Simulation of manufacturing sequences for ...

Simulation of manufacturing sequences for verification of product properties

MATS WERKE

THESIS FOR THE DEGREE OF LICENTIATE OF ENGINEERING STOCKHOLM, SWEDEN 2009

KTH – Dept of Production Engineering SE-100 44 Stockholm SWEDEN TRITA IIP 09 – 02 ISSN 1650-1888 © Mats Werke, March 2009

Printed by Universitetsservice US-AB

Preface The manufacturing sequence may to a large extent influence properties like residual stress and hardness and, as a consequence, the fatigue life and shape accuracy of a component. Numerical simulation tools based on the Finite Element Method (FEM) are today established in a majority of engineering departments for analysis of various design concepts. However, the conceptual analysis does not include the properties induced by the manufacturing processes. By simulating the manufacturing sequence and extracting important accumulated data, like residual stress, hardness and shape, the possibilities of early analysis of a design concept of a component and optimisation of the complete manufacturing sequence may increase. An established methodology has the potential to reduce physical testing, increase the process knowledge and reduce product development time and costs. This thesis suggests principles for simulating manufacturing sequences. The reference knowledge is based on the following case studies: •

A forged steering knuckle including billet manufacturing, pre forming, final forming, trimming, cooling, blasting, heat treatment, turning and deep rolling

•

A forged steering arm including hot shearing, the roughing blow, the finishing blow, trimming, hot calibration, controlled cooling and blasting

•

A forged pinion including roughing blow, finishing blow, controlled cooling, isothermal annealing, turning, case hardening and straightening

•

Two sheet metal-formed components including sheet metal forming, welding and heat treatment.

It is possible to establish virtual manufacturing sequences and connect different commercial simulation softwares into a chain with support from methods for data communication and in process modelling. However, the data communication has pitfalls and also material data, process data and material models for single processes may be insufficient in order to conduct a quantitative analysis of the accumulated properties. Thus, a thorough validation of the sequential simulation results should be performed. Further, engineering simplifications of the sequence are recommended and e.g. a combination of numerical and empirical methods may be preferable in order to predict accumulated properties with high accuracy. Future development concerning standards for data communication as well as meshing, mapping and modelling technique is recommended in order to improve the quality of the accumulated simulation results. Finally, methods for integration of sequential simulation in the overall component design process, including conventional Computer Aided Design (CAD) and Finite Element Analysis (FEA), should be developed. The main advantage of sequential simulation may be conceptual studies of process and material parameter variations and their influence on the final product properties.

Acknowledgements This work was accomplished with support from a number of people. First, I would like to address my gratitude to my supervisor Professor Torsten Kjellberg for his guidance and support throughout this work. I also wish to address my gratitude towards my co-supervisor Professor Mihai Nicolescu for discussions and consultations. Further, I would like to thank all the researchers, who have contributed to this work by means of discussions, co-authoring of papers and reports and active work in the case studies. I especially would like to thank Hans Kristoffersen, Swerea IVF, for simulation of case hardening and straightening in the pinion case, Peter Ottosson, Swerea IVF, for simulation of sheet metal forming, Lars Johanson, Swerea IVF for development of the C6t2 utility, Dr Sven Haglund, Swerea KIMAB, for fatigue discussions and Finite Element analyses etc in the steering arm case. Further I would also like to thank Mikael Hedlind, Magnus Lundgren and Dr Astrid von Euler, KTH/IIP, for discussions concerning astrakan models and the ISO 10303 standard. My gratitude also goes to representatives of companies for contribution with time and resources in various research projects by means of simulation work, material data development, residual stress measurements, fatigue testing, co-authoring of papers and reports, discussions and knowledge contribution. Lennie Svensson, Anders Nord and Jonas Andersson (Bharat Forge Kilsta), Dr Bengt Johannesson (Volvo 3P), Krister Johansson and Eric Sandqvist (Scania CV), Torbjörn Kvist and Dr Henrik Tersing (Volvo Aero Corporation), Dr Anna Bellini (Alfa Laval Lund AB), Hossein Ghotbi (Parker Hannifin), Dr Laurent D’Alvise and Dr Olivier Pierard (Cenaero, Belgium). I also would like to thank Dr Sven Hjelm (Scania CV) for his support of this thesis work. The work was made possible by financial aid from The Swedish Governmental Agency for Innovation systems, The Swedish Vehicle Research programme, the MERA programme and the 5th EU framework programme. Finally I would like to thank my family Birgitta, Fredrik and Sofia for supporting me in all thinkable ways.

List of papers The following three papers are included in this thesis: Paper I. - M. Werke, R. Moshfegh, B. Johannesson, J. Svenningstorp, P. Johansson, E. Stinstra, “Statistical methods for extracting and evaluating manufacturing parameters of significant importance to component robustness”. Presented at the 8th QMOD Conference (Quality Management and Organisational Development) in Palermo, Sicily, 2005. Paper II. - M. Werke, H Kristoffersen, S. Haglund, L. Svensson, A. Nord, “Prediciting residual stresses and hardness of a critical component using a combination of numerical and empirical methods”. Presented at the 12th International Conference, Metal forming 2008, Krakow, Poland, 2008 and published in the Steel Research International Journal, special edition 2008. Paper III. - M. Werke, “Methods and models for shot peening simulation”, Presented at the Swedish Production Symposium, Stockholm, Sweden, 2008. The articles have been reformatted to fit the layout of the thesis.

Contents 1 1.1 1.2

Introduction and research area Introduction Research area

5 5 6

2 2.1 2.2 2.3 2.4

Reference knowledge Steering knuckle Steering arm Pinion Sheet metal-formed components

8 8 9 12 19

3 3.1 3.2 3.3 3.4

Research results Establishment of a sequence In process modelling Data communication Validation of the sequence

26 26 27 29 31

4

Conclusions

32

5

Future research

33

6

References

34

7

Web-links

35

Paper I

Statistical methods for extracting and evaluating manufacturing parameters of significant importance to component robustness

36

Paper II

Predicting residual stresses and hardness of a critical component using a combination of numerical and empirical methods

49

Paper III Metods and models for shot peening simulation

62

1 INTRODUCTION AND RESEARCH AREA

1

Introduction and research area

1.1

Introduction

The main trend in the manufacturing industry sector is the strong demand for shortening of product development time and reducing of costs. Virtual methods invoking CAD models and tools for finite element analysis of component strength and stiffness provide well established support for efficient product development and are compulsory in a majority of the mechanical design offices when designing and optimising components. It is a well-known fact that the manufacturing processes have a large influence on product properties like fatigue strength, shape accuracy and surface quality. In fact, according to a truck company, 47 % of failures due to fatigue depend on the manufacturing of the components. The influence of product properties induced by the manufacturing processes is not yet included in the virtual product development methodology and the design analysis often starts with virgin material data. This means that the engineering development includes several loops of prototyping and physical testing in order to ensure that the demanded properties are reached. This uncertainty also results in a conservative dimensioning philosophy, unnecessarily heavy and overdimensioned components and thereby unnecessarily high safety factors. A virtual design philosophy invoking the effects on component properties from the manufacturing sequence may have good potential to increase efficiency in the product development process and increase the quality of the component and manufacturing processes. Properties like residual stress, hardness and shape accuracy are often not solely results from a single process but may be accumulated through several manufacturing steps. Thus, the complete manufacturing sequence has to be taken into consideration when predicting the final properties. The establishment of a well-functioning methodology may be a foundation for: •

Reducing time in the conceptual product development process

•

Personnel, tool, material, testing, and energy cost reduction as well as environmental impact reduction

•

Evaluation and optimisation of manufacturing processes by virtually tuning process parameters

•

Improvement of the possibilities to cooperate between the design office, material laboratory, physical test laboratory and process planning office as well as between the Original Equipment Manufacturer (OEM) and subcontractor

•

Enhancement of knowledge concerning processes and their influence on product properties. 5


The research questions are defined in the “Research area” Section. The reference knowledge development is based on evaluation of five test cases and is discussed in the “Reference knowledge” Section. The experiences from the test cases are outlined in the “Research result” Section. The conclusions are summarized in the “Conclusion” Section and identified subjects for future research are discussed in the “Future research” Section.

1.2

Research area

The goal is to evaluate possibilities for simulation of manufacturing sequences in order to predict the final product properties induced from the manufacturing processes. Typical output properties (see Figure 1) that are discussed in this work are hardness, residual stresses, the component shape accuracy and fatigue strength. The vision is to establish a seamless chain of “In Process Models” (IPM:s) of separate core processes according to Figure 1 and include the output results from the manufacturing sequence in the overall virtual product development methodology.

Figure 1

Connection of single process simulation to a virtual manufacturing sequence.

The state of the art concerning simulation of manufacturing sequences is still limited but Åström [1] discusses possibilities for welding, heat treatment and cutting. Hyun et al [2] discuss possibilities to simulate a chain of processes using a finite element code with adaptive meshing. Alberg [3] discusses modelling and validation of welding and heat treatment in a sequence. Altinas et al [4] give an overview of modelling possibilities of the machining process and emphasise the necessity of sequential simulation.

