lunds tekniska högskola · 2007. 5. 8. · tig-welded part with complex geometry 61 paper ii...

��

��

��

��

��

��

��

��

��

��

��

��

�� !�" #� ��$�� %��

�� &�'��

��

(��

)�� *+,+ -�. ""#� �� " �� )�� %��

*��

�,��/0 )��/12��3 "��"4" "��12��4

5�-/ "�0 6" ��# �#�6 "5�-/ "70 68# 6" ��# �#�6 8

�©�� '� ��9�� +

:�� +

*�� %��;(� � )�� :-� )��

Abstract

Welding is one of the most important tasks carried out by robots in manufacturing in-dustry. The operator usually performs the programming of the robot manually, i.e. byjogging the robot arm to each coordinate pose in space. Programming can, however, bemade more accurate by the use of simulation, using so called Computer Aided Robotics(CAR). Simulation can also be a powerful tool in the evaluation and control of weldingheat effects, such as unwanted stresses and deformation.

The objective of this thesis is to develop a simulation tool and a method by which robottrajectories, temperature histories, residual stresses and distortion can be analysed and op-timised off-line. This was performed by integrating robot simulation software with finiteelement analysis software. A special interface was created which facilitated informationexchange between the two software programs. To validate the method comparisons weremade between simulation results and measurements during real welding

The method was used to program welding trajectories, both for planar plates and for partswith complex shapes. The welding trajectories were downloaded to the finite elementanalysis software where temperature and residual stress prediction were performed. Goodagreement was found between the programmed robot trajectory, and the actual trajectory,necessitating only minor adjustments. Temperature measurements were performed usingboth thermocouples and infrared imaging. Good agreement was also found between theresults using these two methods, as well as between predicted and measured temperatures.

Predicted residual stress distributions were compared with neutron diffraction measure-ments and fair agreement was found. A specified software architecture was developedwhich allowed full time synchronization between different simulation systems. Finally,weld velocity optimization was performed through a developed algorithm making it pos-sible to minimize distortion.

The research conducted in the present work indicates that the models and computer pro-grams that were developed could be combined to create powerful tools for the evaluationand optimisation of welding processes.

iv Abstract

Acknowledgments

The work on this thesis has been a long and difficult journey and I have been fortunateto have received the benefit of contributions, assistance and support from a number ofpeople. I am extremely grateful for all the help and support that I have received. Some ofthese people deserve particular thanks.

First, I would like to thank my supervisor, Professor Gunnar Bolmsjö, Faculty of Engi-neering, Lund University, for his support and help during this work. I would also like toexpress my indebtedness to my second supervisor, Dr. Per Nylén, University West (formerUniversity of Trollhättan/Uddevalla), for his support, contributions and his enthusiasm forthis project. Without his help, none of this work would have been possible.

My research has mainly been performed at the University West and I would like to thankall members of the VIP and MIA research teams at the Department of Technology, Math-ematics and Computer Science, for contributing to the friendly working and social envi-ronment at the department. In particular, I would like to mention the following people.Stefan Björklund, Kjell Hurtig and Mats Högström for all the valuable discussions wehave had and for all the help they have provided with my experiments in the weldinglaboratory. I would also like to thank Dr. Fredrik Danielsson for his help and for ourdiscussions about c-programming and simulation technology. Fredrik Sikström for theassistance in 3D scanning. John Lorenzon for our discussions regarding finite elementanalysis and MSC.Marc. I must also thank Dr. Anna-Karin Christiansson for her helpand support during my undergraduate degree and, not least, during my time as a PhDstudent. I would also like to thank Alastair Henry, Department of Social and BehaviouralStudies, for his careful linguistic revisions of both my thesis and my articles. Finally, Iwould like to express my thanks to Dr. Martin Friis (a former member at the VIP group)for his friendship and all the great moments we have shared both at the university and inour free time.

During a period of 2004 and 2005 I was seconded to the Volvo Aero Corporation andI would like to thank those people at the Advanced Materials and Manufacturing De-partment for creating such a friendly environment and for their hospitality. I would liketo express my gratitude and appreciation to a number of people at the Volvo Aero Cor-poration. In particular, I would like to thank Per Henrikson, for his help and supportwith temperature measurements, Börje Nordin for his willingness to share his knowledgeabout robotised TIG welding, Peter Johnsson and Jan Lundgren for our many valuable

vi Acknowledgments

discussions and for the support they provided with validation components, Dr. HenrikRunnemalm for the support and the opportunity to work in the department. Finally, Iwould like to thank Dr. Daniel Berglund (former of the Volvo Aero Corporation) forintroducing me to the wonderful world of finite element analysis.

Thanks also go to the research team of the robotic division at Faculty of Engineering,Lund University for all their help and assistance. The project was funded by the Foun-dation for Knowledge and Competence Development, EC Structural Funds and VITAL(EC, Sixth Framework Programme).

Last, but not least, I would like to take this opportunity to express my deepest gratitudeto my parents, Sten and Ingrid, to my brother Stefan and to my wife Anna, for all theirsupport and understanding during my research.

Mikael EricssonMay 2006Trollhättan

Contents

Abstract iii

Acknowledgments v

List of Figures xi

List of Tables xiii

Nomenclature xv

1 Introduction 1

1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Scope and Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.4 Research Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.5 Experimental Equipment . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.6 Outline of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Welding Theory 7

2.1 Principle of Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1.1 TIG Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1.2 Laser Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Heat Effects of Welding . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.1 Temperature Fields . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.2 Residual Stresses and Distortion . . . . . . . . . . . . . . . . . . 17

viii Contents

3 Modelling Techniques 21

3.1 General Principles of the Off-Line Programming of Robots . . . . . . . . 21

3.2 Off-Line Programming in the Present Study . . . . . . . . . . . . . . . . 23

3.2.1 Off-Line Programming of Arc Welding . . . . . . . . . . . . . . 23

3.2.2 Off-Line Programming of Metal Deposition . . . . . . . . . . . . 24

3.3 General Principles of Finite Element Modelling of Welding . . . . . . . . 27

3.3.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . 27

3.3.2 Material Modelling . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.3.3 Material Properties . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.4 FEM-modelling in the Present Study . . . . . . . . . . . . . . . . . . . . 29

3.4.1 Boundary Conditions in the Present Study . . . . . . . . . . . . . 29

3.4.2 Material Properties . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.4.3 Properties for the Thermal-Mechanical Modelling . . . . . . . . . 34

3.5 Principle of the Integration between the Off-Line Programming Modeland the Finite Element Analysis Model . . . . . . . . . . . . . . . . . . . 34

3.5.1 Integration of OLP and FEA Models . . . . . . . . . . . . . . . . 34

3.5.2 Synchronized Time Domain . . . . . . . . . . . . . . . . . . . . 37

4 Model Validation Techniques 41

4.1 OLP Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.1.1 Signature Calibration . . . . . . . . . . . . . . . . . . . . . . . . 41

4.1.2 Tool Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.1.3 Work Cell Calibration . . . . . . . . . . . . . . . . . . . . . . . 42

4.1.4 Robot Position Calibration . . . . . . . . . . . . . . . . . . . . . 43

4.2 Temperature Measurement Techniques . . . . . . . . . . . . . . . . . . . 43

4.2.1 Thermocouple Instrumentation on Plates . . . . . . . . . . . . . 43

4.2.2 Infrared Imaging Measurement Techniques . . . . . . . . . . . . 44

4.3 Residual Stress Measurements Techniques . . . . . . . . . . . . . . . . . 45

5 Results and Summary of Appended Papers 47

5.1 Paper I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5.2 Paper II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

5.3 Paper III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.4 Paper IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

Contents ix

5.5 Paper V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5.6 Paper VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

6 Discussion 49

6.1 Accuracy in the OLP Model . . . . . . . . . . . . . . . . . . . . . . . . 49

6.2 Parameter Sensitivity Study . . . . . . . . . . . . . . . . . . . . . . . . 49

6.2.1 Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

6.2.2 Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

6.2.3 Results Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . 50

7 Summary and Conclusion 55

References 57

Paper I Three-Dimension Simulation of Robot Path and Heat Transfer of aTIG-Welded Part with Complex Geometry 61

Paper II Non-contact Temperature Measurements using an Infrared Camerain Aerospace Welding Applications 75

Paper III Three Dimensional Simulation of Robot path, Heat Transfer andResidual Stresses of a welded Part with Complex Geometry 93

Paper IV Computer Aided Robotics combined with a Finite Element Analysisfor Process Simulation of Welding 115

Paper V Optimization of robot welding speed based on process modeling 127

Paper VI Off-Line Programming or Robots for Metal Deposition 147

x Contents

List of Figures

2.1 Principle of Tungsten Inert Gas welding. . . . . . . . . . . . . . . . . . . 8

2.2 Principle of laser welding. . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.3 Differential control volume, dx, for conduction analysis in Cartesian co-ordinates (Incropera & DeWitt 1996). . . . . . . . . . . . . . . . . . . . 12

2.4 Schematic of the welding thermal model. . . . . . . . . . . . . . . . . . 14

2.5 Temperature [K] contour plots with different welding parameters. Up-per left: welding speed 2.0 mm/s, welding current 100 A, Upper right:welding speed 3.0 mm/s, welding current 100 A. Lower left: weldingcurrent 100 A, welding speed 2.0 mm/s, Lower right: welding current 80A, welding speed 2.0 mm/s. . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.6 Important temperature characteristics for peak temperature and cooling rate. 16

2.7 Example of a distortion that can occur during welding (Cannon 1991). . . 18

3.1 IGRIP model of the experimental setup. . . . . . . . . . . . . . . . . . . 23

3.2 Aerospace component, whole part (top), 113 of the part (bottom) . . . . . 24

3.3 Complete geometry (top), sliced geometry (middle) and generated paths(bottom) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.4 Fe-C diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.5 The non-uniform mesh used in paper one. Note! Higher densities alongthe weld path. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.6 Shell model of a part of the aerospace component. . . . . . . . . . . . . . 30

3.7 a) Cross section of a plane plate mounted in a welding fixture. b) appliedboundary conditions in a heat transfer simulation. . . . . . . . . . . . . . 31

3.8 Heat flux Gaussian distribution with 5% cut off limit. . . . . . . . . . . . 32

3.9 Specific heat for 316L. . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

xii List of Figures

3.10 Conductivity for 316L. Conductivity without considering convection (a)and conductivity when weld pool convection is considered by increasingthe conductivity value above the melting point (b). . . . . . . . . . . . . . 34

3.11 Block chart showing the integration between OLP and FEM. . . . . . . . 35

3.12 Robot pose description for a path. . . . . . . . . . . . . . . . . . . . . . 36

3.13 Input file to the FEA simulation generated by the robot simulation program. 36

3.14 Overview of a cross section from an FEA simulation showing the pene-tration. The colour represents a temperature interval close to the meltingpoint. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.15 The overall architecture of the off-line simulation system. . . . . . . . . . 39

4.1 Schematic diagram of a plate with thermocouples together with selectedmeasurement lines for the IR camera measurements. . . . . . . . . . . . 44

4.2 Principle overview of the VarioScan 3021 high resolution camera. . . . . 45

6.1 Conductivity values used in the sensitivity study. . . . . . . . . . . . . . 51

6.2 Specific heat values used in the sensitivity study. . . . . . . . . . . . . . . 51

6.3 Responses selected from the temperature history. . . . . . . . . . . . . . 52

6.4 The influence of different factors on the peak temperature. . . . . . . . . 52

6.5 The influence of different factors on the width of the temperature. . . . . 53

6.6 Different temperature profiles at 3.6 mm from the center of the weld seam. 54

List of Tables

3.1 Boundary conditions in the heat transfer analysis in paper I . . . . . . . . 31

3.2 Material properties for stainless steel 316L and Greek Ascaloy. . . . . . . 33

6.1 Factors and corresponding values used in the sensitivity study . . . . . . 50

xiv List of Tables

Nomenclature

αq The concentration factor.

η The efficiency factor.

κ [m2/s] Thermal diffusivity of the base metal.

λc [W/(m·K)] Thermal conductivity.

λr Relaxation parameter.

ν [m/s] Welding speed.

ρ [kg/m3] Density.

τ Coefficient to determine thick or thin plate.

ξ [m] Moving coordinate.

Cp [J/(kg·K)] Specific heat.

d [m] Plate thickness.

dx infinitesimally small volume expressed in one dimension.

e The base of the natural logarithm.

E [V] Welding voltage.

h Surface heat loss coefficient.

Hnet [J/m] Net energy input, equal to Pg

ν .

I [A] Welding current.

K0 The modified Bessel function of the second kind, zero order.

L [J/m3] Latent heat of fusion.

lx, ly , lz The direction cosines to the boundary surface.

xvi Nomenclature

Pg [W] Generated effect.

q [W/m2] Heat flow density.

Q [W/m3] Internal heat generation rate.

q0 [W] Heat transferred to the workpiece.

qx [J/s] Condition heat rate in volume x.

R [K/s] The cooling rate at point at the weld centreline.

r [m] The radial distance from the centre of the heat source.

s0 [m/s] Input welding speed.

St [s] The solidification time of the weld metal.

T [K] Temperature.

T (x) [K] Temperature distribution expressed in Cartesian coordinates.

T0 [K] Initial temperature.

T∞ [K] Temperature environment.

Tc [K] Critical temperature for phase changes in the welded metal.

Tmax [K] Maximum temperature at each node.

Tm [K] Melting temperature.

Tp [K] Peak temperature.

x, y, z Cartesian coordinates that denote the welding direction, the transversedirection and the normal direction to the weld.

y [m] The distance from the weld fusion boundary where the peak tempera-ture is calculated.

A1 Temperature curve that determines transformation to perlite.

A3 Temperature curve that determines transformation to ferrite.

Acm Temperature curve that determines transformation to cementite.

AC Alternating Current.

C Carbon.

CAD Computer Aided Design.

CAM Computer Aided Manufacturing.

Nomenclature xvii

CAR Computer Aided Robotics.

CFD Computational Fluid Dynamics.

DC Direct Current.

DoE Design of Experiments.

emf Electromotive force.

Fe Iron.

FEA Finite Element Analysis.

GTAW Gas Tungsten Arc Welding.

HAZ Heat Affected Zone.

IR Infrared.

MAG Metal Active Gas.

MD Metal Deposition.

MIG Metal Inert Gas.

Nd:YAG Nitrogen:Ytterium Aluminum Gardner.

PLS Partial Least Squares.

RP Rapid Prototyping.

RRS Realistic Robot Simulation.

T/C Thermo Couples.

TCP Tool Centre Point.

TIG Tungsten Inert Gas.

xviii Nomenclature

Chapter 1

Introduction

1.1 Background and Motivation

Arc welding equipment was originally designed to be used manually but, during theprocess of industrial evolution and through the introduction of robots into automaticwelding processes in the 1970s, it is today one of the most common tasks carried outby industrial robots. Examples of the driving forces underpinning this automation arethe demands to increase productivity, achieve higher quality standards and the demandfor ever-increasing flexibility (O’Brien 1991), (Olsen, Siewert, Liu & Edwards 1997).Automatic TIG (Tungsten Inert Gas) welding is however still rather rare, since it placeshigh demands on equipment and part geometry accuracy. Within the aerospace industry,which is the application area for this research, manual TIG-welding is one of the mostcommon welding processes. This is due to high product requirements for materials withhigh heat and corrosion resistance, and which must also have good fatigue properties andlow weight. Examples of such materials are Inconel 718 and Greek Ascaloy, which canbe successfully joined by TIG welding resulting in joints with few defects and, in com-parison with other welding processes, low distortion. In recent years the laser weldingprocess has become increasingly popular within the aerospace industry. Compared toTIG welding, laser welding gives a high quality result using less heat input to the mater-ial. It is also, compared to TIG welding, a much faster welding process (O’Brien 1991),(Ion 2005). However, any welding process induces changes in the base material and gen-erates unwanted stresses and deformation due to the heat input. The most common wayto avoid such deformation is to use fixtures to clamp the part to be welded. Unfortunately,these fixtures are difficult to design, time consuming to construct and very expensive.

Another common method to reduce deformation is to optimise the welding sequence toallow a more uniform heat distribution into the part. This welding sequence is howeverhard to find and requires the work of a skilled operator. Consequently, a simulation toolthat can be used to evaluate fixture solutions and to plan welding sequences early in theproduct development stage, would be desirable. Such a simulation tool would reduce

2 Introduction

both the number of welding experiments and the need for welding operator experience.The tool should preferably be able to simulate the welding torch path, have the capacityto detect collisions between the torch and workpiece, and the ability to optimise weldingparameters relevant to penetration and component deformation.

Research in welding simulation can be divided in two main fields; robot simulation, whichis often refereed as CAR (Computer Aided Robotics) and thermomechanical modelling(Olsson 2002), (Bolmsjö, Olsson, & Brink 1996), (Radaj 1992), (Goldak, Chakravarti &Bibby 1984). CAR involves the simulation and programming of a robot task using a vir-tual model of the workcell and the part to be welded. Examples of research in this areaare the integration and development of virtual sensors and the optimisation of welding se-quences and torch trajectories to avoid collisions and to increase productivity (Cederberg,Olsson & Bolmsjö 2002), (Cederberg, Olsson & Bolmsjö 1999), (Ting, Lei & Jar 2002).

Thermal-mechanical modelling involves the modelling of the influence of the processon the component. It includes prediction of temperature histories, microstructure phasetransformations, residual stresses and distortion (Brust, Yang, Y.Dong & T.Jutla 2000).

This thesis addresses both these areas, i.e. CAR and thermomechanical simulations,through the integration of an off-line programming system with a finite element (FEA)modelling system. CAR is used to simulate arc welding torch paths, and to detect col-lisions between the torch and workpiece, as well as for generating robot trajectories forthe actual welding experiments. FEA is used for the prediction of temperature histories,residual stresses and fixture forces.

The industry interest in manufacturing simulation tools has increased substantially in re-cent years, which is why simulation has become an increasingly common tool to testand verify different approaches prior to manufacture. A simulation tool such of the typedescribed in this thesis would therefore be of great benefit to industry.

1.2 Objective

The objective of this research is to develop and validate a simulation methodology anda simulation tool for the welding process. The tool shall be capable of simulating torchpaths, predicting temperature histories, residual stresses and deformation, thus making itpossible to optimise welding sequences and fixture solutions before manufacturing. Ofparticular interest is whether models that are not unduly complex and which can be usedindustrially in the design and production engineering phases can be developed.

1.3 Scope and Limitations

Simulation of welding is a very broad field that encompasses a number of different tech-niques and disciplines. Models are continuously being developed for simulating the robotpath, the arc, the liquid pool, and for temperature, residual stress and deformation predic-tion. The different models involve disciplines such as plasma physics, electromagnetics,

1.4 Research Methods 3

fluid mechanics, solid mechanics, material science and production technology. The dif-ferent models also place different demands as regards time and space resolutions. Timescales range from microseconds to minutes, whilst length scales range from microme-ters up to decimeters and meters. Limitations are therefore necessary when a simulationmodel is to be developed. The limitations relevant to the current work are:

• No model of the arc or of the molten pool. Convective heat transfer within themolten pool is considered by increasing the heat conductivity when the temperatureincreases above the melting point.

• No models of fixtures. Reaction forces from fixtures are considered in the boundaryconditions.

• No development of material models such as phase transformations etc. Luleå Tech-nical University, Sweden and Volvo Aero Corporation, Sweden, have developedmaterial routines used.

• The study is limited to the TIG and laser welding processes, although the results areexpected to be generic and have a general applicability to other welding processes.

• The study is limited to two materials namely, Greek Ascoloy and Stainless Steel316L. The developed methodology is, however, not material dependent.

• The Synchronized Distribution Simulation Protocol described in chapter 3.5.2 wasdeveloped by Dr. Fredrik Danielsson (Danielsson 2002)

A focus is placed on CAR and the integration of CAR with FEA. The thermomechanicalsimulations have been made in collaboration with Luleå Technical University, Sweden(Berglund 2001), (Berglund 2003), (Alberg 2005) and the Volvo Aero Corporation. In thevalidation work, focus is temperature measurements.

1.4 Research Methods

Research has been performed in two areas:

• Applied research, which focuses on the observation, modelling and analysis of realworld phenomena. Results from this type of research can be used directly in prac-tical situations.

• Theoretical research, which deals with the development of new concepts based onproven scientific knowledge, which can subsequently be used in practical situations.

The research in this thesis has mainly been performed in the area of applied research,although efforts have been made to ensure that the models that have been developed have asolid theoretical foundation. The results are thus, primarily, of a practical nature, enablingthem to be used directly in industrial situations.

4 Introduction

The purpose of the research methodology is to identify, formulate and analyse an indus-trial problem in scientific terms. Furthermore, it should provide support in the searchfor and selection of, problem-solving strategies for implementation and validation. Thefollowing research methodology has been selected and used in the research work:

• Investigation: Published research has been evaluated whilst informal discussionswith industrial engineers have taken place with the aim of identifying, understand-ing, and defining industrial research problems.

• Evaluation: The selection of modelling techniques, methods and tools. Within thisphase the following modelling techniques were selected:

i. Robot trajectory generation by the use of CAR software.

ii. Heat transfer modelling within a workpiece by the use of FEA.

iii. Residual stress prediction by the use of FEA.

• Development: Identification of the parts in the selected models, methods and toolswhere software development is needed. The development of simulation softwarein-house or by application programming in commercial software.

• Validation: To use and evaluate the different models in an industrial environment.To disseminate and discuss research results with industrial representatives, at meet-ings seminars and conferences.

• If refinement is required, a repetition of the steps described above.

1.5 Experimental Equipment

Different experimental equipment and software have been used in this research. All weld-ing experiments were carried out with a robotised welding cell consisting of two six-axisrobots (ABB IRB 1400 and ABB IRB 4400) from ABB Automation Technology Prod-ucts AB Robotics, 721 68 Västerås, Sweden. For the TIG welding experiments, the IRB1400 was equipped with a torch and thoriated tungsten electrodes from Abicor BinzelAB, Karlskrona, Sweden. The power source was a TIG Commander 400 AC/DC fromMigatronic A/S, Aggersundvej 33, DK-9690 Fjerritslev, Denmark. The laser welding ex-periments were carried out using 2.3 kW Nd:YAG laser cw2500 from ROFIN-SINARLaser GmbH, NeufeldstraSSe 16, Günding D-85232 Bergkirchen, Germany, which wasmounted onto the IRB 4400 robot.

The robot simulations were performed using the IGRIP commercial software from Das-sault Systemes, Suresnes Cedex, France. The thermomechanical simulations were per-formed using the Marc system from MSC Software Corporation, Santa Ana, USA. Thecalculations were performed using two in-house developed Linux clusters consisting often 1.0 GHz Pentium III processors and ten double 1.7 GHz Pentium IV processors re-spectively.

1.6 Outline of Thesis 5

For temperature measurements, both thermocouples (type K) and infrared imaging wasused. A PC-based data acquisition system was used to sample the signals from the ther-mocouples and to write the data to disk. The infrared camera used was a VarioScan3021-ST high resolution 16 bit camera from Jenoptic GMbH, Jena, Germany.

1.6 Outline of Thesis

In this thesis, a method for off-line optimisation of welding by the use of simulation isproposed. The thesis starts in chapter 2 with an introduction to the theory of TIG weldingand its heat effects, such as residual stresses and distortion. Chapter 3 describes methodsfor robot simulation, as well as modelling techniques for the prediction of temperaturehistories, residual stresses and distortion. In chapter 4 different methods for the valida-tion of temperature, residual stresses and distortion are discussed. Results of publishedpapers follow in chapter 5 and discussions in chapter 6. Finally chapter 7 summaries andconcludes. Published papers are provided in the Appendix.

6 Introduction

Chapter 2

Welding Theory

This chapter provides a brief presentation of the theory of TIG and laserwelding and heat effects on the base material.

2.1 Principle of Welding

Welding is widely used for fabrication, maintenance, and repair purposes. While there aremany methods for joining metals, welding is one of the most convenient and time-efficientmethods available. The term welding refers to the process of joining metals by heatingthem to a plastic or molten state and causing the plastic/molten material to fuse together.There are several different techniques to weld and the most common ones are TungstenInert Gas (TIG), Metal Inert Gas (MIG), Metal Active Gas (MAG) and laser welding(Connon 1991), (Easterling 1992). In TIG, MIG and MAG welding, an arc is formedbetween the electrode and the workpiece. The principle of TIG welding is presented insection 2.1.1. In Laser welding, the workpiece is welded by the use of a concentratedlight beam (Ion 2005). The principle of laser welding is provided in section 2.1.2.

2.1.1 TIG Welding

The TIG welding process was invented during the Second World War due to the demandsof the American aircraft industry for a method of joining magnesium and aluminium.Russell Meredith (Meredith 1942) demonstrated the first TIG process for the weldingof magnesium using a Tungsten electrode and helium gas in the late 1930s (O’Brien1991). TIG welding or GTAW (Gas Tungsten Arc Welding which is the common namein North America) uses a non-consumable tungsten electrode protected by an inert gas.The electrode is either made of pure tungsten or tungsten, mixed with small amounts ofoxides (thoriumoxide, zirconiumoxide) which improves the stability of the arc and makesit easier to strike (Connon 1991). Since the process uses a non-consumable electrode,

8 Welding Theory

filler material is usually added. The principle of the process is schematically presentedin Figure 2.1. The electrical discharge generates a plasma arc between the electrode tipand the work piece to be welded. The arc is normally initialised with a power sourcefrom a high frequency generator, which produces a small spark that provides the initialconducting path through the air for the low voltage welding current. The frequency of thisignition pulse is large, up to several MHz. This frequency, together with a high voltage(several kV), produces strong electrical interference around the welding cell, which isa disadvantage when sensors and measuring equipment are used (O’Brien 1991). Thearc consists of a high-temperature conducting plasma that produces the thermal energyneeded to melt the base and the filler material. The arc temperature spans between 12000K and 15000 K above the pool surface and the temperature of the melted surface spansfrom 1700 K to 2500 K, dependent on the material (Easterling 1992), (Zacharia & Chen1998).

Weld Center Line

Arc

Gas Shield

Tungsten Electrode

Shieldinggas

CurrentConductor

Filler Metal

Solidified Weld Metal

Weld Pool

Figure 2.1: Principle of Tungsten Inert Gas welding.

Three different alternative types of current can be used, namely, direct current (DC) witha positive electrode, DC with a negative electrode or alternative current (AC) (O’Brien1991). AC is mainly used for the welding of aluminium and magnesium since cleaningof the oxide layer on the surface is best achieved using this method. DC with a negativeelectrode is used for most other materials, including thick plates of aluminium. Pulsedand non-pulsed currents can be used. A non-pulsed current is most common. The use ofa pulsed current has some advantages, such as increased penetration. Depending on the

2.1 Principle of Welding 9

thickness of the base material, type of joint and certain other factors, extra filler materialmight be needed. In automatic TIG welding, hot or cold wire can then be used (O’Brien1991), (Olsen et al. 1997). Cold wire is fed in at the front of the melted pool whilst hotwire is fed at the back. The filler material is usually the same as the base material. Aninert gas is used to sustain the arc and to protect the melted pool and the electrode fromatmospheric contamination. Depending on the welding parameters and welding materials,either argon, helium or a mix of the two gases can be used. Argon is commonly used inwelding unalloyed, low alloyed and stainless steels.

However, a mixture of argon and hydrogen or argon and helium can be used for mechani-cal welding. For duplex stainless steel, it is common to mix argon with nitrogen to ensurea correct ferrite/austenite balance. Aluminium and aluminium alloys are usually weldedusing argon. Addition of helium can be used to improve the heat transfer and is thereforesometimes used for the welding of thicker parts. Argon is suitable for welding copper andits alloys, and gives excellent results for thicknesses up to 6 mm. Helium, or a mixtureof helium and argon (up to 35%), are suitable for thicknesses greater then 6 mm. Tita-nium requires an extremely high purity of the shielding gas, usually not less then 99.99%.Either argon or helium can be used here. Argon is the more common shielding gas forthicknesses less than 3 mm, while helium is more commonly used for thicknesses in ex-cess of 3 mm. In stainless steel and other easily oxidised materials, applications of a rootgas can be used to protect the root side of the weld from oxidation. The root gas can be amixture of nitrogen and hydrogen, or pure argon.

