Vous êtes sur la page 1sur 1014

Trends in Welding Research

Proceedings of the
7 International Conference

May 16–20, 2005

Callaway Gardens Resort
Pine Mountain, Georgia, USA

Edited by
Stan A. David
Tarasankar DebRoy
John C. Lippold
Herschel B. Smartt
John M. Vitek

Sponsored by

Published by
ASM International
Materials Park, Ohio 44073-0002
Copyright © 2006
ASM International®
All rights reserved

No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording, or otherwise, without the written permission of the copyright

First printing, August 2006

Great care is taken in the compilation and production of this book, but it should be made clear that NO
THIS PUBLICATION. Although this information is believed to be accurate by ASM, ASM cannot guarantee that
favorable results will be obtained from the use of this publication alone. This publication is intended for use by
persons having technical skill, at their sole discretion and risk. Since the conditions of product or material use are
outside of ASM's control, ASM assumes no liability or obligation in connection with any use of this information. No
claim of any kind, whether as to products or information in this publication, and whether or not based on negligence,
shall be greater in amount than the purchase price of this product or publication in respect of which damages are
NEGLIGENCE OF SUCH PARTY. As with any material, evaluation of the material under end-use conditions prior
to specification is essential. Therefore, specific testing under actual conditions is recommended.

Nothing contained in this book shall be construed as a grant of any right of manufacture, sale, use, or reproduction,
in connection with any method, process, apparatus, product, composition, or system, whether or not covered by
letters patent, copyright, or trademark, and nothing contained in this book shall be construed as a defense against any
alleged infringement of letters patent, copyright, or trademark, or as a defense against liability for such infringement.

Comments, criticisms, and suggestions are invited, and should be forwarded to ASM International.

ISBN-13: 978-0-87170-842-7
ISBN-10: 0-87170-842-6
SAN: 204-7586

ASM International®
Materials Park, OH 44073-0002

Printed in the United States of America

Multiple copy reprints of individual articles are available from Technical Department, ASM International.

Conference Co-Chairmen

Dr. Stan A. David, FASM

Oak Ridge National Laboratory
Oak Ridge, Tennessee

Dr. Tarasankar DebRoy, FASM Dr. Herschel B. Smartt

The Pennsylvania State University Idaho National Engineering and
University Park, Pennsylvania Environmental Laboratory
Idaho Falls, Idaho

Prof. John C. Lippold, FASM Dr. John M. Vitek, FASM

The Ohio State University Oak Ridge National Laboratory
Columbus, Ohio Oak Ridge, Tennessee

International Organizing Committee

Prof. Harry K. D. H. Bhadeshia Prof. Thomas W. Eagar, FASM Mr. Christopher Smallbone
University of Cambridge Massachusetts Institute of Welding Technology Institute of
Great Abington, Cambridge, Engineering Australia
United Kingdom Cambridge, Massachussetts, Lidcombe, Australia
Prof. Horst Cerjak Prof. Yanhong Wei
Graz University of Technology Prof. Toshihiko Koseki Harbin Institue of Technology
Graz, Styria, Austria The University of Tokyo Harbin, China
Tokyo, Japan
Dr. Sidney Diamond Dr. Christopher Weisner
U.S. Department of Energy Prof. Kindo Kou TWI
Washington, D.C., USA University of Wisconsin Cambridge, United Kingdom
Madison, Wisconsin, USA
Dr. Pingsha Dong Prof. Konstantin Yushchenko
Battelle Prof. Erich Lugscheider E. O. Paton Electric Welding
Columbus, Ohio, USA Aachen University of Technology Institute
Aachen, Germany Kiev, Ukraine

Prof. David Olson

Colorado School of Mines
Golden, Colorado, USA

Trends in Welding Research

Recent Advances in Modeling of Solidification Behavior.....................................................1
J.M. Vitek, S.A. David, S.S. Babu, Oak Ridge National Laboratory, Oak Ridge, TN, USA

Relationship between the Formation of Hollow Bead Defects and

Hydrogen Assisted Cold Cracking........................................................................................11
I.H. Brown1, G.L.F. Powell1, V.M. Linton1, A. Kufner2, The University of Adelaide,
Adelaide, South Australia1, F-H Konstanz, Konstanz, Germany2

Achieving Grain Refinement through Weld Pool Oscillation..............................................17

T.J. McInerney1, R.B. Madigan1, P. Xu2, C.E. Cross2, Montana Tech of the University of
Montana, Butte, MT, USA1, Federal Institute for Materials Research and Testing, Berlin, Germany2

Modeling and Analysis of a New Solidification Cracking Test ...........................................23

L. Zhu, S. Bachani, R. Nordstrom, M.V. Li, and J. Devletian, Portland State University,
Portland, OR, USA

Proposal of Independent Two Phase Growth during

Solidification in Austenitic Stainless Steels ........................................................................29
H. Inoue1, T. Koseki2, Nippon Steel Corporation, Japan1, The University of Tokyo, Tokyo, Japan2

A Transport Phenomena Based Model to Prevent Liquation Cracking in

Aluminum Alloy Welds...........................................................................................................35
S. Mishra1, T. DebRoy1, S. Chakraborty1,2, Pennsylvania State University, University Park,
PA, USA1, currently with the Indian Institute of Technology, Kharagpur, India2

Advances in Modeling
Prediction of Mechanical Properties of Steel Spot-Welds ..................................................41
M. Mimer, L.-E. Svensson, Volvo Truck Corporation, Gothenburg, Sweden

Predictions of TIG Weld Depth from a

Unified Electrode-Arc-Workpiece Treatment .......................................................................47
J.J. Lowke1, M. Tanaka2, M. Ushio2, CSIRO Industrial Physics Sydney, NSW, Australia1,
Joining and Welding Research Institute, Osaka, Japan2

Algorithms for Inverse Analysis of Welding Processes......................................................53

S.G. Lambrakos1, D.W. Moon1, J.O. Milewski2, U.S. Naval Research Laboratory,
Washington, DC, USA1, Los Alamos National Laboratory, Los Alamos, NM, USA2
Improving the Understanding of Laser Deposition
Processes through Process Simulation ...............................................................................59
R.P. Martukanitz, A.C. Naber, R.M. Melnychuk, R.W. McVey, Pennsylvania
State University, University Park, PA, USA

A Microstructure Model for Laser Processing of Ti-6Al-4V ................................................65

S.M. Kelly1, S.S. Babu2, S.A. David2, T. Zacharia2, S.L. Kampe3, Pennsylvania
State University, State College, PA, USA1, Oak Ridge National Laboratory, Oak Ridge,
TN, USA2, Virginia Tech, Blacksburg, VA, USA3

Simulation of Effect of Weld Variables on Thermal Cycles during

Twin Wire Welding..................................................................................................................71
A. Sharma, N. Arora, S.R. Gupta, Indian Institute of Technology, Roorkee, India

Transport Phenomena
Three-Dimensional Modeling of Transient Heat Transfer and
Fluid Flow during Orbital Gas Tungsten Arc Welding of Pipes..........................................79
W. Zhang, C. Conrardy, I. Harris, Edison Welding Institute Inc., Columbus, OH, USA

Reliable Modeling of Heat and Fluid Flow in Gas-Metal-Arc Fillet Welds through
Optimization of Uncertain Variables .....................................................................................85
A. Kumar, T. DebRoy, Pennsylvania State University, University Park, PA, USA

Transport Phenomena and Genetic Algorithm Based Window of

Welding Variables to Achieve a Target Gas Metal Arc Fillet Weld Geometry ...................91
A. Kumar, T. DebRoy, Pennsylvania State University, University Park, PA, USA

Quantitative Observations of Surface Flow and Solidification on

Autogenous GTA Weld Pools................................................................................................97
D. DeLapp, G. Cook, A. Strauss, W. Hofmeister, Vanderbilt University, Nashville, TN, USA

Scaling Laws in Welding Modeling .....................................................................................103

P.F. Mendez, Colorado School of Mines, Golden, CO, USA

Alloying Element Vaporization and Liquid Metal Expulsion during

Laser Microjoining of Stainless Steel with Short Pulse ....................................................109
X. He1, T. DebRoy1, J.T. Norris2, P.W. Fuerschbach2, Pennsylvania State University,
University Park, PA, USA1, Sandia National Laboratories, Albuquerque, NM, USA2

Numerical Analysis for Optimization of Aluminum Tube Welding...................................115