6


The following research questions will be analysed: “Establishment of the sequence” Sequential simulation is a complex task and it is difficult to use just one information model that “travels” through the chain of core processes. It is therefore important to find methods in order to simplify the simulation sequence. “In process modelling”. In order to realise a virtual sequence it is necessary to connect single core process models into a framework of models. The single core process models may have different requirements concerning geometry, mesh topology and element types. An important task is to investigate possibilities for adaption of single process models into a chain of models. “Data communication”. In order to establish a seamless framework of models through the sequence it may be necessary to communicate data between various software codes. The reason for this may be the fact that the performance of various codes may be specialised on the simulation of different single core processes. An important research question is therefore to investigate the possibilities to convert the native output data from one code into native input data to the proceeding code and also investigate how to standardise the dataflow through the sequence. “Validation of the sequence”. In order to realise sequential simulation it is necessary to obtain confidence in the results. Especially the knowledge concerning the effects from variations in material data, process data and model quality etc on the accumulated result is essential. An important research question is to investigate the requirements of a validation procedure for sequential simulation.

7

2 REFERENCE KNOWLEDGE

2

Reference knowledge

The reference knowledge was developed in conjunction with R&D work with five practical test cases. The case studies are described in Sections 2.1 – 2.4. Due to confidentiality the described material and process data is limited in the case descriptions.

2.1

Steering knuckle

Project and component: The steering knuckle case, see Figure 2, was developed in the project “IMS-EURobust” with participation from Japan, USA and Europe. The European commission’s 5th framework programme financed the project and the main objective was to develop world-class robust design tools and techniques for European organisations. The project duration was 2002 – 2005 and Swerea IVF managed the project. The steering knuckle was studied in one of the work packages in close cooperation between Swerea IVF, Chalmers, AB Volvo and CQM in Holland. The division of quality science at Chalmers developed the VMEA method and Swerea IVF developed the Finite Element (FE) model for deep rolling described in this case. CQM contributed with methodology for robust simulation. The tough hardened steering knuckle is an important component in the steering mechanism of a truck.

Figure 2

Steering knuckle.

Scope and manufacturing sequence: The aim was to study the robustness in the residual stress profile for the most critical process parameters in the manufacturing sequence. The main steps in the manufacturing sequence are described in Figure 3.

Figure 3

Manufacturing sequence.

Methodology: The methodology is a combination of engineering knowledge, statistical methods and simulation and is described in detail in paper [I]. With the 8


aid of Failure Mode and Effect Analysis (FMEA) the most important single process with strongest influence of component durability was selected. A new methodology Variation Mode and Effect Analysis (VMEA) was then used to extract the most important process parameters from the selected process. Also methodology for simulation of parameter variations according to Design and Analysis of Computer based Experiments (DACE) was evaluated and added to the overall methodology. The steering knuckle was exposed to deep rolling in a critical fillet radius. The deep rolling process was selected from the sequence and a nonlinear finite element model for simulation of the deep rolling process was developed. Material and process parameters: Deep rolling process parameters like the spindle velocity, feed rate and hydraulic pressure were included. Further material parameters, like Young’s modulus, yield strength and tensile strength, and the surface geometrical surface roughness variations after turning were included in the model. Validation: The deep rolling FE model was validated with residual stress measurements with the X-Ray diffraction method. Comments and remarks: •

The methodology is generic and may be applied to arbitrary manufacturing sequences

•

The accuracy concerning the surface residual stresses could be improved and further development of the deep rolling model is recommended.

2.2

Steering arm

Project and component: The steering arm case, see Figure 4, was developed in the project “Simulation of product properties for efficient product development” (ProSim). The Swedish Governmental Agency for Innovation - Vinnova financed the project together with participating companies. The duration time was 2005 – 2008 and Scania, Volvo 3P, Bharat Forge, Alfa Laval Lund, Volvo Aero, Volvo Powertrain, Parker Hannifin and Ovako Steel participated together with Swerea KIMAB and Swerea IVF. Swerea IVF was responsible for the project. A focused study concerning residual stresses after blasting for the steering arm case was performed in the project “Simulation of product properties for forged components”. The Swedish Vehicle Research Programme (PFF) financed this project together with Bharat Forge Kilsta (project manager). The duration time was 2006 – 2008 and Bharat Forge Kilsta AB and Swerea IVF participated in the project. The micro alloyed steering arm is an essential component in the steering mechanism of a truck.

9


Figure 4

Forged steering arm.

Scope and manufacturing sequence: The aim was to develop a method for prediction of fatigue life of a forged steering arm that included the accumulated residual stresses and hardness from the manufacturing sequence. The main steps in the manufacturing sequence are described in Figure 5.

Figure 5


Methodology: Based on a literature study described in [III] it was found that the existing numerical models for shot peening did not manage to describe the complex blasting process where the parts are tumbling against each other at the same time as they are exposed to blasting. An approach with empirical models of blasting was therefore tested. Thus, the sequence was analysed using a combination of numerical and empirical methods. The forging steps were simulated in a sequence using one code, Forge2005, see [5]. Also the controlled cooling was simulated in a sequence with forging using the Forge2005 quenching module. The fatigue life analysis was done using a Finite Element Analysis (FEA) based on a CAD-model of the design concept of the arm. The methodology is described in detail in paper [II]. Material and process parameters: Process parameters for forging (press kinematics, billet and tool temperatures), cooling (cooling time) and blasting (blasting time, number of items in the machine) were included in the sequence analysis. Further material parameters for forging and the thermo metallurgical phase changes during cooling were included. Validation: The predicted fatigue life was validated with rig tests with constant amplitude loads. Comments and remarks: The experiences through the sequence are summarised below and in Figure 6: •

The material segregation during forging was not included in the simulation models. Further the decarburisation effect, mainly in the forging, and cooling operations, that affect the residual stresses and hardness at the surface layers, was not included in the forging and cooling models. Also 10


the mesh topology after the forging and controlled cooling simulation was too coarse for analysing the surface residual stresses and hardness •

The lack of accurate material data may have influenced the accuracy of the results after controlled cooling

•

The initial attempt using a parametric model of the blasting was replaced by measurements of stress ranges. Further development of a parametric model that includes the surface decarburisation effect is recommended.

Figure 6

Methods and tools for the steering arm sequence.

11


2.3

Pinion

Project: The pinion case, see Figure 7, was developed with support from the ProSim project and the project “Model-based component manufacturing” (ModArt). The main objective with ModArt was to develop tools and methods based on information models and IT-support for efficient production development. ModArt was financed by the Swedish MERA programme and managed by KTH/IIP. The duration time was 2006 – 2008. Scania CV, Sandviken Coromant and other companies participated together with KTH/IIP, Swerea IVF and Swerea KIMAB. Bharat Forge simulated the forging, cooling and isothermal annealing processes and Swerea IVF simulated the turning, case hardening and straightening processes. The case hardened pinion is a powertrain component in the final drive system of a truck.

Figure 7

Crown wheel (left) and pinion (right).

Scope and manufacturing sequence: The main objective was to simulate the manufacturing sequence in order to analyse the final stresses and strains of the pinion. The main steps in the manufacturing sequence are described in Figure 8.

Figure 8


After the forging and cooling processes the part is exposed to isothermal annealing, according to Figure 9, in order to improve the machinability.

Figure 9

Pinion isothermal annealing cycle.

After the isothermal annealing the part is exposed to turning and gear hobbing. The gear hobbing was not included in the simulation sequence. After the turning 12


process the pinion is subjected to case hardening. A carrier is loaded with four pinions according to Figure 10 and travels through a “pusher furnace” with heating, carburisation and diffusion zones with different temperature and atmosphere conditions. After the pusher furnace the carrier with pinions is quenched in oil or a couple of minutes. The quenching is performed for 1000 sec and starts at 870 °C where the pinion is 100 % austenitised.

Figure 10 Carrier for the case hardening process. The case hardening may cause shaft misalignment and some parts are therefore exposed to manual cold straightening according to a set-up, shown in Figure 11.

Figure 11 The straightening equipment The straightening force depends on the axial pitch tolerance that is measured by a dial indicator. Residual stresses will be redistributed during straightening and may influence fatigue behaviour. Methodology: The approach was to simulate the chain of processes with different simulation softwares and transfer residual stresses and strains through the sequence. The forging and cooling was simulated in the same way as described in the steering arm case, see paper [II]. Also the material data for forging, cooling and isothermal annealing were retrieved in the same way as for the steering arm but for a case hardening material. The cooling and isothermal annealing simulation were performed in sequence after the forging simulation using the Forge2005 quenching module. The forging output data, like mesh and temperature distribution, was used as input to the cooling step. Only the output mesh from the cooling was used as input to the isothermal annealing step. The material transformations during heating were not possible to simulate in the Forge2005 13


quenching module and were therefore disregarded. Thus, the isothermal annealing started with step 3 in Figure 9 and with a fully austenitic material. Figure 12 shows temperature distribution of the billet before forging, residual stress after cooling and the bainite distribution after isothermal annealing.

Figure 12 Forging, cooling and isothermal annealing simulations. The output data (mesh, stress, strains, deformations, hardness, metallurgical phase composition) from the annealing was stored in an ASCII file (unv-file format, see Section 3.3). Only the output mesh and residual stress field were used as input to the proceeding turning step. The turning modelling was restricted to a study of how the residual stresses before turning, in combination with the turning operation itself, influence the global shape and global stresses. Factors like the local effects due to the cutting tool, the clamping forces and dynamic effects from the machine were disregarded. The global turning simulation (see Section 3.1) was performed with the Finite Element code Morfeo [6] and the simulation utilised the Level Set Method [7] in order to remove elements from the model. The output unv-file from the annealing simulation was translated into a Morfeo-specific input format (Gmsh-file format) with a script in Morfeo. With support from the turning specification two surface meshes for the cutting operations were created using the Gmsh pre-processor. Two global cutting operations were performed in Morfeo. Figure 13 describes the dataflow with mesh and residual stress input from the previous process, cutting surface meshes and output mesh with residual stress field. Two cutting operations, pass 1 and pass 2, were analysed and for each operation a surface mesh was created in order to define the interface for deletion of elements.