2.1.2 Laser Welding

The principle of a laser was first presented in 1917, when Albert Einstein described thetheory of stimulated emission (Olsen et al. 1997), (Ion 2005). The term LASER is anacronym for "Light Amplification by Stimulated Emission of Radiation". In laser weldinga focused high-power coherent monochromatic light beam is used. At the focal point,the metal vaporizes, which produces as deep penetration column, often called "keyhole".This column is surrounded by a liquid pool. As the pulse ends, the liquid metal around the"keyhole" flows back in, solidifying and creating a small "spot" weld (Ion 2005). Severaltypes of laser power sources exist the most common ones are Nd:YAG (Neodymium-doped Yttrium Aluminum Garnet) Laser, Diode Laser and CO2 (Carbon dioxide) Laserand more recently Disc laser and Fibre laser

The main difference is the method by which the laser beam is generated. In the Nd:YAGlaser, the laser beam is generated by sending light through Nd:YAG crystals (Ion 2005).As a light source, both white light flashlamps and laser diodes can be used. The lattermethods are used to produce high quality beams, which can be focused to smaller spots(and therefore produce higher power densities) than the flashlamp pumped lasers (Ion2005).

In a diode laser, the laser beam is generated by the p-n junction (diode) within a multi-layer semiconductor structure. Each element generates a light beam, which has to beoptically processed and combined to a focal point.

10 Welding Theory

The active medium in a CO2 laser is a mixture of carbon dioxide, nitrogen and (generally)helium. It is the carbon dioxide which produces the laser light, while the nitrogen mole-cules help excite the CO2 molecules and increase the efficiency of the light generationprocesses. The helium plays a dual role in assisting heat transfer from the gas created bythe electric discharge used to excite the gas, and also helps the CO2 molecules to returnto their ground state.

There are two ways in which the laser beam can be delivered to the work piece. The firstinvolves the use of "hard optics", whilst the second involves the use of a fiber optic cable(Ion 2005). "Hard optics" basically means that the laser beam is deflected and focusedthrough the exclusive use of mirrors and lenses. Optical fibre means that the laser lightis created nearby the robot and guided through an optical wire to the TCP (Tool CentrePoint) at the robot (Ion 2005). In figure 2.2 a schematic picture of a welding process isshown.

Diods

Lens for focusingthe spot

Plate

Weld pool(key hole)

Laser beam

Figure 2.2: Principle of laser welding.

In order to make the seam as strong and as free from contamination as possible, it is nec-essary to use a shield gas to prevent a reaction in the melting pool (Ion 2005). The choiceof shielding gas depends on the material. Common shielding gases for laser welding areargon and helium.

The beam can be collimated and refocused onto the work piece by the use of mirrors.This method makes it possible to accurately deliver the beam to the required area. Italso enables movement of the focusing optics instead of, or in addition to, moving theworkpiece itself.

2.2 Heat Effects of Welding 11

2.2 Heat Effects of Welding

Heat transfer phenomena play an important role in welding. Heat effects of welding referto temperature fields, residual stresses and distortion that occur during or after welding.Since focus in this thesis is placed on temperature prediction, a thorough description oftemperature fields is provided below.

2.2.1 Temperature Fields

One objective of heat transfer analysis in welding applications is to determine the temper-ature fields in an object resulting from the conditions imposed on its boundaries (Incropera& DeWitt 1996). The quantity that is sought is the temperature distribution, which repre-sents how temperature varies as a function of the time in the object. When this distributionis known, the conduction heat flux calculated at any point in the medium, or at the surface,may be computed using Fourier’s law (Incropera & DeWitt 1996). The temperature fieldsduring welding are highly heterogeneous and transient. The temperature of a componentcan vary from below zero to 3000 centigrade, i.e. the evaporation temperature of themetal. Within this range of phase changes, micro structural transformations and thermalstrains take place, all of which determine the nature of residual stresses and distortion.Fourier’s law states that the heat flow density q [J/m2] is proportional to the negativetemperature gradient δTδn [K/m] by the equation (Incropera & DeWitt 1996)

q = −λc ∂T∂n

(2.1)

where λc [J/(m·s·K] denotes the thermal conductivity and T [K] the temperature. Con-sider a homogenous medium expressed in one dimension xwith a temperature distributionT (x) expressed in Cartesian coordinates with an infinitesimally small control volume dx.The condition heat rate at the control area can thus be expressed as qx. The conditionheat rate at the opposite surface can be expressed as a Taylor series expansion neglectinghigher order terms as (2.2)

qx+dx = qx +∂qx∂x

dx (2.2)

Inside the control medium, see figure 2.3 an energy source term Eg (2.3) and an energystorage term Est (2.4) can be expressed as

Eg = qdx (2.3)

Est = ρCp∂T

∂tdx (2.4)

12 Welding Theory

dx

qx

qx+dx

Est

Eg

Figure 2.3: Differential control volume, dx, for conduction analysis in Cartesian coordi-nates (Incropera & DeWitt 1996).

Using the law of energy conservation equations 2.3 and 2.4 can be substituted to

qx + qdx− qx+dx = ρCp∂T

∂tdx (2.5)

Substitution from equation 2.2 and Fourier’s law the heat diffusion equation can be writtenin a general form, in three dimensional Cartesian coordinates as (Incropera & DeWitt1996),

∂

∂x[λc(T )

∂T

∂x] +

∂

∂y[λc(T )

∂T

∂y] +

∂

∂z[λc(T )

∂T

∂z] = ρCp(T )

∂T

∂t−Q (2.6)

where the Cartesian coordinates x, y and z denote the welding direction, the transversedirection, and the normal direction to the weld, respectively, see figure 2.4. Q [W/m3]stands for internal heat generation rate and the material properties, thermal conductivity,density, and specific heat are denoted by λc, ρ and Cp respectively.

Several possibilities for initial conditions exist. The most common is: T = T0 at t = 0.A general boundary condition can be written as (Olsen et al. 1997):

λc∂T

∂xlx + λc

∂T

∂yly + λc

∂T

∂zlz − q + h(T − T∞) = 0 (2.7)

where h denotes surface heat loss coefficient, lx, ly and lz the direction cosines to theboundary surface. The surface temperature and environment temperature are denoted byT and T∞ respectively.

The heat diffusion equation, (2.6), can be solved both analytically and numerically (in thelatter case, the FEA is commonly used, which is further presented in section 3.3). Theequation can be analytically solved assuming the following conditions (Connon 1991):


• The energy from the welding heat source is applied at a uniform rate.

• The heat source is moving with a constant speed.

• The cross section of the work piece is constant.

• Constant material properties are used.

• The end effects resulting from the initiation and termination of the arc weld areneglected (quasi-static solution).

Different analytical solutions exist depending on the plate thickness and welding posi-tions. The plate can be assumed to be thick, thin and finite, respectively. The dimen-sionless τ is used to determine whether the plate is to be considered as thin or thick, seeequation 2.8 (Connon 1991). The plate is considered to be thin if τ is less the 0.75 and tobe thick if τ is larger than 0.75. In a thick plate the heat flow is considered to be three di-mensional, through the plate thickness and lateral from the weld. The thin plate equationcan be applied where the heat flow is essentially lateral. This means that the differencein temperature between the bottom and top surface is small in comparison to the meltingtemperature.

τ = d

√ρCp(Tc − T0)

Hnet(2.8)

where d denotes the plate thickness. Hnet is net energy input equal to ηPg

ν . Pg denotesthe generated effect, Pg = EI in TIG welding. T0 stands for the initial plate temperature,Tc denotes the temperature at which the cooling rate is calculated.

For an analytical quasi-static solution of the heat transfer model it is assumed that thematerial properties are independent of the temperature, that the metallurgy zones arehomogenous, and that the thermal model is linear in the welding direction. The solu-tion gives the temperature in a specific point if the welding speed (ν), energy heat input(E, I, ν) and the material properties (ρ, λc, Cp) are known. This point is defined by r:

r =√ξ2 + y2 + z2 (2.9)

where ξ denotes a moving coordinate,

ξ = x− νt (2.10)

Here the origin of the moving coordinates (ξ,y,z, se figure 2.4) is fixed at the centreof the heat source see figure 2.4. This means that the coordinates move with the heat

14 Welding Theory

Weld centre line

Welding direction

Movingcoordinate

ξ

yGlobalcoordinate

x

y

Figure 2.4: Schematic of the welding thermal model.

source at the same speed. Solutions are usually derived separately for thin and thickplates respectively.

Analytical solutions were first presented by D. Rosenthal 1935 (Easterling 1992), (Connon1991) and (Olsen et al. 1997).

T − T0 =ηPq

2πλcre[

−ν(ξ+r)2κ ] (2.11)

for a thick plate and

T − T0 =ηPq

2πλcde

−νξ2κ K0(

νr

2κ) (2.12)

for a plate considered to be thin, where κ denotes thermal diffusivity of the base metalgiven by κ = λc

ρCp, T0 denotes the preheat temperature in the base metal, d stands for plate

thickness and K0 denotes the modified Bessel function of the second kind, zero order.This relationship for the temperature heat flow is not accurate close to the welding arc.Since a point or a line source is assumed for thick and thin plates respectively, singularitieswill occur at the source locations where the temperature tends to infinity.

Figure 2.5 shows the influence of welding parameters on surface temperature. The weld-ing speed and welding current have been varied and the temperature distribution has beensolved using equation 2.12. In the upper two figures and in the lower the welding current,the welding speed has been varied. It can be seen that both the welding speed and currenthave a strong influence on the heat distribution. Several welding defects, such as resid-ual stresses and distortion (see section 2.2.2) are dependent on the heat input. If the heatinput can be minimised, whilst still maintaining a full penetration, these defects will de-crease. The thermal condition in, and close to, the weld is very important since it controlsthe metallurgical events in the weld. Parameters requiring control are; the distribution ofpeak temperature in the Heat Affected Zone (HAZ), cooling rates in the weld metal andin the HAZ, and the solidification rate of the weld metal, figure 2.6.


-0.4 -0.2 0 0.1

-0.03

0

0.03400

400

800 1200

1700

-0.4 -0.2 0 0.1

-0.03

0

0.03

400

400

800 1200

1700

-0.4 -0.2 0 0.1

-0.03

0

0.03400

400

800 1200

1700

-0.4 -0.2 0 0.1

-0.03

0

0.03400

400

800 1200

1700

Figure 2.5: Temperature [K] contour plots with different welding parameters. Upper left:welding speed 2.0 mm/s, welding current 100 A, Upper right: welding speed 3.0 mm/s,welding current 100 A. Lower left: welding current 100 A, welding speed 2.0 mm/s, Lowerright: welding current 80 A, welding speed 2.0 mm/s.

The peak temperature close to the weld seam is given by (Connon 1991)

1Tp − T0

=√

2πeρCpdyHnet

+1

Tm − T0(2.13)

where peak temperature, base metal melting temperature, initial base temperature aredenoted by Tp, Tm and T0 respectively. The material properties, density, and specific heatare denoted by ρ and Cp. y denotes the distance from the weld fusion boundary wherethe peak temperature is calculated, d states the plate thickness and e denotes the base ofthe natural logarithm. Hnet is net energy input equal to ηEI

ν (for TIG welding). Thisequation can be used to predict peak temperature at a specific point in the HAZ, the widthof the HAZ, as well as the effect of preheat on the width of the HAZ.

If the cooling rate for a specific point along the weld line is known, a prediction of themetallurgy in the welded area can be made. Cooling rates are important in the weldingof heat-treatable steels. This is due to the formation of martensite in the welded area.In the case of carbon and low alloy steels, the temperature at which the cooling rate is

16 Welding Theory

Peak temperature

Cooling rate

Time (s)

Tem

pera

ture

(K

)

Figure 2.6: Important temperature characteristics for peak temperature and cooling rate.

calculated is not critical. Therefore, the major use of cooling rates is to calculate thepreheating temperature (Connon 1991). The general cooling rate equation can be defined,using the moving coordinate ξ (2.10), which gives ∂ξ

∂t = −ν. Using the chain rule thecooling rate equation can be written as (Olsen et al. 1997)

∂T

∂t= −ν ∂T

ξ(2.14)

An analytical solution of the cooling rate can be defined for both a thick (2.15) and a thinplate (2.16), respectively (Connon 1991),

R =2πλc(Tc − T0)2

Hnet(2.15)

R = 2πλcρCp(d

Hnet)2(Tc − T0)3 (2.16)

where R is the cooling rate at point at the weld centreline just at the moment when thepoint is cooled past the Tc temperature. Tc denotes the critical temperature for phasechanges in the welded metal. The material properties, thermal conductivity, density, andspecific heat are denoted by λc, ρ, and Cp. Hnet is net energy input equal to ηEI

ν (forTIG welding).

The solidification rate can have an important impact on the metallurgical structure, prop-erties, and material response to heat treatment. The solidification time St of the weldmetal, is given by


St =LHnet

2πλcρCp(Tm − T0)2(2.17)

where L is the latent heat of fusion.

2.2.2 Residual Stresses and Distortion

Residual stresses are self-balanced internal stresses which exist in the component withoutany external loads and can be classified as macrostresses and microstresses (Lin 1988),(Connon 1991). The definition of macrostresses is that they are self-equilibrated in a crosssection of the manufactured part. Microstresses can be defined as stresses that are eitherhomogenous or inhomogeneous in a microscale (Lin 1988). They are introduced in thecomponent as results from the manufacturing process.

Since the welding process heats the material locally the temperature distribution is notuniform. In the melted weld pool stresses are released and can be assumed to zero. Dur-ing the solidification of the melted weld pool, the metal starts to shrink and starts to exertstresses on the surrounding weld metal and HAZ. Another source of stresses derives fromthe microstructure transformation during cooling. These stresses remain in the materialafter welding and result in unwanted distortion. During the welding sequence, mechani-cal properties change as a function of the temperature. The main mechanical propertiesare elastic modules, yield strength and Poisson’s ratio. These properties can change dras-tically at temperatures as low as 500◦C-600◦C (Easterling 1992).

Another contribution to the residual stresses is the dilatation due to phase transformation(Easterling 1992). The temperature for phase transformation depends on different factors,such as the grain size, peak temperature and cooling rate. The stress generated by thephase transformation interacts with the thermal stress. It is not necessary that the thermaland transformation stresses generate higher final residual stresses.

The stress level in the solidification area is proportionately low, but the stress level in theweld area increases and can be as high as the yield limit of the base material, which cancause unwanted fractures. Stresses in a welded plate can be divided in two directions;transverse and longitudinal to the weld.

Longitudinal residual stresses can arise from different causes, the most common of whichis the longitudinal contraction of the weld as it cools down. Transverse residual stressesare generated by the transverse contraction of the weld during the cooling phase. It canalso be generated indirectly due to the longitudinal contraction (Radaj 1992).

Different types of distortion can be found in manufactured structures as a result of residualstresses induced in welding (Connon 1991). A typical example of distortion is given infigure 2.7. Longitudinal and transverse shrinkage can cause distortion in the plane of theworkpiece. Plane or axisymmetrical angular shrinkage can cause distortion perpendicularto the plane of the welded component. Another distortion is bending caused by grids withlongitudinal and transverse welds (Connon 1991), (Easterling 1992).

18 Welding Theory

Bending Shrinkage

Longitudinal Shrinkage

Transverse Shrinkage

Angular Shrinkage

Figure 2.7: Example of a distortion that can occur during welding (Cannon 1991).


Several principal factors affect the type and size of the distortion (Lucas 2004). The mainfactors are parent material properties, the amount of restraint, the joint design, the partfit-up and the welding procedure.

Distortion can be restricted or indeed eliminated by making changes to the design andmanufacturing procedures. Changing welding process from TIG to laser will generateless heat in the part material, leading as a result to reductions in distortion. Anotherdesign issue is to change the welding sequence. By alternating the welding sequences orby changing fixture design, residual stresses can be minimized.

20 Welding Theory

Chapter 3

Modelling Techniques

This chapter describes the modelling techniques used in this thesis. A presen-tation of the off-line programming (OLP) of robots is provided first, followedby a presentation of temperature and residual stress prediction by the use ofthe Finite Element Method (FEM).

3.1 General Principles of the Off-Line Programming of Robots

Computer Aided Robotics (CAR) is a graphical tool for manufacturing simulations thatcan be used in several applications, such as the off-line programming of robots, tele-operation of robots and the simulation of general kinematic systems. The off-line pro-gramming of robots using a graphical simulation tool was first demonstrated in the be-ginning of the 1980s. Developments in computer technology have significantly improvedthe precision of off-line programming. The simulation of the production of a componentmakes it possible to test, verify and optimise robot motions and to design fixtures andautomation equipment before the real production process begins (Yong, Gleave, Green,& Bonne 1985), (Ting et al. 2002). Accessibility, collisions and timing can be verifiedbefore expensive machine tools and robots are purchased. Further, production accuracycan be improved by using off-line programming of robots, thus avoiding the complexityinherent in manually programming a robot in three dimensions with a high degree of ac-curacy. The most common method to manually program a robot is still that an operatorjogs the robot arm to each coordinate pose in space, a procedure that is both error-proneand heavily dependent on the experience and skill of the operator. The application ofthree-dimensional computer graphic tools for production planning makes it possible toachieve high degrees of accuracy for complex parts.

The technique of off-line programming is commonly used in industry. Several commer-cial software packages, such as IGRIP, RobCad and Grasp, are used in industry. Allof these systems use essentially the same work order (Bolmsjö, Olsson & Brink 1997),

22 Modelling Techniques

(Yong et al. 1985):

1. Modelling of the work cell

2. Work cell calibration

3. Programming of robot and other optional work cell equipment

4. Down loading of the program to the control system

5. Additional robot programming

6. Test running

In the first step the geometrical model of the work-piece, together with a geometric anda kinematic model of the workcell, including the most important parts such as fixtures,robots and positioners, are created. The geometrical model can either be created in aCAD/CAM software and imported to the robot simulation software, or created directlyin the robot simulation software. Most of the robots currently on the market incorporatepredefined robot simulation softwares, which can readily be retrieved from a software li-brary. If a new design for a robot or a new kinematic device is to be used, a kinematicmodel has to be created. This naturally involves the modelling of different parts of thekinematic devices, such as robot arms. These devices are usually created in succession,starting with the base part and ending with the tool tip, by connecting the parts at joiningpoints and describing movement patterns and boundary conditions for each joint. The sec-ond step is the calibration of the workcell. It can include several sub-steps, such as TCP-(Tool Centre Point), workpiece- and signature calibration. A tool calibration is usuallyperformed by rotating the real robot TCP around a sharp fixed point. Each new position isthen stored and uploaded into the robot simulation program and a "best fit" is calculatedusing statistical regression. Similarly, the position of the workpiece is calibrated by mov-ing the robot to identified positions. To find errors in the geometrical model of the robot,an arm signature calibration can be used. This calibration finds the deviations between thephysical and virtual model in the lengths between the robot joints and in the zero poses forthe joints. Corrections to the virtual model are then made. A more detailed descriptionof calibration is provided in the OLP modelling validation section in chapter 4. In thethird step the programs for all the different devices are written. Three categories of com-mands are commonly used. The first category includes commands for the visualisationof the simulation. Examples of this type of command are graphical commands, such asviewpoint rotation and zooming towards or outwards from an object. These commandsare not included in the robot code and consequently not downloaded to the robot. Thesecond category concerns commands that are used in the simulation, but which are alsoincluded in the robot code. Movement commands for the robot are typical examples ofthis category. For a welding application, this category will also include commands suchas ignition and termination of the arc. The final category of commands concerns com-mands that are not used in the simulation but which are directly downloaded to the robot.Examples of commands in this category are I/O’s, such as a gas protection which can be

3.2 Off-Line Programming in the Present Study 23

set to either ’on’ or ’off’ during welding. Step four is to translate the simulation pro-gram to the specific robot code and to download it to the real robot control system. Stepfive concerns additional robot programming and adjustments that have to be performedon-line at the robot. In step 6 a test run of the program is performed. If the test result issatisfactory, then production can be commenced. Steps 1-6 might, however, have to be re-peated iteratively until the required accuracy is achieved. Accuracy does not only dependon the effectiveness with which steps 1-6 have been performed, but also on how well thesimulation control system emulates the physical control system. A special module calledRRS (realistic robot simulation) (Bernhardt, Schreck & Willnow 1995) can be used toincrease the agreement between the physical and the virtual control system. Using RRS,the original control system software for motion interpolation and transformation can beintegrated into the CAR system.

3.2 Off-Line Programming in the Present Study

3.2.1 Off-Line Programming of Arc Welding

The method used to program robots off-line, described above, is the one that has beenadopted in this study. An in house welding cell, see figure 3.1, was modelled in IGRIP(Interactive Graphics Robot Instruction Program, Deneb Robotics). Figure 3.1 shows asnapshot of the experimental setup.

Figure 3.1: IGRIP model of the experimental setup.

Both plane plates and components with complex geometry were modelled and used astest cases. Both a tool and a workpiece calibration were performed, see section 4.1.2 and4.1.3. A high-level graphical simulation language in IGRIP, GSL, was used for program-ming all devices in the cell. An example of a modelled part was a section of an aerospace


part, namely a turbine component from the V2500 engine produced by Volvo Aero Cor-poration, figure 3.2, top. A 1/13 fragment was cut out of the real part, which originallyconsisted of an inner and outer ring and 13 vanes, figure 3.2, bottom. All componentswere created in the UniGraphics CAD/CAM system and imported using both a directtranslator and the neutral interface IGES. No RRS model was used. As another test case,an experimental aerospace component was used.

Vane

Outer ring

Inner ring

Outer ring

Vane

Outer ring

Figure 3.2: Aerospace component, whole part (top), 113 of the part (bottom)

3.2.2 Off-Line Programming of Metal Deposition

One special type of welding is metal Rapid Prototyping (RP), often referred to as MetalDeposition (MD). Using MD, a new component can be manufactured or a worn part re-paired (J.Mazumder, D.Dutta, N.Kikuchi & A.Ghosh 2000). The technique uses a weld-ing heat source to melt a powder or a wire material, which subsequently solidifies on a

3.2 Off-Line Programming in the Present Study 25

surface, thus enabling a part to be built up, drop-by-drop. To ensure repeatability and highquality, a robot is used. In this work three different methods for programming of the robotwere evaluated:

1. Reversed milling using a CAM module

2. Adapted Rapid Prototyping

3. Adapted Computer Aided Robotics

The general principle for off-line programming for all three methods was 1) creation of aCAD model of the part, see top of figure 3.3, 2) slicing the CAD model into thin cross-sectional layers, see middle of figure 3.3 and 3) definition of geometry paths for eachlayer, see bottom of figure 3.3.

In the first method (Reversed milling using a CAM module), CAD/CAM software wasutilized. In CAD/CAM applications, the Computer Aided Manufacturing (CAM) moduleis normally used to define turning or milling paths. Such a milling path was, in the currentwork, instead used to define the welding path. Application programming was used to slicethe model and to construct robot paths for each layer. The robot paths were then exportedto CAR software where a robot simulation of the movement was conducted. Finally, therobot paths were translated to robot code and downloaded to the robot.

In the second method that was evaluated (Adapted Rapid Prototyping), a dedicated RapidPrototyping software was used to slice the model and to generate the robot paths. Thissoftware is normally designed to create 3D polymer prototypes from CAD drawings.These applications use a much smaller dimension of the additative material (0.03 to 0.1mm) which is why a scaling procedure is needed. This was solved by means of applicationprogramming. Following the generation of robot paths, the paths were then exportedto the CAR software and a simulation was performed. Finally, the scaled paths weredownloaded to the robot.

The most interesting solution for the generation of robot trajectories for Metal Depositionwas exclusive use of CAR software (adaptive Computer Aided Robotics). The biggestdisadvantage with this method, though, is the lack of predefined functions for path gener-ation, which include both slicing and the creation of robot motion for each layer, which iswhy extensive application programming was necessary. The major advantage, comparedto the other two methods, was a better overview over the complete process. In the caseof simple geometries, a stand-alone program was suggested. Such a program functionsinteractively, asking the operator for geometry information, such as component height,width and position. For cases where more generalized shapes were needed, the path wasgenerated by implementing functions directly in the CAR software. Information aboutthe weld path, as well as information about the welding parameters, was finally exportedto the robot.


Figure 3.3: Complete geometry (top), sliced geometry (middle) and generated paths (bot-tom)

3.3 General Principles of Finite Element Modelling of Welding 27

3.3 General Principles of Finite Element Modelling of Welding

Numerical methods have been used since the beginning of the 1970s to simulate weldingprocesses. The focus has been the prediction of thermal histories, residual stresses anddistortion before the real welding process begins. FEA simulations have been the nu-merical method most commonly used and many papers on this topic have been presented(Berglund 2001), (Tosello, Tissot & Barras 1997), (Ueda & Yamakawa 1971) and (Hibbit& Marcal 1973). Large complex simulation models of three-dimensional componentsare, however, still rare, mainly due to the lack of computational power. One reason is that,in order to be able to compute the temperature and residual stress fields in the affectedzone, a very fine discretization of the space variable is required. Another complexity con-cerns the heat transfer between the electrode and the part to be welded (Zacharia, Vitek,Goldak, DebRoy, Rappaz & Bhadeshia 1995), (Zacharia & Chen 1998). This is a com-plex phenomenon with several interaction effects. The phenomenon can be divided intothree different groups, the plasma arc, the weld pool and the solid material. Modelling theplasma arc is complicated since chemical reactions, ionizations, and vaporisation both ofelectrodes and at the surface of the weld pool, have to be considered (Li & Wu 1997). De-veloping a model of the weld pool is also complicated since, as a result of various drivingforces, such as surface tension forces, electromagnetic and buoyancy forces, the meltedmaterial undergoes vigorous circulation (Zacharia et al. 1995). The weld pool surface isalso strongly influenced by a drag force that depresses the surface and induces a surfaceflow. The solidification process in the liquid-solid boundary region is also complicated,and thus difficult to model. Different assumptions and simplifications must therefore betaken into consideration when building an FEA model of the welding of a complex part.Examples of areas that have to be considered are part geometry simplifications, the typeof material models to be used, load conditions, heat transfer and other boundary con-ditions and numerical strategies (Berglund 2003), (Lindgren 2001a), (Lindgren 2001b),(Lindgren 2001c), (Lindgren 2001d). The most important considerations that have to bemade are described in the following sections.

3.3.1 Boundary Conditions

All FEA problems are defined in terms of initial and boundary conditions. A typical typeof initial condition for a welding application is the initial temperature, which, in mostcases, is set to room temperature. Examples of important boundary conditions are fixtureforces, and heat transfer coefficients between the part and its surroundings. Since the heattransfer between the electrode and the part to be welded is usually too complex to beintegrated in the same model, a simplified heat source with empirical parameters has tobe used. Different types of such heat sources have been suggested (Goldak et al. 1984),(Radaj 1992). Three main methods are usually used. The first method is to apply anenergy source within the part to be welded. The size and energy density are then adjustedto retrieve a fusion zone with good agreement with the real weld. The second methodis to use a surface distribution to simulate the arc (Radaj 1992). Here the energy sourceheat flux depends on the distance from the centre of the arc. A very common distribution


used is the Gaussian surface distribution. The final method is to use a double ellipsoidheat source, which was first recommended by Goldak (Goldak et al. 1984). This methodcombines the first two types, since it includes a surface distribution as well as a heatsource within the material. Although this method is the most realistic, it requires moreparameters to be fitted by means of experiments.

3.3.2 Material Modelling

The melting temperature for a pure metal is constant. However a mixed (duplex) metal(such as stainless steel) will have two critical temperatures for the solidification of themelted metal. The first one is the liquidus temperature and denotes the starting tempera-ture for the solidification. The second temperature is the solidification temperature, whichdenotes the temperature when the completely melted metal has changed to solid state. Inthe interval between the liquidus and solidus temperatures, both melted and solid metalco-exist.

Iron (Fe) mixed with carbon (C) follows a phase diagram such as in Figure 3.4. Fortechnical applications, iron can be mixed with up to 6.7% carbon. This mix gives anintermediate phase called cementite.

1 2 3 4 5 6

600

800

1000

1200

1400

1600

727 oC

1148 oC

Melted iron

Austenite, γ

Cementite

A1

AcmA

3

Ferrite, α

δ

% Carbon

Tem

pera

ture

(

o C)

Figure 3.4: Fe-C diagram.

Two specific points, called the eutectoid phase transformation, exist in the Fe-C chart.The first one is at 727 ◦C (austenite transforms to ferrite and cementite) while the secondis at 1148 ◦C (melted metal transforms to austenite and cementite). The temperaturecurve which controls transformation to ferrite is commonly denoted by A3, Acm is thecorresponding curve for transformation to cementite and A1 corresponds to the curve for

3.4 FEM-modelling in the Present Study 29

perlite transformation, see figure 3.4 (Landcaster 1999).