J. Menke1, D.F. Farson1, M.H. Cho1, B. Green2, L. Brown2, Ohio State University,
Columbus, OH, USA1, EWI, Columbus, OH, USA2

Advanced Welding Processes

Laser Plasma Powder Hybrid Welding in Vertical-Up and
Vertical-Down Positions.......................................................................................................121
K. Stelling, Th. Boellinghaus, M. Lammers, H. Schobbert, Federal Institute for Materials
Research and Testing, Berlin, Germany
Experimental Differences between Aluminium Welding of Tee and
Lap Joints by a cw/Nd:YAG Laser ......................................................................................127
L. Dubourg, National Research Council Canada, Quebec, Canada

Hybrid Laser-GMAW Welding of Aluminum Alloys: A Review .........................................133

D. Rasmussen, L. Dubourg, National Research Council Canada, Quebec, Canada

Laser-GMA Hybrid Welding: Process Monitoring and Thermal Modeling.......................143

E.W. Reutzel, S.M. Kelly, R.P. Martukanitz, M.M. Bugarewicz, P. Michaleris,
Pennsylvania State University, University Park, PA, USA

Double-Sided Arc Welding Process....................................................................................149

Y.M. Zhang1, A.T. Male1, B. Losch1, L. Kvidahl2, M. Ludwig3, J. Emmerson4, University of
Kentucky Center for Manufacturing, Lexington, KY, USA1, Northrop Grumman Ship
Systems, Pascagoula, MS, USA2, General Dynamics Bath Iron Works, Bath, ME, USA3,
Magnatech Limited Partnership, East Granby, CT, USA4

Thin Plate Gap Bridging Study for Nd:YAG Pulsed Laser Lap Welds .............................155
J. Norris, R. Roach, P. Fuerschbach, J. Bernal, Sandia National Laboratories,
Albuquerque, NM, USA

Gravitational Effects on the Weld Pool Shape and

Surface Deformation during GTAW and LBW....................................................................161
N. Kang1, J. Kim1, J. Kim1, C. Kim1, J. Singh2, A.K. Kulkarni2, Korea Institute of Industrial
Technology, Incheon, S. Korea1, Pennsylvania State University, University Park, PA, USA2

Cored Wires for MIG-Welding of Aluminum .......................................................................167

S. Baumgartner1, H. Cerjak1, G. Posch2, Graz University of Technology, Graz, Austria1,
Böhler Schweißtechnik Austria, GmbH, Kapfenberg, Austria2

FSW: Tools, Machines, and Equipment

Tool Design in Friction Stir Processing: Dynamic Forces and Material Flow.................173
D.E. Clark, K.S. Miller, C.R. Tolle, Idaho National Laboratory, Idaho Falls, ID, USA

Experimental Characterization of Tool Heating during Friction Stir Welding .................179

J.L. Covington, W. Robison, B.W. Webb, Brigham Young University, Provo, UT, USA

Microstructural Characteristics and Mechanical Properties of

Friction Skew-Stir Welded Lap Joints in 5083-0 Aluminium.............................................185
G.M.D. Cantin1, S.A. David2, E. Lara-Curzio2, S.S. Babu2, W.M. Thomas3, CSIRO
Manufacturing and Infrastructure Technology, Adelaide, Australia1, Oak Ridge National
Laboratory, Oak Ridge, TN, USA2, TWI, Cambridge, United Kingdom3

Recent Advances in Welding of Aluminum Alloys Using a

Self Reacting Pin Tool (SRPT) Approach with Application Examples .............................191
R. Edwards, G. Sylva, MTS Systems Corp., Eden Prairie, MN, USA

A Novel Tool Design for Friction Stir Spot Welding ..........................................................201

M. Valant, E. Yarrapareddy, R. Kovacevic, Southern Methodist University, Richardson, TX, USA
Friction Stir Spot Welding of Automotive Lightweight Alloys..........................................207
S. Lathabai, M.J. Painter, G.M.D. Cantin, V.K. Tyagi, CSIRO Manufacturing and
Infrastructure Technology, Woodville, Australia

Design and Testing of a Friction Stir Processing Machine for

Laboratory Research............................................................................................................213
K.S. Miller, R.J. Bitsoi, E.D. Larsen, H.B. Smartt, Idaho National Labroatory,
Idaho Falls, ID, USA

Complete Inspection of Friction Stir Welds in Aluminum Using

Ultrasonic and Eddy Current Arrays...................................................................................219
A. Lamarre, O. Dupuis, M. Moles, R/D Tech, Quebec, Canada

MWM®-Array Inspection for Quality Control of Friction Stir Welded Extrusions ............227
D. Grundy1, V. Zilberstein1, N. Goldfine1, J. Green2, I. Stol2, JENTEK Sensors, Inc.,
Waltham, MA, USA1, Alcoa Technical Center, Pittsburgh, PA, USA2

FSW: Process Physics and Modeling

Friction Stir Welding and Processing: A Sprinter's Start, A Marathoner's Finish ..........233
M.W. Mahoney, Rockwell Scientific, Thousand Oaks, CA, USA

Torque and Plunge Force during the Plunge Phase of Friction Stir Welding .................241
A. Nunes1, J. McClure2, R. Ávila2, NASA, Huntsville, AL, USA1, University of Texas at
El Paso, El Paso, TX, USA2

Development of a Heat Input Model for Friction Stir Welding ..........................................247

J.W. Pew, J.H. Record, T.W. Nelson, C.D. Sorensen, Brigham Young University,
Provo, UT, USA

Quantifying the Material Processing Conditions for an Optimized FSW Process ..........253
J.A. Schneider1, A.C. Nunes, Jr.2, Mississippi State University, Mississippi State, MS, USA1,
NASA, Huntsville, AL, USA2

Metal Cutting Theory and Friction Stir Welding.................................................................257

L.N. Payton, Auburn University, Auburn, AL, USA

3D Modelling of Thermofluid Flow in Friction Stir Welding ..............................................261

E. Feulvarch1, Y. Gooroochurn1, F. Boitout1, J.-M. Bergheau2, ESI Group, Cedex 03,
France1, ESI North America, Bloomfield Hills, MI, USA1, Laboratoire de Tribologie et
Dynamique des Systèmes, Cedex 02, France2

Eulerian Elasto-Plastic Formulations for Residual Stress Analysis of

Friction Stir Welding.............................................................................................................267
J. Song, P. Michaleris, Pennsylvania State University, University Park, PA, USA

Modelling Temperature Histories in Friction Stir Welding

Including Material Convection Effects................................................................................273
A. Simar, T. Pardoen, B. de Meester, Université Catholique de Louvain,
Louvain-la-Neuve, Belgium
Simulation of HSLA-65 Friction Stir Welding .....................................................................279
D. Forrest1, J. Nguyen1, M. Posada1, J. DeLoach1, D. Boyce2, J. Cho2, P. Dawson2, Naval
Surface Warfare Center, West Bethesda, MD, USA1, Cornell University, Ithaca, NY, USA2

Simulation of Friction Stir Weld Microstructures in Steel: Preliminary Studies .............287

S.J. Norton, J.C. Lippold, Ohio State University, Columbus, OH, USA

Repair Weldability of Service-Exposed, Heat-Resisting Alloys—
Austenitic Stainless Steel Castings: HP45Nb, HP50Nb, and 20-32Nb .............................293
S. Shi1, J.C. Lippold1, J. Ramirez2, Ohio State University, Columbus, OH, USA1,
Edison Welding Institute, Columbus, OH, USA2

Weldability of Boron Alloyed High-Temperature

Resistant 9% Chromium Casting Steel COST-CB2A .........................................................299
P. Mayr, A. Schalber, E. Letofsky, H. Cerjak, Graz University of Technology, Graz, Austria

Weldability of a High Strength Engineering Steel with High Carbon Equivalent ............305
L. Hasselrot, P. Olund, Ovako Steel AB, Hofors, Sweden

Prediction of Maximum Crack Length in Longitudinal Varestraint Testing ....................313

C.V. Robino1, M. Reece1, G.A. Knorovsky1, J.N. DuPont2, Z. Feng3, Sandia National
Laboratories, Albuquerque, NM, USA1, Lehigh University, Bethlehem, PA, USA2,
Oak Ridge National Laboratory, Oak Ridge, TN, USA3

Linear Inspection of Welds Using Ultrasonic Phased Arrays...........................................319

M. Moles, S. Labbé, R/D Tech, Quebec, Canada

Ductility-Dip Cracking Susceptibility of

Filler Metal 52 and 52M Ni-Base Filler Metals.....................................................................327
N.E. Nissley, J.C. Lippold, Ohio State University, Columbus, OH, USA