14


Figure 13 Turning in Morfeo. Both case hardening and straightening were simulated using the FE software Sysweld. A new FE model of half the pinion was created in Sysweld, in the Visual Mesh pre-processor, with symmetry conditions. The heating process was disregarded and only the quenching process was simulated. The cooling conditions may vary on the carrier and around each pinion and thereby impact the straightness of the shaft. In order to study the influence of different quenching conditions around the periphery of a pinion a calculation was set up where half the pinion model was quenched with a high level of intensity and the other half of the pinion model with a low level of intensity (low heat transfer coefficient setting according to flow rate 0.14 m/s and high heat transfer coefficient setting according to flow rate 0.52 m/s) see Figure 14.

Figure 14 Cooling rate in oil tank and estimated flow rate for the pinions at the carrier. The carburisation effect will also have an impact on strains and residual stress, especially in the pinion surface region. In order to study this a layer method 15


available in the Sysweld package was used. When using the layer method different meshing layers, according to Figure 15, were assigned with material properties according to a varying carbon content. The properties assigned are the ones necessary for a thermo-metallurgical-mechanical calculation and are reflected in the FE-model. The material data was achieved by using both dilatometer measurements and calculations with tools like the software JMatPro. Material data like phase transformations, thermal dilatation and yield strengths was then implemented in the material data file in Sysweld.

Figure 15 Carbon layers in the FE model The typical output data from the case hardening simulation was residual stresses, deformations, hardness and phase distribution. Figure 16 describes the axial deformations after case hardening.

Figure 16 Displacements after case hardening (red=0.3mm, blue=-0.05 mm). The output mesh, stresses and strains from the case hardening simulation were used as input for the straightening simulation. The straightening load was increased from 0 to full load (400 kN) during 10 sec and the unloading was also performed during 10 sec. The material model was elastic-plastic with isotropic hardening. The load was distributed onto several elements. Figure 17 describes the FE model with loads and boundary conditions and calculated plastic strains. 16


Figure 17 FE model and plastic axial strains after unloading (red colour=1 % tension, blue colour= 0.5 % compression). Material and process parameters: The process parameters involved in the study were forging parameters (press kinematics, billet and tool temperatures), cooling (cooling time), isothermal annealing parameters (annealing temperatures and times), turning parameters (cutting paths), case hardening parameters (cooling conditions, temperatures and time) and straightening parameters (straightening force). Further material data for forging and thermo metallurgical data for cooling were included. Also material data that reflected the variations in the surface carbon content during case hardening were included in the case hardening model. Validation: The validation of the results was done using engineering judgement. Comments and remarks: The experiences from the case study are summarised below and in Figure 18: •

In order to get better results from the turning simulation a new FE model with finer mesh density should be developed and data from the previous step should be mapped onto the new mesh

•

It is known by experience that the residual stresses from the turning operation are small and do not have a significant impact on the final shape after the complete sequence. Hence, the case hardening simulation started with a new Finite Element Model and no output data from the previous turning step was mapped onto the new model.

•

In order to improve the output results from the straightening simulation the output mesh from the case hardening should be adapted to straightening conditions and re-meshed in a future analysis

•

Further validation of the stresses and strains after straightening with e.g. X-ray diffraction measurements is recommended.

17


Figure 18 Tools and data flow for the pinion sequence.

18


2.4

Sheet metal-formed components

Project and components: The sheet metal-formed components were evaluated within the frames of the ProSim project. The first component is a sheet metal formed and welded test part of a heat exchanger in stainless steel according to Figue 19. Alfa-Laval Lund simulated the forming and Swerea IVF performed the data transfer. The Technical University of Denmark (DTU), which was subcontracted by Alfa-Laval Lund, performed the welding simulation.

Figure 19 Test part of a heat exchanger in stainless steel. The second test part is a sheet metal formed and welded test part in titanium alloy according to Figure 20. Swerea IVF simulated the forming and performed the data translation. Volvo Aero Corporation simulated the welding and heat treatment.

Figure 20 Test part in titanium alloy. Scope and manufacturing sequences: The main purpose was to study the accumulated residual stresses and strains after forming, welding and heat treatment, see Figure 21.

19


Figure 21 Manufacturing sequence. For the stainless steel part the sheet metal formed plates of stainless steel were welded together along sealing paths. The material thickness of a stainless steel sheets is 0.6 mm. The process setup for the forming process is shown in Figure 22.

Figure 22 Blank, punch, die and final form. After the forming process two details are laser-welded according to Figure 23. The two plates are clamped together during the welding process by a wheel that rolls over the plate in order to create contact before the laser welding (IPG YLR 6000 Fiber laser).

Figure 23 Process setup for welding and two connected test plates. The titanium alloy part is a U-shaped profile with the outer dimensions 86x30x42 mm. The nominal thickness of the initial sheet material in titanium alloy is 2.0 mm. During forming, the punch is moving downward to form the part into the cavity of the die tool. In the final stage of closing the dies, the blank is pressed between both of the tool halves giving the part its final shape. The forming temperature was 400 ˚C. Two profiles are welded with TIG-welding (Tungsten Inert Gas). The welding is done to join two formed pieces of the Ushaped test specimen together. The supplied welding power is 550 W and the welding tip is moving with the velocity of 90 mm per minute with respect to the work piece. During the proceeding heat treatment in a vacuum heat treatment 20


furnace, the heat is transferred to the welded part by radiation. In this stress relieve heat treatment the heating rate is 54°C/min with a holding time of two hours at the target temperature 540°C. The work piece is put on a metal grid during the heat treatment. During cooling both radiation and convection at 20 W/m2K transport the heat until equilibrium prevails. Methodology for the stainless steel part: The sheet metal forming simulation was performed with a forming step using the LS-Dyna explicit solver and a subsequent springback simulation step using the LS-Dyna implicit solver. The blank was modelled with shell elements. A material model including anisotropy was used (Barlat/Lian model according to [8]). The welding process was simulated with Abaqus. The LS-Dyna output data from the forming process, like the mesh, strains and the stress field was translated to Abaqus input data with the C6t2 utility developed in the ProSim project (see Section 3.3). It was decided to use solid elements for welding simulation and the shell elements from the forming simulation were translated to Abaqus input solid elements. The C6t2 software also mapped the output stresses and strains from the forming simulation in LS-Dyna on the new Abaqus input solid elements. In the first thermal simulation the heat source moved along the plates and the temperature field was established. In the second step a mechanical simulation was performed. It included pre-loading of the two plates due to clamping, introduction of thermally induced stresses due to the temperature field calculated in the thermal simulation, unloading of the clamping, cooling down and thereby obtaining residual stresses and distortions. A longitudinal strain plot shows shrinkages in the middle according to Figure 24.

. Figure 24 Longitudinal strains (red= 0.4 % tension, blue = 2 % shrinkage). Methodology for the titanium alloy part: The blank was modelled with solid elements. Normally FE simulation of sheet metal forming is performed with shell elements. Here, however, the result was used in a consecutive process i.e. welding simulation. For this FE simulation to be accurate, it is preferred to use solid elements. An alternative option would be to transform the model and results from shell elements to solid elements before doing the welding FE simulation according to the stainless steel case. An anisotropic material model including visco-plasticity and pure isotropic hardening was used according to [8]. The process set-up for the FE model and forming simulation can be seen in Figure 25. 21


Figure 25 Process setup and forming simulation. Since the forming was simulated using LS-Dyna (explicit forming and implicit springback according to the stainless steel case) and the welding simulation was done using MSC.Marc, the data was transferred from the output format used by LS-Dyna to input format possible to read by MSC.Marc. To facilitate this, the C6t2 utility was used and it was possible to transform LS-Dyna solid 4 node tetrahedrons, 6 node pentahedrons and 8 node hexahedrons to the corresponding MSC.Marc elements. The element stress field, and equivalent plastic strain were transferred. In order to prepare a model for the welding simulation the Digital Mock-Up (DMU) realised by FE simulation of the forming operation was copied and translated for proper positioning and the elements that constitute the joint were created, see Figure 26. These virgin elements must follow the geometry from FE simulation with its inherent effects from springback in terms of distortions. The preparation was done in the pre-processor Hypermesh.

Figure 26 Preparation for welding simulation. FE simulation of welding processes embodies thermal, metallurgical, and mechanical phenomena that are interconnected. The coupling can be tackled in different fashions. One way is to set out from the fact that the coupling between thermal and mechanical processes is weak. Then the calculation is performed with respect to the temperature and the thermal influence on the material first. Once this solution is obtained the mechanical problem is solved based on the result 22


from the thermal calculation. This approach was used in the stainless steel case. Another alternative is the so-called “Staggered approach” according to [9] which involves iterations between solving the thermal problem at the actual time step and the mechanical solution from the previous time step. The staggered approach was used to account for the coupling between mechanical and thermal effects. During welding the weld pool is continuously supplied with filler material that adds to the base material in the seam area and builds the joint. To account for this successive addition of material, that forms the bonding between the parts, a technique with activation of elements was utilised. All elements in the joint existed from the beginning of the simulation, but their stiffness was significantly reduced, i.e. they were virtually not there. Once the welding tip was passing the elements were activated, based on either the temperature falling below a certain value under the solidus temperature or on the time since the tip passed a certain element. A material model adapted for heat treatment, including creep, was used for the heat treatment simulation in MSC.Marc. Heat was applied by the use of film boundary conditions to model the radiation transfer of the heat in the vacuum heat treatment furnace. Figure 27 describes Von Mises residual stress results after heat treatment. The stresses were relieved to about 1/3 of the stresses after welding.