To obtain a desired final phase composition according to the Fe-C phase diagram andto allow phase changes to take place, a long period of time and a slow cooling rate arerequired. When a faster cooling rate is desirable, different types of mixed microstruc-tures can be generated. Austenitic can transform to perlite with a slow cooling, whereasbainite requires somewhat more rapid cooling and martensite requires very rapid cooling(Landcaster 1999).

3.3.3 Material Properties

The material properties that have to be included when temperature simulations are to beperformed are specific heat, heat conductivity, density, and liquidus and solidus temper-atures. The conductivity is usually temperature dependent. Weld pool convection is acomplex phenomenon that is difficult to simulate. This convection is therefore simulatedby multiplying the thermal conductivity by a certain factor when the temperature exceedsthe liquidus temperature. Different factors have been proposed in the literature (Goldak& Akhlaghi 2005), (Lindgren 2001b), (Michaleris & DeBiccari 1997). Common valuesare eight and ten. The specific heat for most welding materials are strongly tempera-ture dependent (Zhu & Y.J.Chao 2002), (Goldak & Akhlaghi 2005). The value increasessignificantly during fusion due to the latent heat of transformation.

3.4 FEM-modelling in the Present Study

The welding paths, including initial weld velocities, were exported from the CAR modelto the finite element software where predictions of temperature histories, residual stressesand fixture reaction forces were performed. The same CAD/CAM model as in the CARmodel was imported to and meshed with the FEA software. The commercial FEA pro-gram MARC, from MSC Software, was used. In paper I, a model with solid elements wascreated. Since this type of model is computationally very expensive, this necessitated theselection of only a small part of the component for modelling. The model was divided bya non-uniform mesh with higher densities close to the weld path (where the highest tem-perature gradients were assumed to occur), figure 3.5. Eight-node brick elements wereused. In paper III a shell model of the whole section of the aerospace part was created,see figure 3.6.

3.4.1 Boundary Conditions in the Present Study

Different boundary types have been used. The boundary conditions used in paper I aregiven in table 3.1. Since the model had to be restricted to a subsection of the part, a "metal-metal" boundary condition was introduced which models the heat conduction throughthese interfaces, table 3.1. Different heat transfer coefficients were used for the surfaceswith natural and forced convection i.e. forced convection on surfaces where root gas(see section 2.1) was applied. A typical set up for a plate is illustrated in figure 3.7a.


Weld path

Figure 3.5: The non-uniform mesh used in paper one. Note! Higher densities along theweld path.

Weld path

Flange

Figure 3.6: Shell model of a part of the aerospace component.


The applied boundary conditions for the plate are shown in figure 3.7b. A "metal-meal"convection boundary condition, which models the clamping of the plate to the fixture, isused in group 1. Natural convection is used for group 2 and forced convection is used forgroup 3. The heat input from the arc, group 4 simulates the heat source.

12 4

21

22

1 3 1

Weld poolPlate

Fixture Weld gun

Arc

Weld pool

Shielding gas

Air

Plate

Air

Air Air

Fixture

a)

b)

Figure 3.7: a) Cross section of a plane plate mounted in a welding fixture. b) appliedboundary conditions in a heat transfer simulation.

Natural convection was only used as a heat transfer boundary condition between the partand the surrounding environment in paper III. The flanges on the inner and outer ringswere assumed to be clamped, since no fixture was used (the inner and outer ring werewelded on a steel plate). This was simulated by locking the FEM elements in all direc-tions.

Table 3.1: Boundary conditions in the heat transfer analysis in paper I

Type of condition Interface Value Correspondingnumber in figure 3.7

Face film Metal-Metal 1000 · 10−6 1Face film Metal-Air 20 · 10−6 2Face film Metal-Gas 200 · 10−6 3

Argon as root gas

As a heat source, a Gaussian surface distribution was used in all of the simulations. Thisdistribution was selected since it requires fewer parameters to be calibrated and since


the plates that were welded could be considered to be thin. User subroutines had to bedeveloped to simulate the moving heat source. The heat flux was expressed according to(Radaj 1992)

q = q0 e−αqr

2(3.1)

q0 =ηEIαqπ

(3.2)

q =ηEIαqπ

e−αqr2

(3.3)

where q0 denotes the heat transferred to the workpiece, E the voltage, I the current, ηthe efficiency factor, αq the concentration factor, and r the radial distance from the centreof the heat source. Figure 3.8 shows a typical Gaussian heat flux distribution with a 5 %cut off limit. This means that the distribution is truncated when the heat flux reaches fivepercent of the maximum heat flux permitted i.e. qmin = 0.05qmax. This truncation wasproposed by Radaj (Radaj 1992) and was used in all FEA simulations.

-6 -4 -2 0 2 4 6

5

10

15

20

25

Radial Position (mm)

Hea

t flu

x (W

/mm2 )

Figure 3.8: Heat flux Gaussian distribution with 5% cut off limit.

3.4.2 Material Properties

Two stainless steels, namely Greek Ascaloy and SS316L, have been used. Tempera-ture dependent properties were used for thermal conductivity and specific heat (Zhu &


Table 3.2: Material properties for stainless steel 316L and Greek Ascaloy.

Nomenclature Symbol SS 316L Greek ascaloy unit

Density ρ 7.3·10−6 7.92·10−6 kg·mm−3

Latent heat of fusion ΔH 290000 338308 J/mm−3

Solidus temperature Tsol 1523 1753 Kevin (K)Liquidus temperature Tliq 1673 1853 Kevin (K)

Specific Heat Cp see figure 3.9Thermal conductivity λc see figure 3.10

Y.J.Chao 2002). The properties were taken from (Choo, Szekely & Westhoff 1992),(Tosello et al. 1997) and (Choong 1975) and are given in table 3.2.

The specific heat for Greek Ascaloy and 316L are strongly temperature dependent. Thevalues used for 316L are presented in figure 3.9 (Choong 1975). Simulations were per-formed with and without the consideration of weld pool convection. Figure 3.10 presentsconductivity values used for 316L (Choong 1975)

500 1000 1500 2000 2500 3000400

450

500

550

600

650

700

750

Cp (

J/(k

g K

))

Temperature (K)

Figure 3.9: Specific heat for 316L.

The fusion interval selected in this study is in accordance with analyses conducted byToselo et al. (Tosello et al. 1997).


500 1000 1500 2000 2500 30000

0.01

0.02

0.03

Temperature (K)l

(W/(

mm

K))

500 1000 1500 2000 2500 30000

0.1

0.2

0.3

Temperature (K)

l (W

/(m

m K

))

Figure 3.10: Conductivity for 316L. Conductivity without considering convection (a) andconductivity when weld pool convection is considered by increasing the conductivity valueabove the melting point (b).

3.4.3 Properties for the Thermal-Mechanical Modelling

The initial microstructure of Greek Ascaloy consists of a mixture of ferrite and pearlite.In the numerical model in paper III, the ferrite/pearlite to austenite transformation wasassumed to occur only if the highest temperature experienced by the material was greaterthan a limit temperature, see paper III for further details. A thermal-elastoplastic modelbased on von Mises’s theory was used (Berglund 2001), (Berglund 2003). It was assumedthat no creep strains occur during welding since the material is exposed to a high tempera-ture for a very short period of time. The hardening behaviour of the material was assumedto be isotropic and piecewise linear. Transformation plasticity was not accounted for inthe model. The principles that underpin the thermal-mechanical modelling are furtherdescribed in paper III.

3.5 Principle of the Integration between the Off-Line Program-ming Model and the Finite Element Analysis Model

3.5.1 Integration of OLP and FEA Models

Since two different softwares were used in the Off-line programming and in the FEAwork, an interface between the softwares had to be developed. Figure 3.11 shows a blockchart of the principle. The same part geometry was used in both softwares. The principleis that the part to be manufactured is created in a CAD/CAM software and is then, using

3.5 Principle of the Integration between the Off-Line Programming Model and the FiniteElement Analysis Model 35

either a direct translator or a neutral interface, such as IGES or Step, imported to therobot simulation program. Here the engineer plans the production, following the stepsdescribed in section 3.1 above. A robot path is generated which includes the weldingparameters to be used. The welding program is then exported to the FEA program wherethe thermal history, residual stresses and distortion are predicted. An optimisation canalso be performed to reach a full penetration weld with minimised distortion (furtherdescribed below). If such an optimisation is performed, a new weld parameter is generatedby adjusting the robot speed. This new speed, together with the simulation program forthe robot motion, is then translated to a complete robot program. The robot program isfinally downloaded to the robot control system and a test weld can be performed.

Robot simulation

Geometrye.g. IGES

CAD/ CAM

FEATranslator

Welding path

Thermal historyResidual stresses

Simulation program for the robot motion

IRB Controller

Complete robot code

Full penentration weld with low distortion

Weld velocity (wv)wv

Figure 3.11: Block chart showing the integration between OLP and FEM.

The information exchange between the different softwares is based on the method ofadding attributes to a pose, the same principle as that used in the ABB operative systemS4. Examples of attributes are robot speed and welding data, such as welding current andwelding speed. Figure 3.12 shows three robot poses used in a welding application. Whenthe robot passes a pose it will use the attributes belonging to the next pose, for examplewhen the robot passes pose P20 towards P30 the welding speed is increased to 3.0 mm/s.

The information exchange between the softwares is based on text files using pose co-ordinates (right-handed Cartesian coordinate system (Standard 1999)) together with theattribute’s welding speed and welding current. Figure 3.13 shows an example of an inputfile to the FEA simulation generated in the OLP software. The columns denote x, y andz coordinates, welding speed and welding current respectively.

If an optimization is to be performed, each node in the finite element mesh along the weld-ing path is considered as a welding pose. During the calculation, a text file is generated


P10V

w=2.0mm/s

I=100A

P20V

w=2.0mm/s

I=100A

P30V

w=3.0mm/sI=100A

Figure 3.12: Robot pose description for a path.

-75.0-45.1 0.0 45.5 75.0

0.00.00.00.00.0

0.00.00.00.00.0

2.52.52.53.32.0

100.0100.0110.0 80.0100.0

Figure 3.13: Input file to the FEA simulation generated by the robot simulation program.


with node numbers and corresponding node temperatures for each time step. Figure 3.14shows a typical temperature cross section of a welded plate. The nodes along the weldpath at the top and bottom surfaces are marked with dots.

Node on root side

Node on top side

Figure 3.14: Overview of a cross section from an FEA simulation showing the penetra-tion. The colour represents a temperature interval close to the melting point.

A stand-alone Matlab program was constructed that reads this file and suggests a newweld speed. This weld speed is then used in a new input file for a new simulation. Theoptimisation algorithm used is presented in equation 3.4 where the input welding speed isdenoted by s0. Here λr is a relaxation parameter, Tm the liquidus temperature and Tmaxthe maximum temperature at each node.

si = s0(1 + λrTmaxi

− TmTm

) (3.4)

The weld speed is iteratively adjusted by the program until the temperature on the rootside is sufficiently close to a target temperature. This target temperature is usually setto the liquidus temperature. When an optimal velocity vector i.e. the velocity vectorthat maximizes the speed while keeping full penetration, is found, the velocity vector isexported to the weld program.

3.5.2 Synchronized Time Domain

In virtual design and manufacturing it can be of interest to have the opportunity to in-tegrate process simulations (FEA), robot simulations (CAR) and other manufacturingequipment, such as process control systems. However, different simulation systems areoften running in different time domains. The solution for this is to use a synchronisa-tion mechanism that ensures a global common time domain for all sub-simulations. Theproposed simulation system is based on four different components, namely:

• Computer Aided Robotics for path and robot simulation

• Finite Element Analysis for simulation of heat transfer and component temperature


• An off-line arc-current control system for adaptive process control

• A simulation engine for integration and time synchronisation.

Even if this specific system is based on four different sub-systems, there is no limit to thenumber of sub-systems that the simulation engine can handle. The principle of CAR sim-ulations has been previously discussed in chapter 3.1 and 3.2 whilst FEA was discussedin chapter 3.3 and 3.4. Thus only the simulation engine and arc-current control systemwill be described here.

To manage synchronisation and distributed simulations, a protocol, the SynchronisedDistribute Simulations Protocol (SDSP) was developed (Danielsson 1999), (Danielsson2002). SDSP consist of two parts, a server module and a client module. The server mod-ule handles the common information, such as simulation time, simulation parameters, etc.In paper IV common data consists of welding parameters and welding electrode positions.The client module is an application program interface (API) which all client applicationsuse.

The Arc welding controller obtains the temperature from the server. This value is thencompared with a target temperature. The arc-current is then adjusted to obtain the targettemperature.

For each time step, FEA software reads the new weld gun position and a new arc-currentfrom the SDSP server. The model is then updated with new parameters for the heatsource, i.e., the co-ordinates x, y, z, provided by CAR software and the updated arc-current provided by the arc-current controller. The following steps describe the overallsimulation principle.

1. The component to be manufactured is created in a 3D-CAD software.

2. The component model, in step one, is imported to the CAR software and the FEAsoftware.

3. In the CAR software, a model of the work cell is created. A weld path (includingrobot motion, weld speed, etc.) is developed. In the FEA program the part to bemanufactured is meshed and boundary conditions are applied.

4. The simulation starts. Initial conditions are set in both systems.

5. In the beginning of a time step, all systems are allowed to write data to the server.In our case, the CAR system writes the present Tool Centre Point (TCP) value forthe weld gun, together with the current process parameters to the server.

6. Calculations for current time step are performed for all sub-systems. The FEA sys-tem predicts the temperature distribution. The arc-current control system calculatesa new arc-current value based on the comparison between the desired and predictedtemperature.

7. At the end of each time step, all of the systems read new values from the server.


8. Steps 5, 6 and 7 are performed repeatedly until the completion of the simulation.

The overall architecture of the off-line simulation system is presented in figure 3.15

FEA

Time

Current

Heat SDSP

ServerCAR

Time

-

Hea

t

Cur

rent

Arc position(x, y, z)

Moving heat

source

Heat model Common

data

Time synchronization

Arc - currentcontroller

Arc position

(x, y, z)SD

SPC

lient

Int

erfa

ce

nter

face

SDSP

Clie

nt I

Robot model

Arc path planning

Figure 3.15: The overall architecture of the off-line simulation system.

Chapter 4

Model Validation Techniques

This chapter describes validation techniques for the off-line programming ofrobots, for temperature predictions and for residual stress predictions. Sincea central part of off-line programming is calibration, a more extensive de-scription of calibration techniques is presented first.

4.1 OLP Calibration

Several processes of calibration have to be performed in order to increase the accuracy ina CAR application (K. Schröer 1995). A CAR model consists of several different com-ponents that have to be calibrated, such as the robot, fixtures, manipulators and externalpositioning equipment, see chapter 3.1. All of these components have to be modelledwith the required accuracy and positioned in accordance with their locations in the realwork cell. The calibration of the cell can be divided into three different groups, namelysignature calibration, tool calibration and work cell calibration (Bolmsjö et al. 1997),(Bernhardt & Albright 1993). In this thesis, tool calibrations and work cell calibrationshave been performed.

4.1.1 Signature Calibration

The purpose of signature calibration is to increase the accuracy in the robot arm’s kine-matic chain, both for the real robot and for the modelled version. The signature calibrationcan be divided into three different levels, namely joint level calibration, the kinematiccalibration of the robot and non-kinematic parameter calibration. The selected level ofcalibration depends on the type of robot and the process. In a welding application with alow weight robot and, comparatively to other processes, a low robot speed, the joint levelcalibration is the most important signature calibration method. In the joint level calibra-tion, the joint values of the physical robot are compared with the corresponding values

42 Model Validation Techniques

for the robot model. A common way to perform this calibration is to rotate the robotarm around a measuring stick located at several positions in the work cell. The tool tip isrotated around the stick with different joint configurations and the joint values stored in arobot program. The model and the real robot values are then compared.

4.1.2 Tool Calibration

Probably the most important part of the overall calibration process is the tool calibration.When such a calibration has been performed the robot arm, together with the tool tip,can be used as a measuring tool with a high degree of accuracy. The tool calibration isusually performed using a measuring stick with a sharp point that is positioned in thework cell. The tool tip of the robot is moved towards this point in different directions.When the tool tip makes contact with the edge of the stick, the position is stored in arobot program. It is important to move as many joints as possible during this positioningprocess. For each positioning, at least five locations have to be recorded to achieve therequired accuracy. The external axes in the same kinematic change, such as servo trackor position equipment, must not be moved. These axes have to be calibrated separately.In a tool calibration for a welding application, the rotation of the tool is of critical impor-tance since penetration and weld quality are strongly dependent on the angle between theelectrode and the object to be welded. A second tool point calibration is therefore usuallyperformed using a long tool-tip in the welding direction. This type of calibration can alsobe used to calculate the orientation of the tool. An alternative to using the robot to performthe tool calibration is to use external measuring equipment, such as a theodolite or a laserinterferometer (Bernhardt & Albright 1993), (Ryberg, Christiansson & Eriksson 2006),(Schofield 2002). Most of the common CAR softwares have a pre-defined function to cal-culate the tool tip position based on this calibration. In IGRIP, this operation is performedusing statistical regression.

4.1.3 Work Cell Calibration

The position for each object in the work cell has to be determined in the virtual world.This can be achieved using the robot arm or by using external measuring equipment, suchas measuring tape or a theodolite system. Objects that require less position accuracy,such as walls that are not critical for collisions with a kinematic device, are located mosteasily by using a measuring tape. Objects that need high position accuracy, however,i.e. those that are critical for collision or in a need of precise positioning, such as weldpaths, are usually measured using the calibrated robot arm. A work cell calibration isaccomplished by selecting critical points in the work cell. Points on the work-piece aretypical examples. Using a calibrated tool, the robot is moved to these points. The co-ordinates of each point are stored in a program and uploaded in the CAR software. In theCAR model the same points have been identified. By comparing measured and modelledco-ordinates, the software calculates a best fit of the position of the parts using statisticalregression.

4.2 Temperature Measurement Techniques 43

4.1.4 Robot Position Calibration

One important relationship to calibrate in the welding cell was the distance between thetwo robots. A tool calibration was performed on both of the robots, see 4.1.2. One ofthe robots was considered to be the reference point, which in this work, was the IRB4400. The robot arm was moved to critical points on an object, see 4.1.3 for more details.The object was then calibrated in the CAR application. This object was then used as areference position for the next robot. The second robot (IRB 1400) was then moved to thesame positions on the object, which was uploaded to the CAR application. The positionof the second robot was then calibrated using the object as a reference.

4.2 Temperature Measurement Techniques

Temperature can be measured by the use of a wide range of sensors. All of them mea-sure the temperature by sensing a change in a physical characteristic. The most commonmethods are thermocouple, resistance temperature devices (RTD’s and thermistors), in-frared radiators, bimetallic devices, liquid expansion devices and change of state devices(Morris 1993). In this work, both thermal couple- and infrared imaging measurementshave been used to measure thermal time histories on plates and on complex shaped parts.The main purpose of the thermocouple measurements was to obtain reference data bywhich the infrared imaging measurements could be calibrated (Morris 1993).

4.2.1 Thermocouple Instrumentation on Plates

Thermocouple is the most common method used to measure temperature. This is due tothe fact that the thermocouples are cheap, interchangeable and can measure a wide rangeof temperatures. A thermocouple consists of two wires, each of which is derived from adifferent type of metal, which are joined at one end (Maldague 2001). A change in thetemperature at the connection of the two wires will induce a change in the electromotiveforce (emf) between the other ends. As the temperature changes, the emf will change.Often the thermocouple wires are located inside a metal or ceramic shield that protectsthem. The most commonly used thermocouple type is type K (Maldague 2001). It hasone nickel-chromium wire and one nickel-aluminium wire. The contact point of the ther-mocouple is spot welded on the plate at the desired position where the temperature historyis to be recorded (Bolton 1996). Thermocouples of type K with a wire diameter of 0.11mm have been used in all experiments on the plates in this work. Six thermocouples werepositioned perpendicular to the weld seam. The first gauge was mounted as closely aspossible to the melted zone at a distance of 4 mm from the centre of the weld. The re-maining gauges were located at 4.5, 5, 6, 7 and 8 mm respectively from the weld centreline, figure 4.1. A PC-based data acquisition system was used to sample the signals fromthe thermocouples and to monitor the signals on a screen. The measurement data wassimultaneously written to disk. The complete measurement system was calibrated in thetemperature range 0◦ - 1350◦C


Welding direction

Line scan

Weldcenter

line

Figure 4.1: Schematic diagram of a plate with thermocouples together with selectedmeasurement lines for the IR camera measurements.

4.2.2 Infrared Imaging Measurement Techniques

The infrared (IR) temperature sensor technique is a non-contact measuring method. Itmeasures the temperature by recording the IR energy emitted by the object (Maldague2001). As the temperature increases in the object, the amount of infrared radiation alsoincreases. Different materials radiate different amounts of IR energy at the same tem-perature. This efficiency factor is called the emissivity, which is defined as the frac-tion of radiation emitted by an object as compared to the radiation emitted by a per-fect radiator, known as the blackbody, at the same temperature (Siegel & Howell 2002),(Maldague 2001). The emissivity may vary from close to 0 (highly reflected mirror) toalmost 1 (for a blackbody). The problem with the emissivity is that it can vary with wave-length, component curvature, component surface roughness, viewing angle, and also asresult of surface film effects (Siegel & Howell 2002). An accurate temperature can notbe measured if the object’s emissivity is unknown. The infrared camera used in this workis a VarioScan 3021 high resolution 16 bit Stirling cooled camera produced by JenopticGmbH. The camera has two scanning mirrors to image an object on a point detector ofMCT type (HgCdTe), see figure 4.2. The camera operates in the wavelength range 8-12μm and has an image resolution on 360(h)× 240(v) pixels.

Temperature measurements have been performed in full frame mode (1 Hz) and in linescan mode (270 Hz). The temperature measurements using the line scan mode were per-formed in combination with thermocouple measurements, see figure 4.1. The selected

4.3 Residual Stress Measurements Techniques 45

Horizontal scanner

Vertical scanner

Detector

Lens

Figure 4.2: Principle overview of the VarioScan 3021 high resolution camera.

measuring line was then scanned continuously at a rate of 270 lines/s. The position ofeach thermocouple was registered after welding using a microscope. These positionswere used to make comparisons with the IR recordings where the corresponding imagepixels were selected. Different techniques for surface treatments were evaluated in orderto overcome the problem with surface emissivity variations due to oxidation. Soot froman acetylene flame was found to give a highly temperature resistant black surface witha constant emissivity value (Micron Instrument Company, USA 2002). Using this tech-nique, reliable measurements could be performed on the complete part, with the exceptionof the region where the actual fusion had occurred.

4.3 Residual Stress Measurements Techniques

After welding, residual stresses can be measured. Both destructive and non-destructivemeasurement techniques exist. The techniques can be divided into three main groups,namely stress-relaxation, X-ray or neutron diffraction, and cracking methods. Measure-ments of residual stresses using the neutron diffraction method were performed in thiswork. The stress-relaxation method is a destructive method. It can be divided in two sub-groups. The first group is based on mechanical or electrical strain gauge measurements.The second group uses no electrical or mechanical strain gauges. The residual stressesare, instead, measured by estimation of the elastic strain release. By cutting the test ob-ject in several pieces or by drilling and removing a piece of the part, the residual stressescan be relaxed and measured. Stress relaxation is the most commonly used method since


reliable, quantitative data can be obtained. The x-ray and neutron diffraction methods arevery similar. Both methods measure crystal lattice parameters in the welded material andproduce interference phenomena that are related to the interplanar spacing of the lattice.The residual stresses are then determined from the change in the interplaner spacing viaBragg’s law and Hooke’s law and compared to the stress-free state. The difference be-tween the methods is that neutron diffraction uses neutrons scattered by a nuclear powersource, whilst X-rays are used to scatter the electrons in the latter method. The neutrondiffraction method provides deeper penetration depths, approximately 30 mm in steel,as compared to X-ray penetration which achieves depths of about 10 μm. The crackingresidual stress measurement method determines residual stresses by studying the cracksin the melted zone. The cracks may have been introduced by hydrogen or stress cor-rosion. The method is particularly useful for the analysis of components with complexstress distributions.

Chapter 5

Results and Summary of AppendedPapers

5.1 Paper I

In this paper, a CAR software package was used to simulate welding operations and toprogram robot motions off-line for plates, as well as for an aerospace component. Theoff-line programming results revealed a high degree of accuracy and very little post-calibration fine-tuning was required. The method seems to be a powerful tool, espe-cially in cases of small batch production. FEA was used in the same paper to predicttemperature-time histories. Good agreements between predicted and measured tempera-tures, both in peak temperatures and in cooling histories, were found. The overall conclu-sion from the simulations conducted was that the model predicted the thermal cycle verywell.

5.2 Paper II

Temperature measurements were performed using both thermocouples (T/C) and Infrared(IR) camera measurements both on plates of Greek Ascaloy and on Stainless Steel 316L.An Acetylene/Oxygen soot deposition method was evaluated making the IR measure-ments emissivity independent. Good agreement between IR and T/C temperature mea-surements was found. It was concluded that IR imaging can function as a useful non-contact method to validate predicted temperature histories on complex shaped parts.

48 Results and Summary of Appended Papers

5.3 Paper III

In this paper, residual stress distributions were predicted using FEA. Predictions wereevaluated along three different sampling lines perpendicular to the weld seam. The stresseswere also compared with neutron diffraction measurements. The longitudinal stress levelwas shown to be very similar along all three sampling lines. The transversal stress com-ponents between the three sampling lines were, however, discovered to be significantlydifferent between the sampling lines. Reaction forces were also predicted.

5.4 Paper IV

In this paper, a software architecture for off-line programming, simulation, and controlof robotic arc welding was proposed. The key components in the system are the use ofFEA software for the evaluation of thermal histories, coupled to an off-line arc-currentcontrol system to minimize heat effects. Information exchange between the softwareswas controlled by means of a dedicated distributed server. All simulations were timesynchronized i.e. were run in the same virtual real time. The proposed simulation systemprovides a useful tool to optimise temperatures on complex shaped parts and, thereby,minimizing heat effects such as distortion.

5.5 Paper V

In this paper, a simulation method for welding parameter optimization was proposed.A CAR software was used for simulation of the robot motion and for the off-line pro-gramming of the robots. FEA software was used for the simulation of the heat effects.A stand-alone matlab program was developed for weld velocity optimization. Resultsshowed that after just 10 iterations, a target temperature close to the weld seam can bereached within ±30 K for a part with smooth variation in thickness. Weld penetration canthus be controlled for by minimizing the heat effects. A predicted cross-section of theweld seam showed a good agreement with corresponding measurements.

5.6 Paper VI

In this paper, an evaluation of three different methods for path generation for Metal De-position was performed. All three methods were shown to be capable of generating pathsfor both simple and complex geometries. The most flexible method was shown to be theadaptation of Computer Aided Robotic (CAR) software. Using this method, a fully auto-mated path generation seems possible, even for complex-shaped parts. Another advantagewith this solution was that only one system was needed for both path generation and robotsimulation.

Chapter 6

Discussion

This chapter discusses the different modelling assumptions and the need forfurther model development. A sensitivity study is provided thus making it pos-sible to estimate errors. Limitations and possible extensions are discussed.

6.1 Accuracy in the OLP Model

High degrees of accuracy, both in the kinematic model as well as in the control logic, arenecessary in order to fully utilize the potential of off-line programming. As yet, there isstill not a 100% transparency between the simulation control systems and the physicalsystem. The development of specific modules for more accurate descriptions of controlsystems, such as RRS, (Bernhardt et al. 1995) will ensure more accurate motion interpola-tion in the CAR system. No RRS model was, however, used in the present work. Anotherinteresting possibility is the development of a control system emulator. The advantagesof such an emulator is described in (Danielsson 2002).

6.2 Parameter Sensitivity Study

One drawback in FEA modelling work is the need to perform experiments to find ap-propriate parameter values (Berthiau 2001). A sensitivity study designed to consider theaccuracy of various modelling assumptions was therefore performed. The aim was tofind the main parameters that influence the thermal history. A number of factors, suchas the heat concentration factor α and the arc efficiency η (specified in section 6.2.1),were selected for the analysis. To limit the number of simulations, some parameters werekept constant throughout all of the simulations. Examples of such parameters are weldingvoltage, welding current and the surrounding ambient parameters.

50 Discussion

6.2.1 Factors

Design of experiments, DoE was utilised to be able to consider interaction effects betweenparameters. In table 6.1 minimum, center and maximum values in the DoE are given foreach factor, together with a reference setting.