Cracking Behavior in Nickel-Based Single Crystal Superalloy Welds.............................333

J.M. Vitek1, S.S. Babu1, S.A. David1, J.-W. Park2, Y. Hu3, W. Hehmann3, Oak Ridge
National Laboratory, Oak Ridge, TN, USA1, Samsung Electro-Mechanics Co., Suwon,
South Korea2, Honeywell Aerospace Services, Greer, SC, USA3

Thermomechanical Behavior of Nickel Base Single Crystal Superalloy towards

Understanding of Weld Hot Cracking .................................................................................339
J.M. Vitek1, S.A. David1, S.S. Babu1, M. Murugananth2, Oak Ridge National Laboratory,
Oak Ridge, TN, USA1, Nanyang Technological University, Singapore2

Liquation Cracking in Full-Penetration Aluminum Welds:

A Necessary Condition for Crack Susceptibility ...............................................................345
C. Huang, G. Cao, S. Kou, University of Wisconsin, Madison, WI, USA

Liquation Cracking in Partial-Penetration Aluminum Welds:

Effect of Welding Conditions ................................................................................................353
C. Huang, S. Kou, University of Wisconsin, Madison, WI, USA
New Approaches to Evaluation of Weldability of Materials ..............................................361
K.A. Yushchenko, V.V. Derlomenko, E.O. Paton Electric Welding Institute, Kiev, Ukraine

Peculiarities of A-TIG Welding of Stainless Steel ..............................................................367

K.A. Yushchenko, D.V. Kovalenko, I.V. Kovalenko, E.O. Paton Electric Welding
Institute of the National Academy of Science of Ukraine, Kyiv, Ukraine

FSW: Application to Alloys

Process-Property Relationships and Microstructure Evolution in
Friction Stir Welded Thin Sheet 2024-T3 Aluminum Alloy ................................................377
A.K. Shukla1, W.A. Baeslack III2, Rensselaer Polytechnic Institute, Troy, NY, USA1,
Ohio State University, Columbus, OH, USA2

Transverse Tensile Properties of AA2524 Friction Stir Welds:

Quasi-Static and High Rate Loading...................................................................................383
A.P. Reynolds, J. Pohlman, University of South Carolina, Columbia, SC, USA

Effect of Microstructure on Postweld Formability in

Friction Stir Welded Al Alloy 5052 ......................................................................................387
Y.S. Sato, Y. Sugiura, Y. Shoji, S.H.C. Park, H. Kokawa, K. Ikeda, Tohoku University,
Sendai, Japan

Effect of Heat and Plastic Deformation on the Texture of a

Friction Stir Processed 6061-T6 Aluminum Alloy: A Neutron Diffraction Study.............393
W. Woo1,2, H. Choo1,2, P.K. Liaw1, Z. Feng2, S.A. David2, C.R. Hubbard2, D.W. Brown3,
M.A.M. Bourke3, University of Tennessee, Knoxville, TN, USA1, Oak Ridge National
Laboratory, Oak Ridge, TN, USA2, Los Alamos National Laboratory, Los Alamos, NM, USA3

Investigation of Properties of Friction Stir Welds in

Age Hardenable 7xxx Aluminium Alloys ............................................................................401
V.M. Linton, S. Renc, I.H. Brown, University of Adelaide, Adelaide, Australia

Parametric Study on High Speed-Friction Stir Welding of

Dissimilar Aluminum Alloys ................................................................................................407
Gangadhar Bhat K, G. Talia, Wichita State University, Wichita, KS, USA

A Study on Dissimilar Friction Stir Welds between Al and Mg Alloys .............................413

R. Zettler, J.F. Dos Santos, A. Blanco, A. da Silva, GKSS-Forschungszentrum
Geesthacht, Geesthacht, Germany

Mechanical Properties and Corrosion Resistance of

Friction Stir Welded AZ31B-H24 Magnesium Alloy ...........................................................421
M. Pareek, A. Polar, F. Rumiche, J.E. Indacochea, University of Illinois at Chicago,
Chicago, IL, USA

Modifying AM60B Magnesium Alloy Die Cast Surfaces by

Friction Stir Processing .......................................................................................................427
M. Santella1, Z. Feng1, C. Degen2, T.-Y. Pan3, Oak Ridge National Laboratory,
Oak Ridge, TN, USA1, South Dakota School of Mines and Technology,
Rapid City, SD, USA2, Ford Motor Company, Dearborn, MI, USA3
Friction Stir Welding of Copper:
Metallurgical Characterization and Corrosion Resistance...................................................431
A. Polar, F. Rumiche, M. Pareek, J.E. Indacochea, University of Illinois at Chicago,
Chicago, IL, USA

Microstructures and Performance of Welded Joints of Red Copper and

Brass by Friction Stir Welding ............................................................................................437
X. Liu, J. Yan, N. Yang, Northwestern Polytechnical University, Xi' an 710072, China

Sigma Phase Formation in Friction Stirring of Iron-Nickel-Chromium Alloys ................441

C.D. Sorensen, T.W. Nelson, Brigham Young University, Provo, UT, USA

Welding Processes and Consumables

Trends in Microwelding........................................................................................................447
G.A. Knorovsky1, D.O. MacCallum1, E.A. Holm1, J.R. Michael1, V.V. Semak2,
B.M. Nowak-Neely3, Sandia National Laboratories, Albuquerque, NM, USA1,
Pennsylvania State University, Kittanning, PA, USA2, University of New Mexico,
Albuquerque, NM, USA3

Microwelding Using a Modified SEM ..................................................................................453

D. MacCallum1, G. Knorovsky1, B. Nowak-Neely2, Sandia National Laboratories,
Albuquerque, NM, USA1, University of New Mexico, Albuquerque, NM, USA2

Prediction of Dilution in Universal Gas Metal Arc Cladding

Using Response Surface Methodology ..............................................................................459
A.S. Shahi1, S. Pandey2, Sant Longowal Institute of Engineering and Technology,
Longowal, Punjab, India1, Indian Institute of Technology, Delhi, India2

Weld Repair of Heat Recovery Steam Generator Tubing ..................................................465

D. Gandy, G. Frederick, K. Coleman, EPRI, Charlotte, NC, USA

Design of Process-Material-Shielding Combinations for

Hard Coatings Using Laser Surface Alloying ....................................................................473
S.S. Babu1, S.M. Kelly2, R.P. Martukanitz2, M. Murugananth3, Oak Ridge National
Laboratory, Oak Ridge, TN, USA1, Applied Research Laboratory, State College,
PA, USA2, Nanyang Technological University, Singapore3

Influence of Procedure Variables on C-Mn-Ni-Mo ANSI/AWS

A5.29-98 E111T5-K3 Metal Cored Wire Ferritic All-Weld Metal.........................................479
E.S. Surian1, N.M. Ramini de Rissone1, H.G. Svoboda2, L.A. de Vedia3, National
Technological University, Buenos Aires, Argentina1, University of Buenos Aires,
Buenos Aires, Argentina2, National University of San Martín, Buenos Aires, Argentina3

Development of Filler Metals for Super Austenitic Stainless Steels................................485

T.D. Anderson, M.J. Perricone, J.N. DuPont, Lehigh University, Bethlehem, PA, USA

Effect of Arc Welding Electrode Temperature on Vapor and Fume Composition ..........491
N.T. Jenkins1, P.F. Mendez2, T.W. Eagar3, Massachusetts General Hospital,
Charlestown, MA, USA1, Colorado School of Mines, Golden, CO, USA2,
Massachusetts Institute of Technology, Cambridge, MA, USA3
Thermoelectric Power Assessment of Weld Microstructure,
Phase Stability, Residual Stress, and Properties ..............................................................497
Y.D. Park1, D.L. Olson2, B. Mishra2, A.N. Lasseigne2, Hyundai Motors, Kyunggido,
South Korea1, Colorado School of Mines, Golden, CO, USA2

Hydrogen in Rutile Wire Welds ...........................................................................................507

W. Mazur, CSIRO Division of Manufacturing and Infrastructure Technology,
Woodville, Australia

Characterization of E6010 and E7018 Welding Fume........................................................513

J.W. Sowards1, J.C. Lippold1, D.W. Dickinson1, A.J. Ramirez2, Ohio State University,
Columbus, OH, USA1, Brazilian Synchrotron Light Laboratory, Campinas SP, Brazil2