Figure 27 Von Mises residual stress after heat treatment (Blue=0 MPa, yellow=475 MPa). Material and process parameters: The process parameters involved in the study were the forming parameters (punch kinematics), welding parameters (heating source effect, clamping and heat source movements) and heat treatment parameters (heating and cooling time). In the stainless steel case the anisotropic material parameters for the forming simulation along rolling direction, across rolling direction and in between rolling and across directions were obtained from tests. For the titanium alloy case the material was considered as isotropic since the experimental data was limited. Further, the visco-plastic capabilities were not considered due to lack of material data. For the stainless steel case the welding material data, like stress-strain curves, Young’s modulus, thermal conductivity 23


and thermal capacity, for various temperatures were obtained from literature and tests. For the titanium alloy case the material data was obtained from tests. Validation of results: For the titanium alloy case the geometry after simulation of forming, forming+welding and forming+welding+heat treatment was compared to measurements of test pieces with the Atos geometry scanning system and accompanying tool for calculating geometry deviations. Also the simulated forming punch force was compared to the experimental forming punch force. Further the data transfer from LS-Dyna to MSC.Marc was evaluated by comparing the LS-Dyna output element stresses with the MSC.Marc input element stresses. For the stainless steel case the single forming process was evaluated with experiments. The forming+welding process was evaluated with temperature measurements and measurements of geometry deviations, with support from the Aramis equipment according to Figure 28. Further the data transfer from LS-Dyna to Abaqus was evaluated by comparing element stresses in the same way as for the titanium alloy case.

Figure 28 Deformation measurements with Aramis equipment. Comments and remarks: The experiences from the case studies are summarised below. The stainless steel case is also described in Figure 29: •

Data transformation and data mapping between different commercial softwares have pitfalls to be aware of and must be used with care. Even though a successful data transfer is performed there might still be problems with the digital mock-up not being in equilibrium. This can e.g. depend on the fact that the element formulations, material models and/or material data do not entirely map between the software in question. In addition to this, there might be simplifications used inside the software that can be difficult to avoid even when known to the analyst

•

Reliable material data is compulsory and not easily maintained

•

There is a need for further development of the welding simulation methodology for the stainless steel case. 24


Figure 29 Methods and tools for the stainless steel sequence.

25

3 RESEARCH RESULTS

3

Research results

3.1

Establishment of a sequence

The commercial numerical softwares for process simulation are up to now limited to the macro level aspects like prediction of global deformations, residual stresses and hardness. However, the establishment of seamless simulations through a manufacturing sequence may also involve micro and meso structural phenomena and e.g. Milesi [10] discusses possibilities for simulation of the micro structural changes during forging. Based on the complexity described above and limitations in the available commercial simulation softwares it is essential to find simplifications and engineering short cuts when analysing a complete sequence. This was obvious in the steering arm case where it was difficult to find accurate numerical models for the blasting process, see paper [III], and a combination of numerical and empirical models through the sequence was used. A manufacturing sequence may be categorised according to see Minisadram [11], see Figure 30, and simplifications will be discussed accordingly.

Figure 30 Classification of manufacturing sequence. 1. Analyse the complete manufacturing chain, including the melting/solidification processes, forming processes and supplementary processes. 2. Predict the accumulated effects from a subset of the manufacturing chain. 3. Select and analyse a single process from a manufacturing chain with the highest influence of product properties. The first approach, to realise a complete sequence, starting with the melting of pellets to ingots, billet manufacturing, forming and then supplementary processes like heat treatment and mechanical surface treatment is very complex and may be a subject for further research. The second approach, to extract a subset of the manufacturing chain, may be more engineering adaptable and in all the case studies, except for the steering knuckle case, the sequences consisted of subsets with forming and various supplementary processes. The engineering knowledge about the sequence must guide the selection of processes and tools like Failure Mode and Effect Analysis (FMEA) 26

3 RESEARCH RESULTS

may be a good support, see paper [I]. Further, different subsets should be selected based on the purpose of the analysis. As an example, in the pinion case, one could select different parts of the sequence depending of the purpose. If, for instance, the purpose was to study the machinability for the turning process the selected subset should be forging-cooling-isothermal annealing. However, if the purpose was to investigate the final stresses and strains it could be sufficient to study the case hardening- straightening sequence. The third approach, to select the most critical single process out of a sequence, is the most elementary and was utilised in the steering knuckle case. Also, a single process analysis may be sufficient if the purpose is to optimise the equipment for a single process. For instance, in the pinion case, the case hardening simulation may be used solely for optimisation of the quenching equipment and straightening simulation may be used solely for optimisation of the straightening equipment. Thus in order to establish a sequence for analysis a subset should be selected based on the purpose of the analysis and with support from engineering knowledge. Further a combination of numerical and empirical tools may be used through the sequence in order to improve the accuracy of the results.

3.2

In process modelling

Different processes require various geometry, mesh topologies, and element types depending on the physical phenomena that the models are intended to reflect, see Figure 31.

Figure 31 Different models and mesh topologies for different processes. The experiences from the case studies concerning in process modelling are summarised below: Local and global models: When analysing local effects, like surface residual stresses and hardness, it may be sufficient with a model of a small substrate of the 27

3 RESEARCH RESULTS

material near the surface. This type of models may be utilised in deep rolling simulation, according to paper [I], and shot peening simulation, according to paper [III]. However when analysing global effects on a component from the manufacturing sequence, like the final shape of the component, the complete geometry should be taken into consideration. For some processes both the local and global phenomena may be of interest and e.g. turning processes may be analysed with both global and local models. Here local modelling may be used for analysis of the influence of turning parameters on the surface residual stresses and hardness and global modelling may be used for prediction of component deformations due to clamping forces. In the case studies there was no communication of data between local and global models through the sequences. This may be a question for future investigation. Mesh topology: The requirements of mesh topology may vary depending on the physical effects that are involved in the single process. For instance for case hardening, in the pinion case, the mesh was modelled fine below the surface in order to reflect the variations in the carbon content while for the forging processes the mesh was guided by adaptive algorithms through the forging simulations. The experiences from the case studies shows that there is a need to create different mesh topologies for different processes within a sequence and mapping of data between various mesh topologies may therefore be necessary. It is of course possible to maintain the same mesh through the sequence but if this is the case special modelling considerations should be taken into account. As an example in the sheet metal formed cases the mesh topology for the sheets were the same through the sequence but the mesh topology of the blanks were adjusted in order to be more adaptable for the next welding simulation. Element type: Sometimes it may be necessary to utilise various element types for various processes and e.g. in the stainless steel case shell elements were used for sheet metal-forming while solid element were used for the welding process. Here the switch from shells to solids was done automatically with support from the C6t2 utility, see section 3.3, and also the mapping of data from the shells to the solids was done with support from the C6t2 utility, see Figure 32 below.

Figure 32 Transformation of shell elements to solid elements and mapping of stresses (Von Mises) on the new solid mesh.

28

3 RESEARCH RESULTS

Thus, it is obvious that the requirements of geometry, mesh topologies and element types may vary through the chain of processes and mapping of data between models may be necessary in order to establish a chain of processes.

3.3

Data communication

The input decks for commercial simulation softwares are often represented in different software-specific ASCII-file formats. It is not so common to export the output data in an ASCII-format and use it as input to another commercial simulation software. However it is possible to save data from a majority of the most common simulation softwares in native ASCII-formats. The simulation results may, during post-processing, be saved in various file formats like image formats (jpg-, tiff-, and gif- formats), animation formats (avi-format) and 3D formats (vrml-, vtf- and unv- formats). The Virtual Reality Modelling Language (vrml) is an open standard for visualisation of 3D models on the Internet [12]. The View Tech file format (vtf-format) was originally developed by the company Ceetron (former ViewTech) and is a standard file format for post-processor applications like Glview Inova and Glview Express (used by e.g. Forge2005). The Ideas universal file format (unv-format) was originally developed by the Structural Dynamics Research Corporation (SDRC) in order to facilitate data transfer between computer aided design (CAD), computer aided test (CAT) and computer aided engineering (CAE). In the Pinion case study the data transfer from Forge2005 to Morfeo was performed through the unv-format. Another future possibility may be to utilise the generic ISO 10303 (STEP) standard for data communication. One of the advantages with STEP is that the standard includes all aspects concerning data transfer between different engineering disciplines including CAD, CAM and CAE. STEP contains several applications protocols (AP:s) dedicated for specific engineering tasks. Especially AP 209 (Composite and metallic structural analysis and related design) is suitable for Finite Element Analysis (FEA) and has possibilities to communicate both CAD-models and FE-models, see [13]. ISO10303 –209 is accessible for some simulation softwares, like MSC.Nastran, but is limited to linear FEA. However within AP209 there is a common integrated application resource, Part 104 (Finite Element Analysis), which according to [14] may be adapted to various numerical disciplines according to Figure 33 and the possibilities to adapt AP209 towards non-linear process simulation may be a subject for future investigation.