Table 6.1: Factors and corresponding values used in the sensitivity study

Factor Reference Minimum Center Maximum Unit

α 0.1 0.06 0.195 0.33η 0.9 0.6 0.8 1.0

hgas 8.0 · 10−5 6.0 · 10−5 8.0 · 10−5 1.0 · 10−4 (W/(Kmm2)hair 1.9 · 10−7 6.7 · 10−9 2.0 · 10−7 4.0 · 10−7 (W/(Kmm2)

hmetal 14.27 · 10−3 13.96 · 10−3 14.75 · 10−3 15.53 · 10−3 (W/(Kmm2)L 247000 230000 260000 290000 (J/kg)ρ 7.0 · 10−6 5.8 · 10−6 6.9 · 10−6 8.0 · 10−6 (kg/(mm3)

λ(T ) See figure 6.1Cp(T ) See figure 6.2

Note: High and low values of thermal conductivity (λ(T )) and specific heat (Cp(T )) aregiven in figure 6.1 and 6.2.

6.2.2 Responses

Thermal history results were selected as responses in the analysis. Each simulation wasanalysed using the peak temperature and six measurements on the width of the peak.Three widths at (1/4, 1/2 and 3/4 of the maximum peak temperature) were selected, to-gether with three locations (3/4, 1/2 and 1/4 of the maximum peak temperature) in thecooling sequence, see figure 6.3.

6.2.3 Results Sensitivity

Coefficient plots of peak temperature results are provided in figure 6.4. The bar for eachfactor corresponds to the derived regression coefficient when the specific factor is variedfrom its low to its high value, see table 6.1. The peak temperature is mainly influenced bythe arc efficiency (μ) and only slightly by the mass density (ρ), as can be seen in figure6.4. The multivariable Partial Least Squares (PLS) was used (Umetrics AB 2005). Thevertical line intervals in figure 6.4 represent 95% confidence intervals.

6.2 Parameter Sensitivity Study 51

500 1000 1500 2000 2500 3000

0.1

0.2

0.3

0.4

Temperature (K)

λ (W

/(m

m K

))

MinimumCenterMaximumReference

Figure 6.1: Conductivity values used in the sensitivity study.

500 1000 1500 2000 2500 3000

450

500

550

600

650

700

750

800

Temperature (K)

Cp (

J/(k

g K

))

MinimumMaximumCenter and reference

Figure 6.2: Specific heat values used in the sensitivity study.

52 Discussion

10 15 20 25 30 35

400

700

1000

1300

1600

1/4a

1/2a

3/4a 3/4b

1/2b

1/4b

Tmax

Time (s)

Tem

pera

ture

(K

)

Figure 6.3: Responses selected from the temperature history.

-50

0

50

100

150

200

α η L ρ hgas

hair

hmetal

Cpλ

Influ

ence

on

T max

(K

)

Figure 6.4: The influence of different factors on the peak temperature.

6.2 Parameter Sensitivity Study 53

It was shown that all widths (1/4, 1/2, 3/4) in figure 6.3) were influenced by several factors.A typical result showing the width of the cooling part (width 1/2b in figure 6.3 coefficientplot for the η, ρ and α have the major influence of the width. However the influence ofthese factors have to be considered as fairly low.

-0.2

0

0.2

0.4

α η L ρ hgas

hair

hmetal

Cpλ

Influ

ence

on

wid

th 1

/2B

Figure 6.5: The influence of different factors on the width of the temperature.

The peak temperature was shown to be very sensitive to choice of the right values of η,ρ, α as can be seen in figure 6.6. The temperature corresponds to a point located 3.6 mmfrom the center of the weld seam. Experimental trials are thus necessary to find appriatevalues for these parameters. The need of such experiments can be reduced by numericalmodelling of the shape of the molten pool by the use of Computation Fluid Dynamics(CFD) analysis. Indeed, a research project with this particular aim was recently started atthe University West (Edstorp & Eriksson 2005).

Several extensions of the modelling work described in this thesis are possible. To ex-tend the work to include filler wire, pulsed currents, as well as considering tack-weldingprocesses, would all be valuable extensions. The inclusion of transformation plasticity inthe material model would lead to improvements in the prediction capability. In the questto find good validation methods, it becomes essential to evaluate the proposed model as-sumptions. The sooting technique used in paper II was shown to be successful, althougha method where the surface does not has to be modified would be of greater industrialinterest. To develop a good non-contact measurement method for distortion predictionswould be another area of interest. Research on this subject is currently being conductedat the University West. A development of the proposed simulation tool to automaticallyupdate robot trajectories based on the predicted deformation would also be of interest.Such a system would decrease the need for sensors i.e. seam tracking, in the productionprocess. To develop a methodology for simulation-based fixture design using the sim-ulation method developed here would again be an interesting development. Of specificinterest here would be the question of how the FEA models can be simplified to enablethe rapid evolution of different fixture design concepts.

54 Discussion

10 20 30

200

600

1000

1400

1800

Slight influenced by:ρ, η, C

p, α, L and

hmetal

influenced by η,and slight ρ and α

Time (s)

Tem

pera

ture

(K

)

Figure 6.6: Different temperature profiles at 3.6 mm from the center of the weld seam.

The overall conclusion is that numerable extensions to the modelling work conducted herecould be made. The functionality of models with different levels of detail is discussedin (Lindgren 2001a), (Lindgren 2001b), (Lindgren 2001c), (Lindgren 2001d), (Goldak2005). However, it should be kept in mind that the purpose of the models in the currentthesis has been to develop a tool for the production engineer. With an extension of themodels, a balance between computational speed, complexity, and model accuracy, mustbe considered.

Chapter 7

Summary and Conclusion

In this thesis a methodology and a simulation tool for welding process simulations weredeveloped and evaluated. Robot trajectories were defined and both thermal and resid-ual stress distributions were predicted for parts with complex shapes. The method wasevaluated on a piece of an aerospace component where robot weld paths were defined off-line and automatically downloaded to a Finite Element Model, where temperatures andresidual stress distributions were predicted. The temperature predictions were comparedwith experimental measurements using both thermocouple and infrared emission mea-surements and good agreements were found. Residual stress distributions were validatedusing neutron diffraction and showed fairly good agreement. A software architecture wasproposed enabling the time synchronization of different simulation systems. An opti-mization algorithm enabling weld penetration control was developed and validated. Themethod described provides a powerful means for the construction and optimisation oftorch trajectories and process parameters off-line. Using this method, production can bemaximised whilst still minimising heat effects. The work can be regarded as providinga foundation for a simulation tool via which robot trajectories, welding sequences andfixture designs can be optimized prior to manufacture.

56 Summary and Conclusion

Bibliography

Alberg, H. (2005), Simulation of welding and heat treatment, modeling and validation,Doctoral Dissertation, Department of Mechanical Engineering, Luleå University ofTechnology, Sweden.

Berglund, D. (2001), Simulation of welding and stress relief heat treating in developmentof aerospace components, Licentiate in Engineering Thesis, Department of Mechan-ical Engineering, Luleå University of Technology, Sweden.

Berglund, D. (2003), Validation of models for welding and post weld heat treatment inproduct development of aerospace components, Doctoral Dissertation, Departmentof Mechanical Engineering, Luleå University of Technology, Sweden.

Bernhardt, R. & Albright, S. (1993), Robot Calibration, Chapman & Hall, 2-6 BoundaryRow, London SE1 8HN, United Lingdom.

Bernhardt, R., Schreck, G. & Willnow, C. (1995), ‘Realistic robot simulation’, Computingand Control Engineering Journal pp. 174–176.

Berthiau, J. (2001), Calibration of heat transfer parameters of tig-welding simulation,Master’s thesis, University of Trollhättan/Uddevalla, Sweden.

Bolmsjö, G., Olsson, M., & Brink, K. (1996), Programming GMAW robotic systemsusing advanced graphical simulation tools, in ‘Proceedings of the 6th InternationalConference on Computer Technology in Welding’, Lanaken, Belgium. paper no 28.

Bolmsjö, G., Olsson, M. & Brink, K. (1997), ‘Off-line programming of GMAW roboticsystems - a case study’, Int J. for joining of Materials 9, pp. 86–93.

Bolton, W. (1996), Mesurement and Instrumentation systems, first edn, ButterworthHeinenmann Ltd., Oxford.

Brust, F. W., Yang, Y. Y., Y.Dong & T.Jutla (2000), ‘Weld process modeling and its im-portance in a manufacturing enviroment’, Advances in computational engineeringand science 2000 1, pp. 714–719.

58 Bibliography

Cederberg, P., Olsson, M. & Bolmsjö, G. (1999), A generic sensor interface in robotsimulation and control, in ‘Proceedings of Scandinavian Symposium on Robotics99’, Oulu, Finland, pp. 221–230.

Cederberg, P., Olsson, M. & Bolmsjö, G. (2002), ‘Virtual triangulation sensor develop-ment, behavior simulation and car integration applied to robotic arc-welding’, Jour-nal of Intelligent and Robotic Systems 35(4), pp. 365–379.

Choo, R., Szekely, J. & Westhoff, R. C. (1992), ‘On the calculation of the free surfacetemperature of gas-tungsten-arc weld pools from the first principles - part ii: Mod-elling the weld pool and comparison with experiments’, Metallurgical Transaction23B, pp. 376–384.

Choong, S. K. (1975), Thermophysical properties of stainless steel, Doctoral Dissertation,Argonne national laboratory, Illinois, United States.

Connon, L. P., ed. (1991), Welding Handbook - Welding Technology, Vol. 1, AmericanWelding Society.

Danielsson, F. (1999), A distributed system architecture for optimizing control logic incomplex manufacturing systems, in ‘Proceedings of the ISCA 12th internationalConference’, pp. 163–167.

Danielsson, F. (2002), Off-line programming, verification and optimisation of industrialcontrol system, Doctoral Dissertation, De Montfort University, Leicester, UnitedKingdom.

Easterling, K. (1992), Introduction to the Physical Metallurgy of Welding, second edn,Butterworth Heinenmann Ltd., Oxford.

Edstorp, M. & Eriksson, K. (2005), Modelling and simulation of moving boundaries andconvective heat transfer in non-autogeneous fusion weld pools using femlab 3.1, in‘Proceedings of FEMLAB CONFERENCE 2005’, www.comsol.se/femlab2005.

Goldak, J. (2005), Challenges for computational weld mechanics research, in ‘The In-ternational Conference on Advanced Welding/Joining Technology for 21st Century2005’, pp. pp. 44–49.

Goldak, J. A. & Akhlaghi, M. (2005), Computational Welding Mechanics, Springer Sci-ence + Business Media Inc.

Goldak, J., Chakravarti, A. & Bibby, M. (1984), ‘A new finite element model for weldingheat sources’, Metallurgical Transaction 15B, pp. 299–305.

Hibbit, H. D. & Marcal, P. (1973), ‘A numerical thermo-mechanical model for the weldingand subsequent loading of a fabricated structure’, Comp. ‘I&’ Struct 3, pp. 1145–1174.

Incropera, F. P. & DeWitt, D. P. (1996), Fundamentals of heat and mass transfer, fourthedn, John Wiley & Sons, New York.

Bibliography 59

Ion, J. C. (2005), Laser Processing of Engineering Materials, Elsvier Butterworth-Heinemann, United Kingdom.

J.Mazumder, D.Dutta, N.Kikuchi & A.Ghosh (2000), ‘Closed loop direct metaldeposi-tion: art to part’, Optics and Lasers in Engineering 34, pp. 397–414.

K. Schröer, R. Bernhardt, S. A. H. W. S. K. D. v. A. (1995), ‘Calibration applied to qualitycontrol in robot production’, Control Eng. 3(4), pp. 575–580.

Landcaster, J. F. (1999), Metallurgy of Welding, sixth edn, Abington Publishing.

Li, Z. & Wu, C. (1997), ‘Analysis of the transport phenomena in the interfacial regionbetween tig arcs and weld pools’, Computational Materials Science 8, 243–250.

Lin, R. (1988), On residual Stresses and Fatigue of Laser Hardened Steels, Doctoral Dis-sertation, Department of Mechanical Engineering, Lidköpings University, Sweden.

Lindgren, L.-E. (2001a), ‘Finite element modeling and simulation of welding part 1: In-creased complexity’, Journal of Thermal Stresses 24, pp. 141–192.

Lindgren, L.-E. (2001b), ‘Finite element modeling and simulation of welding part 2: Im-proved material modeling’, Journal of Thermal Stresses 24, pp. 195–231.

Lindgren, L.-E. (2001c), ‘Finite element modeling and simulation of welding part 3: Ef-ficiency and integration’, Journal of Thermal Stresses 24, pp. 305–334.

Lindgren, L.-E. (2001d), Modelling for residual stresses and deformation due to weld-ing ’knowing what isn’t neccessary to know, in ‘Proceedings of 6th Int. SeminarNumerical Analysis of Weldability’, Graz, Austria.

Lucas, B. (2004), ‘Controlling distortion in arc welding - a guid to best practice’, TWILtd, United Kingdow .

Maldague, X. P. V. (2001), Theory and Practice of Infrared Technology for NondestructiveTesting, John Wiley & Sons.

Meredith, R. (1942), ‘2,274,631’, U.S. Patent.

Michaleris, P. & DeBiccari, A. (1997), ‘Prediction of welding distortion’, Welding Jour-nal pp. 172–181.

Micron Instrument Company, USA (2002), Table of various surfaces. Availablehttp://www.mikroninst.com.

Morris, A. S. (1993), Principles of Measurement and Instrumentation, first edn, PrenticeHall, New Jersey.

O’Brien, R. L., ed. (1991), Welding Handbook - Welding Processes, Vol. 2, AmericanWelding Society.

60

Olsen, D. L., Siewert, T. A., Liu, S. & Edwards, G. R., eds (1997), ASM handbook, Volume6, Welding, Brazing and Soldering, eight edn, Library of Congress Cataloging-in-Publication Data.

Olsson, M. (2002), Simulation and execution of autonomous robot systems, PhD the-sis, Division of Robotics, Department of Mechanical Engineering, Lund University,Sweden.

Radaj, D. (1992), Heat effects off welding, Springer Verlag, Berlin.

Ryberg, A., Christiansson, A.-K. & Eriksson, K. (2006), Accuracy investigation of a lowcost vision based system for pose measuring, in ‘submitted to 2006 IEEE Interna-tional Conference on Robotics and Automation ICRA’.

Schofield, M. (2002), ‘Does your robot really know where it is?’, Industrial Robot: AnInternational Journal 29(5), pp. 407–411.

Siegel, R. & Howell, J. R. (2002), Thermal Radiation Heat Transfer, fourth edn, Taylor& Francis, New York, United States.

Standard, A. N. (1999), For industrial robots and robot system, point-to-point and staticperformance characteristics - evaluation, Technical report, American National Stan-dard.

Ting, Y., Lei, W. I. & Jar, H. (2002), ‘A path planning algorithm for industrial robots’,Computers & Industrial Engineering 34(4), pp. 299–308.

Tosello, I., Tissot, F. X. & Barras, M. (1997), ‘Modelling of the weld behaviour for thecontrol of gta process by computer aided welding’, Matehematical Modelling ofWeld Phenomena 4, pp. 80–103.

Ueda, Y. & Yamakawa, T. (1971), ‘Analysis of thermal elastic-plastic stress and strainduring welding by finite element method’, Trans JWRI 2, pp. 90–100.

Umetrics AB (2005), Users guide to modde. Available http://www.umetrics.com.

Yong, Y. F., Gleave, J. A., Green, J. L., & Bonne, M. C. (1985), Off-line programming ofrobots, Handbook of Industrial Robotics, first edn, John Wiley & Sons., New York.

Zacharia, T. & Chen, Y. (1998), ‘Modelling of fundamental phenomena in gas tungstenarc welds’, Int. J. of Materials and Product Technology 13(1/2), pp. 77–88.

Zacharia, T., Vitek, J. M., Goldak, J. A., DebRoy, T. A., Rappaz, M. & Bhadeshia, H. K.D. H. (1995), ‘Modeling of fundamental phenomena in welds’, Modelling Simula-tion Material, Sci. Eng. 3, pp. 265–288.

Zhu, X. & Y.J.Chao (2002), ‘Effects of temperature-dependent material properties onwelding simulation’, Computers and Structures 80, pp. 967–976.

Paper I

Three-Dimension Simulation of Robot Path andHeat Transfer of a TIG-Welded Part with Complex

Geometry

Mikael Ericsson, Per Nylén, Gunnar Bolmsjö

SME Technical Paper AD02-292 (Dearborn, Mich.: Society ofManufacturing Engineers, 2002). Also published in 11th

International Conference on Computer Technology in Welding,Colombus, Ohio, USA December 5-6, 2001.

Three-Dimension Simulation of Robot Path andHeat Transfer of a TIG-Welded Part with Complex

Geometry

Mikael Ericsson, Per Nylén, Gunnar Bolmsjö

Abstract

The applications of commercial software (OLP) packages for robot simulation,and programming, us interactive computer graphics, provide powerful tools for cre-ating welding paths off-line. By the use of such software, problems of robot reach,accessibility, collision and timing can be eliminated during the planning stage. Thispaper describes how such software can be integrated with a numerical model that pre-dicts temperature-time histories in the solid material. The objective of this integrationis to develop a tool for the engineer where robot trajectories and process parame-ters can be optimized on parts with complex geometry. Such a tool would decreasethe number of weld trials, increase productivity and reduce costs. Assumptions andprinciples behind the modeling techniques are presented together with experimentalevaluation of the correlation between modeled and measured temperatures.

1 Introduction

The metallurgical structure of a metal, which determines its mechanical properties, isa function of its chemical composition, its initial structure and the thermal effects of thewelding process. Theoretically, if both the thermal events and the response of the materialto the thermal process is known, the resulting changes in microstructure and propertiescan be predicted. Several papers have been published concerning numerical modeling ofthermal histories, residual stresses, and distortion (Ref. 1-7). Mainly two-dimensionalstudies have been performed. Three-dimensional studies are still restricted to simplershapes such as plates and pipes. The use of robots for arc welding started in the early 70t’sand is now extensively used in the MIG/MAG processes. Using robots for TIG (GTAW)welding is however still rare. One of the reasons is the increased demand for preciseprogramming and control. Programming of welding robots is usually done manually bythe jog teach method. Using this method the robot is off-line, the part stationary, and therobot arm jogged through the program under reduced power and at reduced speed, via ajoystick. Generating a path by hand in this way can be time consuming. On a complexgeometry, it is virtually impossible for a programmer to maintain constant gun velocity,distance from, and orientation to, the part. However, by using computer simulation thisproblem can be overcome. Using this method, the programming is moved away fromthe robot to a graphical computer system, often referred to as a "off-line programming"

64

system (OLP). The technology in this area is well established and has been a research area(Ref. 8-11) for some ten years. Despite these extensive investigations, the two differentsimulation techniques (numerical process modeling and OLP) seems only to have beenstudied separately. The need for a better simulation tool for arc welding was the startingpoint for a research program at the University Trollhättan/Uddevalla in collaboration withVolvo Aero Corporation. The objective of this program is to provide temperature -timehistories and metallurgical- and mechanical-properties predictions on robot welded parts.The program is divided into four parts:

1. to off-line program parts with complex shapes,

2. to numerically predict the shape of the molten pool by the use of ComputationalFluid Dynamics (CFD) techniques,

3. to numerically solve the energy equations in the solid material with sufficient ac-curacy that metallurgical predictions can be made, as well as to link the off-lineprogramming model with this numerical model, and

4. to empirically establish relationships between temperature-time history and metal-lurgical and mechanical properties

This paper is concerned with parts 1 and 3 above; namely methods of programming robotsoff-line and of predicting temperature-time histories on parts with complex shapes.

2 Principle of Off-Line Programming (OLP)

Several commercial software packages for off-line programming of robots exist (CATIA,IGRIP, Robcad GRASP etc.). A brief description of the methodology using such systemsis given below. A more detailed description is given in (Ref. 8). The methodology ofOLP includes the following steps (Ref. 8):

1. modeling of the work cell,

2. modeling of the work-piece,

3. calibrating the work-cell,

4. adjusting and fine-tuning up and down loading of programs,

5. programming, and

6. test runs and macro programming enhancements

The first step to model the work cell concerns the construction of a geometric and a kine-matic model of the robot, positioner etc. This demands access to design drawing of thecell together with measurements of critical dimensions in the cell. The workcell model

65

is usually constructed directly in the OLP system. The IGRIP (Interactive Graphics Ro-bot Instruction Program, Deneb Robotics) system was used in this study. In the secondstep a geometrical description (CAD data) of the part to be welded is generated either ina CAD/CAM software or in the off-line programming software (OLP). If this model iscreated in a CAD system the data is imported to the OLP software either using a neutralinterface (for instance IGES) or a direct reader. The accuracy of the modeled workcell isusually not high and the third step is therefor to make a calibration by measuring differ-ent points in the physical welding cell. This procedure might include several sub-stepsdepending on the complexity of the workcell. In this work a tool calibration and a cali-bration of the workpiece were performed. Tool calibration is performed to determine thetool center point and to determine the orientation of the weld torch. The procedure usedin this study was to have a measuring arrow in a fixed position in the work cell and tomove the robot to this position in different directions. The positions from the real robotcell were then uploaded to the OLP software and a "best fit" was performed by the system.The calibration of the workpiece was performed similarly by moving the robot to clearlyidentified positions on the workpiece. These positions were recorded and uploaded to theOLP software where the difference between model and measurements was calculated andan adjustment of the model using least squares fitting was done. The motion of the robotis then programmed in steps four and five, either in a high level programming language(for instance GSL, which is the graphical simulation program in IGRIP) or in a specificrobot language (such as RAPID, which is the program language for ABB robots). A ro-bot trajectory is then defined by a set of coordinate frames specifying locations and gunorientation. After that, the motion may be simulated to check the results on the computer.High level languages are then translated and the program finally downloaded to the robotcontroller where in the final step, test runs are performed. Figure 1 shows a screen-captureduring the simulation in the OLP system.

3 The Computational Heat Transfer Model

The primary aim of the numerical finite element model is to predict the temperature evo-lution outside the molten zone on a part with complex geometry. Here, the software ICEMCFD, HEXA, which is a commercial pre-processor for CFD and structural applications,was used to mesh the part. The commercial FEM program Marc from MSC Software wasused in the heat transfer predictions. User subroutines had to be developed to simulate themoving heat source. A Gaussian surface distribution was used to model the heat sourcefrom the weld. This distribution was preferred to a volumetric one (Ref. 4) since it re-duces the number of parameters (unknown variables) to be fit and because the plates to bewelded were considered thin (<1.5mm) The heat flux was expressed as (Ref. 4):

q = q0 e−αqr

2(1)

q0 =ηEIαqπ

(2)

66

Figure 1: OLP model.

q =ηEIαqπ

e−αqr2

(3)

where q denotes the heat transferred to the workpiece, E the voltage, I the current, νthe efficiency factor, q the concentration factor, and r the radial distance from the centerof the heat source. The distribution was truncated in the radial direction, at a cut offlimit of 5% of the maximal heat input, as proposed by D. Radaj (Ref. 4). Temperaturedependent properties as well as phase change was included in the analysis. The thermalconductivity was increased by a factor 10 when the temperature reached the liquidustemperature to account for connective heat transfer in the melted zone (Ref. 4). Thissimplified model was used instead of more advanced CFD models to simulate the physicsin the molten zone, since the latter methods are too computationally demanding. The partwas divided into a non-uniform mesh with higher mesh densities close to the weld path(where the highest temperature gradients were assumed to occur), Figure 2. Eight-nodebrick elements were used. Six elements were defined in the thickness direction. Gridsensitivity trials were made by successively refining the mesh. The final number of nodesin the model was 181500.

The boundary conditions for the analysis are given in Table 1. The model had to berestricted to a subsection of the part to be welded since otherwise to long computationaltimes would have been required. To compensate for this simplification a "Metal-Metal"boundary condition was introduced which symbolizes pure heat conduction through theseinterfaces, Table 1.

The simulations were to time consuming to be run on a single workstation why parallel

67

Weld path

Figure 2: The non-uniform mesh. Note the higher densities along the weld path.

Table 1 : Boundary conditions in the heat transfer analysis.

Type of condition Interface Value

Face film Metal - Metal 1000·10−6

Face film Metal - Air 20·10−6

Face film Metal - Gas 20·10−6

Flux Metal - Heat source user subroutine(q above)

Argon as root gas

68

computing on a Linux cluster was used. By the use of the parallel computing option inMarc, the part is subdivided into several domains. For each domain, subsolutions arethen calculated in parallel on different processor, and an iterative procedure assembles thesubsolutions to the global solution. Ten 800 MHz processors were used in the presentstudy.

4 Integration of the Heat Transfer and Off-Line ProgrammingModels

By the integration of the heat transfer model above with the off-line programming systemIGRIP, a powerful, yet efficient tool for temperature prediction and optimization may beobtained. To that end, an interface translating the data; robot coordinates, welding speedsand currents between the two softwares was developed. This interface calculates a linearmotion between each robot point, which controls the moving heat source in the finiteelement calculation. The two softwares (IGRIP and Marc) have to be installed on thesame workstation since communication between different operating systems has not beenconsidered. The overall architecture of the simulation system is given in Figure 3.

Robot simulation

Geometrye.g. IGES

CAD/ CAM

FEATranslator

Welding path



IRB Controller

Complete robot code



Figure 3: The overall architecture of the simulation system.

5 Experimental

TIG (GTAW) welding was performed using an in-house robotised welding cell. The torchused is from Binzel AB and is linked to a six-axis robot from ABB, IRB1400. The power

69

source is a TIG Commander 400 AC/DC from Migatronic AB. Throughout all experi-ments, thoriated tungsten electrodes were used. Welding experiments were performed onboth plane plates and on a section of an aerospace part (in the following referred to asthe production part), namely a turbine component from the V2500 engine provided fromVolvo Aero Corporation. 1/13 was cut out of the real part, which originally consisted ofan inner, an outer ring and 13 vanes, see Figure 4. The reasons for the experiments onthe plane plates were twofold: to calibrate the heat source parameters used in the simu-lations, and to determine the emissivity in the infrared temperature measurements. Boththe plane plates and the production part were made of Greek Ascoloy with a thickness of1.25 mm. To avoid oxidation, argon was used as root gas in all weld trials. The types ofwelds performed were bead on plates, and no filler material was used. All plates were tac-welded together before performing experiments so that no gap or misalignments could beintroduced. Two special fixtures were designed, one for the plane plates and one for theproduction part, Figure 4.

Vane

Inner ring

Outer ring

Figure 4: Fixture for plane plate (left) and fixture with component (right).

To obtain temperature measurements, both thermocouples and high-resolution infrared(IR) emission measurements were used. Six thermocouples were positioned perpendicu-lar to the welding direction. The first gauge was positioned as close to the melted zoneas possible at a distance of 4 mm from the center of the weld. The second and third werepositioned 0.5 mm radially from the previous one. The remaining gauges were positioned1.0 mm radially from the preview one. The sampling frequency for all thermocoupleswas 270 Hz. The IR-camera used is a VARIOSCAN High Resolution, from JENOPTIK,Laser, Optik, Systeme Gmbh, which works in the IR radiation spectrum of 8 - 12 μm.The camera was used both in a line scan mode with a scanning frequency of 270 Hz aswell as in a full-frame mode with a frequency of 1 Hz. The analysis of the IR measure-ments were made using the IRBIS Plus software provided by JENOPTIK. A comparisonbetween the IR results with the thermocouple was made. Different techniques to soot theplane plates were evaluated to reduce emissivity -dependency in the IR-measurements.Finally a method using an acetylene flame was selected. This technique was also used onthe production part.

70

6 Results and Discussion

The results of the robot programs made off-line showed a high accuracy and very littlefine-tuning after calibrations had to be made. The method appears as a powerful tool,particularly in small batch production such as within the aerospace industry. The com-putational time from the parallel calculation was 34 hrs for the production part. Thepredicted fusion zone was 5.0 mm on the top-side and 4.8 mm at the root-side whichagreed well with measured widths. The predicted and IR-measured temperature historiesin the point located 4 mm and 7 mm from the center of the weld are given in Figure 5.The IR-measurements were performed twice, corresponding to the captions IR1 and IR2in the Figure 5. There is an excellent agreement between predictions and measurementsfor the 4.0 mm case both in peak temperature and in the cooling history. The agreementfor the 7.0 mm case is not as good as for the 4 mm case but still good. An example ofthe comparison between the thermocouple and the IR-measurements are given in Figure6. There is a very good agreement, which implies that that the technique to soot workedwell. The reason why data is lacking in the IR-curve is that the camera-system collectsdata during a maximum time interval of 20 seconds. This data then have to be writtento disk before a new sampling sequence can be gathered. A more extensive evaluationbetween thermocouple and IR measurements is planned.