Root Bead Welding of Duplex Stainless Steel Pipeline Girth Welds without
Backing Gas..........................................................................................................................519
M. Boring1, N. Ames1, M. Collins2, D. Fetzner3, Edison Welding Institute, Columbus,
OH, USA1, ConocoPhillips Alaska, Inc., Anchorage, AK, USA2, BP Alaska, Inc.,
Anchorage, AK, USA3

Investigation of a Possible Health Hazard from

Asbestos Fibers Used in Welding Electrodes....................................................................525
A.A. Johnson, R.J. Storey, Metals Research Inc., Louisville, KY, USA

Residual Stresses and Distortion

Predicting Distortion and Residual Stress in
Complex Welded Structures by Designers ........................................................................531
J. Goldak1, J. Zhou1, S. Tchernov1, D. Downey1, S. Wang2, B. He2, Goldak Technolgies Inc.,
Ottawa, Ontario, Canada1, Carleton University, Ottawa, Ontario, Canada2

Evaluation of a Decoupled Plastic Strain Method for

Welding Distortion Prediction .............................................................................................541
L. Zhang1, P. Michaleris1, J. Sun1, P. Marugabandhu2, Pennsylvania State University,
University Park, PA, USA1, Maglev Inc., Monroeville, PA, USA2

Finite Element Modeling of Vibration Stress Relief after Welding ...................................547

Y.P. Yang1, G. Jung1, R. Yancey2, Edison Welding Institute, Columbus, OH, USA1,
Altair Engineering, Inc., Irvine, CA, USA2

MPI Implementation of the FETI-DP-RBS-LNA Algorithm and

Its Applications on Large Scale Problems with Localized Nonlinearities .......................553
J. Sun, P. Michaleris, Pennsylvania State University, University Park, PA, USA

Residual Stress in Thick Sectioned Highly Restrained Steel Welds................................559

S. Pearce1, V. Linton1, G. Sloan2, The University of Adelaide, Adelaide, Australia1,
ASC, Adelaide, Australia2

Residual Stress Measurements in Side Bonded Resistance Welds ................................565

P.S. Korinko, E.G. Estochen, G.J. McKinney, M.J. Pechersky, Savannah River
National Laboratory, Aiken, SC, USA
Distortion Control in Precision Weldment by Penetration-Enhanced GTAW..................571
S. Khurana, N. Ames, W. Zhang, Edison Welding Institute, Columbus, OH, USA

Welding with a Trailing Heat Sink: How to Optimise the Cooling Parameters................577
E.M. van der Aa1,2, I.M. Richardson1,2, M.J.M. Hermans2, Netherlands Institute for
Metals Research, Delft, The Netherlands1, Delft University of Technology, Delft,
The Netherlands2

Development of Compressive Residual Stress in

Structural Steel Weld Toes by Means of Weld Metal Phase Transformations ................583
F. Martinez1, S. Liu2, Caterpillar Technical Center, Moosville, IL, USA1, Center for
Welding, Joining, and Coatings Research, Golden, CO, USA2

Numerical Analysis and Measurement of Residual Stresses in

Multi Pass Welding—Influence on Fatigue Root Crack Propagation...............................589
Z. Barsoum, Royal Institute of Technology, Stockholm, Sweden

Fatigue and Residual Stress Responses of Welded Piping Joints ..................................595

A.E. Humphreys1, T. Hassan1, C.R. Hubbard2, North Carolina State University,
Raleigh, NC, USA1, Oak Ridge National Laboratory, Oak Ridge, TN, USA2

Vibratory Stress Relief: Methods Used to Monitor and Document Effective

Treatment—A Survey of Users and Directions for Further Research..............................601
B.B. Klauba1, C.M. Adams2, J.T. Berry3, Airmatic Inc., Philadelphia, PA, USA1,
Annapolis, MD, USA2, Mississippi State University, Mississippi State, MS, USA3

AH-Adaptive FE Scheme for Welding Distortion Analysis in Large Structures..............607

S.H. Tsau, P. Michaleris, Pennsylvania State University, University Park, PA, USA

Measurement of Welding Residual Stresses and

Redistribution due to Cyclic Loading .................................................................................617
C. Sanger, P. Kurath, University of Illinois Urbana-Champaign, Urbana-Champaign,

Automation, Robotics, Sensing, and Control

Design of a Robotic Welding System .................................................................................623
H.B. Smartt, E.D. Larsen, D.P. Pace, R.J. Bitsoi, C.R. Tolle, Idaho National Laboratory,
Idaho Falls, ID, USA

Off-Line Programming of Robots for Metal Deposition.....................................................629

M. Ericsson1, F. Danielsson1, H. Carlsson1, P. Nylén1,2, University of
Trolhättan Uddevalla, Trolhättan, Sweden1, Volvo Aero Corporation, Trolhättan, Sweden2

Process-Planning Models for Welding Using Bayesian Network.....................................635

M. Kristiansen, O. Madsen, Aalborg University, Aalborg, Denmark

Planning of Dynamic Trajectories for Weld Process Control Variables by

Finite Element Simulation and Iterative Learning..............................................................641
P.V. Jeberg1, S. Lambaek1, H. Holm2, Odense Steel Shipyard, Odense, Denmark1,
Aalborg University, Aalborg, Denmark2
Fabrication of Net-Shape Metallic Parts by Overlapping Reinforced Weld Beads .........647
K.P. Cooper, S.G. Lambrakos, Naval Research Laboratory, Washington, DC, USA

Effect of Welding Parameters on Formation of Toe Imperfections in

Tandem Gas Metal Arc Welding ..........................................................................................653
M. Farajian-Sohi1, N. Järvstråt1, M. Thuvander2, University of Trolhättan/Uddevalla,
Trolhättan, Sweden1, ESAB, Göteborg, Sweden2

Development of a Model to Predict Weld Bead Geometry during the

GMAW Process Using High Speed Camera .......................................................................659
E.J. Lima II, C. Castro, A.Q. Bracarense, M.M. Campos, Federal University of
Minas Gerais, Minas Gerais, Brazil

Weld Pool Surface Monitoring and Depth Extraction Using a

Calibrated CCD Sensor ........................................................................................................665
G. Saeed, Y.M. Zhang, C. Jaynes, University of Kentucky, Lexington, KY, USA

A Compact Sensor for Welding Process Control ..............................................................671

G. Saeed, Y.M. Zhang, S. Cook, University of Kentucky, Lexington, KY, USA

Arc Acoustic Feedback in GMA Welding............................................................................677

J. Tam, J. Huissoon, University of Waterloo, Waterloo, Ontario, Canada

Advanced Welding Control and Monitoring Systems Developed for

Thick-Section Narrow Groove Welding ..............................................................................683
J. Kikel1, S. Menicos1, B. Grut1, D. Schwemmer2, BWX Technologies Inc., Barberton,
OH, USA1, AMET Inc., Rexburg, ID, USA2

Model Based Metal Transfer Control ..................................................................................687

J.S. Thomsen, Aalborg University, Aalborg, Denmark

Nonlinear Interval Model Control of the Double Sided Arc Welding Process .................693
B. Losch, Y.M. Zhang, Center for Manufacturing Systems, Lexington, KY, USA

Effects during the Starting Period of the MIG Process .....................................................697

G. Huismann, University of the Federal Armed Forces, Hamburg, Germany

Dynamic Response of Electrode Extension to Step Changes in

Average Current for Pulsed GMAW ....................................................................................705
P.G. Krepp, R. Crawford, G.E. Cook, A.M. Strauss, Vanderbilt University, Nashville, TN, USA

Microstructure and Properties of Weldments

Microstructure and Mechanical Properties of Keyhold Gas Tungsten
Arc Welds in Titanium and Titanium Alloys .......................................................................711
S. Lathabai, K.J. Barton, L.K. Green, V.K. Tyagi, CSIRO Manufacturing and
Infrastructure Technology, Woodville, Australia

Residual Strength in Integral (Welded) Al Fuselage Structures:

Understanding Failure Peculiarities Enables Amazing Strength Values.........................717
F. Palm, EADS Corporate Research Center Germany, Ottobrunn, Germany
Correlating Temperatures, Structures, and Hardness in A36 GMA Welds......................725
D.W. Moon, C.R. Feng, S.G. Lambrakos, E.A. Metzbower, US Naval Research
Laboratory, Washington, DC, USA

Effects of Enhanced Convection on the Microstructure of Dissimilar Welds .................731