29

3 RESEARCH RESULTS

Figure 33 Possibilities with ISO 10303-104 according to [14]. The transfer of output data from one simulation code to input data to the next simulation code may also be tailored between the native file formats of specific codes. In the ProSim project the C6t2 utility was developed in order to change the output specific format from one commercial simulation code to input specific data to the proceeding simulation software in the sequence. The utility was used in the sheet metal formed cases in order to transfer data from LS-Dyna to MSC.Marc and LS-Dyna to Abaqus. The C6t2 can also read and write in ASCII formats to and from other softwares and file formats according to Figure 34.

Figure 34 Data transfer functionality in C6t2. The C6t2 utility is written in C++ and runs at a command prompt or from a script file. Besides from the reformatting possibilities it is also possible, based on specific requests on the C6t2 command prompt, to create a solid element mesh from a shell mesh and map the results from the shell mesh onto the solid mesh.

30

3 RESEARCH RESULTS

Table 1 below describes what is supported for the different file formats. Table 1

Supported features related to implemented file formats. Read solid Write solid

Read shell Write Initial Initial Effective Temperature shell stress plastic plastic strain strain

Abaqus

-

-

Deform LD-Dyna

Tet4, Hex8 Tet4, Hex8 Tet4, Tet4, Tri3, Penta5, Penta5, Quad4 Hex8 Hex8

Tri3, X Quad4

X

X

-

X Tri3, X Quad4

X

X

X X

-

-

X

-

X

-

Tet4, Hex8 Tet4, Hex8 -

-

-

-

-

X

MSC.Marc -

VTF

Tet4, Penta5, Hex8

Tet4, Penta5, Hex8

During the development and evaluation of the C6t2 software several pitfalls were notified. Several issues varied in the input and output formats like node and integration point ordering, distribution of integration points, results described in either a local or global coordinate system etc. Based on the discussions above the conclusion is that one should carefully evaluate the data transfer between different file formats with tests before industrial implementation.

3.4

Validation of the sequence

According to the experiences from the case studies it is obvious that the numerical simulation of various single processes suffers from limitations concerning data communication and in process modelling. Further all the case studies have identified problems in finding accurate material data for the processes involved in the sequences. It is therefore of outmost important to establish a validation technique in order to tune and modify models and data. The validation methodology should include sensitivity analysis where some parameters are changed in a certain range and the effects on the final result is investigated. A generic validation procedure for sequential simulation should include validation of each single process, validation of the data transfer between processes and also validation of the chain of processes.

Figure 35 Validation of the sequence.

31

4 CONCLUSIONS

4

Conclusions

It is possible to establish virtual manufacturing sequences and connect different commercial simulation softwares into a chain. However, the data communication has pitfalls and also material models for single processes may be insufficient in order to conduct a quantitative analysis of the accumulated properties. Further, all the case studies showed difficulties in retrieving accurate material data and also the mapping of data on various of mesh topologies through the sequence may be a source of error. Thus a thorough validation of the virtual sequence should be performed. Also, simplifications of the sequences have been found useful, like the selection of a subset of the sequence and occasionally replacing numerical models with empirical ones through the sequential framework of models. In order to realise sequential simulation that involves several simulation softwares the following steps should be taken into consideration, see Figure 36. •

Select a subset of the sequence based on the purpose of the analysis

•

Select the data to be transferred through the sequence

•

Develop and implement routines that reformats output data to software specific input data

•

Create In Process Models

•

Develop and implement mapping routines

•

Simulate the sequence

•

Validate the virtual sequence and modify models and data if necessary

Figure 36 Flow chart of activities for sequential simulation. The main advantage of sequential simulation may be conceptual studies of process and material parameter variations and their influence on the final product properties. Also sequential simulation may be a good support in order to increase the knowledge of the effects on product properties from various processes in the sequence.

32

5 FUTURE RESEARCH

5

Future research

The practical work with the case studies has identified several research questions concerning in process modelling, data communication, validation and conceptual design methodology. In process modelling: The experiences from the case studies indicate that here is a need for further development concerning single process simulation. The validation of the deep rolling simulations in the steering knuckle case shows that there is a need for further development in order to improve the accuracy. The forging simulations in the steering arm and pinion case show that there is a need to include the effects from material segregation in the simulation model. Also the possibilities to include surface decarburisation effects during forging simulation should be investigated. The experiences from the steering arm case concerning blasting indicate a need for further development of numerical and empirical models. The pinion case indicate a need for further development of modelling technique concerning turning and straightening. The sheet metal-formed case studies indicate a need for development concerning welding simulation. Further, in order to establish a seamless chain of models it is essential to map output data on the proceeding mesh and further investigation of mapping algorithms is recommended. Data communication: The experiences from the case studies indicate a need for standards for data communication and STEP has been identified as a possible generic standard. Further investigations concerning the adaption of STEP towards non-linear Finite Element simulations is recommended. Validation: There is a need for development of a thorough validation methodology that includes studies of how various sources of modelling and material errors may impact the final result. Conceptual design: Today the design engineer uses tools like Computer Aided Design (CAD) and linear Finite Element Analysis (FEA) in order to optimise the geometry and material, a methodology that is formalised in e.g. ISO 10303-209. A future research question may be to study possibilities for integration of sequential simulation in the overall CAD/FEA methodology according to Figure 37.

Figure 37 Integration of manufacturing sequence in the CAD/FEA methodology

33

6 REFERENCES

6 [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

References P. Åström, “Simulation of Manufacturing processes in Product Development”, Thesis, Luleå University of Technology, 2004 S. Hyun, L-E Lindgren, “Simulation of a chain of manufacturing processes using a finite element code with adaptive meshing”, Finite Elements in Analysis and Design 40, 2004, pp 511 – 528. H. Alberg “Simulation of Welding and Heat Treatment, Modelling and Validation”, Thesis, Luleå University of Technology, 2005 Y. Altinas, C. Brecher, M. Weck, S. Witt “Virtual Machine Tool”, CIRP No55 Turkey 2005, Vol 54 No2, 2005, pp 651 – 674. Forge3 Reference guide, Release 6.3, Transvalor S.A, 2004. Morfeo user manual, Cenaero, 2008 T. Belytschko, N. Moes, S. Usui, C. Parimi “Arbitrary discontinuities in finite elements”, International Journal for Numerical Methods in Engineering 50, 2001, pp 993-1013 LS-DYNA Keyword User’s Manual, 2006 version 971, Livermore Software Technology Corporation L-E Lindgren, “Finite Element Modelling and Simulation of Welding Part”, Journal of Thermal Stresses 24, 2001, pp141-192 M. Milesi, Y. Chastel, M Bernacki, R Loge’, P-O Bouchard, “Explicit Microscopic Fatigue Analysis of Forged Components” Computer Methods in Materials Science Vol. 7 No 4, 2007, pp383-388 R.S. Minisandram, S.K Srivatsa, ”Deformation modeling applications in the aircraft industry”, Simulation of Materials Processing: Theory, Methods and Applications, Mori, 2001, pp385 – 390 M. Werke, “Internet based distribution of simulation results”, EVEN conference, 2003, Trinity College, R. I. Campbell, N. O. Balc, Loughborough, 2003, pp 4 - 13 International Organisation for Standardisation, “Application protocol Composite and metallic structural analysis, ISO 10303-224”, 2001 International Organisation for Standardisation, “Integrated application resource - Finite element analysis, ISO 10303 -104”, 2000

34

7 WEB-LINKS

7

Web-links

The list below summarises the supplier of softwares, standards and experimental equipments used on the different occasions in this study. •

Abaqus

www.simulia.com

•

Atos and Aramis

www.gom.com

•

Deform

www.deform.com

•

Forge2005

www.transvalor.com

•

Gmsh

geuz.org/gmsh

•

HyperMesh

www.altair.com

•

JMatPro

www.thermotech.co.uk

•

LS-Dyna

www.ls-dyna.com

•

Morfeo

www.cenaero.be

•

MSC.Marc

www.mscsoftware.com

•

Sysweld

www.esi-group.com

•

Universal file format

www.sdrl.uc.edu

•

View Tech file format

www.ceetron.com

•

Xstress 3000

www.stresstech.fi

35

Statistical methods for extracting and evaluating manufacturing parameters of significant importance to component robustness Mats Werke(*), Ramin Moshfegh(*), Bengt Johannesson(**), Johan Svenningstorp(***), Per Johansson(****), Erwin Stinstra(*****) (*)

IVF, (**) Volvo Truck Corporation, (***) Volvo Technical Development, (****) Volvo Power Train, (*****) CQM

[email protected], [email protected], [email protected], [email protected], [email protected], [email protected] Abstract The paper presents a methodology developed within the EURobust project (G1RD-CT-2002-00833 and IMS-Robust 97009, www.eurobust.net). Durability and robustness of components are combined effects of external loads, geometrical design, selected materials and manufacturing sequence. The influence of the manufacturing sequence has not yet been explored and implemented in the virtual product development process and this paper presents a new and unique method to help the engineer in invoking the effect of the manufacturing sequence on product properties. The methodology extracts and evaluates manufacturing parameters with a major influence on component durability. The method is a combination of engineering judgement, Variation Mode and Effect Analysis (VMEA), design of experiment plans for computer-based experiments, automatisation of Finite Element simulations and visualisation of results. The paper describes how to use the method when estimating the robustness of a forged steering knuckle where the deep rolling process was identified as the most important manufacturing process in the manufacturing sequence, using Failure Mode and Effect Analysis (FMEA). The impact on residual stresses, surface quality and deformation hardening from the deep rolling parameters such as tool diameter, hydrostatic pressure, circumferential spindle velocity, feed rate, etc was analysed using VMEA. A simulation model of the deep rolling process was developed and the variations of residual stresses as a function of the manufacturing parameters were simulated and analysed using graphical response surfaces. The method will help engineers to better control the design of components taking all aspects, including the manufacturing sequence, into consideration at the early design phase. The method will also increase the knowledge of how the manufacturing processes affect component durability. 1.