The overall conclusion from the simulations is that the model predicts the thermal cyclevery well. However, further research is needed until welded structures can be optimizedwithout experiments. The temperature predictions are naturally dependent on the heattransfer coefficients (boundary conditions in Table 1). To determine these values, exper-iments are required since the heat flow can be convection dependent, specifically if theworkpiece and fixturing are small. Also, the heat source parameters (in the expressionfor q above) have to be calibrated by experiments. The on-going work at the laboratoryat University Trollhättan/Uddevalla to numerically model the magnetohydrodynamics ofthe arc and to predict the shape of the molten pool by the use of Computational FluidDynamics (CFD) techniques seems as a promising tool to compute the heat source para-meters without experiments. Such a model will establish a direct relationship between thewelding current, speed, voltage and the shape of the molten zone, which can be used asa boundary condition in the temperature predictions in the solid region. Such model willalso increase the knowledge of the stirring of the weld pool, the weld pool surface shapeand the physics of the arc. Several extensions of the modeling work described in this arti-cle are possible. The simulations can be extended to compute residual stresses, distortionand in the longer term to predict fracture strength and fatigue life of a structure. To extendthe modeling work to include filler wire and pulsed current would also be valuable.

7 Summary and Conclusions

An engineering method and a simulation tool to define robot trajectories and to predictthermal histories on parts with complex geometries have been developed. The methodwas evaluated on a part with a complex shape where robot weld paths were defined off-

71

0 5 10 15 20 25 30 35

250

750

1250

Time (s)

Tem

pera

ture

( o C

)

SimulationIR 1IR 2

0 5 10 15 20 25 300

250

500

Time (s)

Tem

pera

ture

( o C

)

SimulationIR 1IR 2

Figure 5: Predicted and measured temperature-time histories, 4mm (top) and 7mm (bot-tom) from the weld centerline.

72

0 5 10 15 20 25 30 35

250

750

1250

Time (s)

Tem

pera

ture

( o C

)

SimulationIR 1IR 2

Figure 6: Comparison between thermocouple and IR measured temperatures.

line, and automatically downloaded to a FEM-model where transient temperatures werepredicted. These predictions were compared with experimental measurements using boththermocouple and infrared emission measurements and good agreement was found. Thedescribed method provides a promising means to construct and optimize torch trajecto-ries and process parameters off-line. Using this system, thermal histories can be predictedon complex shaped parts and thereby resulting changes in microstructure and mechani-cal properties be estimated. The models used may after further development enable theoptimization of welding processes, thus increasing productivity and reducing the need ofweld trials.

8 Acknowledge

The authors wish to acknowledge the guidance in the temperature measurements of PerHenrikson (Volvo Aero Corporation), and the assistance in the laboratory by XavierGuterbaum (University of Trollhättan/Uddevalla) and Börje Nordin (Volvo Aero Corpora-tion). Appreciation is expressed to Peter Jonsson of Volvo Aero Coropration for providingsamples for this research and to Anita Hansbo of University of Trollhättan/Uddevalla forlinguistic revision. The work was funded by the Foundation for Knowledge and Compe-tence Development and EC Structural Founds.

9 References

1. Eagar T. W., Tsai, N. S. 1983. Temperature Fields Produced by Traveling Distrib-uted Heat Sources. American Welding Society Journal 62(12) 346-s to 355-s.

73

2. Gu, M.; Goldak ,J.; Hughes, E. 1993. Steady state thermal analysis of welds withfiller metal addition. Canadian Metallurgical Quarterlv. 32 (1): 49-s to 55-s.

3. Kou, S.; Le Y. 1983. Three-dimensional heat flow and solidification during Au-togenous GTA Welding of Aluminum Plates. Metallurgical Transactions A. 14A.:2245-s to 2253-s.

4. Radej, D. 1992 Heat Effects of Welding: Berlin: Springer Verlag.

5. Goldak, J.; McDill, M.; Oddy, A.; House, R.; Chi, M.; Bibby, M. 1987. Computa-tional Heat Transfer for Weld Mechanics. Proc. of Int. Conf. on Trends in WeldingResearch, Advances in Welding Science and Technology. Eds S. A. David: 15-20.Metals Park ASM Int.

6. Jonsson, M.; Karlsson, L; Lindgren, L.E. 1985. Deformation and Stresses in ButtWelding of Large Plates with Special References to the Material Properties, J. ofEng. Mat. And Tech. 107: 265-s to 270-s.

7. Lindgren, L.E.; Karlsson, L. 1988. Deformation and Stresses in welding of ShellStructures. Int. J. for Numerical Methods in Eng. 25: 635-s to 655-s.

8. Bolmsjö, G.; Olsson, M.; Brink, K. 1997. Off- line programming of GMAW roboticsystems - a case study. Int. J. for the Joining of Materials, 9 (3): 86-s to 93-s.

9. Buchal, R.O.; Cheras, D.B.; Sassani, F.; Duncan J.P. 1989. Simulated Off-LineProgramming of Welding Robots. Int. J. of Robotics Research 8 (3): 31-s to 43-s.

10. Bolmsjö, G. 1999. Programming robot welding system using advanced simulationtools. Proc. of the International Conf. on the Joining of Materials JOM-9, 284-291.May 16-19, 1999, Helsingør, Denmark.

11. Walter S. 1994. Simulation and Calibration for Off-line Programming of IndustrialRobots. Proc. of Computer Technology in Welding: Paper 54. Paris 15-16 June.

Paper II

Non-contact Temperature Measurements using anInfrared Camera in Aerospace Welding Applications

Per Henrikson and Mikael Ericsson

Presented 6th International Conference on Trends in WeldingResearch, Pine Mountain, Georgia, USA, April 15-19, 2002

Non-contact Temperature Measurements using anInfrared Camera in Aerospace Welding Applications

Per Henrikson and Mikael Ericsson

Abstract

This paper describes the application of infrared (IR) thermal imaging and temper-ature measurements in welding applications, both on single plane plates and on anaero engine turbine component with complex geometry. Temperature profiles weremeasured on the plates using thermocouples (T/C) in combination with an IR camerasystem, and the results were compared. The IR camera was used both in line scanmode (270 Hz scan frequency) and in full frame mode (1 Hz frame rate). Differentmethods of surface treatments have been tested to handle the problem of the surfaceemissivity variations due to oxidation during welding. Results from measurementsusing thermocouples and IR camera is presented in the paper as well as temperaturemeasurements using the IR camera on an turbine exhaust case (TEC) engine compo-nent.

1 Introduction

Experimental temperature data is required in welding research and development, and forvalidation of numerical simulations of the welding process (Ref. 1). Temperature mea-surements near or in the weld pool are of special interest, and this is a challenge in selec-tion and application of measurement methods and for assessment of measurement quality.

1.1 Thermocouple Measurement in Welding.

Temperature measurements using T/C type K (NiCr-NiAl) is commonly used in weldingresearch. They are relatively inexpensive and can be used at the high temperatures nearthe weld pool. There are however some issues that need to be addressed concerning mea-surement quality when using T/C Type K in welding applications (Ref. 2). Measurementresponse time is critical, especially when the T/C is installed close to the weld. Generally,it is not possible to state the response time for a single T/C in this situation, since it isthe whole measurement system response time that is measured. The T/C wire diameteris an important parameter to consider when discussing response time, but also the attach-ment method is an important factor, which effects the response time. Another issue ismeasurement system calibration (including T/C wire), which must be performed over thewhole measurement range. Measurement inaccuracy needs to be addressed when usingT/C Type K at high temperatures in the range 1250 ◦C - 1372 ◦C, which is above the

78

recommended range of use. There are however other T/C types that can be consideredto be used in welding experiments, for example T/C type B (Pt-30%Rh-Pt-6%Rh) whichcan be used at higher temperatures, but can be more difficult to install.

1.2 Infrared Radiation Measurement in Welding.

Infrared radiation measurement of temperature has several advantages compared to T/Cmeasurement. Using a pyrometer, non-contact temperature measurement can be done ona single point on the object (Ref. 3). A scanning mirror IR-camera with a photon detec-tor is an attractive device in welding research and applications. It allows both full fieldtemperature measurement, as well as high speed measurement using line scanning. Theuse of IR cameras for quantitative temperature measurements in TIG (GTAW) welding isin many cases limited by the difficulties to handle the surface emissivity variations andthe reflected radiation from the electrode during welding. When performing spectral radi-ance temperature measurements, the spectral radiance temperature measured is not onlydepending on the object emissivity, but also on the wavelength spectral band used for themeasurement (Ref. 4). The emissivity used in the measurement must then be spectralemissivity used for the detection device. A number of issues must be addressed when per-forming IR radiation temperature measurements in the weld pool. One important factoris the uncertainty in the surface emissivity in the melt at the measurement wavelength,and the contribution from electrode background reflection (Ref. 5). Another factor toconsider is the occurrence of slag in the melt (Ref. 6). The melted steel and the slaghave different radiance due to their emissivity, which are different. Slag has in generalhigher emissivity than steel and it appears hotter compared to the melted steel at the sametemperature. There is also an uncertainty in the IR measurement due to the change insurface emissivity during material solidification and the effect of the surface oxidationin the cooling phase. The surface emissivity is assumed to increase during cooling andoxidation, but it is difficult to measure actual emissivity during the cooling phase (Ref.7). A well known fact in radiation measurements is that when there are uncertainties inthe emittance of a surface it is general best to do measurement at as short wavelength aspossible. This is due to that the spectral radiance as a function of temperature increasesvery rapidly towards shorter wavelengths. A given uncertainty in emissivity then leads tosmaller uncertainties in temperatures at shorter wavelength.

2 Instrumentation

2.1 Thermocouple Instrumentation on Plates.

Type K thermocouples were used in all experiments on plates. Six T/C were positionedperpendicular to the welding direction. The first gauge was positioned as close to meltedzone as possible at a distance of 4 mm from the center of the weld. The rest of theT/C were positioned at 4.5, 5, 6, 7 and 8 mm from the center of the weld. The T/Cwere coupled to a signal conditioning unit and the T/C signals were amplified to give acalibrated and linearised output from 0 to 5 volt. A PC based data acquisition system was

79

used for sampling of the T/C signals, and the measurements were presented on-line onthe PC monitor and measurement data was written to disk. The complete measurementsystem (including T/C wire) was calibrated over the whole measurement range. The T/Cused in the test had a wire diameter of 0.11 mm and surrounded by ceramic cement andInconel protection for lead out. The T/C wires were attached to the plate using spotwelding.

2.2 Infrared Camera.

The IR camera used is a Varioscan 3021-ST high resolution 16 bit Stirling cooled camerafrom Jenoptik GMbH, Germany. The camera uses scanning mirrors to image the mea-surement object on a point detector. The camera resolution is 360(h)x240(v) pixels, andthe operating wavelength range is 8 μm - 12 μm. The camera detector is of MCT type(HgCdTe). The camera has four filters for different calibrated temperature ranges, and atthe highest measurement range, the system is calibrated in the temperature interval from200 ◦C -1700 ◦C. For higher temperatures linear extrapolation of the Planck black bodyradiator is used. The camera was used both in line scan mode with a horizontal line scan-ning frequency of 270 Hz, as well as in full frame mode with a frequency of 1 Hz. Themeasurements and analysis of the IR tests were made using the IRBIS Plus software fromJenoptik.

2.3 Surface Preparation.

Different techniques of surface treatment have been tested to handle the problem with sur-face emissivity variation on the metallic surface due to oxidation outside the weld joint.The surface treatment should ideally have a low emissivity variation over a wide tem-perature range and should be insensitive to emissivity variations due to oxidation duringwelding. Diffuse black high temperature paint for engine exhaust pipes was tested. Initialtests showed that the paint could be used up to about 650 ◦C. In order to find a surfacetreatment that could be used at higher temperatures, different kind of soot deposition tech-niques were tested. From the experiments it was found that by using an acetylene/oxygenflame, a thin high temperature resistant soot layer could be produced. The optimal gasmixing is reached by starting from a pure acetylene flame, and then gradually increasethe oxygen gas until no soot is visible in the flame. An emissivity value of 0.96 has isreported in Ref. 8 for soot applied to a solid in the range 50-1000 ◦C.

3 Experimental Setup

TIG (GTAW) welding was performed using an in-house robotised welding cell. Thetorch used is from Binzel AB and is linked to a six-axis robot from ABB, IRB1400. Thepower source is a TIG Commander 400 AC/DC from Migatronic AB. Throughout allexperiments thoriated tungsten electrodes were used. A special fixture designed to avoiddistortion has been used during the welding of the plates. The aero-engine component,

80

a part of a V2500 engine turbine exhaust case (TEC) from Volvo Aero Corporation, wasTIG welded using the robotised welding cell. A segment of 1/13 was cut out of the TEC,which originally consists of an inner and outer ring and 13 vanes, see figure 1. The TECis made of Greek Ascoloy with a vane thickness of 1.25 mm. The vane was spot weldedbetween the outer and inner ring.

Figure 1: Overview of the aero engine component and the IR-camera setup.

4 Measurements

Welding experiments were performed on both plane plates and on the turbine component.The purpose of the T/C measurements was to get reference data against which the IRmeasurements could be calibrated. Initial test measurements were performed on T/Cinstrumented stainless steel plates. For comparison of the IR measurements against T/Cmeasurements, the acetylene/oxygen sooting technique was used on the plates. Using thistechnique, all but the weld joint remained sooted during and after the weld experiment.This allowed quantitative temperature measurement on the outside the weld joint using theIR camera. In fig. 2 the TEC component is shown after a weld test. In the experiments,the Greek Ascoloy plates and the TEC vane had a thickness of 1.25 mm. The stainlesssteel plate thickness was 2.0 mm. To avoid oxidation on the backside during welding,Argon gas was used as root gas in all weld trials. The types of welds performed werebead on plates, and no filler material was used.

4.1 Measurement on Plates.

Figure 3 are showing temperature measurements on a Greek Ascoloy plate with 6 T/Cinstalled.

81

MiddleMiddle

FrontFront

BackBack

Figure 2: Weld test on the Turbine exhaust case (TEC).

The T/C were spot welded to the plate and positioned in radial direction to the weld. Thefirst T/C was mounted as close as possible to the weld (fig. 4). The position of the T/Cwas measured in a microscope after welding. These T/C positions were later used in theanalysis of the IR line scan temperature images, for which the corresponding pixels wereselected and used for comparison of temperatures.

When using the IR-camera in line scan mode, two different acquisition modes can beused. In one mode, 5600 lines are scanned continuos at a rate of 270 lines/s and then thedata is saved do disk. The scanning time and the readout time is both 21 s, which meansthat only a part of the whole temperature cycle will be measured, and this can be seen infig. 16. In the other line scan acquisition mode, images like those in fig. 5 are captured,with an image size of 360(h)x240(v) pixels. After a camera readout time of approx. 0.1 s,which is indicated in the fig. 5 by the white field, another line scan image is taken, and soon until the end of the weld test. During the IR camera measurements a macro lens wasused with a working distance of 100 mm to the plate, see figure 15. Due to the viewingangle, only a small section of the plate will be in focus. In the camera software, a cameraline perpendicular to the welding direction (se fig. 4) is selected at the focused position onthe plate and scanned at 270 Hz. The optical magnification in the IR camera system givesa spatial resolution of 7 pixels/mm on the object during line scan. Care has been taken toselect the correct pixel in IR scan lines that correspond to the T/C position on the plate.In fig. 6 radial temperature profiles representing different line scans have been plotted.

As can be seen from fig. 6, the radial temperature profile for the sooted stainless steel platehas three peaks. For the curve going through the maximum temperature in the weld pool,the soot is probably attached to the surface outside the weld pool for temperatures untilthe curve drops on each side of the weld. For comparison, temperatures measured withT/C positioned at 4, 5 and 6 mm from the center of the weld are also plotted in figure 6.The agreement between IR- and T/C measurements is good in this region, as can be seen

82

10 20 30 40 500

250

500

750

1000

1250

Time (s)

Tem

pera

ture

( o C

)

4mm4.5mm5mm6mm7mm8mm

Figure 3: Temperature profiles (T/C) on Greek-Ascoloy, plate thickness 1.25 mm, I=75A, v=2.5 mm/s.

4.0 mm

150 mm

50 m

m

AB Weld line

Wel

ding

di

rcet

ion

Line

sc

an

Figure 4: The drawing shows the measurement positions on a plate for T/C and IR cam-era line scan.

in figure 6. Over the melting temperature, the soot layer has disappeared and the surfaceemissivity changes instantly. This can be seen as a sudden temperature drop at the edgeof the weld pool. At the center of the weld, the high temperature peak is due to reflectionof radiation from the electrode, and this is explained from the results in figure 14 and 15.The line plotted 6.7 s after the maximum temperature profile shows that the temperaturewave has propagated far out on the plate, and that temperature in the center of the weld isin the same range as at the edge of the weld (where the soot is still attached). The lowertemperature seen at the weld joint is due to a different surface emissivity compared tooutside the weld. The surface emissivity of the weld is changing (increasing) over timedue to material solidification and surface oxidation. During the IR camera measurements,a constant emissivity value of ε =0.99 was used at every pixel in the image.

83

Figure 5: Two 360 (h)x240(v) infrared line scan temperature images taken during weld-ing. One image represents 0.9 s, and the welding direction is from right to left in thepictures.

4.2 Measurements on the Aero-Engine Component.

Several welding experiments were performed on the component and temperatures weremeasured using the IR camera at 1 Hz full frame rate and the 270 Hz line scan mode.No thermocouples were used in the experiments on the TEC component. In fig. 7 isshown a full frame (1 Hz) IR temperature measurement during welding on the sootedTEC (compare the weld path and the component set-up in fig. 2 and fig. 1). In thisexperiment the filter range was up to 800 ◦C, therefore temperatures higher than 800 ◦Care shown in black in the image. On the outer ring of the component, there are two rigidsupports (which can be seen in fig. 1), and the cooling effect of these supports can beclearly seen in fig. 7.

Line scan temperature measurements were also performed during the weld test, and acontour plot of the temperature distribution of the middle section is seen in fig. 8. Thecooling effect of the support is clearly also in this graph.

Figure 9 shows line scanned temperature profiles on the front part of the vane. All profilesare measured to the right side of the weld as seen in fig. 7 and in the contour plot in fig. 8.The temperature profiles in figure 9 correspond to the radial positions for which the T/Cwere instrumented on the weld tests on the Greek-Ascoloy plane plates. Figure 10 showsfour temperature profiles from the middle part of the TEC. The highest temperature in thispart of the component is much lower compared to the front and back part due to the bigheat sink on the outer ring support. The measured IR line scan temperature profiles forthe back part of the TEC is shown in figure 11. Here the wall thickness in the vane andthe outer ring can be assumed to be homogenous, like the front part.

84

50 100 150 200 250

500

1000

1500

2000

Pixel

Tem

pera

ture

( o C

)

Max+3.0s+4.8s+6.7sT/C

Figure 6: IR line temperature profiles and T/C measurements in radial direction from theweld for a stainless steel plate.

Figure 7: Full frame thermal image of turbine component showing the heat propagationin the range 100-800 ◦C

85

0 100 200 3000

1000

2000

3000

4000

5000

250

500

750 10

00

Weld line

Pixel

Lin

e nu

mbe

r

Figure 8: Temperature contour plot of the middle section of the turbine exhaust case(TEC) measured using line scanning.

5 10 15 200

250

500

750

1000

1250

Time (s)

Tem

pera

ture

(o C)

4 mm5mm6mm7mm

Figure 9: Infrared line scan temperature profiles measured on the front part of the TEC.

86

5 10 15 20200

400

600

800

Time (s)

Tem

pera

ture

(o C)

4mm5mm6mm7mm

Figure 10: Infrared line scan temperature profiles measured on the middle part of theTEC.

5 10 15 20200

450

700

950

1200

Time (s)

Tem

pera

ture

( o C

)

4mm5mm6mm7mm

Figure 11: Cross section of a welded plate.

87

5 Discussion

5.1 Thermocouple Measurements.

Close to the weld, the temperature gradient is very high and the spot welded T/C installa-tion will effect the transient response of the T/C, and this may cause a significant error inthe measurement. Comparison of T/C and IR measurements indicates that there may bea significant difference in peak temperature between the T/C temperature measurementand the IR measurement near the weld joint. The spot-welded T/C had a diameter about0.7 mm on the plate. Measurements were done to study the effect in transient responseand measured peak temperature, using a smaller spot weld diameter, approx. 0.56 mm.In order to study the potential lag effects of different spot sizes, four T/C were installedin pairs on two plates, see fig. 4. One plate had T/C separated at the radial distance 3.8mm (C1 and C2) and 4.3 mm (D1, D2) from the center of the weld. The T/C pairs atthe same radial distance were separated 3 mm in the axial direction. The other plate hadthe T/C separated at the radial distances of 4.3 mm (A1, A2) and 4.8 mm (B1, B2). Thetwo plates were welded with different welding currents. The reason for this is that at thefirst plate, the first T/C pair was at the very edge of the weld seam. On the other plate thewelding current was lowered so the T/C should be about 1-1.5 mm away from the edgeof the weld pool. The two measurements have been plotted in the same graph, se fig. 12.Experiments were also done with smaller spot weld sizes, but these did not survive theweld test.

25 30 35 40 45850

950

1050

1150

1250

1350

Time (s)

Tem

pera

ture

(o C)

A1A2B1B2C1C2D1D2

Figure 12: Comparison of T/C peak temperature difference at the thermal gradient nearthe weld for different T/C spot weld sizes and radial positions.

As can be seen, the T/C pair (C1, C2) closest to the weld, and at a distance of 3.8 mm fromthe center of the weld, is showing a relatively large difference in peak temperature. TheT/C pair (D1, D2) at a distance of 4.3 mm shows a smaller difference in peak temperature.In the second test, the weld width was smaller, and the T/C pairs (A1, A2) and (B1,B2)was positioned 4.3 and 4.8 mm from the center of the weld. As can be seen in fig. 12,

88

there is a small peak temperature difference for T/C pair (A1, A2), but for T/C pair (B1,B2) the two different spot weld sizes shows identical temperatures. During the passingof the TIG weld temperature transient, no difference in transient response time due tothe different spot weld sizes could be measured during the test, only a difference in peakresponse temperature. The peak T/C temperature difference due to surface attachmentsize is significant only very close to the weld. Using a non-contact fast response IRdetector in this region, it can be expected to give even higher peak temperatures (if thesurface emissivity and the background reflection are known).

5.2 Infrared Image Data Processing.

In order to generate temperature profiles at different radial distances from the weld, thecenter of the weld has to be defined in the IR image. Two different methods based onpixel averaging and peak temperature detection have been used to accomplish this, andboth have been found to give the same result, within ± one pixel. In fig. 13 a temperaturecontour plot from a weld test on a sooted stainless steel plate is shown. In fig. 14 the hightemperature region of fig. 13 is contoured.

0 100 200 3000

1000

2000

3000

4000

5000

250

500

500

500

500

750

750

750

1000

1000

Pixel

Lin

e nu

mbe

r

Figure 13: IR line scan temperature contour plot from a weld test on a sooted stainlesssteel plate.

By counting the number of lines between the two high temperature peaks in figure 14, andusing the line sampling frequency, f =270 Hz, together with the welding speed (v=2,5mm/s), gives that the weld torch has traveled the distance of 3.22 mm between the twopeaks. Using the electrode to plate distance 1.5 mm gives the angle =25 ◦ in figure 15,which is close to the viewing angle during the experiments

In fig. 16, temperature measurements using T/C and IR are shown from a welding exper-iment on a sooted Greek-Ascoloy plate. The measurement position is just at the edge ofthe weld, 4 mm in radial direction from the center of the weld. The explanation for thehigh IR temperatures up to 500 ◦C in the beginning of the heat transient is due to the weldtorch that comes into the field of view of the camera. The next part of the curves shows

89

145 150 155 160 165

185019001950200020502100215022002250 75

0

1000

1000

1250 12

50

1500 15

00

1750 1750

2000

2000

Pixel

Lin

e nu

mbe

r

Figure 14: Temperature contour plot showing the peaks where the electrode positionis right above the measurement position (line 1844) and the peak due to reflection ofelectrode radiation (line 2192).

TorchTorch

αα

IR

camera

Welding direction

Figure 15: Optical and geometrical calculation gives torch positions of maximum re-flected radiation in the IR-image.

the temperature peak response. It is seen that the IR measurement reach a higher peakvalue (TIR=1418 ◦C) compared to the T/C measurement (TT/C=1243 ◦C), and this canalso be seen in the measurements in fig. 6 for stainless steel. From the T/C response tests,it was stated that an IR measurement is expected to reach a higher peak temperature veryclose to the weld, compared to a T/C measurement. In all experiments, the T/C positionand the IR scan line is not taken at the same position on the plate, see fig. 4, but are aseparated by a few millimeters. Due to process variations, the IR measurement positionmay be inside the melt, explaining the sudden drop in temperature as a result of a suddenemissivity variation. As can be seen in figure 17, during the cooling phase, the IR and T/Ctemperature curves show good agreement. This means that surface emissivity is close tothe value set in camera, ε = 0.99, indicating that the surface oxidation during the coolingphase results in a high emissivity. It should be pointed out that the agreement between theT/C and IR measurements increases with the radial distance from the weld, and this canbe seen in fig. 17.

90

10 20 30 40 50 60 700

250

500

750

1000

1250

Time (s)

Tem

pera

ture

( o C

)

T1IR 1

Figure 16: Comparison of T/C and IR line scan temperatures on a sooted Greek-Ascoloyplate at 4 mm from the weld line.

An example of this is shown in figure 17 for a stainless steel plate, showing good agree-ment between T/C and IR measurements at 6 mm and 7 mm from the weld line.

10 15 20 25 30 35200

400

600

800

Time (s)

Tem

pera

ture

(o C)

TC 6mmIR 6mmTC 7mmIR 7mm

Figure 17: T/C and IR temperatures on a sooted stainless steel plate 6 mm and 7 mmfrom the weld line.


Temperature measurements have been successfully performed on an aero engine turbinecomponent using an infrared camera system. Both full field temperature images and timeresolved line scan profiles have been measured and analyzed. By deposition of a sootlayer on the metal surfaces to be welded, a surface with high emissivity was producedthat made it possible to handle the emissivity variation due to surface oxidation outside

91

the weld joint, and to suppress reflected radiation. Thermocouple- and infrared measure-ments have been performed on plane plates made of stainless steel and Greek Ascoloy,and comparative analysis has been made of the results. Infrared radiation temperaturemeasurements has also been made in the weld pool and during solidification and the cool-ing phase, and the results have been analyzed and problem areas have been identified thatpromote further work in this field.

7 Acknowledgements

The authors would like to thank Xavier Guterbaum (University of Trollhättan/Uddevalla),for the assistance and valuable discussions in the laboratory. The authors also would liketo thank Dr. Per Nylén for comments and suggestions in the preparation of the paper.This work was made in collaboration between Volvo Aero Corporation and Universityof Trollhättan/Uddevalla. The work done by the University of Trollhättan/Uddevalla wasfunded by the Foundation for Knowledge and Competence Development and EC Struc-tural Founds.

8 References

1. T.Zacharia, S.A. David and J.M. Vitek, Effect of Evaporation and Temperature-Dependent Material Properties on Weld Pool Development, Metallurgical Transac-tions B, Volume 22B, 233-240 (1991)

2. Manual on the use of Thermocouples in Temperature Measurement, ASTM SpecialTechnical Publication 470B (1981)

3. D. Farson, R. Richardson and X. Li, Infrared Measurement of Base Metal Temper-ature in Gas Tungsten Arc Welding, Welding J., Vol 77(9), 396-401 (1998)

4. D.P. DeWitt and G.D. Nutter, Theory and Practice of Radiation Thermometry, J.Wiley & Sons, Inc. (1988)

5. H.G. Kraus, Experimental Measurement of Stationary SS 304, SS 316L and 8630GTA Weld Pool Surface Temperatures, Weld. J., vol. 68(7), 269-279 (1989)

6. G.R. Peacock, Thermal Imaging of Liquid Steel and Slag in a Pouring Stream, Proc.Of SPIE Vol 4020, Thermosense XX22, 50-60 (2000)

7. P.D. Jones and E. Nisipeanu, Spectral-Directional Emittance of Thermally Oxidized316 Stainless Steel, Int. Journal of Thermophysics, Vol. 17, No4, 967 - 978 (1996)

8. Table of Emissivity of Various Surfaces, www.mikroninst.com, Mikron InstrumentCompany, USA.

Paper III

Three Dimensional Simulation of Robot path, HeatTransfer and Residual Stresses of a welded Part

with Complex Geometry

M. Ericsson and P. Nylén and D. Berglund and R. Lin-Peng

Int. J. for the Joining of Materials 17(2) 2005

Three Dimensional Simulation of Robot path, HeatTransfer and Residual Stresses of a welded Part

with Complex Geometry

M. Ericsson and P. Nylén and D. Berglund and R. Lin-Peng

Abstract

In this article a simulation system is presented that combines computer aided ro-botics software used to define the welding operation, with a finite element modelthat predicts temperature-time histories and residual stress distributions for weldingapplications. The objective is to develop a tool for engineering processes in whichrobot trajectories and welding process parameters can be optimized off-line on partswith complex geometries. The system was evaluated on a stainless steel gas turbinecomponent. Temperature dependent properties and phase change were included inthe analysis. The turbine component was welded using an in-house TIG welding cell.The assumptions and principles that underpin the modeling techniques are presentedtogether with predicted temperature histories, residual stresses, and fixture forces.Predicted residual stresses were compared with neutron diffraction measurements.