T. Gandhi, D.K. Aidun, Clarkson University, Potsdam, NY, USA

Welding Procedures and Type IV Phenomena...................................................................737

J.A. Francis1, W. Mazur1, H.K.D.H. Bhadeshia2, CSIRO Manufacturing and Infrastructure
Technology, Adelaide, Australia1, University of Cambridge, Cambridge, United Kingdom2

Tritium Effects on Fracture Toughness of Stainless Steel Weldments ...........................743

M.J. Morgan, G.K. Chapman, M.H. Tosten, S.L. West, Westinghouse Savannah
River Company, Aiken, SC, USA

Welding Development for the World's Strongest Pipeline: X120 .....................................749

D. Fairchild1, M. Macia1, N. Bangaru2, J. Koo2, A. Ozekcin2, H. Jin2, ExxonMobil
Upstream Research Co., Houston, TX, USA1, ExxonMobil Research and Engineering Co.,
Annandale, NJ, USA2

Structural Integrity of Submarine Pressure Hull Welds ....................................................755

G.W. Sloan, ASC Pty Ltd., South Australia, Australia

A Microstructural Study of Liquated Grain Boundaries in

Heat Affected Zone of Welded Inconel 738LC Superalloy ................................................761
O.A. Ojo, N.L. Richards, M.C. Chaturvedi, University of Manitoba, Winnipeg,
Manitoba, Canada

Auger Spectroscopy Results from Ductility Dip Cracks Opened under

Ultra-High Vacuum ...............................................................................................................767
T.E. Capobianco, M.E. Hanson, Lockheed Martin, Schenectady, NY, USA

Welding of Single-Crystal Nickel-Based Superalloys:

How to Avoid Stray Grains and Why ....................................................................................773
J.M. Vitek, S.S. Babu, S.A. David, Oak Ridge National Laboratory, Oak Ridge, TN, USA

The Effects of Processing Parameters on the

Microstructural Evolution and Mechanical Properties of
Inertia Friction Welded 21Cr-6Ni-9Mn .................................................................................781
J.D. Puskar1, B.P. Somerday1, D.K. Balch1, J.A. Brooks1, C.H. Cadden1, J.R. Michael2,
Sandia National Laboratories, Livermore, CA, USA1, Sandia National Laboratories,
Albuquerque, NM, USA2

Effect of Penetration-Enhancing Compounds on the

Weld Metal Microstructure of Super Duplex Stainless Steels ..........................................787
N. Ames1, J. Lippold2, M. Johnson3, EWI, Columbus, OH, USA1, Ohio State University,
Columbus, OH, USA2, Los Alamos National Labs, Los Alamos, NM, USA3

Effect of Austenitic Fillers on Mechanical Properties of

AISI 410 Martensitic Stainless Steel Weldments ...............................................................793
A.V. Kumar, N. Muthukrishnan, Sri Venkateswara College of Engineering, Tamilnadu, India
Advanced Joining
Techniques for Determination of Interstitial Nitrogen Content in
Nitrogen-Strengthened Austenitic Stainless Steel Alloy 1.4565 ......................................799
A.N. Lasseigne1, D.L. Olson1, Th. Boellinghaus2, Colorado School of Mines, Golden,
CO, USA1, Federal Institute of Materials Research and Testing, Berlin, Germany2

Spectroscopic Monitoring of Hydrogen in Welding Arcs .................................................807

J. Chandler, G. Edwards, S. Liu, Colorado School of Mines, Golden, CO, USA

Analysis of Double-Electrode Gas Metal Arc Welding ......................................................813

C.S. Wu1, G.X. Xu1, K.H. Li2, Y.M. Zhang2, Shandong University, Jinan, China1,
University of Kentucky Center for Manufacturing, Lexington, KY, USA2

Experimental Investigations of the Arc in MIG-MAG Welding ..........................................819

S. Zielińska1, K. Musioł1, N. Pellerin2, S. Pellerin2, Ch. de Isarra2, F. Briand3, F. Richard3,
T. Opderbecke3, Jagellonian University, Krakow, Poland1, Centre Universitaire de Bourges,
Cedex 2, France2, CTAS-Air Liquide, Cedex, France3

Silicon Surface Modification due to CO2 and Free Electron Laser Radiation .................825
D.W. Neat, D.R. DeLapp, J.A. Kozub, G.E. Cook, A.M. Strauss, Vanderbilt University,
Nashville, TN, USA

Laser Cladding of Low Pressure Turbine Blades ..............................................................831

M. Brandt1, J. Harris1, A. Bishop2, Swinburne University of Technology, Melbourne,
Australia1, Welding Technology Institute of Australia, Melbourne, Australia2

Risk of Skin Cancer from Arc Welding ...............................................................................835

A. Dixon1, B. Dixon2, Skincanceronly, Victoria, Australia1, Defence Science Technology
Organisation, Victoria, Australia2

In-Process Monitoring of Pinch Welding:

An Investigation into a Bond Quality Metric ......................................................................839
D.A. Hartman1, M.G. Smith1, M.J. Cola1, V.R. Dave1, P.S. Korinko2, S.H. Malene2,
N.R. Tolk3, J.P. Miller4, W.H. King5, Los Alamos National Laboratory, Los Alamos,
NM, USA1, Savannah River National Laboratory, Aiken, SC, USA2, Honeywell FM&T,
Albuquerque, NM, USA3, Honeywell FM&T, Kansas City, MO, USA4, William H. King
Consulting, Higganum, CT, USA5

Friction Model Parameter Optimization for Friction Welding

Simulation of Ti-6246............................................................................................................849
J.P. Åström, M.O. Näsström, Luleå University of Technology, Luleå, Sweden

Process Optimization for Linear Friction Welding of Ti6Al4V ..........................................855

P. Wanjara, C. Booth-Morrison, E. Hsu, M. Jahazi, National Research
Council-Institute for Aerospace Research, Montreal, Quebec, Canada

Effect of Substrate Grain Size and Orientation on the Transient

Liquid-Phase Bonding of Ferritic Oxide Dispersion Strengthened Alloys ......................861
V.G. Krishnardula, N.I. Sofyan, J.W. Fergus, W.F. Gale, Auburn University,
Auburn, AL, USA
Transient Liquid Phase Joining of a Current Generation Gamma TiAl Alloy—
Gamma Met PX .....................................................................................................................867
D.A. Butts, W.F. Gale, Auburn University, Auburn, AL, USA

Joining of Inconel 617 Sheets in TLP Bonding Using Ni-P and Ni-Si-B Alloys...............873
F. Jalilian1, R.A.L. Drew1, M. Jahazi2, McGill University, Montreal, Quebec, Canada1,
Aerospace Manufacturing Technology Center, Montreal, Quebec, Canada2

Wide-Gap Transient Liquid Phase Bonding of Single Crystal to

Polycrystalline Nickel-Base Superalloys: Microstructural Development and
Mechanical Properties..........................................................................................................879
R. Aluru, N.I. Sofyan, J.W. Fergus, W.F. Gale, Auburn University, Auburn, AL, USA

Solid-State Diffusion Bonding of MA956 and PM2000 ......................................................885

V.G. Krishnardula, R. Aluru, N.I. Sofyan, J.W. Fergus, W.F. Gale, Auburn University,
Auburn, AL, USA

Diffusion Bonding of Alloy 690—Initial Studies.................................................................889

P.W. Hochanadel, M.J. Cola, V.R. Dave, A.M. Kelly, R.S. Casey, R.D. Bramlett,
D.W. Rendell, Los Alamos National Laboratory, Los Alamos, NM, USA

Analysis of Cold Bonding at Gold-Gold Contact in a Thermal Switch.............................895

L. Li, Utah State University, Logan, UT, USA

Phase Transformations
Advanced Techniques for In-Situ Monitoring of Phase Transformations during
Welding Using Synchrotron-Based X-Ray Diffraction.......................................................901
J.W. Elmer1, T.A. Palmer1, W. Zhang2, T. DebRoy2, Lawrence Livermore National
Laboratory, Livermore, CA, USA1, Pennsylvania State University, University Park, PA, USA2

Microstructures of Inertia Friction Welds in Fe-Ni-Cr Ternary Alloys..............................911

J.D. Puskar, J.A. Brooks, N. Yang, Sandia National Laboratories, Livermore, CA, USA

Phase Selection Phenomena during Low-Alloy Steel Weld Solidification ......................917