Introduction

The manufacturing processes have a large influence on product durability and robustness and will affect residual stresses, material hardness, microstructure of 36

the material, probability of notches and surface quality. The vision of the future is to quantify the effects of component durability caused by the accumulated manufacturing sequence with support from computer-based simulations. A lot of research remains to be done before this is possible. However a simple and robust way is to select and evaluate one manufacturing process that influences the product durability more than the others. This paper presents a methodology that selects one process and extracts, simulates and analyse the influence on product properties from the most important manufacturing parameters of the selected manufacturing process. The industrial use of the Finite Element Method (FEM) in the simulation of manufacturing processes varies a lot. The use of FEM in sheet metal forming is quite established and the status of solid forming processes, like forging and casting, is similar to that of sheet metal forming, maybe somewhat less established. Then there is a line of processes like heat treatment, welding, grinding, turning and mechanical surface treatment that are less common due to more difficult modelling and/or numerical problems. This paper also presents a new model for simulation of deep rolling which is a process for mechanical surface treatment. 2.

Methodology

The method should be used in the development phase and should be performed in collaboration between design engineers, process specialists and testing engineers. Firstly, the engineer identifies the failure area, mainly based on engineering judgement and knowledge. Secondly, the core process with strongest influence on component life is identified. Thirdly, critical parameters from the selected core process are defined and evaluated using Variation Mode and Effect Analysis (VMEA). Fourthly, a simulation model is created and in a fifth and a sixth step an experiment plan is designed and the critical parameter variations are simulated. In the seventh and eighth steps a regression model is developed and the response surfaces/curves are visualised/analysed. The method is illustrated in Figure 1.

37

Figure 1. Methodology 2.1

Identify potential risk areas

The identification of possible failure areas of the product concept should be done based on engineering knowledge and experience from physical testing in the field and in test rigs. 2.2

Analyse manufacturing sequence and select process

The manufacturing sequence for the potential risk areas should be discussed together with process-, material- and design engineers. Each manufacturing operation in the sequence should be analysed and the critical manufacturing process with the strongest influence of component durability should be selected. Engineering support tools such as Failure Mode and Effect Analysis (FMEA) according to Lundberg (1991) can be used in order to structure the selection process. 2.3

Analyse process and select parameters

A Variation Mode and Effect Analysis (VMEA) according to Chakhunashvili et al. (2004) should be conducted in order to analyse the critical manufacturing process. Key Product Characteristics (KPC) of component durability and Sub Key Product Characteristics (Sub-KPC) should be decided according to engineering knowledge. Sources of variations should be defined as well as their sizes of variation. A variation risk priority number for each Sub-KPC should be calculated and thereby the most important Sub KPC derived. Based on the selected Sub-KPC a simulation model should be established and the defined sources of variations for the selected Sub-KPC should be translated as input parameters for the simulation model.

38

2.4

Build simulation model

Physical effects from manufacturing processes that affect product properties are material hardening, cold work deformation, residual stresses and the occurrence of micro notches. Material hardening depends mainly on effects from heat treatment like a change in microstructure and the chemical composition of the material. Cold work deformation is a result of mechanical surface treatment, like shot peening and deep rolling, and increases the hardness mainly as an effect of an increase of material dislocations as well as a change of grain size. The occurrence of micro notches as a result of manufacturing processes has a significant influence on component life. Finite Element simulation (FE-simulation) of material hardening and cold work deformation is an area open for future research and theory for analysing crack propagation is a field currently researched that might be adaptable for FE-simulation. Manufacturing processes such as heat treatment, grinding, turning and mechanical surface treatment cause residual stresses in the product and according to Totten et al. (2002) residual stresses have a significant influence on component life. Residual stresses are well suited to analyse with FEM and Section 3.4 describes a model for simulation of compressive stresses after deep rolling. 2.5

Design experiment plan

In order to predict the influence of input parameter variations on the response it is necessary to combine simulation technique with statistical methods, i.e. design a simulation experiment plan. The problem of choosing an experiment plan is called Design of Experiments (DoE). Designing and analysing plans for computer-based experiments is also called Design and Analysis of Computer Experiments (DACE) according to Sacks et al. (1989). The main difference between traditional DoE and DACE is that in physical experiments measuring errors might occur which is not possible in DACE. Also the cost of executing many test points in computer experiments is limited only to the CPU-time. In traditional DoE, where the purpose is to design a plan for physical experiments, usually only high and low values of each input parameter are selected and combined. To accomplish a design plan for computer simulation experiments Stehouwer et al. (1999) recommend a plan with several test points between the low and high values. The plan is a so-called space filling scheme with a Latin Hypercube Design (LHD) according to Figure 2. First the design parameters are identified and secondly a simulation scheme according to the LHD methodology is created where each design parameter is divided into a number of equidistant levels. Based on the LHD a simulation scheme is created. The simulation scheme should be organised as an input file to the simulation software.

39

Figure 2. LHD design 2.6

Simulate parameter variations

The simulation scheme according to LHD may consist of hundreds of design points and in order to simulate the variations it is necessary to automate the simulation procedure. The procedure should be a script that in a sequence reads the design points from the LHD, simulates the response and stores the results. This could be invoked in a loop where the response from all the test points should be stored as an output file for the regression analysis. 2.7

Fit a regression model

In order to analyse the response of arbitrary combinations of manufacturing parameters a regression model on the output results should be generated. Typical model types include first or second order polynomials and Kriging models. These models are often referred to as compact models, approximating models or metamodels. Such models can be used to predict the outputs in new settings for the input parameters, thus eliminating the need for more FE-simulations. In some cases it is preferable to create response curves based on linear or multiple regression instead of surfaces.

Figure 3. Response surface based on LHD 2.8

Visualise and analyse response surfaces/curves

Since the response model is analysed instead of the simulation model, this phase does not need large amounts of CPU-time. Different software’s for visualising response surfaces/curves such as CQM:s Compact model viewer or MatLab can be used as engineering tools. More information about software from CQM can be 40

found at www.cqm.nl and more information about MatLab can be found at www.mathworks.com. Based on the response surfaces a sensitivity analysis can be performed and conclusions can be drawn concerning the impact of component durability for different manufacturing parameters. 3.

Estimation of robustness of a forged steering knuckle

The method was evaluated on a forged steering knuckle in heat treatment steel material. 3.1

Identify potential risk areas

After discussions with testing- and design engineers at Volvo Truck the possible failure area was identified to be a fillet radius between the axle and the rest of the knuckle body according to Figure 4. 3.2

Analyse manufacturing sequence and select process

In order to select the most important core process in a failure risk context a process FMEA with a customer perspective was performed. Operations such as billet manufacturing, pre forming, final forming, trimming, cooling, blasting, heat treatment, turning and deep rolling were analysed. The deep rolling process received the highest risk factor and was chosen as the core process for further studies.

Figure 4. CAD model of steering knuckle and deep rolling of fillet radius 3.3

Analyse process and select parameters

A VMEA of the deep rolling process was performed together with engineers from Volvo. The KPC was decided to be component fatigue. The KPC was divided into three Sub-KPCs: surface quality, deformation hardening and residual stresses. Several sources of variations were identified for each Sub-KPC. With support from articles such as Kloos et al. (1987) variations in residual stresses were identified as the most important Sub-KPC with the highest influence on fatigue. Since the axial external loads have a strong impact on component durability the axial residual stresses were selected to be studied. Table 1 shows typical sources of variations of Sub-KPC residual stresses. After the VMEA the identified sources of variation were transferred to input parameters for the FE-model.

41

Table 1. VMEA on residual stress

The VMEA resulted in a total of 9 parameters for the simulation model according to Table 2. Material parameters, such as phase, grain size and chemical composition were translated to Young’s modulus, yield strength and tensile strength. Geometry after turning was translated to surface roughness and the remaining manufacturing parameters according to Table 2 were directly translated as inputs to the Finite Element model (FE-model). 3.4

Build a simulation model of the deep rolling process

Deep rolling is a proven process to enhance fatigue strength of dynamically loaded components. The enhancements are based on the combination of a deep layer of residual compressive stress in the surface of the component, a strength increase through cold working deformation and elimination of micro notches. The effects are obtained by rolling a ball with high pressure at the surface of the component. The operation can be done after turning and in the same machine or in a separate machine. Figure 5 describes the principles of the process and typical compressive stress profiles under the surface according to Ecoroll. Complementary studies of FE-simulations of deep rolling have been done according to Jung (1995) and Schaal (2002). More information about the deep rolling process can be found at Ecorolls web page www.ecoroll.de.