Keywords: Off-line programming, Finite element analysis, Residual stress measur-ments

1 Introduction

A large number of aerospace components have complex shapes and are manufactured inseveral stages, often including joining operations performed by robots. Joining by weld-ing, however, induces changes in the microstructures of the base metals and can generateunwanted stresses and deformation. To avoid deformations, expensive and complex fix-tures often have to be used. Furthermore, the planning of optimal welding sequences inaerospace component welding is difficult and requires highly experienced operators. Pro-gramming of welding robots is usually performed manually by means of the ’jog teach’method. This method requires that the robot is off-line i.e. not used in production andthat the part is stationary. The robot arm is jogged through the program under reducedpower and at reduced speed, via a joystick. Generating a path by hand in this way can betime consuming. On a complex geometry, it is virtually impossible for a programmer tomaintain a constant gun velocity and a constant distance from and orientation to, the part.Correct welding parameters i.e. parameters which generate a full penetration of the part,are usually derived by means of experimentation. Therefore a simulation tool that couldbe used for the evaluation of features such as structural behavior, welding sequences and

96

fixture solutions during the design stage would be desirable. The use of such a tool wouldmake it possible to evaluate manufacturing processes in the early stages of product de-velopment and would reduce both the number of welding experiments required and theneed for welding operator experience. The tool should preferably be capable of simulat-ing a number of facets of the process including the welding torch path, the detection ofcollisions between the torch and workpiece, and the optimization of welding parametersand fixture solutions relating to penetration and component deformation behavior. Thus,a combination of the simulation techniques, finite element analysis (FEA) and computeraided robotics (CAR) is necessary.

FEA simulation of thermal history, residual stresses and distortion has been performedsince the early 1970s and several papers have been presented [1-6]. Large complex simu-lation models of three-dimensional components are however still rare, mainly due to thelack of computational power. The underlying reason is that, in order to be able to computetemperature and residual stress fields in the affected zone, a very fine discretization of thespace variable is required to properly accommodate sharp gradients.

The CAR technology is also well known and has been an established research area formany years [7, 8, 9]. Using this technique, the programming of the robot is transferredfrom the workshop to a computer system where the programming can be performed with-out disturbing production. The technique can be used to simulate the welding torch path,to detect torch-workpiece and torch-fixture collisions, and to control torch orientation aswell as electrode distance.

In this project an integrated approach, using two commercial CAR and FEA codes, mak-ing it thus possible to optimize torch trajectories, weld parameters and fixture solutionsoff-line, is described. The best welding parameters, i.e., the parameters that generate thelowest component deformation while keeping full penetration, can thus be found. Theobjective of this project is to describe the principles upon which the system is based andto demonstrate some of its capabilities. The principle of the OLP-FEA integration whichpreviously has been presented in [10] is summarized here for the sake of completeness.

2 Integration of CAR and FEA

The overall principle of the integrated system is given in Figure 1. The following stepsdescribe the procedure:

1. The component to be manufactured is created in a CAD/CAM system.

2. The model in step one is imported to the OLP software and the FEA software,either using a direct translator or by using a neutral file format, such as IGES orSTEP. In the CAR software a model of the work cell is created. A welding program(including robot motion, weld speed, and arc current, etc.) is developed. Checksfor collisions are also conducted.

3. The welding parameters, i.e., robot poses, weld speed and arc current, are exportedfrom the CAR software to the FEA software using an interface developed as part of

97

this project.

4. Thermal histories and residual stresses are predicted in the FEA software. An opti-mization of weld velocity is performed to generate the lowest component deforma-tion while keeping full penetration.

5. The optimized welding parameters are exported to the robot motion program. Atranslation of the program to a specific robot manufacturer language is carried out.

6. The final program is downloaded to the manufacturing equipment (robot and weldcontroller systems).

Robot simulation

Geometrye.g. IGES

CAD/ CAM

FEATranslator

Welding path



IRB Controller

Complete robot code



Figure 1: Simulation system architecture.

A more detailed description of the principles of the robot simulation (CAR), step twoabove, and the FEA simulation, step four, is provided in the following sections.

3 Computer Aided Robotics

Several commercial software packages for CAR exist (GRASP, IGRIP and Robcad etc.).The procedure using these systems can be summarized by means of the following steps.A more detailed description can be found in [7].

1. Modeling of the work cell


98

3. Programming of robot, positioner and other optional work cell equipment

4. Downloading of the program to the robot controller


6. Test running

The first step involves the construction of a geometrical model of the workpiece, includingboth a geometrical and a kinematic model of the work cell. These models are constructedusing drawings, CAD/CAM models, or if such not are available, by obtaining measure-ments of all the critical equipment positions in the cell. If this model is created in a CADsystem the data is imported to the CAR software either using a neutral interface (for in-stance IGES) or a direct reader. The work cell model might, alternatively, be constructeddirectly in the CAR system. Normally, kinematic models as well as geometric models ofrobots are pre-defined by the software manufactures.

In the second step a calibration of the model with the real cell is performed. This step caninclude several sub steps, such as tool point, work piece and signature calibration. [7].Tool calibration is performed in order to determine the tool center point and to determinethe weld torch orientation. A procedure using a measuring arrow in a fixed position in thework cell and moving the robot to this position in different directions is usually used. Thepositions from the real robot cell are then uploaded to the CAR software and a “best fit”is found using regression. The calibration of the workpiece and other critical componentssuch as fixtures is performed similarly by moving the robot to identified positions on thecomponent. These positions are recorded and uploaded to the OLP software where asimilar adjustment of the model is made. To find errors in the geometrical model of therobot, an arm signature calibration can be used. This calibration finds any deviation in thelength of the robot joints and in the zero points for the joints. A more accurate calibrationof the work cell can be obtained using an advanced external measuring system, such as alaser measuring system.

In step three the robot motion is programmed using either a high level language or aspecific robot language. If a high level language is used, the program is translated to thespecific robot language and downloaded (step four) to the robot controller system.

Usually equipment specific additional programming is needed (step five), which is per-formed manually at the robot. Validation of the program by test runs is finally performedin step six.

The IGRIP (Interactive Graphics Robot Instruction Program, Deneb Robotics) system wasused in this study. A tool calibration and a workpiece calibration were performed using therobot arm. Calibration of other objects, such as the welding table were performed usingmeasuring tape. The high level language GSL, which is the graphical simulation languagein IGRIP, was used for programming all of the devices in the cell. The programmed partwas a section of an aerospace part, a turbine component from the V2500 engine providedby the Volvo Aero Cooperation. A 1/13 sized piece was cut out of the original part, whichoriginally consisted of an inner and an outer ring and 13 vanes, see Figure 2a. Both theouter and inner ring were welded onto a steel plate, which acted as a fixture. The vane was

99

tack-welded between the outer and inner ring using manual TIG welding. The weldingpaths, including initial weld velocities, were then exported to the finite element softwarewhere predictions of temperature histories, residual stresses and fixture reaction forceswere performed. The principle of this model is described in the next section.

Vane

Outer ring

Inner ring

Outer ring

Vane

Outer ring

Figure 2: a) Turbine component consisting of an inner- and an outer ring and 13 vanes.b) experimental component (1/13 of the turbine component).

100

4 Finite Element Analysis

Numerical methods have been used since the beginning of 1970’s to simulate the weldingprocess [1-6]. The main focus has been to predict thermal histories, residual stressesand distortion. Different assumptions and simplifications have to be considered whenbuilding an FEA model. Examples of areas that have to be considered are part geometrysimplifications, the type of material to be used in the model, load conditions, heat transferand other boundary conditions and, in addition, numerical strategy. In this project thecommercial FEA program MARC from MSC Software was used. The model contained3056 shell elements and 3182 nodes, see Figure 3. A staggered approach was used for thecoupled thermal-mechanical simulation. This means that the updating of the geometryused in the thermal calculation is lagging one time step behind.

Weld path

Flange

Figure 3: Shell model of experimental component.

4.1 Boundary Conditions

All FEA problems are defined in terms of initial and boundary conditions. A typicaltype of initial condition for a welding application is the initial temperature that, in mostcases, is set to room temperature. Examples of the most important boundary conditionsare fixture forces and heat transfer coefficients between the part and its surroundings. Amodel of the arc between the electrode and the part is usually too complex to be integratedin the same model. The most common method is to use a moving heat source. User

101

subroutines were therefore developed to simulate a moving Gaussian surface distribution[4]. This distribution was preferred over a volumetric one since it reduces the number ofparameters to be fit and because the plates to be welded were considered thin (<1.5mm).The heat flux was expressed as [4]:

q =ηEIαqπ

e−αqr2

(1)

where q denotes the heat transferred to the workpiece, E the voltage, I the current, η theefficiency factor, q the concentration factor, and r the radial distance from the center ofthe heat source. The distribution was truncated in the radial direction, at a cut off limit of5% of the maximal heat input, as proposed by D. Radaj [4].

Natural convection was only used as the heat transfer boundary condition between the partand the surrounding environment. The flanges on the Inner- and Outer ring, see Figure2a, were assumed to be clamped in the model since no fixture was used. The Inner- andOuter ring were instead welded on a steel plate, see Figure 2b.

4.2 Material properties

The initial microstructure of the material consists of a mixture of ferrite and pearlite.In the numerical model the ferrite/pearlite to austenite transformation was assumed tooccur only if the highest temperature experienced by the material was greater then theAe3 temperature, Figure 4. Since the cooling rate of any austenite phase formed duringwelding is always higher than 0.3 ◦C/s between the Ae3 and martensite start temperature,it can safely be assumed that all austenite is transformed to martensite irrespective ofcooling rate. This assumption is supported by the results of the thermal dilatation forspecimens heated at 100 ◦C/s and then cooled at 10 ◦C/s, and at 0.3 ◦C/s, see Figure 4.

Both cases gave the same amount of martensitic transformation. It should be noticed thatit is assumed that pure martensite is formed. It should also be noticed that the martensitestart temperature, Point 1 in Figure 4, decreases when the cooling rate is increased. Thishas not been accounted for in the numerical model. The thermal dilatation for reheatedmartensite follows the martensite curve, Figure 4.

A thermo-elastoplastic model based on von Mise’s theory was used. It was assumed thatno creep strains occur during welding since the material is exposed to a high temperaturefor a very short period of time. The hardening behaviour of the material was assumed tobe isotropic and linearly piecewise. Transformation plasticity was not accounted for inthe model. The temperature dependent Young’s modulus and Poisson’s ratio are shown inFigure 5. Figure 5 also shows the virgin yield limit for the initial ferrite/pearlite mixture,σyf , and the yield limit of pure martensite, σym.

The yield limit of the material changes due to the phase transformation from ferrite/pearliteto martensite. If the peak temperature during welding had been higher than 850 ◦C (Ae3)the yield limit would have followed the curve for martensite when the material is cooled.The curve denoted σyf in Figure 5 is followed during cooling if the peak temperature had

102

0 200 400 600 800 1000-0.004

-0.002

0

0.002

0.004

0.006

0.008

0.01

Martensite curve

Ae3

1

Temperature ( oC)

ε th

0 200 400 600 800 1000-0.004

-0.002

0

0.002

0.004

0.006

0.008

0.01

Temperature ( oC)

ε th

Figure 4: Thermal dilataion, εth vs. temperature for, (above) a cooling rate of 10 ◦C/s,(below) a cooling rate of 0.3 ◦C/s.

been lower than the Ae3-temperature. To avoid convergence problems in the numericalcalculations, the minimum yield limit was set to 20 MPa and the maximum Poisson’sratio to 0.45. The temperature-dependent thermal properties are shown in Figure 6. Thethermal conductivity was increased by a factor of 4.7 when the temperature reached theliquidus temperature to account for connective heat transfer in the melted zone [4]. This

103

200 600 1000 14000

50

100

150

200

E

ν

Temperature (oC)

E [G

Pa]

200 600 1000 14000

0.1

0.2

0.3

0.4

0.5

ν

0 500 1000 15000

200

400

600

800

1000

σym

σyf

Temperature ( oC)

σ y (MPa

)

Heating, if Tpeak

>850 oC

Cooling,if T

peak<850 oC

Cooling, if Tpeak

>850 oC

Figure 5: Temperature dependent mechanical properties, (above) Young’s modulus, E,and Poisson’s ratio, ν, (below) Temperature dependent yield limit for the ferrite/pearlitephase, σyf , and yield limit for the martensitic phase, σym.

simplified model was used instead of more advanced CFD models to simulate the physicsin the molten zone, since the latter methods are computationally too demanding. The la-tent heat of melting was set to 338 kJ/kg, Tsolidus to 1480 ◦C and Tliquidus to 1600 ◦C.The emissivity factor used for the radiation boundary condition is shown in Figure 6.

104

0 500 1000 1500

100

300

500

700

900

c

λ

Temperature (oC)

C(J

/kg

⋅o C)

0 500 1000 1500

20

40

60

80

100

120

140

160

λ (W

/m⋅o C

)

0 500 1000 1500

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Temperature ( oC)

ε

Figure 6: (left) Temperature dependent conductivity λ, and heat capacity C. (right)Emissivity.

5 Experimental

The component was TIG welded using an in-house robotised welding cell. The torch usedis from Binzel AB (thoriated tungsten electrodes) and lined to a six-axis robot, ABB IRB1400. The power source is a TIG Commander 400 AC/DC from Migatronic AB. Thevane was spot-welded between the outer and inner ring prior to the welding of the seam.

105

Argon gas was used, both on the topside and on the root side to avoid oxidation of thecomponent. No filler material was used during the welding. A welding speed of 2.5 mm/sand a welding current of 80 A were used during the experimental.

The residual stresses measurements were performed on the dedicated neutron diffrac-tometer REST [11] at the Studsvik Neutron Research Laboratory, Sweden. The principleof strain measurement by neutron diffraction is shown in Figure 7. In a typical neutrondiffraction experiment a collimated neutron beam, of wavelength λ, is diffracted by thepolycrystalline sample and passed through a second collimator to the detector. The slitsof the two collimators define the gauge volume.

Knowing the wavelength λ of the monochromatic neutron beam and measuring the 2θ an-gle corresponding to the maximum value of the intensity in the obtained diffraction peak,the interplanar distance dhkl (where h, k and l are the Miller indices of the investigatedlattice plane) can be evaluated using Bragg’s law:

λ = 2dhkl sin θ (2)

The corresponding lattice strain is defined as:

εhkl =dhkl − d0

d0(3)

where d0 is the interplanar distance in a stress-free material. For an isotropic material,the residual stresses are calculated using Hooke’s law which, in case of triaxial (x, y, z)measurement is written as:

σi =E

(1 + ν)(1 − 2ν)[(1 − ν)εi + ν(εj + εk)] (4)

with i, j, k corresponding to x, y, z and where E is Young’s modulus and ν is Poisson’sratio. If the stress along the specimen normal, σ33, can be assumed to be zero, a so-calledsin2 ψ technique [12] can be employed for residual stress analysis. For such a bi-axialstress state, the in-plane stress parallel to φ (Figure 7), σφ, is correlated to the measuredstrain along a sample direction (φ, ψ) by the following equation [12].

dφψ − d0

d0=

(1 + ν

E

)σφ sin2 ψ − ν

E(σ11 + σ22) (5)

where σ11 and σ22 are the stress components parallel to the in-plane specimen co-ordinates.For the definition of (φ, ψ ), see Figure 7. The stress can be calculated from the slope ofdφψ vs. sin2 ψ distribution. The d0 can be replaced by the measured interplanar spacingalong ψ = 0◦ without introducing significant error. More details on stress analysis bymeans of diffraction measurement can be found in [12].

The Fe-(211) plane was used for the transverse stress measurement. With a nominalwavelength of 1.7 Å, the diffraction angle was found to be about 2θ = 93.5◦. The slit

106

Figure 7: A schematic representation of the strain measurement by neutron diffractionand the specimen co-ordinate system used for stress analysis with L3 indicating the direc-tion of measured strain.

size was 0.5 mm (width) by 10 mm (height) with the length of the gauge volume parallelto the weld. The diffraction peak of martensite is inherently broad and, to avoid a peakclipping effect due to the use of a small slit width, the position sensitive detector wasscanned at 15 positions to map a diffraction peak. The peak centre was determined byfitting a Gaussian function to the measured diffraction data. As the residual stress statecan reasonably be assumed to be bi-axial, the in-plane residual stress was calculated usingEquation (6) with a diffraction elastic constant, 1+ν

E = 5.81 · 10−6 MPa−1 [13]. Due tothe complex component geometry, measurements could only be made at ψ = 0◦ and 90◦,

107

respectively. The uncertainty of sample positioning was estimated to be ±0.2 mm for thetransverse direction and less than to ±0.1 mm for the normal direction.


The results of the robot program conducted off-line revealed a high degree of accu-racy and very few adjustments after the calibration were required. A validation studyof temperature predictions for the specific part using thermocouples and infrared camerameasurements had been performed on a prior occasion. A typical example of predictedtemperature-time histories, located at 4.2, 6.3 and 8.4 mm respectively from the weldcenterline, is given in Figure 8.

50 100 150 200

250

500

750

1000

1250

Time (s)

Tem

per

atur

e ( o C

)

4.2mm6.3mm8.4mm

Figure 8: Temperature histories in three points along a perpendicular line to the weldseam.

Residual stresses were evaluated along three lines, located at the front, in the middleand at the back of the weld, Figure 9. The predictions are summarized in Figures 10and 11 along each line. Figure 10 shows the calculated longitudinal stresses (i.e. in theweld direction) and Figure 11 the stresses perpendicular to the weld direction. Since ashell model was used, the stresses in the thickness direction are zero. The length axis inFigures 10 and 11, represents the distance from the weld center and the stresses shownwere recorded after 200 s from the start time of the welding. The longitudinal stresslevel and distribution are similar along all sampling lines and the maximum longitudinalstress is about 800 MPa. The stress components perpendicular to the weld direction atdifferent reference line locations show large differences. Compressive stress is generatedboth along the middle- and back reference line but this is not the case for the front line.

108

The reason for these differences is due to the difference in stiffness of the componentalong the seam.

Figure 9: Lines where evaluations were performed.

0 20 40 60 80 100 120

-200

0

200

400

600

800

Length (mm)

Stre

ss (M

Pa)

FrontMiddleBack

Figure 10: Predicted longitudinal stress component along the three lines in Figure 9.

Figure 12 show the diffraction peak with as Full Width at Half Maximal intensity (FWHM),measured in both normal direction, as a function of increasing distance from the weld cen-ter line. The distribution of the peak width indicates that microstructure in the weld and

109

0 20 40 60 80 100 120

-150

0

150

300

Length (mm)

Stre

ss (M

Pa)

FrontMiddleBack

Figure 11: Predicted stress component (perpendicular to the weld) along the three linesin Figure 9.

the heat affected zone (HAZ) is essential martensite and its volume decreases rapidly atabout 5 mm from the weld. This agrees well with the simulation result. The derivedresidual stresses are plotted in Figure 13.

0 5 10 15 20 25 30

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

Distance from weld centre (mm)

Resi

dua

l str

ess

(MPa

)

0 5 10 15 20 25 30

20

40

60

80

100

Mar

tens

ite

(Vol

%)

TransverseNormalMartensite

Figure 12: Comparison of the measured distribution of diffraction peak width (FWHM)and simulated volume of martensite.

The residual transverse stress distribution derived from the measured diffraction peak

110

0 5 10 15 20 25 30

93.2

93.4

93.6

93.8


2θ (

°)

0 5 10 15 20 25 30

20

40

60

80

100

Mar

tens

ite

(Vol

%)

NormalTransverseMartensite

Figure 13: Variation of diffraction peak centre line with increasing the distance fromweld centre for the normal and transverse direction, respectively.

0 5 10 15 20 25 30-200

-150

-100

-50

0

50

100

150

200


Resi

dua

l str

ess

(MPa

)

0 5 10 15 20 25 30

20

40

60

80

100

Mar

tens

ite

(Vol

%)

StressMartensite

Figure 14: Measured transverse stress distribution.

centre (Figure 13) is plotted in Figure 14. Figure 15 compares the predicted and measuredresidual stresses are compared in Figure 14. The discrepancy between the measured andpredicted results shown in Figure 14 has a number of possible explanations that relate tothe finite element model:

• A simplified model for the phase transformation has been used

• The model did not include a transformation induced due to plasticity

111

• The model did not include plasticity

0 5 10 15 20 25 30

-125

-75

-25

25

75

125


Stre

ss (M

Pa)

0 5 10 15 20 25 30

25

50

75

100

Mar

tens

ite

(Vol

%)

StressMartensite

Figure 15: Comparison between simulation and measurement residual stresses.

One other reason for the discrepancy can be due to the repositioning of the componentbetween the transverse and the normal measurement. This may lead to errors, particularlyin the heat affected zone, where the peak centre, and thus the lattice parameter, changerapidly. However this error can not explain the observed oscillation. Due to the large scat-ter in the transverse stress measurements in the heat affected zone, no specific conclusioncan be arrived at. There is a fairly good agreement outside this zone.

Reaction forces in a rectangular coordinate system located in a point along the weld be-tween the outer ring and steel plate, see Figure 2b, are shown in Figure 16. The forces arean example of a result that can be used to evaluate fixture solutions on a real part. An-other example of an application of the model given in Figure 17, where an optimizationof the weld velocity to generate the lowest component deformation while keeping fullpenetration, was conducted.

Several developments of the FEA models are possible. One simplification in the presentmodel is that the tack welding (performed before the main weld) was not considered.This tack welding will most probably affect the stress level. The development of a newsolid model, instead of a shell model, including these tack welds is planned. Includingtransformation plasticity in the material model will also lead to a change in the stress state[6].

The present model can, however, function as a powerful tool to qualitatively evaluatedifferent weld parameters and fixture designs off-line.

112

0 50 100 150 200

-8000

-6000

-4000

-2000

0

2000

4000

6000

8000

10000

Distance (mm)

Forc

e (N

)

XYZ

Figure 16: Reaction forces along curve c, Figure 9, between the outer ring and the steelplate.

0 50 100 1500

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Distance (mm)

Spee

d (m

m/s

)

Speed 1Speed 2

Figure 17: Weld velocity before (solid) and after (dashed) optimization.


A simulation tool to define robot trajectories and to predict thermal histories and resid-ual stress distributions on parts with complex geometries has been developed. The tool

113

was evaluated on a part with a complex shape where robot weld paths were defined off-line, automatically downloaded to an FEA-model, where transient temperatures, residualstresses and fixture reaction forces were predicted. Residual stresses on the componentwere measured using neutron diffraction and compared with predicted residual stressesand fairly good agreement was found. The method described seems to provide a power-ful tool for the construction and optimization of torch trajectories and process parametersoff-line.

Acknowledgements

The authors wish to acknowledge the assistance in the development of the material modelsby Andreas Lundbäck of LuleåUniversity of Technology. Mr. Alastair Henry of Univer-sity of Trollhättan/Uddevalla for linguistic revision. Mr Xavier Guterbaum of Universityof Trollhättan/Uddevalla for all our valuable discussions in the robot laboratory. Thework was funded by the Foundation for Knowledge and Competence Development andEC Structural Funds.

8 References

1. Y. Ueda and T. Yamakawa, Analysis of thermal elastic-plastic stress and stress dur-ing welding by finite element method, Trans. JWRI, Vol. 2 (90-100, (1971).

2. Y. Ueda and T. Yamakawa, Thermal stress analysis of metals with temperaturedependent mechanical properties, Proc. of Int. Conf. on Mechanical Behavior ofMaterials, Vol. III,10-20, (1971).

3. H.D. Hibbit and P.V Marcal., A numerical thermo-mechanical model for the weld-ing and subsequent loading of a fabricated structure, Comp. & Struct., Vol. 31145-1174 (1973).

4. Radej, D. Heat Effects of Welding, p33, Springer Verlag, Berlin (1992)

5. T.W Eagar, N.S. Tsai, American Welding Society Journal 62(12) 346-s to 355-s.(1983)

6. F.G. Rammerstorfer, D.F Fischer, W. Mitter, K.J Bathe and M. D. Snyder, Onthermo-elasto-plastic analysis of heat-treatment processes including creep and phasechanges, Comput. Struct., Vol 13, 771-779. (1981).

7. G. Bolmsjö, M. Olsson, K. Brink, Off-line programming of GMAW robotic systems- a case study. Int. J. for the Joining of Materials, Vol. 9 (3), 86-93, (1997).

8. S Walter Simulation and Calibration for Off-line Programming of Industrial Robots,paper 54, Proc. of Computer Technology in Welding, Paris, June 1997.

9. R.O. Buchal, D.B. Cheras, F. Sassani, J.P. Duncan, Int. J. of Robotics Research 8(3): 31-43 (1989).

114

10. M. Ericsson, P. Nylén, G. Bolmsjö, Three-Dimensional Simulation of Robot Pathand Heat Transfer of a TIG-welded Part with Complex Geometry, 11th InternationalConference on Computer Technology in Welding, Colombus, Ohio, Dec. 2001.

11. R. Lin and K. Sköld, The Neutron Diffraction Facility for Residual Stress Measure-ments in Studsvik, in Proc. of the 4th European Conf. on Residual Stresses, Clunyen Bourgogne, France, pp. 145-152, 1996.

12. I. C. Noyan, J. B. Cohen, Residual Stress. Measurement by Diffraction and Inter-pretation. Springer-Verlag, 1987.

13. B. Eigenmann und E. Macherauch: Röntgenographische Untersuchung von Span-nungszuständen in Werkstoffen, Mat.-wiss.u. Werkstoffech. 27, 426-437 1996.

Paper IV

Computer Aided Robotics combined with a FiniteElement Analysis for Process Simulation of Welding

Fredrik Danielsson, Mikael Ericsson

Presented at 35th International Symposium on Robotics,Paris, France, March 23-26 2004

Computer Aided Robotics combined with a FiniteElement Analysis for Process Simulation of Welding

Fredrik Danielsson, Mikael Ericsson

Abstract

In this paper an off-line simulation system has been created with the aim of in-creasing the quality of robotic arc welding. The particular approach examined hereis to combine different simulation techniques to form an overall simulation of a ro-botic weld process. The application of this approach can result in more accurate robotpaths and increases in quality in the weld process. The proposed off-line simulationsystem is mainly based on two simulation techniques, namely, Computer Aided Ro-botics and Finite Element Analysis. A key feature in the integration of these twosimulation techniques is a proposed time synchronisation mechanism that ensures aglobal common time domain. The different components in the simulation system andthe total architecture are presented. The validation of the proposed simulation systemwas performed by comparing simulation results against experiments. The simulatedtemperature predictions agreed fairly well with the IR-camera measurements.

1 Introduction

A large number of components have complex shapes and are manufactured in multi-stepprocesses that often include joining operations. Joining by welding, however, induceschanges in the microstructure of base metals and can generate unwanted stresses anddeformation. To avoid deformations, expensive and complex fixtures often have to beused. Furthermore, the planning of optimal robotic welding sequences for complex com-ponents is difficult and requires highly experienced operators. Consequently, a simulationtool used for the evaluation of features such as structural behaviour, welding sequencesand fixture solutions during the design stage would be desirable. The use of such a toolwould make it possible to evaluate manufacturing processes in the early stages of productdevelopment and would result in the reduction of both the number of welding experimentsand the need for specialised welding operator experience. The tool should, ideally, be ca-pable of simulating the robotic welding torch path, detecting collisions between the torchand work piece, and optimising welding parameters as regards penetration and compo-nent deformation behaviour. Thus a combination of Finite Element Analysis (FEA) andComputer Aided Robotics (CAR) simulation techniques is necessary.