S.S. Babu1, S.A. David1, J.M. Vitek1, J.W. Elmer2, T.A. Palmer2, M.A. Quintana3,
Oak Ridge National Laboratory, Oak Ridge, TN, USA1, Lawrence Livermore National
Laboratory, Livermore, CA, USA2, Lincoln Electric Company, Cleveland, OH, USA3

Simulation of Dissolution and Coarsening in the HAZ of

6061 Al-Alloy during Laser Welding ...................................................................................923
A.D. Zervaki, G.N. Haidemenopoulos, University of Thessaly, Volos, Greece

Structure and Mechanical Properties of Pulsed-GMA Welded Al-Li Alloy ......................929

G. Padmanabham1, S. Pandey2, M.K. Schaper3, Ministry of Science & Technology,
New Delhi, India1, Indian Institute of Technology Delhi, Delhi, India2, Dresden
University of Technology, Dresden, Germany3

Development of Intergranular Corrosion Resistant 316

Austenitic Stainless Steel by Grain Boundary Engineering .............................................935
M. Michiuchi, H. Kokawa, Z.J. Wang, Y.S. Sato, K. Sakai, Tohoku University,
Sendai, Japan
Ferrite Formation in a Duplex Stainless Steel during Controlled Heating: X-Ray
Diffraction and Modeling Using Para-Equilibrium Diffusion Kinetics..............................939
W. Zhang1, T. DebRoy1, T.A. Palmer2, J.W. Elmer2, Pennsylvania State University,
University Park, PA, USA1, Lawrence Livermore National Laboratory, Livermore, CA, USA2

Direct Observations of Sigma Phase Growth and

Dissolution in 2205 Duplex Stainless Steel........................................................................945
T.A. Palmer1, J.W. Elmer1, S.S. Babu2, E.D. Specht2, Lawrence Livermore National
Laboratory, Livermore, CA, USA1, Oak Ridge National Laboratory, Oak Ridge, TN, USA2

Numerical Simulation of Microstructural Evolution during

Welding of Duplex Stainless Steels ....................................................................................951
C.M. Garzón1, J. Gomes1, A.J. Ramirez1, S.D. Brandi2, Brazilian Synchrotron Light
Laboratory, São Paulo, Brazil1, University of São Paulo, São Paulo, Brazil2

Kinetics of the Delta Ferrite—Sigma Phase Transformation in

22Cr-13Ni-5Mn Multi-Pass Gas Tungsten Arc Welds ........................................................957
D.K. Balch1, J.D. Puskar1, B.P. Somerday1, D.F. Susan2, Sandia National Laboratories,
Livermore, CA, USA1, Sandia National Laboratories, Albuquerque, NM, USA2

Microstructure and Properties of Post Weld Heat

Treated 2.25Cr1Mo Weld Metal............................................................................................963
E.-L. Bergquist, L. Karlsson, M. Thuvander, E. Keehan, ESAB AB, Gothenburg, Sweden

Understanding Mechanical Properties of

Novel High Strength Steel Weld Metals through
High-Resolution Microstructural Investigations................................................................969
E. Keehan1, L. Karlsson1, H.O. Andrén2, H.K.D.H. Bhadeshia3, ESAB AB, Gothenburg,
Sweden1, Chalmers University of Technology, Gothenburg, Sweden2, University of
Cambridge, Cambridge, England3

A New Methodology for Studying Phase Transformations in

High Strength Steel Weld Metal...........................................................................................975
B.T. Alexandrov, J.C. Lippold, Ohio State University, Columbus, OH, USA

Effect of Steel Composition on the Dispersion of Oxide

Inclusions in Steel Melts ......................................................................................................981
T. Suzuki, T. Koseki, The University of Tokyo Hongo, Tokyo, Japan

Investigation of Factors Affecting Ferrite Transformation from

Steel-Oxide Interface ............................................................................................................987
M. Tsutsumi, H. Kato, T. Koseki, The University of Tokyo Hongo, Tokyo, Japan

A Dual-Mesh Strategy for Microstructure Development in a

Macroscopic Heat Affected Zone: Studies on AISI316L and AISI1005 ............................993
R.G. Thiessen1, I.M. Richardson1, J. Sietsma2, Netherlands Institute for Metals Research,
Delft, The Netherlands1, Delft University of Technology, Delft, The Netherlands2

Genetic Algorithm Based Optimization of Johnson Mehl Avrami Equation

Parameters for Ferrite to Austenite Transformation in Steel Welds ..............................1001
S. Mishra1, A. Kumar1, T. DebRoy1, J.W. Elmer2, Pennsylvania State University,
University Park, PA, USA1, Lawrence Livermore National Laboratory, Livermore, CA, USA2
Prediction and Measurement of Phase Transformations,
Phase-Dependent Properties and Residual Stresses in Steels......................................1007
M. Becker, C. Jordan, S.K. Lachhander, A. Mengel, M. Renauld, Lockheed Martin,
Schenectady, NY, USA

Author Index .......................................................................................................................1013

Keyword Index ....................................................................................................................1019


The traditional Appreciation Barbeque at Callaway Garden’s West Beach

ASM International® proudly sponsored the 7th International Conference on Trends in Welding Research (TWR) May
16–20, 2005 at the Callaway Gardens Resort in Pine Mountain, Georgia (USA). The five-day event, endorsed and
supported by the American Welding Society and fifteen International organizations, was attended by 275 delegates
from twenty countries around the world. As is the trademark of TWR, the ideal weather and relaxing venue provided
the prefect atmosphere for the week’s excellent technical programs, networking, and learning opportunities.

The conference continued to earn its well-deserved reputation as the only event of its kind to attract the world’s
leading welding researchers. There were five keynote presentations by world-renowned experts in the field of
welding science and technology, and a host of professional speakers from around the globe. It is our ability to attract
professionals of this caliber that makes TWR a globally recognized event for the welding research community.

TWR 2005 consisted of thirty oral technical sessions and one poster technical session. Topics covered included
transport phenomena, solidification of welds, modeling, microstructure and properties, phase transformation,
weldability, residual stresses, advanced characterization techniques, properties of welds, advanced joining processes
including hybrid welding, friction stir welding, and sensing control and automation. Some of the highlights of the
conference included five technical sessions devoted to friction stir welding and sessions on advanced neutron
diffraction studies and in-situ synchrotron, diffraction investigations to quantitatively understand evolution of phases
and phase transformation kinetics during welding. The presentations described recent and innovative developments
in the field of welding science and technology and were characterized by the integration of theory, modeling, and
experiments across a number of disciplines yielding a more comprehensive picture of the welding process.

Plans are already underway for the 8th Trends In Welding Research Conference to be held in the Spring of 2008.
Please visit our Website at www.asminternational.org/trends in the coming months for dates, location and
conference details. We hope to see you all in 2008.

As a final note, the 2005 Conference Chairmen would like to recognize the planning and organizational efforts of
the staffs of ASM International (Deborah Porter and Jennifer Arnold) and ORNL (Shirin Badlani) and their
dedication to making each Trends In Welding Research event a memorable experience for our delegates.

Regards, your TWR 2005 Chairmen

Dr. Stan A. David, FASM, and Dr. John M. Vitek, FASM Oak Ridge National Laboratory
Dr. Tarasankar DebRoy, FASM Pennsylvania State University
Dr. Herschel B. Smartt Idaho National Laboratory
Dr. John C. Lippold, FASM The Ohio State University

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Recent Advances in Modeling of Solidification Behavior