Figure 5: Principles for the deep rolling process and compressive stress profiles after shot peening and deep rolling (HG) 3.4.1 Finite Element model The main effect of deep rolling is restricted to a thin layer below the surface (0 – 0.1 mm). A FE- model that describes the effects on a small geometric volume on a thin layer below the pressure ball was developed according to Figure 6. Three models with various surface roughness (4, 10 and 20 micrometer) and three models with various ball diameters (4,6 and 8 mm) were created in HyperMesh. 42

The mesh topology was zoomed with a higher density close to the surface and at the centre of the strip. Each model consists of 16300 elements with shell elements defining the ball and solid elements defining the thin layer below the surface. The software LS-Dyna was used as a solver. The contact between the surface and the ball was defined using LS-Dyna contact algorithms and the material model was elastic plastic with kinematic hardening. The analysis was done in the following four steps: - The ball was applied in contact with the surface of the strip with a ramped force (hydraulic pressure) and then forced to roll (spindle velocity) on the strip – explicit analysis - Springback analysis after the first deep rolling phase - Implicit analysis. - Same as the first step but the ball has been displaced in x-direction (feed rate) – explicit analysis. - Springback after the second deep rolling phase – Implicit analysis. The four steps were invoked in one LS-Dyna input file. In order to speed up the calculations the mass was increased to 727 times the actual mass. With this mass scaling the CPU time for one cycle was 34 minutes on a Linux cluster with two parallel CPU:s. The Finite Element results (stresses, strains, deformations) from elements just below the surface, according to Figure 7, were analysed and post processed with HyperView. The distance from the surface to the nodal analysis points was 0, 25, 50, 100, 150, 200 and 275 micrometers. Information about HyperMesh and HyperView can be found at www.altair.com and information about LS-Dyna can be found at www.lstc.com.

Figure 6. FE-model of deep rolling and element topology at surface

Figure 7. Analysis zone with residual stresses on the model 43

3.4.2 Validation of the Finite Element model The model was validated concerning variations in boundary conditions, geometrical size of the strip, mesh density of the ball and mass scaling. With no mass scaling the accuracy may improve somewhat but the CPU time for executing one cycle would be 31.5 hours. No major deviations occurred during the model validation. The results were also evaluated and compared with residual stress measurements with X-ray diffraction method at both Volvo and IVF according to Figure 8. The impact of the etching process, that is part of the X-ray measurement procedure, was evaluated with a FE-model according to Figure 9. No deviations occurred as a result of the etching process.

Figure 8. Residual stress measurements with Xstress3000 equipment at IVF

Figure 9. FE-analysis of the etching process 3.5

Design experiment plan

The upper and lower bounds of the input parameters were decided according to material specifications and specifications for the deep rolling process for the steering knuckle according to Table 2. An LHD experiment plan was designed. Parameters except the ball diameter and surface roughness were varied continuously according to the LHD. Distinct values were set for ball diameter and surface roughness.

44

Table 2. Input parameters for simulation model Youngs modul 2.06E5 – 2.11E5 Mpa Yield strength 690 – 1200 Mpa Tensile strength 862 – 1500 Mpa Spindle velocity 190-240 m/min Feed rate 0.15-0.2 mm/rpm Hydralic pressure 200 – 250 bar Friction 0 – 0.1 Ball diameter 4, 6 and 8 mm Surface roughness 4, 10 and 20 micrometer A script for simulating the variations according to the LHD with LS-Dyna was developed according to below: 1.

Read a parameter set from an Excel file

2.

Simulate the parameter set in LS-Dyna according to cycle in subsection 3.4.1

3.

Store the results for selected elements in an Excel file

4.

Loop 1 – 3 for all the combinations in the LHD and store the results in an excel file.

90 combinations were simulated and the total CPU-time was 51 hours with the actual mass scaling. 3.5

Fit a regression model

Response surfaces with maximal axial residual stresses for each parameter set were created based on polynomial functions such as the Kriging function. Also 2D curves of the residual stress profile beneath the surface were created and evaluated based on multiple regression analysis according to Figure 11. 3.6

Visualise and analyse response surfaces/curves

The response surfaces were studied with the CQM-software “Compact viewer” according to Figure 10 below. Two arbitrary manufacturing variables were set on the x- and y-axis and the residual stress response was displayed on the z-axis. With support from the Compact viewer it was possible to vary the remaining manufacturing parameters (with spacebars at the left in Figure 10) and visualise the change in the response surface of the residual stresses.

45

Figure 10. Visualising the results with Compact Viewer Based on the simulated 90 stress profiles a multiple regression analysis of a subset of parameters (ball diameter 6 mm and surface roughness 10 micrometer) was performed in Matlab. A sensitivity analysis according to Figure 11 was performed and the conclusion was that variations in the material parameters (yield and tensile strength) influence the residual stresses more than the other manufacturing parameters (spindle velocity, feed rate, friction, hydraulic pressure, etc).

Figure 11. Residual stress sensitivity to material and manufacturing parameters According to Thomas et al. (2003) low plasticity in combination with high compressive stress have a positive impact on fatigue properties, especially when the structure is overloaded. A scatter of all the results was plotted according to Figure 12 where each point represents a defined parameter set and the responses are maximal compressive stress and maximal plastic strain for each set. The scatter can be used in order to find optimal parameter sets with high compressive stress in combination with low plastic strain.

46

Figure 12: Scatter plots for finding parameter sets with maximal compressive stress and minimal plastic strain 4.

Conclusions

The new method is a combination of engineering knowledge, statistical methods and simulation technique and has been tested on a manufacturing sequence including forging. The method is generic and can be applied to other manufacturing sequences including sheet metal forming, casting, etc. The method will help engineers to better control the design of components taking all aspects, including the manufacturing sequence, into consideration at the early design phase. The method will also increase the knowledge of how the manufacturing processes affect component durability. The FE model for simulation of the deep rolling process and the procedures for automating the simulation of variations as well as analysing response is a good way to get a survey of all possible responses in the deep rolling process, find optimal design points, investigate the fatigue sensitivity for different manufacturing and material parameters and thereby increase the knowledge of the deep rolling process. The number of test points in the LHD should be increased in order to get a better resolution in the Kriging response surface and also the number of parameters in the design plan should be discussed as well as their variations. More validations with X-ray measurements and material tests should be performed in order to tune the FE model. Also more work should be done in order to find criteria for optimal deep rolling parameter settings.

47

5.

Aknowledgements

This method was developed as a part of the EURobust project (G1RD-CT-200200833 and IMS-Robust 97009, www.eurobust.net), financed by the 5th EU framework programme, Vinnova and participating companies such as Volvo Truck, SKF, Carl Zeiss, etc. The authors wish to thank the Volvo production facility in Köping for supplying IVF with steering knuckles for measuring residual stresses with X-Ray diffraction method. References Chakhunashvili, Johansson, Bergman (2004), “Variation Mode and Effect Analysis” Proceedings of Annual Reliability and Maintainability Symposium. Los Angeles, USA Ecoroll, ”Deep rolling, versatile and efficient against Fatigue”, lecture 6090e, lecture material Jung (1995), “FEM-simulation of the deep rolling process” Proceedings from the Marc European users conference. Kloos, Fuchsbauer, Adelmann (1987), “Fatigue properties of specimens similar to components deep rolled under optimized conditions”, Int J Fatigue 9 No 1, page 35 - 42 Lundberg (1991), “Feleffektanalys (FMEA) av smidesprocessen” IVF-skrift 91815, IVF Mölndal Sacks, Welch, Mitchell, Wynn (1989), “Design and Analysis of Computer Experiments“, Statistical Science Vol 4 No 4, 409 – 435Schaal (2002), “FEMsimulation des Festwalzens und Dauerfestigkeitsberechnung mit Methoden der linear-elastischen Schwingbruchmechanik“, Dissertation, Shaker Verlag, Aachen 2002 Stehouwer, Hertog (1999); “Simulation-based design optimization: methodology and applications”, Proceedings of the first ASMO UK/ISSMO Conference in Engineering Design Optimization, Ilkley, UK Thomas, Jacobs, (2003) ”Low plasticity burnishing: An affordable Effective, Means of Surface Enhancement”, ASM international, proceedings from the 1st international surface engineering Congress. Columbus Ohio, page 151 – 159 Totten, Howes, Inoue (2002), Handbook of Residual Stress and Deformation of Steel, ASM international

48

PREDICTING RESIDUAL STRESSES AND HARDNESS OF A CRITICAL COMPONENT USING A COMBINATION OF NUMERICAL AND EMPIRICAL METHODS Mats Werke1, Hans Kristoffersen1, Sven Haglund2, Lennie Svensson3, Anders Nord3 1

Swerea IVF AB Swerea KIMAB AB 3 Bharat Forge Kilsta AB 2

ABSTRACT This paper describes the experiences gained when using numerical and empirical methods in order to predict the accumulated surface characteristics for a safety component after several forging steps, controlled cooling and blasting. The forging steps were simulated in a sequence using one Finite Element (FE) code. The output forging mesh was used as input to the cooling simulation but was too coarse in order to reflect surface characteristics. The decarburisation effect during cooling that may influence the surface characteristics was not included in the cooling model. An attempt to create a parametric model of the blasting machine with output residual stresses and hardness as a function of input residual stresses, hardness and process parameters indicated the need of further investigation concerning the physical phenomena during blasting in the machine. A new method was developed for analysing the influence of the blasted surface texture on the stress intensity. The measured residual stresses and hardness span caused by variations in the blasting process were successfully used together with the stress intensity factor as input to a fatigue strength analysis. In order to establish a seamless chain of models through the manufacturing sequence further development concerning cooling and blasting models is required. Keywords: Sequential simulation, manufacturing chain, fatigue properties 2.