Numerical methods have been used since the beginning of the 1970’s to simulate weldingprocesses. Their focus has been to predict thermal histories, residual stresses and distor-tion. Finite Element Analysis (FEA) simulations have been the most common numerical

118

method and numerous papers have been presented [Toselo, Berglund, Ueda and Hibbit].

Computer aided robotics (CAR) technology is also a well-established field of research[Bolmsjö and Walter]. By using this technique, the programming of the robot is trans-ferred from the workshop to a computer system where it can be performed without in-terrupting on going production. The technique can be used to simulate the welding torchpath, to detect torch-work piece collisions and to control both torch orientation and elec-trode distance.

The aim of this research study is to investigate the integration of different simulation toolsto control a robotic arc weld process, paying regard to both robot and material qualities.The integration technique used is to allow all simulation models to run in parallel and toshare common data. However, this is only possible if all of the sub-simulations can be ex-ecuted in the same time domain. A key feature of this study is thus a time synchronisationmechanism that ensures a global common time domain for all sub-simulations.

One conclusion from the evaluation of the system is that the finite element analysis modelpredicts the temperature-time history fairly well. The evaluation also shows that both thesimulated robot path and process time agree well with the real robot.

1. to off-line program parts with complex shapes,

2. to numerically predict the shape of the molten pool by the use of ComputationalFluid Dynamics (CFD) techniques,

3. to numerically solve the energy equations in the solid material with sufficient ac-curacy that metallurgical predictions can be made, as well as to link the off-lineprogramming model with this numerical model, and

4. to empirically establish relationships between temperature-time history and metal-lurgical and mechanical properties

This paper is concerned with parts 1 and 3 above; namely methods of programming robotsoff-line and of predicting temperature-time histories on parts with complex shapes.

2 Components in the Simulation System

This research study focuses on the integration of different simulation tools to control arobotic arc weld process. The proposed simulation system is based on four differentcomponents, namely:

1. Computer Aided Robotics for path and robot simulation,

2. Finite Element Analysis for simulation of heat transfer and component tempera-tures,

3. Off-line arc-current control system for adaptive process control and

119

4. A simulation engine for integration and time synchronisation.

Even if this specific system is based on four different sub-systems, there is no limit to thenumber of sub-systems that the simulation engine can handle. Each tool is described ingreater detail in the following sections.

3 Computer Aided Robotics

The rapid advance of robotics during the last fifteen years has propelled forward the de-velopment of robot simulation packages. Computer Aided Robotics systems embody 3Dgraphical systems with kinematics emulation capacities and are mainly intended for thesimulation of industrial robots and machine systems [Oscarsson]. Computer Aided Ro-botics applications can be equipped with collision detection algorithms, checks for singu-lar position, off-line programming capabilities, checks for reachability etc. [McKerrow].Several commercial software packages for Computer Aided Robotics are currently avail-able, such as GRASP, RobotStudio, IGRIP and RobCad. In this research study IGRIP(Interactive Graphics Robot Instruction Program) system [www.delmia.com] was used,Figure 1.

Figure 1: OLP model in IGRIP.

4 Finite Element Analysis

Finite Element Analysis simulations of thermal history, residual stresses and distortionhave been performed since the early 70s and numerous papers have been presented [Berglund,

120

Hibbit and Ueda]. The Finite Element Analysis, also named the Finite Element Method(FEM), is an approximate mathematical method for solving problems which can be de-termined by differential equations such as heat transfer problems. The main idea of FiniteElement Analysis is to deconstruct a complicated problem consisting of an initial stateand boundary conditions into smaller pieces (elements) of a finite size. Each element isconsidered to be part of the main problem, thus connected to other closely allied nodes.Forces (such as thermal load in a heat transfer process) which act on the boundary of thecomponent can be simplified as simply acting at a number of discrete nodes. Overall, thisresults in an extensive system of linear equations that can be easily solved by a computerfor any given time frame. The construction of a FEA weld process model contains a num-ber of necessary simplifications. The areas to be considered are the following: geometry,material, loads, boundary conditions and numerical strategy. For a more in-depth discus-sion see [Berglund]. Large, complex simulation models of three-dimensional componentsare however still quite rare, mainly due to a lack of computational power. The reason isthat, to be able to compute temperature and residual stress fields in the affected zone, avery fine discretisation of the space variable is required in order to properly accommodatethe sharp gradients.

5 Off-Line Arc-Current Control System

At the present time very few commercial on-line process control systems for the arc weldprocess exist due to the difficulties in developing robust sensor systems, for instance tomeasure component temperature close to or within the weld pool. However, the off-linesimulation system proposed here makes it possible to estimate the temperature duringsimulated welding. Therefore, a control system was added to the simulation system. Thecontroller iteratively adjusts the arc-current until the part temperature in a selected movingco-ordinate is sufficiently close to a target temperature. Typically is a point close to theweld pool chosen, see Figure 2 and the target temperature set to value that ensures a fullpenetration weld. Figure 2 shows a typical temperature cross section of a welded platewith the nodes where the temperature is evaluated. The main idea is to use the output i.e.the optimised time dependent arc-current from the simulation to control the real processon-line.

6 Synchronised Distributed Simulation Protocol

The integration of different simulation techniques requires a time synchronisation mech-anism. However, a distributed solution has been chosen in order to make computationallydemanding simulations possible. To manage synchronisation and distributed simulations,a protocol, Synchronised Distribute Simulations Protocol (SDSP) was developed. SDSPwas originally developed to handle the synchronisation of an emulator for a control sys-tem [Danielsson 1999 and Danielsson 2002]. SDSP consists of two parts, a server moduleand a client module. The SDSP server handles all the common information for the sys-tem, such as common data, simulation parameters, model status, etc. The server is also

121

Figure 2: Cross section of a welded plate.

responsible for the simulation clock in the system (i.e. synchronisation between all sys-tems). The system as a whole can be described as a discrete simulation with a time step,even if a client- module/simulation part can and will be continuous. SDSP is currentlybased on the TCP/IP communication protocol, which is accessible either via LAN or theInternet. The client part of SDSP is an application program interface that both the serverand all the simulation clients use. The client part can be compiled on several platformse.g. IRIX, Solaris, Linux, and NT.

7 System Description - Implementation

This section describes the developed implementation of four sub-systems above and thearchitecture of the integrated system. All of the subsystems are time-dependent simula-tions. Therefore, a key feature of the overall simulation system is the time synchronisationmechanism where the entire system with the robot simulation, the control system and theprocess model are forced to run in the same time domain. This time synchronisationmechanism is managed and driven by the SDSP server [Danielsson 1999]. The SDSPserver is also responsible for all global simulation data - examples of global data beingtime, weld gun position and arc-current. The overall architecture of the integrated systemis given in Figure 5.

The Computer Aided Robotics software IGRIP was used to simulate the robot controlsystem and the robot motion. Prior to the simulation a geometrical model of the work-piece, a geometric and a kinematic model of the work-cell including the most importantparts such as fixtures, robots and positioners were created. The high level language GSL,which is the graphical simulation language in IGRIP, was used for programming all of thedevices in the cell. For each time step during the simulation, IGRIP calculates a new weldgun position (x, y and z). This new position is then sent to the SDSP server.

The commercial FEA program MARC, from MSC Software, was used to simulate thethermal model of the arc welding process. In order, to simulate a moving heat source(i.e., the arc) a moving Gaussian surface distribution was used, see [Radej]:

122

q = q0 e−αqr

2(1)

q0 =ηEIαqπ

(2)

q =ηEIαqπ

e−αqr2

(3)

where q denotes the heat transferred to the work piece, E the voltage, I the current, η theefficiency factor, q the concentration factor, and r the radial distance from the centre ofthe heat source. The FEA model consisted of 36 480 solid elements and 43 659 nodes,see Figure 3 and was based on the assumptions:

• room temperature, 20 degrees Celsius, at each boundary

• initial plate temperature 20 degrees Celsius

• isotropic material

• A thermal conductivity increase by a factor 10 when the temperature reached theliquidus temperature to account for fluid flow i.e. convective heat transfer in themelted zone.

• a constant welding voltage was used in each experiment, see table 1

• a constant efficient factor, η = 0.85, was used

For each time step, MARC reads the new weld gun position and a new arc-current fromthe SDSP server. The model is then updated with new parameters for the heat source,i.e., co-ordinates x, y, z, provided by IGRIP and the updated arc-current provided by thearc-current controller. The following steps describe the overall simulation principle.

1. The component to be manufactured is created in a 3D-CAD software.

2. The component model, in step one, is imported to the CAR software and the FEAsoftware.

3. In the CAR software a model of the work cell is created. A weld path (includingrobot motion, weld speed, etc.) is developed. In the FEA program the part to bemanufactured is meshed and boundary conditions applied.

4. The simulation starts. Initial conditions are set in both systems.

123

Figure 3: The non-uniform mesh. Note the higher densities along the weld path.

5. In the beginning of a time step all systems are allowed to write data to the server.In our case, the CAR system writes the present Tool Centre Point (TCP) value forthe weld gun together with the current process parameters to the server.

6. Calculations for current time step are performed for all sub-systems. The FEA sys-tem predicts the temperature distribution. The arc-current control system calculatesa new arc-current value based on the comparison between the desired and predictedtemperature.

7. At the end of each time step, all of the systems read new values from the server.

8. Steps 5, 6 and 7 are performed repeatedly until completion of the simulation.

8 Experiment

The validation of the proposed simulation system was performed by comparing simulationresults against measured data. TIG (GTAW) welding was performed using an in-house ro-botic welding cell. The torch used is from Binzel AB and is linked to a six-axis robot fromABB, IRB1400. The power source is a TIG Commander 400 AC/DC from MigatronicAB. Throughout all experiments, thoriated tungsten electrodes were used. The weldingexperiments (bead on plates) were performed on 2 mm thick, 150x100 mm stainless steelplates. No filler material was used. The arc-current was kept constant in each experiment.Five different arc-current levels were used, Table 1.

Temperature measurements were carried out by high-resolution infrared (IR) emissionmeasurements [www.jenoptic.de]. Usually, there is a degree of uncertainty in the IR-camera measurement due to the change in surface emissivity during material solidificationand surface oxidation. Surface treatment has therefore to be applied. The surface should

124

Table 1: Experimental welding parameters.

Arc-current Measured voltage

Low 85.0 A 9.1 VNormal 1 97.5 A 9.8 VNormal 2 110.0 A 10.1 VNormal 3 122.5 A 10.2 V

High 135.0 A 10.5 V

ideally have a low emissivity variation over a wide temperature range. From earlier ex-periments [Ericsson] it was found that by using an acetylene/oxygen flame a thin, highlytemperature resistant, soot layer could be produced. A more detailed description of theprinciple of the measurements is given in [Henrikson]. The predicted and IR-measuredtemperature histories in a point located 4mm from the centre of the weld are given inFigure 4. No calibrations of the simulation parameters against these measurements wereperformed.

0 6 12 18200

400

600

800

1000

Time (s)

Tem

pera

ture

( o C

)

IRMarc

Figure 4: IR-measurements against simulation result.


In this paper a tool for off-line programming, simulation, and control of robotic arc weld-ing has been proposed. The approach is described with an emphasis on the integrationof different simulation tools. The implementation of the different parts of the system

125

has been completed, and tentative tests been performed. Key features of the proposedsimulation tool include:

• Finite element analysis simulation to evaluate thermal histories

• Computer aided robotics simulation to manage off-line programming and robotmotion simulation

• Off-line arc-current control system to minimise heat effects.

• a distributed server based solution to handle large applications

• a time synchronised non-real time simulation model running in virtual real time

This simulation system can be used as an advanced off-line programming and verificationtool for a robotic arc welding. One of its primary areas of usage could be to predict and op-timise process temperatures on complex shaped parts and, thereby, minimise heat effectssuch as changes in microstructure and induced distortion. The conclusion from the tem-perature simulation was that the model predicts the thermal cycle fairly well but furtherfine-tuning is needed. These predictions are naturally dependent on assumed boundaryconditions such as heat transfer coefficients. Further experiments are needed to establishthese. The simulated robot path and process time agreed well with experiments.

10 Acknowledgements

The authors greatly acknowledge Dr Per Nylén of University of Trollhättan/Uddevallafor fruitful comments on the manuscript. The work was funded by the Foundation forKnowledge and Competence Development and EC Structural Founds.

FEA

Time

Current

Heat SDSP

ServerCAR

Time

-

Hea

t

Cur

rent

Arc position(x, y, z)

Moving heat

source

Heat model Common

data

Time synchronization

Arc - currentcontroller

Arc position

(x, y, z)

SDSP

Clie

nt I

nter

face

nter

face

SDSP

Clie

nt I

Robot model

Arc path planning

Figure 5: The overall architecture of the off-line simulation system.

126

References

Berglund D. Simulation of welding and stress relief heat treating in development ofaerospace components. Licentiate in Engineering Thesis, Luleå University of Tech-nology, Department of Materials and Manufacturing Engineering, Division of Man-ufacturing Enginering, 2001.

Bolmsjö, G., Olsson, M. and Brink, K.. Off-line programming of GMAW robotic systems- a case study. Int. J. for the Joining of Materials, Vol. 9 (3), 86-93, 1990.

Danielsson F. A distributed system architecture for optimizing control logic in complexmanufacturing systems Proceedings of the ISCA 12th international Conference,Atlanta, USA, pages 163-167, ISBN: 1-880843-30-7(Dearborn, Mich.: Society ofManufacturing Engineers, 2002)

Danielsson F. A distributed system architecture for optimizing control logic in complexmanufacturing systems Proceedings of the ISCA 12th international Conference,Atlanta, USA, pages 163-167, ISBN: 1-880843-30-7(Dearborn, Mich.: Society ofManufacturing Engineers, 2002)

Ericsson M. Simulation of robotic TIG-welding. Licentiate in Engineering Thesis, LundUniversity, Department of Mechanical Engineering, Division Robotics, 2003.

Henrikson P. and Ericsson M. Non-contact Temperature Measurements using an InfraredCamera in Aerospace Welding Applications Trends in Welding Research: Pro-ceedings of the 6th International Conference, pp 930-936, Pine Mountain, Georgia,USA., 2002

Hibbit, H. D. and Marcal, P. A numerical thermo-mechanical model for the welding andsubsequent loading of a fabricated structure Comp. & Struct, Vol 3, pp 1145-1174,1973

McKerrow, P., J. Introduction to Robotics , Sydney, Addison-Wesley, 1991

Oscarsson J. Enhanced virtual manufacturing: advanced digital mock-up technology withsimulation of variances , PhD thesis, De Montfort University, Faculty of Computingand Engineering Sciences, UK, 2000

Radej, D. Heat Effects of Welding , p33, Springer Verlag, Berlin, 1992

Toselo, F., Tissot, X. and Barras M. Modelling of the weld behaviour for the controlof GTA process by computer aided welding , Matehematical Modelling of WeldPhenomena 4, pp 80-103, 1997

Ueda, Y. and Yamakawa, T. Analysis of thermal elastic-plastic stress and strain duringwelding by finite element method , Trans JWRI, Vol 2, pp 90-100, 1971

Walter, S. Simulation and Calibration for Off-line Programming of Industrial Robots ,paper 54, Proc. of Computer Technology in Welding, Paris, June, 1997

Paper V

Optimization of robot welding speed based onprocess modeling

Mikael Ericsson, Per Nylén

Submitted for publication in Welding Journal

Optimization of robot welding speed based onprocess modeling

Mikael Ericsson, Per Nylén

Abstract

Simulation tools to search for optimal process parameters are of great interest toreduce the number of experiments and thereby reduce cost and production time. Inthis paper robot simulation has been used in combination with finite element sim-ulations to optimize robot speed in order to minimize distortion while keeping fullpenetration. In an earlier work performed by the authors, a finite element model wasdeveloped to predict heat transfer and residual stresses of parts with complex shapes.An interface between a robot simulation model and a finite element analysis modelwas also constructed. In this paper an iterative method for robot speed optimizationhas been developed using MATLAB. The algorithm is designed to maintain full pene-tration while maximizing productivity by utilizing the fastest weld speed. The methodmakes it possible to optimize the heat input to the component and thereby minimizecomponent deformation for parts with complex shapes.

The system was evaluated on stainless steel plates with varying thickness. Robotweld paths were defined off-line and automatically downloaded to the finite elementsoftware where the optimization was performed. Simulations and experimental vali-dations are presented.

1 Introduction

CAD based path planning of robot welded parts is an elegant technique. Using thismethod the programming is moved away from the robot to a graphical computer sys-tem often referred to as an "off-line programming" system (OLP). This method makesit possible to maintain constant velocity, distance from, and orientation with respect to,a part with complex shape. This would be virtually impossible using manual program-ming. The OLP technology is well established in industry and has been an active researcharea [1-4] for some ten years. There is however a need for a computer aided processplanning tool by which process parameters could be defined and optimized off-line. Thisfunctionality does not exist in commercially available OLP tools today. Such a systemshould be capable of optimizing process parameters such as welding speed and powerdue to variations in part geometry (thickness variation), material and part temperatures(heat sources). Of specific interest is to determine an optimal weld speed i.e. the speedthat generates the lowest component deformation while keeping full penetration. Such aprocess-planning tool could be developed by a combination of robot simulation and finiteelement simulations. Finite Element Analysis (FEA) for welding process simulations on

130

fairly simple shaped parts is a well established technique [7-11]. It is usually used toinvestigate structural behavior, usually to predict residual stresses. Manufacturing simu-lations to plan welding sequences and to optimize process parameters or fixture designsare still rare specifically simulations of complex three-dimensional parts.

In earlier works performed by the authors, integration between a robot simulation modeland a Finite Element Analysis (FEA) model was proposed [12-14]. This model was de-veloped to predict heat transfer residual stresses and fixture forces considering parts withcomplex shapes. In the present study work a MATLAB implementation of an iterativemethod to optimize weld speed and thereby minimize component deformation is de-scribed. Simulations and optimizations on plates with varying thickness are presented.A validation of the temperature predictions is performed by comparing the predictionswith thermocouple and IR measured temperatures. A brief description of the OLP-FEAintegration as well as the process model are also summarized. A more detailed descriptionof these models can be found in [12-14].

2 Principle of Off-Line Programming (OLP) and Integration withthe FEA Model

The overall architecture of the simulation system is given in Figure 1. The program-ming of the robot motion is based on a simulation of the process by the IGRIP systemof Deneb Inc. The model consists of two main parts: a) a geometric, kinematic and dy-namic model of the robot and b) a model of the workpiece to be welded. The workpiecemodel is usually first constructed in a CAD/CAM system and afterwards exported to theOLP system. The geometrical as well as the kinematic model of the work cell are usuallymade directly in the OLP system. In this system a weld trajectory is also generated bydefining torch locations and orientations. This trajectory is then simulated and checks forcollisions between work-piece and the weld gun are made. Checks for and elimination ofrobot singularities are also made. A calibration of the model with the real cell is thereafterdone; this can include several sub steps such as tool point- work- piece- and signature-calibration [1]. A translation of the program to a specific robot manufacturer language ismade and the robot co-ordinates, welding speeds and process parameters are finally ex-ported from the OLP model to the FEA model where a heat and residual stress predictionis made. The principle of this FEA model is given in the next section.

3 The Heat Transfer Model

A finite element analysis model was used to predict the temperature evolution outside themolten zone. The FEA program Marc from MSC Software was used. User subroutineswere developed in earlier work to simulate a moving heat source [12-14]. A Gaussiansurface distribution was used. This distribution was preferred to a volumetric one [8]since it reduces the number of parameters (unknown variables) to be fit. The surface heatflux distribution was expressed as [11]:

131

Robot simulation

Geometrye.g. IGES

CAD/ CAM

FEATranslator

Welding path



IRB Controller

Complete robot code



Figure 1: The overall architecture of the simulation system..

q = q0 e−αqr

2(1)

q =ηEIαqπ

e−αqr2

(2)

where q denotes the heat transferred to the workpiece, E the voltage, I the current, η theefficiency factor, αq the concentration factor, and r the radial distance from the centre ofthe heat source. This distribution was truncated in the radial direction, at a cut off limitof 5% of the maximal heat input, as proposed by D. Radaj [11]. The parameter αq in theheat flux distribution was set to achieve a fusion zone fitting experimental data obtainedby measuring the top side and root side widths of cross-sections of welds. Experimen-tal trials were made on plane plates to find an appropriate value of αq. A value of 0.1was selected which gave good agreement between predicted geometry of the fusion zoneand corresponding measured zone. This parameter fit was considered necessary since asemi-empirical approach such as proposed by [15] was not possible due the short elec-trode distance (1.5 mm), which made photographing of the welding arc not possible. Theefficiency factor η was estimated experimentally using the method proposed by [16]. Theelectrode was kept still at a distance 1.5 mm from the plate. A very high efficiency wasdetermined η = 0.90. Both αq and η were kept constant through all simulations. Thisassumption was considered justifiable since the electrode distance and current were keptconstant. Convection boundary conditions were applied to the free surface dissipating

132

energy as well as at the contact surfaces between fixture and plates. Figure 2 shows theapplied boundary conditions. A flow of argon gas, see table 1, was used to protect the rootside of the weld. The heat transfer coefficients were set to 2 · 10−5 W/m2 at the topsideof the plate (number 2 in figure 2b), and to 2 · 10−4 W/m2 at the root side of the weld(number 3 in figure 2b), since forced cooling by argon was applied at the root side of theplates. The contact surfaces between the plates and the fixtures were assumed to have aheat transfer coefficient of 10−3 W/m2 (number 1 in figure 2b). The location of the arcis indicated with number 4 in figure 2b.

Fixture Weld gun

Arc

Shielding gas

Air Air

Air Air

Fixture

a)

b)

Plate Weld pool

22

1 3 1

12 4 2 1

Weld poolPlate

Figure 2: Principle outline of applied boundary conditions in the FEA model.

The material properties used are given in table 1 below.

Table 1: SS 316L physical properties [17, 18, 19]

Nomenclature Symbol Value Unit

Density ρ 7.3 · 10−6 kg/mm3

Latent heat of fusion DH 2.47 · 10−5 J/kgSolidus temperature Tsol 1673 K

Liquidus temperature Tliq 1773 KThermal conductivity k Seefigure3

Heat capacity Cp Seefigure4Initial temperature T0 293 K

Temperature dependent properties such as thermal conductivity and specific heat wereused, see Figures 3 and 4. Phase change was included in the analysis. Weld pool convec-tion has been shown to strongly affect the heat transfer in the weld pool. This convection

133

has however to be artificially treated in a solid mechanic model by multiplying the thermalconductivity by a certain factor when the temperature exceeds the liquidus temperature.This method has been commonly used [20, 21, 22]. An intensive circulation was notedand a factor of 10 was selected. The same factor has also been used in earlier work[12-14], Figure 33

500 1000 1500 2000 2500 30000

0.01

0.02

0.03

Temperature (K)

l (W

/(m

m K

))

500 1000 1500 2000 2500 30000

0.1

0.2

0.3

Temperature (K)

l (W

/(m

m K

))

Figure 3: Conductivity for a SS 316L Top: Original values [18], Bottom: Increased witha factor 10 above the liquids temperature.

134

500 1000 1500 2000 2500 3000400

450

500

550

600

650

700

750

Cp (

J/(k

g K

))

Temperature (K)

Figure 4: Specific heat for SS 316L [18].

The computational domain was discretized by a non-uniform mesh with higher densitiesin regions close to the weld path as well as where steep thickness variations were present.Eight-node brick elements were used, see Figure 5 and 6.

To verify the proposed optimization method two different geometries were defined: a)a two dimensional plate (referred to as part A) with continuously varying thickness ac-cording to Figure 5, and b) a three dimensional plate (referred to as part B) with stepwisevarying thickness, Figure 6 and 7. Grid sensitivity trials were made for part B. The finalmesh for this part consisted of 144000 elements. A constant time step of 0.05 s was used.

Figure 5: Cross section and computational mesh of two dimensional part. Dimensions inmm.

135

Figure 6: Cross-section and computational mesh of the three dimensional part. Not thehigh mesh density along the weld path and close to the steps

6.0

3.0

60.0

1.02.0

150.0

a)

b)

Figure 7: Profile of the three dimensional part. A) Along the weld path. B) Perpendicularto the weld path. All dimensions in mm.

3.1 Robot Speed Optimisation

Once the robot path and the desired root side temperature are chosen, the robot speedmight be optimized. The liquidus temperature was a natural choice for input for theoptimization since the main purpose was to control penetration. The following algorithmwas used, starting from a given robot speed s0 along the trajectory: 1) Compute themaximum temperature Tmaxi

along the trajectory by simulating the weld using the speedsi 2) Update the speed along the trajectory using the iteration

si+1 = si(1 + λTmaxi

− TmeltTmelt

) (3)

Here λ is a relaxation parameter, Tmelt the liquidus temperature and Tmax the maximumtemperature at each node. The iteration corresponds to increasing the robot speed whenthe temperature becomes too high. As the computational cost of one iteration is very lowcompared to the temperature calculation, each iteration is cheap. It should however benoticed that the proposed method is not an optimization method in the usual sense seams itdoes not always converge to a local or global optimum. Iterations are therefore performed

136

until the error (ε) no longer decreases. The principle of the overall optimization is givenin Figure 8. An initial robot speed is defined in IGRIP and downloaded to Marc where thetemperature calculation is performed. The root side temperatures are compared with theliquidus temperature and a new robot speed vector is calculated. The calculations continueiteratively until an optimal velocity vector is found, i.e. the velocity vector that maximizesthe speed while keeping full penetration. This velocity vector is finally exported back toIGRIP for final process simulation.

Figure 8: Principle of optimization. Loop until ε > εmin.

4 Experiments

Gas Tungsten Arc Welding (GTAW) was performed on plane plates in order to validatethe temperature predictions and to be able to determine the concentration factor (αq inEq. 1) using an in-house robotized welding cell. The torch used was from Binzel ABand was mounted on to a six-axis robot from ABB, IRB1400. The power source is aTIG Commander 400 AC/DC from Migatronic AB. Throughout all experiments, thoriatedtungsten electrodes were used. The process parameters are shown in Table 2.

Both thermocouples and high-resolution infrared (IR) emission measurements were usedfor the temperature measurements. Six thermocouples were positioned perpendicularlyto the welding direction. The first gauge was positioned as close as possible to the meltedzone at a distance of 4 mm from the centre of the weld. The rest of the thermocouples werepositioned 0.5 mm radially from the previous gauge along the radial direction. The sam-pling frequency was 270 Hz for each thermocouples. The IR-camera was a VARIOSCANHigh Resolution, from JENOPTIK, Laser, Optik, Systeme GmbH, that works in the IRradiation spectrum of 8 - 12 μ m. The camera was used both in a line scan mode with ascanning frequency of 270 Hz, as well as in a full-frame mode with a frequency of 1 Hz.The analysis of the IR measurements was made using the IRBIS Plus software providedby JENOPTIK. A comparison between the IR results with the thermocouple was made.The plates were sooted before welding in order to reduce the emissivity-dependency inthe IR-measurements. A more detailed description of the sooting technique and the IR

137

Table 2: Process parameters used.

Parameter Value

Current 85AVoltage 8.6V

Weld Velocity Optimised mm/sRoot gas flow rate 20l/min

Shielding gas 17l/minArc length 1.5mm

Filler metal none

measurement principle can be found in [23].


Thermocouple and IR measured temperature histories in a point located 7mm from thecentre of the weld are given in Figure 9.

5 15 25 35 45 55 650

200

400

600

800

Time (s)

Tem

pera

ture

(

ο C)

IR cameraThermocouplle

Figure 9: Thermocouple and IR measured temperatures.

There is a good agreement between the two techniques. The predicted and correspond-

138

ing IR measured temperatures at location B, see Figure 17, are given in Figure 10 and11 respectively. Due to soot evaporation close to the weld seam relievable temperaturemeasurements could not be made at 5.5 mm, see Figure 11. The conclusion from thiscomparison was that the model was capable of predicting the thermal cycle well.

10 20 30 40 50

250

500

750

1000

1250

Time (s)

Tem

pera

ture

(K)

5.3 mm7.5 mm8.9 mm10.5 mm

Figure 10: Predicted temperature-time histories.

5 10 15 20 25 30 35

300

500

700

900 Emissivity problem

Time (s)

Tem

pera

ture

(K

)

5.5 mm7.9 mm9.1 mm10.9 mm

Figure 11: Measured temperature-time histories.