J. M. Vitek, S. A. David
Oak Ridge National Laboratory, Oak Ridge, Tennessee, U. S. A.

S. S. Babu
formerly Oak Ridge National Laboratory, now Edison Welding Institute, Columbus, Ohio, U. S. A.

Modeling and simulation are ideal tools for developing a
Modeling of weld solidification behavior has made significant fundamental understanding of weld solidification behavior.
advances in recent years. Solidification theory has been Since weld conditions vary with location, controlled
applied to produce models that identify phase selection during solidification experiments that may be useful for studying
solidification, phase transformation behavior after traditional casting processes have more limited value with
solidification, and microstructural evolution in general. In regard to weld solidification. Therefore, modeling is needed to
addition, the use of computational thermodynamics allows for consider the entire range of conditions that may exist.
a robust evaluation of phase stability in multicomponent
systems that represent real alloy systems. Several examples of Solidification theory has advanced significantly over the last
currently available models for solidification behavior are few decades. Rather than concentrate on the theoretical
presented. The expanded use of models can lead to better developments, this paper will be directed at advances in
optimization of weld procedures and weld alloys at relatively modeling techniques that allow for the application of the
small cost since extensive trial and error experiments can be advanced theories to study weld solidification. Furthermore,
avoided. Further success in modeling will rely, to a large this paper will not deal with advances in heat and fluid flow
extent, on the ability to integrate existing models. modeling. This subject is treated elsewhere in this conference
as well as in earlier proceedings in this series [2]. Instead, this
Introduction paper will focus on the following subjects: thermodynamic,
kinetic and phase transformation modeling, interface response
A thorough knowledge of solidification behavior during function models, modeling of grain structure development
welding provides the necessary solid foundation for during welding of single crystals, and application of phase
understanding and proper interpretation of material field modeling to welding. Examples of these advanced
performance. Solidification behavior directly impacts the techniques will be presented in an effort to demonstrate the
weldability of an alloy, and it controls the solidification wide range of phenomena that can be successfully modeled
microstructure, which in turn, controls the properties and with present-day techniques.
performance. Much progress has been made in the
advancement of solidification theory for castings and the same Computational Thermodynamics, the Backbone
theory is applicable to welding since welds represent small of Advanced Models
castings [1]. However, there are many unique features of
welding that must be taken into account. Welding conditions A through understanding of phase stability is essential in order
typically lead to high growth rates, cooling rates, and thermal to model microstructural development. This is especially true
gradients. Welding is also associated with vigorous fluid flow. in welding, where solidification segregation, dilution effects
Solidification often does not involve nucleation since epitaxial between filler and base metals, and dissimilar welding all lead
growth conditions prevail. Finally, solidification conditions to significant composition variations and related
(growth rate, thermal gradients, and cooling rate) vary microstructural variations, on both micro and macro scales.
significantly with position in the weld pool. For example, the This is shown in Figure 1, which is a phase field simulation of
highest thermal gradients but lowest growth rates are found at competition between ferrite and austenite solidification. Figure
the fusion line while the lowest thermal gradients and highest 1a shows the solidification of a primary dendritic phase (grey)
growth rates are present at the weld centerline. into a melt (black) and the simultaneous formation of a

Copyright © 2006 ASM International® 1

potential for forming non-equilibrium phases during
(a) solidification but also is useful when predicting the phase
stability during subsequent solid state cooling, heat treating,
and service. Both the composition and quantity of phases can
be calculated. CT software became commercially available
within the last 15 to 20 years. Over the years, the databases
that are used to describe the free energy functions of phases
have been expanded to cover more alloy systems and the
accuracy of these databases has steadily improved. Three
examples will be presented to demonstrate the wide range of
calculations that can be made based on CT.

Example 1: Simulation of Solidification under Scheil

During welding, local solute segregation effects on phase
stability can be simulated by considering the evolution of
stable phases as a function of fraction solid under so-called
Scheil solidification conditions [4,5]. An example of such a
(b) calculation for an IN718 alloy composition is shown in Figure
2. It is noteworthy that the calculation is done for an 8-
component system, and therefore is representative of the
behavior in the real alloy. Such a calculation is far more
accurate and realistic than simplified calculations based on
binary or even ternary phase diagrams. In Figure 2a, the
overall fraction solid is plotted versus temperature. Figure 2b
shows the fraction solid versus temperature for individual


Figure 1: (a) simulated solidification of ferrite (grey) and

austenite (white) into liquid (black) for an Fe-1 at % C- 1 at %
Mn alloy. (b) corresponding compositional map for carbon
showing segregation in and around the dendrites
secondary phase (white) in the interdendritic regions. The
secondary phase is stabilized by the local solute segregation
that takes place, as shown in Figure 1b. Solute concentrations
(for one component) are depicted by the intensity of the grey (b)
coloring and it is clear that significant solute segregation to the
interdendritic regions takes place during solidification. The
development of computational thermodynamics (CT) allows
one to address phase stability for multicomponent systems as a
function of temperature and therefore provides the critical
basis for understanding phase stability in weldments. Several
commercial packages are currently available for CT
calculations [3]. They all are based on the same fundamental
approach of modeling the free energy of a multicomponent
system as a function of temperature. This approach allows one
to identify phase stability under both equilibrium and non-
equilibrium conditions. With CT software, one can calculate
the extent of solute segregation as well as the phases that will Figure 2: Results from Scheil solidification analysis for alloy
appear as a result of such segregation. A thorough description IN718. (top) total fraction solid versus temperature, (bottom)
of the extent of solute segregation can be used to identify the fraction solid for each of 4 minor phases versus temperature.

minor phases and it is clear that many different phases can be secondary dendrites is ignored in such a calculation).
expected to form during the latter stages of solidification. Furthermore, by relating the compositional profile in the liquid
Many of these are non-equilibrium phases and are a direct to the phase stability of liquid versus solid, one can derive an
result of the solute segregation that takes place during estimate of the extent and spatial distribution of the liquid
solidification. In many cases, current CT software can also undercooling. This is shown in Figure 4b for the same
account for the effect of solid state diffusion. conditions shown in Figures 3 and 4a. Such calculations can
be used to quantitatively evaluate the extent of interdendritic
Example 2: Diffusion Kinetics Modeling of Solidification undercooling and could, in theory, be used to estimate dendrite
Using CT as the foundation, diffusion kinetics calculations for arm spacings. For example, if the spacing used in the
multicomponent systems can also be carried out. One such calculations is too large, then large undercoolings will be
software package that is commercially available is Dictra, found, indicative of an unstable condition that would lead to
which interfaces with the commercial ThermoCalc software the stabilization of more dendrites and to the reduction in
[6]. In such simulations, equilibrium conditions imposed at the spacing. Similarly, if the calculated extent of undercooling
interface between two phases act as the driving force for phase was small, then one could expect the dendrite arm spacings to
transformation. When applied to solidification behavior, the increase as a result of the balance between undercooling and
effects of both solid and liquid diffusion can be calculated dendrite surface energy.
directly, without any need for the simplifying assumptions
used in the Scheil solidification analysis. While these Example 3: Calculation of Ferrite Number Distribution
calculations can be made for multicomponent systems, the The last example in this section does not deal with welding but
example shown in Figure 3 considers only a simple binary Al- it demonstrates the power of CT-based kinetics calculations in
4 wt % Cu system. The solute profiles are shown as a function understanding as-solidified microstructure. The example
of distance and time. Solidification proceeds from the left into considers the distribution of ferrite in stainless steel
the liquid at the right and both the solid and liquid solute continuous castings. Experimental results revealed that the
profiles are displayed. The fact that the solute profile in the center of continuous-cast stainless steel billets had a higher
solid does not change with time indicates that diffusion in the ferrite number than the billet surface. This behavior could not
solid during solidification is negligible. However, such be explained by standard qualitative arguments. The
calculations can be extended to reveal much more than simply remaining ferrite represents as-solidified ferrite that did not
the solute profiles and whether Scheil solidification conditions transform to austenite during cooling after solidification. One
exist or not. For example, if one assumes a thermal gradient might expect that the ferrite content at the surface, where the
along the length of the primary dendrite, then one can apply cooling rates are higher, might be largest since the ability to
the results from Figure 3 to derive an approximate dendrite transform as-solidified ferrite to austenite during cooling
shape, as shown in Figure 4a (note that the formation of would be diminished. However, this is counter to the
experimental findings. Alternatively, one might expect a lower
ferrite content at the surface if the primary mode of
solidification changed from primary ferrite formation to
primary austenite formation. Once again this explanation is
not correct because the alloy in question solidified as primary
austenite at all cooling rates. Kinetics calculations were used
to explain the results. Approximate thermal histories for
time continuously cooled billets were available in the literature [7].
These thermal histories should be applicable to the continuous
castings in question and they are shown in Figure 5a. The
curves show that the cooling rate during solidification was
larger at the surface than at the billet center. However, cooling
rates in the range of 1300 to 1100°C were comparable at the
surface and center, and even were smaller at the surface over
part of the temperature range. Calculations were carried out
for the ferrite to austenite transformation using the calculated
thermal histories and setting the dendrite spacing proportional
to the inverse square root of the cooling rate during
solidification. The results of the calculation are shown in
Figure 5. Although the kinetics calculations did not predict the
final ferrite numbers at the surface and center exactly, they
clearly showed a difference between the two locations, with a
Figure 3: Concentration profiles from diffusion kinetics
higher ferrite content in the center, in agreement with
simulation of solidification in an Al-4 wt % Cu alloy as a
experimental results.
function of time. Solid grows from left into liquid on right.