INTRODUCTION

The obvious trend in the automotive industry sector is to shorten the product development time and reduce costs with support from virtual tools. The manufacturing sequence may have a large influence on product properties and in order to fully explore the potential of simulation there is a need to invoke the manufacturing sequence in the virtual product development process. However, the vision to predict the accumulated properties like residual stresses and hardness solely by means of numerical simulation in a chain may be difficult to accomplish 49

with high accuracy. The reason for this is the fact that there is a lack of mature numerical tools for analysing all the aspects of material changes like the effects of anisotropy, grain size and decarburization through various manufacturing processes like forging, heat treatment and mechanical surface treatment. Thus, this paper investigates the possibility to use a combination of numerical and empirical methods for analyzing the evolvement of properties through the sequence. The overall idea was to integrate CAD-models, numerical models and different concepts of analytical and empirical models into a framework in which the different models should be able to communicate and exchange data between one another. This may improve load analysis, product specification development, shape analysis and manufacturing sequence analysis according to Figure 1.

Figure 1. Framework methodology. The analyzed manufacturing sequence for a safety component, a forged steering arm according to Figure 2, consisted of several forging steps followed by controlled cooling and blasting.

Figure 2. Forged steering arm. The forging and cooling steps were simulated in a sequence according to Sections 2 and 3. The blasting process was analysed using a combination of empirical measurements and numerical modelling according to Section 4. In connection to the blasting process a new method was developed to predict the surface stress intensity caused by the blasted surface topology. The blasting output like the residual stress and hardness span and the surface stress intensity were used as input to a fatigue strength analysis of the component according to Section 5. The practical experience from the test case is discussed in Section 6. 2.

FORGING PROCESSES

Cad models of the different forging tools were created and used as input for the simulation models. Also a mesh of the billet was created with support form the 50

CAD software Unigraphics. Hot shearing, shape rolling, roughing blow, fine blow, trimming and hot calibration were then simulated in a sequence using Forge2005 on a 9 node PC cluster where the output database from the previous step was used as input to the next forging step. The press kinematics for the different forging steps were used as input process parameters for each step. The coefficients for the Hansel Spittel rheology law were selected from a similar material, 46MnSi, since no data was found for the component material, (micro alloyed steel, V2904 according to Table 1). The input billet temperature for the sequential simulation was set to 1265ºC, tool temperature to 200ºC and environment temperature to 50ºC. The heat exchange coefficient with air was set to 10 W/m2ºC and the thermal exchange coefficient with the rigid die was set according to instructions in [1]. Table 1. Chemical composition of V2904. C 0.4

Mn 1.35

Si 0.6

Cr 0.2

V 0.085

N 0.02

During hot shearing in the 140 ton shearing press machine the FE model of the billet was separated by the Latham Cockroft fracture criterion [2] according Figure 3. The mesh was refined at the cutting surface region in order to get a good geometry after the operation.

Figure 3. Hot shearing simulation. The virtual part in process was re-meshed after hot shearing in order to get a good element topology for the proceeding steps. The output results, like nodal temperatures, plastic strains etc, were mapped on the new mesh in Forge2005. After hot shearing the part was exposed to a two-step shape rolling process followed by roughing blow and finishing blow in a sequence in a 2500 ton press according to Figure 4.

Figure 4. Shape rolling, roughing- and finishing blow.

51

The elements defining the flash after the fine blow were deleted using a trimming die operation in Forge2005. After the trimming operation the virtual part in process was calibrated in a calibration tool according to Figure 5. The model size after each operation is shown in Table 2.

Figure 5. Hot calibration. Table 2. FE model size. Operation Hot shearing Shape rolling Roughing blow Finishing blow Trimming Hot Calibration

Number of elements 140484 101880 276934 183952 169037 95962

The part in process after the forging chain (geometry, mesh topology, temperatures, etc.) was used as input to the cooling simulation. 3.

COOLING PROCESS

The controlled cooling simulation was performed in the Forge2005 quenching module. The output model from the forging sequence was complemented with a load carrier with thermal contact with the part in process according to Figure 6. The carrier temperature was set to 70 ºC, the air temperature to 20 ºC, the heat exchange coefficient with air to 10 W/m2ºC and the emission coefficient to 0.8.

Figure 6. FE model of the arm and carrier with nodal temperature distribution. The simulation in the Forge 2005 quenching module included a coupled metallurgy, thermal and mechanical computation [1]. Since no material data was found for V2904 a similar material, 38MnSiV5, according to Table 3 was used. 52

The Forge2005 TTT-Base module was used in order to retrieve mechanical material data (phase elastic-, phase plastic- phase thermal-, phase expansion parameters, etc) based on the chemical composition and the ASTM grain size. The metallurgy data was created based on a CCT-diagram from the steel subcontractor that was converted to a TTT-diagram according to Figure 7. The mechanical and metallurgical data was included in the material input file. Table 3. Chemical composition of 38MnSiV5. C 0.4 Cu 0.1

Mn 1.39 S 0.033

Si 0.82 Mo 0.03

Cr 0.19 Ti 0.003

V 0.11

N 0.122

Figure 7. TTT-diagram for 38MNSiV5. The simulation results after one hour of cooling (see Figure 9) were analysed and compared to residual stress and hardness measurements. The residual stress measurements according to Figure 8 were performed with the X-Ray Diffraction method (XRD) using the Xstress3000 equipment.

Figure 8. Measured axial residual stress distribution after cooling.

53

Figure 9. Axial stress distribution after one hour cooling. The simulated surface residual stresses were in the same range as residual stress measurements. However, the coarse surface mesh topology did not reflect the local surface phenomena according to Figure 8. Thus a finer surface mesh may be needed in order to transfer surface data to the next step. The simulated hardness results were in the same range as hardness measurements (~300 HV). A Light Optical Microscope investigation (LOM) revealed Ferrite in the surface layer according to Figure 10. This indicated surface decarburisation, an effect that was not included in the cooling model.

Figure 10. Ferrite surface layer (white spots). 4.

BLASTING PROCESS

In order to remove the slag from the surfaces of the parts they were exposed to blasting in a Gutmann Jatoflex MWJ 12/12/24 machine according to Figure 11. The round steel balls had a diameter of 0.6-0.9 mm with a hardness of 450 – 520 HV and velocity = 80 m/s. It is known according to [3] that the surface residual compressive stresses will be improved when exposed to blasting.

54

Figure 11. Gutmann blasting machine, MWJ 12/12/24. The state of the art concerning numerical models for blasting is still on a basic level with one ball shot or a pattern of impacts against an ideal material [4]. No other effects from the blasting machine such as the tumbling of pieces against each other are included. Consequently an attempt was made to create a parametric model for the Gutmann machine based on measurements on a test series of eight different materials with different input residual stresses and hardness. The ambition was to define the output residual stresses and hardness as a function of surface residual stresses and hardness after cooling, blasting time and number of items in the machine. However, it was difficult to detect any major trends in the output results. This indicates the necessity of further investigation concerning the physical phenomena during blasting. Thus, instead of creating models of the blasting process the output residual stress span and hardness span for the V2904 material were analyzed and used as input to the load analysis. The residual stresses were measured with the XRD equipment and the hardness was measured with a Knoop diamond (200g). The forged test parts were exposed to blasting with different times in the machine (175, 350, 700 and 1400 sec) and also with various numbers of details in the machine (50, 100 and 150 details). The nominal blasting time was 700 sec with nominal 150 parts in the machine. Figure 12 shows measured axial residual stresses as a function of time at a depth of 35 µm. The differences between the measured axial and tangential residual stresses for the V2904 material were small. Thus the axial stresses were used for the further analysis.

Figure 12. Residual stress as a function of time at a depth of 35 µm. 55

The nominal compressive residual stress according to Figure 12 was detected to 447 MPa. The stress span is shown in Table 4. Table 4. Residual stress span for V2904. 700 sec 175 -1400 sec

Min stress -413 MPa -292 MPa

Max stress -468 MPa -478 MPa

Figure 13 shows the measured hardness as a function of time at a depth of 100 µm. For the nominal blasting conditions the hardness was detected to 324 Knoop [HK0.2]. The hardness span for the V2904 test series is shown in Table 5.

Figure 13. Hardness as a function of time at a depth of 100 µm. Table 5. Hardness span for V2904 700 sec 175 -1400 sec

Min 324 HK[0.2] 295 HK[0.2]

Max 341 HK[0.2] 377 HK[0.2]

The conversion factor between HK[0.2] and Vickers was approximated to 1.0 for the proceeding analysis. The influence of stress intensity from the blasted surface texture was investigated. The surface topology was measured with a confocal microscope according to Figure 14.

Figure 14. Surface topology measurement with a confocal microscope. The measured points and surface polygons were transferred to Matlab where a subset of points was selected and transferred as a surface mesh to the FE software Abaqus [5]. A FE mesh of a small surface substrate block was created and a unit 56

tension load applied at the edge surface of the volume mesh according to Figure15.

Figure 15. Surface stress intensity analysis using Abaqus. Based on the calculated maximum groove stress, the stress intensity factor Kt = 1.46 was predicted. 5.

LOAD ANALYSIS

The measured hardness was in the range of 300-400 HV. For hardness levels