The predicted temperatures and weld velocities for the first 10 iterations for part A are

139

given in Figures 12 and 13, respectively. Figure 14 shows the temperature close to thetarget temperature. The temperatures correspond to values predicted along the root sidesymmetry curve, i.e. the weld centerline. The target temperature for the simulation wasset to 1700 K, which corresponds to a full penetration weld. The optimization algorithmconverges quickly. After 5 iterations, the temperature discrepancy has already reached±1100 K, and after 10 iterations this discrepancy goes down to ±30 K. The weld velocitywas initially set to 3 mm/s and it varies between 0.7 mm/s to above 3.5 mm/s after 10iterations. The maximum difference in velocity between iteration 10 and 11 is 0.0167mm/s. Further optimization is not of interest since this velocity is comparative with therobot accuracy.

0 20 40 60 80 100 120 140

800

1000

1200

1400

1600

1800

Distance (mm)

Tem

pera

ture

(K

)

Targetorgrun 1-9run 10

Figure 12: Predicted temperatures at the root side in the centre of the weld. The targettemperature is 1700 K.

The predicted temperatures and weld velocities for the first 10 iterations for part B aregiven in Figures 15 and 16 respectively. The temperatures correspond to values predicted0.1 mm radial to the weld centre line at the root side. This off-set was selected to guar-antee full penetration. Also in this case the target temperature was set to 1700 K. Thetemperatures calculated at the first iteration are unrealistically high, but approach soonthe target temperature. The convergence is more slow in this case than for case A. Atemperature peak of about 2010 K still exists in the 10th iteration. This peak is due tothe step change in thickness. The weld velocity also shows a more dramatic variation forthis plate, with values in the range 2-20 mm/s. The weld velocity was initially set to 3.0mm/s. The average difference in velocity between iteration 10 and 11 is 0.0815 mm/s

It was not considered of interest to optimize the velocity further i.e. to try to eliminatethe peaks in Figures 15 and 16 since the case was selected mainly to demonstrate thetechnique. The step change in thickness would in practice demand a change in size of themelt pool. Part B was welded using the parameters given in table 2. Figure 17 shows the

140

0 20 40 60 80 100 120 140

1650

1700

1750

Distance (mm)

Tem

pera

ture

(K

)

Targetorgrun 1-9run 10

Figure 13: Close up predicted of temperatures around 1700 K at the root side at thecenter of the weld.

IR measuring position (B) and locations where cross sections were evaluated. There wasa fairly good agreement between measured and predicted fusion, see table 3. Predictedvalues are in general somewhat larger except for location E. This might be due to thelocation of E close to the end point of the weld seam. Another possible explanation forthis discrepancy was distortion which increased the electrode distance.

Table 3: Measured and predicted fusion zones. Dimensions in mm.

Location Measure Measure Predicted PredictedWt Wr Wt Wr

A 3.29 0.79 3.71 1.67B 4.79 1.47 5.24 2.44C 6.15 1.0 7.1 2.24D 5.16 1.12 5.24 1.26E 4.09 3.25 3.11 0.51

Figure 18 shows a welded cross section of part B at location D, see Figure 17. Corre-sponding cross section from the simulation is given in Figure 19.

The overall conclusion from the optimization was that although simple, the proposed op-

141

0 20 40 60 80 100 120 140

1

1.5

2

2.5

3

3.5

Distance (mm)

velo

city

(m

m/s

)

Startrun 1-10run 11

Figure 14: Calculated weld velocities for the first 11 iterations. The velocity was set to 3mm/s in the first iteration.

timization algorithm performed very well. Several extensions of this method are possible.It would be of interest to include residual stresses or deformation for instance. Differentwelding sequences could also be automatically evaluated. It would also be valuable toextend the process model to including welding wire and pulsed current.

142

-60 -40 -20 0 20 40 601000

1500

2000

2500

3000

3500

4000

4500

Distance (mm)

Tem

pera

ture

(K

)

TargetStartrun 1-9run 10

Figure 15: Predicted temperatures at the root side in the centre of the weld. The targettemperature is 1700 K.

-60 -40 -20 0 20 40 60

5

10

15

20

Distance (mm)

velo

city

(m

m/s

)

Startrun 1-10run 11

Figure 16: Calculated weld velocities for the first 10 iterations. The velocity was set to 3mm/s in the first iteration.

143

Start welding

A B C D E

Stop welding

Figure 17: Cutting position on part B.

Figure 18: Welded Cross section.

Figure 19: Predicted cross section.

144

6 Conclutions

A simple but powerful optimization method to optimize weld speed has been successfullyimplemented. The proposed method allows optimizing the heat input to the componentand thereby minimize component deformation for parts with complex shapes. The processmodel was initially validated comparing temperature predictions with experimental mea-surements, and a good agreement was found. The optimization algorithm was evaluatedfor two different test cases, a two dimensional plate with continuously varying thicknessand a three dimensional plate with stepwise varying thickness. The temperature con-verged quickly for the two dimensional case and reached a variation of ±30 K around thetarget temperature within 10 iterations. For the second test case, a temperature peak ofabout 2010 K still existed in the 10th iteration due to the a discrete variation in thickness.The weld velocity also showed a more dramatic variation for this plate, with values in therange 2-20 mm/s.

The proposed method to integrate robot simulation, finite element analysis and numericaloptimization provides a promising and powerful tool for construct and optimize off-linerobot torch trajectories and process parameters. The method can also be an efficient tool inearly product development to evaluate different design concepts. The purposed optimiz-ing algorithm was shown computational efficient putting less demand on computationalpower thus making industrial usage possible.

7 Acknowledgment

The authors wish to acknowledge the assistance in calculation by Benoit Ripaud of Uni-versity of West and the assistance in the laboratory by Kjell Hurtig and Mats Högströmof West and to Al Henry of University West for linguistic revision. The work was fundedby the Foundation for Knowledge and Competence Development

8 References

1. Bolmsjö G., Olsson M., Brink K., 1997, Off- line programming of GMAW roboticsystems - a case study. Int. J. for the Joining of Materials, 9 (3): 86-93.

2. Buchal R.O., Cheras D.B., Sassani F., Duncan J.P., 1989, Simulated Off-Line Pro-gramming of Welding Robots. Int. J. of Robotics Research 8 (3): 31-43.

3. Bolmsjö G., 1999, Programming robot welding system using advanced simulationtools. Proc. of the International Conf. on the Joining of Materials JOM-9, 284-291

4. Walter S., 1994, Simulation and Calibration for Off-line Programming of IndustrialRobots. Proc. of Computer Technology in Welding: Paper 54.

5. Eagar T. W., Tsai N. S., 1983, Temperature Fields Produced by Traveling Distrib-uted Heat Sources. American Welding Society Journal 62(12) 346-s to 355-s.

145

6. Gu M., Goldak J., Hughes E., 1993, Steady state thermal analysis of welds withfiller metal addition. Canadian Metallurgical Quarterlv. 32 (1): 49-s to 55-s.

7. Kou S., Le Y., 1983, Three-dimensional heat flow and solidification during Auto-genous GTA Welding of Aluminum Plates. Metallurgical Transactions A. 14A.:2245-s to 2253-s.

8. Goldak J., McDill M., Oddy A., House R., Chi M., Bibby M., 1987, ComputationalHeat Transfer for Weld Mechanics. Proc. of Int. Conf. on Trends in WeldingResearch, Advances in Welding Science and Technology. Eds S. A. David: 15-20.Metals Park ASM Int.

9. Jonsson M., Karlsson L., Lindgren L.E., 1985, Deformation and Stresses in ButtWelding of Large Plates with Special References to the Material Properties, J. ofEng. Mat. And Tech. 107: 265-s to 270-s.

10. Lindgren L.E., Karlsson L., 1988, Deformation and Stresses in welding of ShellStructures. Int. J. for Numerical Methods in Eng. 25: 635-s to 655-s.

11. Radej D., 1992, Heat Effects of Welding: 33 Berlin: Springer Verlag.

12. Ericsson, M., Bolmsjö, G., Nylen, P. Three-Dimensional Simulation of Robot Pathand Heat Transfer of a TIG-Welded Part with Complex Geometry. SME TechnicalPaper AD02-292 (Dearborn, Mich.: Society of Manufacturing Engineers, 2002).2001, Proc 11th International Conference on Computer Technology in Welding

13. Ericsson M., Nylén P., Berglund D.,Lin-Peng R., 2005, Three Dimensional Sim-ulation of Robot path, Heat Transfer and Residual Stresses of a welded Part withComplex Geometry, Int. J. for the Joining of Materials 17(2)

14. Ericsson M. Simulation of robotic TIG-welding. Technical Licentiate Thesis. ISBN91-628-5702-9 2003-05-15

15. Connon, L. P., 1991, Welding Handbook - Welding Technology, American WeldingSociety, Vol. 1.

16. Bisen K.B., Arenas M., El-Kaddah N., Acoff V.L., 2003, Computation and Valida-tion of Weld Pool Dimensions and Temperature Profiles for Gamma TiAl, Metal-lurgical and Materials Transactions, Vol. 34 A, pp 2273-2279.

17. Choo R.T.C., Szekely J. and David S.A., 1992, On the calculation of the free sur-face temperature of gas-tungsten-arc weld pools from the first principles - Part II:Modelling the weld pool and comparison with experiments, Metallurgical transac-tion B, Volume 23B, pp. 371-384

18. Choong S.K., 1975, Thermophysical properties of stainless steel, ANL-75-55, Ar-gonne, Illinois

19. Toselo I., Tissot F.X., Barras M., Modelling of the Weld Behaviour for the Con-trol of the GTA Process by Computer Aided Welding, Commissariat a l’Energieatomique, Centre d’Etudes et de Recherche sur les Materiaux, Gif sur Yvette

146

20. Lindgren L-E., 2001, Finite Element modeling and simulation of welding part 2:Improved material modeling, Journal of Thermal Stresses, Volume 24, pp. 195-231

21. Goldak J., Akhlaghi M., 2005, "Computational Welding Mechanics", Springer Sci-ence + Business Media Inc

22. Michaleris P., DeBiccari A., 1997, Prediction of welding distortion, Welding Jour-nal, April, pp. 172-181

23. Henrikson P., Ericsson M., 2002, Non-Contact Temperature Measurements usingan Infrared Camera in Aerospace Welding Research, Proc. 6th International Con-ference on Trends in Welding Research, 930-935

Paper VI

Off-Line Programming or Robots for MetalDeposition

Mikael Ericsson, Per Nylén, Fredrik Danielsson, HenrikJohansson

Presented 7th International Conference on Trends in WeldingResearch, Pine Mountain, Georgia, USA, May 16-20, 2005

Off-Line Programming or Robots for MetalDeposition

Mikael Ericsson, Per Nylén, Fredrik Danielsson, Henrik Johansson

Abstract

Metal Deposition (MD) is a rapid prototyping technique to build parts by deposit-ing metal in a required fashion. When a complex-shaped part is to be built, a sim-ulation tool is needed to define robot trajectories. Three different simulation-basedmethods for robot trajectory generation are introduced and compared in this study.The methods are; reversed milling, adapted rapid prototyping and application pro-gramming in a computer aided robotics software. All methods were shown capableof creating robot paths for complex shapes, with the CAR software approach beingthe most flexible. Using this method, the geometry to be built is automatically slicedinto layers and a robot path is automatically generated. The method was tentativelyevaluated and appears to provide a powerful technique in the design and optimisationof robot paths for MD. Experiments showed that it is possible to manufacture fullydense parts using an Nd-Yag laser.

1 Introduction

Many of the complex-shaped components, especially those used in the aerospace and au-tomobile industries, are manufactured by casting or forging. Components made by theseprocesses require extensive machining and finishing operations before they can actuallybe used. If the production volume is low then the production cost for each componentis likely to be rather high. Manufacturing by moulding is also a rather inflexible methodsince it is necessary to change the complete mould in order to effect a change in com-ponent geometry. The development of alternative manufacturing methods is therefore ofsignificant interest.

One interesting alternative is metal Rapid Prototyping (RP) often referred to as MetalDeposition (MD). MD has the potential to be used to manufacture a new part, to addfeatures to an existing part, or to repair worn parts. The technique uses a welding heatsource to melt a powder or a wire material which solidifies on a surface, thus enablinga part to be built drop-by-drop. An example of such a method is the Laser-EngineeredNet Shaping (LENS) process developed at Sandia National Laboratories and StanfordUniversity [1]. In the LENS system a laser beam melts the top layer of the part in areaswhere material is to be added. Materials that can be used include 316 stainless steel,Inconel 625, tungsten, and titanium carbide cermets The LENS process produces fully-dense parts with functional mechanical properties. However, at present, the process can

150

only be applied for parts with simple and uniform cross-sections.

Full flexibility and usability of the MD process technique can however be obtained if a ro-bot is used. The robot can then either be programmed manually, or it can be programmedusing Computer Aided Robotics (CAR). Manual programming is usually carried out us-ing the teach-in method. In this method the robot arm is jogged through the programunder reduced power and at reduced speed, via a joystick. The manual generation of apath in this way is very time-consuming and is a tedious task. Robot programming us-ing computer simulation is an interesting alternative which, moreover, is necessary whencomplex shapes are to be built. Using this method the actual process can be simulatedprior to manufacturing. There is, however, to the authors’ knowledge, no commercialsoftware currently available for the simulation and generation of robot paths for MD. Dif-ferent methods could be used to develop this type of software. The aim of this paper is tointroduce and compare three different possible methods.

2 Geometries

Two different part geometries were considered within this project; a solid rectangular box,see Figure 1 and a solid box with a more complex shape, see Figure 2.

The principle for off-line programming that was evaluated was the same for both namely1) creation of a CAD model of the part, 2) slicing the CAD model into thin cross-sectionlayers and 3) definition of geometry paths for each layer.

Figure 1: A solid box.

3 Methods for Robot Trajectory Generation

Several methods can be developed for the generation of the robot trajectory. One methodis to use a CAM module in commercial CAD/CAM software. This method is, in thispaper, denoted as "reversed milling" since it is based on a milling path which is mirrored.

151

Figure 2: Complex geometry.

This transformation of the path is made by a developed postprocessor. An alternativemethod is to use software developed for the rapid prototyping of polymer parts. Sincesuch software is developed for polymer applications, adaptations have had to be madewhich are further described in the section headed Adapted Rapid Prototyping, below. Thefinal method that is evaluated is application programming in Computer Aided Roboticssoftware (CAR). This method is, in this article, denoted as adapted CAR. In all three casesthe part to manufactured is created within CAD software. All three methods also requiresimulation and verification of the robot path using CAR, which is why an initial briefdescription of this technique is necessary.

3.1 Computer Aided Robotics

Several commercial software packages for CAR are currently available (GRASP, IGRIPand RobCad etc). The procedure for the generation and simulation of robot paths usingthese systems can be summarized in terms of the following six steps [2, 3].

1. Modelling of the work cell


3. Programming of robot and other optional work cell equipment

4. Down loading of the program to the control system


6. Test running

The first step involves the construction of a geometrical and a kinematic model of thework cell. This geometric model can either be constructed using a CAD/CAM system, or

152

constructed in the CAR system. In the second step a geometric calibration of the modelwith the real cell is performed. This step can include several sub-steps, such as tool-point,work-piece and signature calibration. [2, 4]. Tool calibration is performed to determinethe tool centre point and to determine the weld torch orientation. A procedure using ameasuring arrow in a fixed position in the work cell and moving the robot to this positionin different directions, is usually used. The positions from the real robot cell are thenuploaded to the CAR system and a "best fit" is found using regression. Calibration ofthe workpiece is performed similarly by moving the robot to identified positions on theworkpiece. In step three the robot motion (and other possible motions) is programmedeither using a high level language or a specific robot language. If a high level languageis used, the program is translated to the specific robot language and downloaded (stepfour) to the robot controller system. The programming of the robot path is, in this study,substituted by the different methods described beneath. Usually, following the initial pro-gramming, additional equipment-specific programming is also needed (step five), whichis performed manually at the robot. Validation of the program by test runs is finally per-formed in step six. Both IGRIP (Interactive Graphics Robot Instruction Program, DenebRobotics) and ROBCAD (of UGS Corp.) systems were used in this study. Both a toolcalibration and a workpiece calibration were performed. In Figure 3, an IGRIP snapshotfrom the experimental setup is shown. A more detailed description of CAR can be foundin [2].

Figure 3: Snapshot from IGRIP.

3.2 Robot Path Generation through Reversed Milling

Unigraphics (UG), of UGS Corp. is a 3-D graphical tool for computer-aided design(CAD).The software is available in both Microsoft Windows and UNIX versions. A spe-cific Computer Aided Manufacture (CAM) module is available. It is used to define up

153

to five axis milling paths, a so called NC-code. In this study the NC code was insteadused to generate robot paths for MD. The procedure is based on three steps. The first ofthese is the generation of the robot path in the CAM module. The slicing of the part ismade at this initial stage and a robot path is constructed for each layer. The second step isthe importation of the path to the CAR software, where a robot simulation of the path ismade. Finally, in the last step the robot motion is downloaded to the physical robot whereMD is performed, see Figure 4.

CAR Welding robot

UniGraphics /CAD/CAM

Figure 4: Schematic principle of the "reversed milling" procedure.

The CAM module generates an NC code which is easy to understand for a person famil-iar with NC machines. The code mainly consists of coordinates and tool data. Since it isdeveloped for milling, it describes how to remove material. In the MD process, materialis added. Thus the need here is to reverse the path so that the last coordinates in the NCcode actually become the start coordinates for the MD process. This mirroring can eitherbe performed by introducing a fictive surface located at the upper surface on the part tobe manufactured, or by the development of a postprocessor that automatically makes thistransformation. This latter method was the approach taken in this study. In the CAR soft-ware this transformed path is verified. A simulation is also performed, verifying the robotorientation and thus ensuring that no collisions between the welding torch and the partoccur. A typical robot path for one layer of a rectangular shape is shown in Figure 5. Thedeposition is started from the lower right corner and a zig-zag movement up to the upperleft corner has been defined. While performing experiments it was shown that it is a majoradvantage if the starting point for each layer could be varied, and if an individual thick-ness could be defined for each layer. The advantages with individual layer thicknessesand different starting points for each layer agree with the findings of [5]. The start loca-tion for each layer and individual layer thicknesses could, however, not be automaticallyvaried. This makes the path generation more cumbersome. Another drawback with thismethod was that a counter support path was needed for each layer in order to maintainthe shape of the part. This counter path also had to be defined manually for each layer.These drawbacks, notwithstanding, the method was shown capable of creating complex3D shapes, thus enabling robot paths for the test geometries to be constructed.

3.3 Adapted Rapid Prototyping

The term rapid prototyping (RP) refers to a class of technologies that can automaticallyconstruct physical models from CAD data [6]. Although several RP techniques are avail-able, all seem to employ the same procedure. The steps to create a path can be summarizedas [6]:

1. Creation of a CAD model of the design

154

Figure 5: Robot path for one layer.

2. Conversion of the CAD model to STL format

3. Slicing the STL file into thin cross-sectional layers

One example of RP equipment manufacturer that employs this procedure is Stratasys, ofEden Prairie, MN, a company that has developed a number of different RP machines.The machines can create complex-shaped 3D polymer prototypes directly from a CAD-drawing. The CAD-drawing is pre-processed in Stratasys’ own Insight software whichimports a STL-file [5], which is the most commonly used file format that in RP. TheSTL-files can be generated by most of the commercially available CAD software. Insightautomatically slices the geometry in layers and then creates tool paths for a specific ma-chine. The tool paths are rotated for each layer i.e. starting each layer in a new corner asdesired. Since the path is generated for a specific machine and for additive polymers onlythin layers (0.03-0.1 mm) can be defined. When metal RP is to be applied, a larger partdimension is necessary and a scaling procedure is thus needed. Another adaptation thathas to be made is that the Insight software exports not coordinates, but pulses to the ma-chine. These pulses must subsequently be translated to coordinates. A specific translatorhad therefore to be developed. A postprocessor was created that automatically translatesthe instructions in the SML-file (Insight file format) to Rapid (ABB robot language). Theprogram also ensures the rotation of the starting point for each layer. In order to obtainthe same tool orientation for each layer, transformations also had to be made to some ofthe tool paths. The path generated by this program is thereafter imported to the CAR soft-ware where the robot motion is simulated and collision checks are made. An advantage inusing the adapted RP method compared to the "reversed milling" method is that counter

155

paths are automatically generated using this method. Drawbacks using this method arethat constant layer thicknesses are generated and that the scaling procedure could be apossible cause of errors. It is necessary that the operator knows what scale factor to use,and this can vary from part to part. A common drawback between this method and thereversed milling method is that two different software systems are needed; one softwaresystem for robot path generation, and another for robot simulation. The final method ofthe three suggested alternatives uses the same software for both purposes.

3.4 Adapted Computer Aided Robotics

The most interesting solution for the generation of robot trajectories is to make sole useof CAR software, see Figure 6.

CAR Welding robot

Figure 6: Robot path generation and simulation using computer aided robotics.

The major advantage with this method is that, compared to the reversed milling and adap-tive RP methods, it gives the operator a better overview of the whole process. The majordisadvantage, though, is the lack of automatic path generation. Application programmingis thus necessary, both for slicing the part into layers, as well as for the creation of themovement in each layer. The program has to calculate robot paths in a general way, in-dependent of part geometry. Two different ways of generating paths were evaluated. Inthe case of simple geometries a standalone program was suggested that interactively asksthe operator for geometry information such as component height and width and positioni.e. the location where the part is to be manufactured. For cases where more generalizedshapes are to be made, such as the example in Figure 2 above, the path is preferably gen-erated by implementing functions directly in the CAR software. Information about theweld path, as well as information about the welding parameters, is then exported to thephysical robot. An example of a side view showing the robot path coordinate systems isshown in Figure 7.

4 Experiments

Experiments were performed using both a robotised TIG welding cell and a Nd-YAG lasercell. The TIG cell consists of a six-axis robot ABB IRB 1400 with a torch from Binzel AB(thoriated tungsten electrode) which is connected to a TIG Commander 400 AC/DC, fromMigatronic Inc. Argon gas was used on the top side to avoid oxidation of the component.The laser welding experiments were carried out with 2.3 kW Nd:YAG laser cw2500 from, Rofin, German witch was linked to the IRB 4400 robot. The material used was stainless

156

Figure 7: Complex geometry with robot poses.

steel 316L. Figure 9 shows a snapshot from the TIG process. An arc of plasma is formedbetween the electrode and the base plate. The process was shown to be rather sensitivewith regard to the process parameters needed to obtain the desired shape of the depositedmetal. The weld torch had to be inclined 20 degrees from the vertical axis. The depositdirection is from left to right in Figure 8. The base plate had to be clamped very firmly toa fixture in order to avoid deformation and to enable efficient heat transportation.

Figure 8: Metal Deposition by TIG welding.

In the TIG experiments Automatic Voltage Control (AVC) was used to maintain an opti-mum electrode distance of 1.5 mm. The process parameters used for the welding opera-tion can be listed as follows:

• Weld speed : 3.0 mm/s

• Wire feed : 1.1 mm/s

• Weld current : 120 Amp

157

Experiments in which a rectangular solid box was manufactured, were performed line byline. The part to be manufactured was 40.6 mm wide, 100 mm long and 18mm high. Itwas initially made by 14 layers without any overlap of the weld seam i.e. the distancebetween the centres of each seam was chosen to be equal to the width of the seam. Itwas, however, shown that an overlap between seams is necessary. New deposit trials wereperformed by taking a weld seam overlap of 2.7mm. The number of seams and layerswas reduced to five and four respectively. Better results were achieved this time as thelayers were attached to each other and no discontinuities were observed. It was, however,observed that at some locations of the deposit the seams were not straight but deviated ina zigzag fashion. It was shown that the process was sensitive to slight deviations in thewire feed angle. Another observation was that edges occurred at the boundaries of theseams, Figure 9. This was shown to be due to partial oxidation of the deposited metal andhence the adjacent seams did not join perfectly.

Experiments were then performed using the Nd-Yag laser cell. Figure 10 shows the solidbox part. Each layer had an individual starting corner and a zig-zag movement from thiscorner had to be performed. A counter path was also used for each layer. It was alsoshown that it was necessary to adjust the wire feed rate and the laser power continuouslyduring the deposit. A sensor-based control system would be necessary to be able to fullyautomate the process. Figure 11 illustrates the necessity of these continuous adjustmentswhere a rectangular shape has been made. The left shape was performed without anyadjustments of the process, whilst the right shape was performed with continuous manualadjustments.

Figure 9: Edges along the seam after four layers.

Visual inspection of the cross-section revealed a fully dense material without any pores,Figure 12. Further work is planned to develop a process window in which a stable processis ensured. Microstructure evaluations and the mechanical testing of material propertieswith the aim of securing the functional performance of the parts are also planned.

158

Figure 10: The solid box geometry manufactured using a Nd-Yag laser cell.

Figure 11: Deposit without continuous process control (left) and with continuous processcontrol (right).

Figure 12: Cross section of the solid box geometry.

159


Three different methods for the off-line programming of robot paths were evaluated inthis paper. Two different geometries were considered; a rectangular solid box and a morecomplex geometry. Computer programs were developed to generate optimal robot pathsfor these geometries. All three methods were shown to be capable of generating the pathsfor the geometries. The major drawback encountered in the use of a CAM module forpath generation is that the software has been developed for milling, which means that itsfunctionality for Metal Deposition is limited. Individual paths for each deposit are needed,thus making the method more time-consuming for complex parts. The method using poly-mer Rapid Protyping software had a higher functionality since different starting pointsfor each layer were automatically defined. A drawback with this method, though, wasthat individual layer thicknesses could not be defined and that scaling procedures werefound to be necessary. The most flexible method seems to be the adaptation of ComputerAided Robotic (CAR) software. Using this method, a fully automated path generationseems possible, even for complex-shaped parts. Another advantage with this solution isthat only one system is needed for both path generation and for robot simulation. Furtherwork is however needed to evaluate the use of the method for more complex-shaped parts.The welding experiments showed that it was possible to manufacture fully dense parts,although the continuous control of process parameters is a requirement. Development ofa sensor-based control system is therefore under way. The present study seems to providean efficient way of manufacturing parts by means of the process of metal deposition. Thefinal quality of the product in terms of accuracy, finish and mechanical properties can beimproved further by examining different aspects of the process, such as heat transfer andmicrostructure, thus enabling more optimized weld parameters. The development of asimulation tool that can predict part temperature and distortion during deposition wouldalso be a worthwhile innovation. Optimal robot speeds, and trajectories related to shrink-age and distortion can thus be automatically defined. Such a simulation tool softwaresystem can by accomplished by linking the CAR software with Finite Element softwarein the same manner as was done in previous work regarding welding [7].

6 Acknowledgement

The authors wish to acknowledge the assistance in the experiments by Mr Kjell Hurtig andMr Mats Högström of University Trollhättan/Uddevalla and Mr. Peter Jonsson and Mr.Ingmar Fransson of Volvo Aero Corporation, Trollhättan for their help and encourage-ment. Mr. Alastair Henry of University of Trollhättan/Uddevalla for linguistic revision.The authors also wish to acknowledge the initial simulation work performed by Mr ManuSinghal. The work was funded by the EC Structural Founds and Innovatum Teknik.

160

References

1. M. L. Griffith, D. M. Keicher, C. L. Atwood, J. A. Romero, J. E. Smugeresky, L. D.Harwell, D. L. Greene, Free Form Fabrication of Metallic Components using LaserEngineered Net Shaping (LENSo), proceedings of the Solid Freeform FabricationSymposium, pp. 125. August 12-14, 1996, Austin, TX

2. G. Bolmsjö, M. Olsson, K. Brink, Off-line programming of GMAW robotic systems- a case study, Int J. for joining of Materials, 9, pp. 86-93, 1997

3. Y. F. Yong, J. A. Gleave, J. L. Green, and M. C. Bonne, Off-line programming ofrobots, Handbook of Industrial Robotics, New York, John Wiley & Sons, (1985)

4. R. Bernhardt, Robot Calibration, pp. 3-55, London: Chapman & Hall, (1993)

5. Zhang Y., Chen Y, Li P, Alan T. and Male A, Weld deposition-based rapid prototyp-ing: a preliminary study, Journal of Materials Processing Technology 135 (2003)347-357

6. Pennsylvania State University/Mechanical Engineering Department. http://www.me.psu.edu/lamancusa/ rapidpro/ index.htm [2005-04-18]

7. M Ericsson, Simulation of robotic TIG-welding, pp 5-9, licentiate thesis, LundsUniversity (2003)

lunds tekniska högskola · 2007. 5. 8. · tig-welded part with complex geometry 61 paper ii...

Documents