas four different flow rates are shown. In Figure 7(a), the relation between the measured mass flow rates upstream and downstream the turbine are shown and as expected the mass flow variation during the pulse cycle increases for increasing flow rate. In Figure 7(b) phase shifting is applied for maximum correlation between the upstream and downstream side and now it is clear that the in and out flow follow each other, although the outflow amplitude is damped with about a factor of two. In Figure 7(c,d) the flow rates are normalized with the average flow rate and it is clearly seen that in this format the curves more or less overlap, hence the behaviour is independent of the flow rate in this flow range. Finally, in Figure 8, the turbo map for these different flow rates are plotted together. In Figure 8(a) the maps are plotted without any time shift whereas in Figure 8(b) all data are shifted to correlate with p01. There is no large difference between the maps on the upstream side between Figure 8(a) and (b), since the upstream mass flow rate and p01 are measured almost at the same position, whereas the map using the downstream mass flow changes significantly. The steady response is plotted as the dotted line and can be compared with the mean values obtained from the pulsating flow given as symbols. As can be seen the steady points agree well with the mean of the pulsating flow. In Figures 8(c) and (d) the same data are plotted but normalized with the mean pressure ratios and mass flow rates, respectively, showing that the path followed in the turbine map is fairly independent of the flow rate. 0.25

(b) m ˙ 2(t − τ ) [kg/s]

0.2 m ˙ 2(t) [kg/s]

0.25

55 g/s 80 g/s 105 g/s 130 g/s

(a)

0.15 0.1 0.05 0 0

0.2 0.15 0.1 0.05

0.05

0.1 0.15 m ˙ 1(t) [kg/s]

0.2

0 0

0.25

2

0.2

0.25

(d) 1.5  m ˙ 2(t − τ )

1.5  m ˙ 2(t)

0.1 0.15 m ˙ 1(t) [kg/s]

2

(c)

1

0.5

0 0

0.05

1

0.5

0.5

1  m ˙ 1(t)

1.5

2

0 0

0.5

1  m ˙ 1(t)

1.5

2

Figure 7. Different flow rates at 60 Hz pulse frequency. The average mass flow rates are 55 g/s (solid thin), 80 g/s (dotted), 105 g/s (dashed) and 130 g/s (solid thick). (a) Instantaneous. (b) Shifted. (c) Instantaneous normalized with time average. (d) Shifted normalized with time average.

413

0.25

0.25

(a)

(b) 0.2

m ˙ [kg/s]

m ˙ [kg/s]

0.2 0.15 0.1 55 g/s 80 g/s 105 g/s 130 g/s

0.05 0 1

1.5

2

0.15 0.1 0.05 0 1

2.5

1.5

2

(d) 1.5

˙ m

1.5

˙ m

2.5

2

(c)

1

0.5

0 0.8

2 p 01/p 2

p 01/p 2

1

0.5

0.9

1  /p 2 p 01

1.1

1.2

0 0.8

0.9

1  /p 2 p 01

1.1

1.2

Figure 8. Turbine map at mean mass flow rate 55, 80, 105 and 130 g/s (increasing line thickness), at 60 Hz pulse frequency. Solid line is upstream flow, dashed line is downstream flow. The mean is given as symbols, filled circles are obtained from the upstream mass flow measurements and open circles from the downstream. Black dashed line is turbine map from steady measurements. (a) Instantaneous. (b) Phase shifted, (c) and (d) same as (a) and (b) but normalized with mean values.

5 CONCLUSIONS AND FUTURE WORK In the present work, time resolved measurements of pressure, temperature and velocity were obtained both upstream and downstream a turbine of a turbocharger. A newly developed mass flow meter based on wavelet analysis of the vortex shedding behind a cylinder is shown to be able to accurately describe the pulsating mass flow rate. In this work it is demonstrated that the pulse shape to a high degree is preserved after passing the turbine. Although the characteristic looping curves dominate in the pulsating flow case, the time averaged value agrees fairly well with the corresponding steady flow points. The mass flow rate and expansion ratio were phase shifted in order to test the quasi-steadiness of the flow through the turbine, however even for the shifted values the hysteresis plots in the turbine map is not a single valued function, as would be expected if the flow is quasi-steady. Thus, the study confirms that the time dependent mass storage in the turbine case makes the quasi-steady assumption invalid, at least for the turbine as a whole. The present data form a unique experimental database of the flow through a turbine that can be used for checking present and future turbine models.

414

ACKNOWLEDGEMENT This work has been carried out within KTH CCGEx. Kim Karlström and Göran Rådberg are acknowledged for their skilful work in setting up the flow rig.

REFERENCE LIST 1.

2. 3. 4.

5. 6. 7. 8. 9. 10. 11. 12. 13.

Capobianco, M. Gambarotta, A. Cipolla, G. 1989 Influence of the pulsating flow operation on the turbine characteristics of a small internal combustion engine turbocharger. Proc. 2nd Int. Conf. Turbochargers and Turbocharging, 4-5 April 1989, London, IMechE, pp. 63-70. Baines, N.C. 2010 Turbocharger turbine pulse flow performance and modelling – 25 years on. In 9th Int. Conf. on Turbochargers and Turbocharging, 19-20 May 2010, London, ImechE, pp. 347-362. Laurantzon, F. 2010 Flow measuring techniques in steady and pulsating compressible flows. Licentiate thesis, KTH Mechanics, TRITA-MEK 2010:09, ISBN 978-91-7415-824-3. Marelli, S. & Capobianco, M. Turbocharger turbine performance under steady and unsteady flow: test bed analysis and correlation criteria, In 9th Int. Conf. on Turbochargers and Turbocharging, 19-20 May 2010, London, ImechE, pp. 373-385. Laurantzon, F., Örlü, R., Segalini, A. & Alfredsson, P.H. 2010 Time-resolved measurements with a vortex flowmeter in a pulsating turbulent flow using wavelet analysis. Meas. Sci. Technol. 21, 123001. Costall, A., Szymko, S. & Martinez-Botas, R.F. 2006 Assesment of unsteady behaviour in turbocharger turbines. ASME Turbo Expo 2006: Power for Land, Sea and Air. Aymanns, R., Scharf, J., Uhlmann, T. & Lückmann, D. 2011 A revision of quasi steady modelling of turbo-charger turbines in the simulation of pulse charged engines. 16th Supercharging Conf. 29-30th Sept. 2011, Dresden. Laurantzon, F., Tillmark, N. & Alfredsson, P.H. 2010 A pulsating flow rig for analyzing turbocharger performance. In Proc. 9th Int. Conf. on Turbochargers and Turbocharging, 19-20 May 2010, London, ImechE, pp. 363-372. Norberg, C. 1994 An experimental investigation of the flow around a circular cylinder: influence of aspect ratio. J. Fluid Mech. vol. 258, pp. 287-316. Fey, U. Konig, M. & Eckelmann, H. 1998 A new Strouhal-Reynolds-number relationship for the circular cylinder in the range 47