(a) (b)

Distance (µm)
Distance (µm)

Distance (µm) Distance (µm)

Figure 4: (left) calculated dendrite shape corresponding to growth in Figure 3 and using an imposed thermal gradient. (right)
corresponding map of extent of undercooling, calculated by relating liquid composition profile (Figure 3) and associated liquidus
temperature to thermal gradient used in the calculation.
Interface Response Function Models the bottom right where the secondary phase (white) is shown
to be growing ahead of the primary phase (grey). This is a
Solidification theory can be used to calculate the extent of consequence of the undercooling that is present (non-uniform
undercooling at the solidification front as a function of the in the schematic figure) and the solute segregation to the
growth rate, morphology (dendritic versus cellular versus remaining liquid that results from solute partitioning during
planar growth) and the phase that is growing [5]. The results solidification. These interface response function models have
of such calculations can be used to generate interface response been applied to Fe-C-Al-Mn alloys [9]. The calculations were
functions which identify the most stable phases and compared to in-situ experimental results that identified the
morphologies as a function of growth velocity [8]. By primary solidification phase under relatively slow growth and
comparing the undercooling for different phases and high growth weld solidification conditions [9,10]. It was found
morphologies at a given growth rate, one can predict which that the model calculations could reproduce the observed
phase will solidify and in what morphological mode at that transition from primary ferrite formation to primary austenite
growth rate (phase or morphology with minimum formation at high growth rates, but only if some model
undercooling will prevail). Such calculations can be readily parameters are adjusted (Figure 6). The results show that
combined with thermodynamic calculations that provide solidification models can effectively simulate competitive
critical parameters such as the solute partitioning as a function growth among phases during welding. However the
of composition to predict the growth characteristics under high calculations also indicate that the model results can be
growth rate conditions such as those found during welding. In sensitive to the parameters used in the models and more work
this way, competitive growth between different phases can be is needed to identify reliable and accurate model parameters.
evaluated. Such competitive growth is shown in Figure 1a at




Figure 6: Calculated solidification front temperatures for

dendritic growth versus growth rate for both primary ferrite
and primary austenite formation in a low carbon steel. (a)
calculation using standard parameters showing no stable
austenite formation, contradicting experiment and (b)
calculation with modified parameters showing stable
austenitic growth at high growth rates, in agreement with in-
situ experimental observations.

Modeling of Stray Grain Formation in Single

Figure 5: (a) thermal histories at the edge and center of Crystal Welds
continuous castings (taken from reference [7]), (b and c)
calculated percent ferrite versus temperature during cooling Models have been developed that describe the evolution of
after solidification showing decrease in ferrite content due to grain structure in weldments [11-14]. These have been used to
transformation from ferrite to austenite. (b) ferrite content at explain the change in grain morphology from columnar to
center of ingot and (c) ferrite content at edge of ingot,
showing less ferrite at the edge, in agreement with experiment.
equiaxed when going from the weld fusion line to the weld
centerline. The grain morphology has a direct impact on the
weldability of some alloys; equiaxed centerline grain
structures are more resistant to center line cracking than
columnar grain structure morphologies.

More recently, much work has been directed at studying the

grain structure evolution in welds made on single crystals.
Early work on model Fe-15Cr-15Ni alloys showed that such
welds can maintain the single crystal grain structure as a result
of the epitaxial growth that prevails during weld solidification.
However, for the case of single crystal nickel-based
superalloys, retaining the single crystal grain structure is
significantly more difficult and yet, of significantly more
interest The loss of the single crystal structure in these alloys
often results in weld cracking [15,16] and is likely to lead to a
loss in elevated temperature creep resistance. Successful
retention of the single crystal grain structure during welding is
desirable because it could allow for weld repair of single
crystal turbine components. Recent work has focused on
evaluating the grain structure development during welding of
single crystal nickel-based superalloys and identifying
conditions under which successful welding could be
accomplished [17]. Such modeling requires the integration of
several models. First, a thermal model must be used to identify
the weld pool shape, and the thermal gradients that are present
at the solidification front as a function of position in the weld
pool. The results from such a model must be combined with a
geometric model that describes the active dendrite growth
directions in the single crystal as a function of solidification
front orientation and base metal crystallographic orientation
[18,19]. Finally, these two models must be combined with a
model that describes the extent of undercooling ahead of the
solidification front and the extent of nucleation and growth of
new grains that would destroy the single crystal grain structure
[13,14,20]. Results from modeling studies that combined these
three different submodels have been published recently
[13,14,17]. Representative results are shown schematically in
Figure 7, where the value of Φ, which depicts the area fraction
of new grains, is shown as a function of position in the weld
pool. The results show that Φ varies significantly with welding
conditions and position in the weld pool. Furthermore, results
for asymmetric base metal orientations show asymmetric
distributions of new grains can be expected. These results
were compared with experimental observations and they
agreed perfectly with the experimental findings [16,21,22].
This type of analysis can also identify optimum welding
conditions for avoiding new (stray) grains and results are
shown in Figure 8. The calculations indicate that high welding
speeds and low weld powers are preferred in order to avoid
stray grain formation, and these predictions have been Figure 7: Calculated tendency to form stray grains as a
supported by recent experiments [23]. function of position in the weld pool for three different weld
conditions. Light color represents low Φ (no stray grains) and
dark indicates high Φ (high probability of stray grains). (top)
low speed symmetric orientation weld, (middle) high speed
symmetric weld, (bottom) low speed asymmetric weld.

[110], (001) Orientation
Power (W)


0.3 0.1
0.5 0.45
1000 0.4 0.35 0.3 0.25 0.2 0.15

4 5 6 7 8 9 2 3 4 5 6 7 8 9 2 3 4 5
0.01 0.1
Weld Speed (ms )
Figure 8: Contour plot showing the calculated extent of stray grain formation (Φ) as a function of weld speed and power. Low Φ
indicates low area fraction of stray grains and is the desired condition. Calculations show low Φ is achieved with low power and high
Phase Field Modeling and Potential Applications analysis considered the nucleation of five ferrite grains (the
to Welding equilibrium primary solidification phase) and allowed for the
nucleation of fifteen austenite grains. The nucleation of the
In the last decade a new modeling method, phase field austenite grains was controlled by the local composition and
modeling, has been developed and is the focus of much temperature. The results at an intermediate stage of
ongoing activity. This approach can be used to describe the solidification are shown in Figure 1. The ferrite grains are grey
spatial variation of composition, grain orientation, phases, etc and austenite grains are white. The simulation over time
and has been adapted to study many complex problems. A clearly shows primary ferrite formation and the dendritic
commercial package has recently become available that microstructure evolution (including formation or
integrates the phase field approach with CT to allow for disappearance of primary and secondary ferritic dendrites).
multicomponent kinetics calculations to be carried out [24]. The solute segregation can be readily followed, for both
Phase field modeling is very computationally intensive, and substitutional (Mn) and interstitial (C) solutes. As
requires the identification of many parameters that may not be solidification proceeds, formation of secondary austenite in
known very well. Nonetheless, its ability to handle the interdendritic regions is observed. Furthermore, where the
multicomponent systems and characterize the spatial variation solidification front lags behind (far right) due the absence of
in two or three dimensions provides many unique capabilities. primary dendrites (only five were allowed), the liquid
For example, in contrast to the kinetics calculations described enrichment and higher undercooling stabilize primary
earlier, phase field modeling adds the ability to model the austenite formation. As the solidification simulation is
spatial distribution in addition to the composition and extent of continued to longer times, the primary austenite at the bottom
transformation derived in the kinetics models. As computer right is found to grow at a rapid rate and eventually overtakes
power improves, the application of phase field models will the ferrite completely, as shown in Figure 9. Thus, the
undoubtedly grow and the insight derived from these models simulation reproduces competitive growth and corresponds
will be immense. directly with the results from models using interface response
functions (described earlier). Since phase field models
Preliminary results that demonstrate the power of the phase consider the spatial distribution of phases and composition,
field modeling approach were obtained for a three component virtual composition scans across the microstructure can be
Fe-C-Mn system. Solidification was modeled under a linear readily made, as shown in Figure 10. Both the carbon and
thermal gradient (from bottom to top in Figure 1). The manganese concentration maps are shown. It is impossible to

Figure 9: Same phase field simulation shown in Figure 1 but
at a later stage of solidification showing the overtaking of Figure 10: Composition maps for (top) carbon and (bottom)
primary ferrite solidification by austenite growth. Away from manganese corresponding to the microstructure shown in
the solidification front, as-solidified ferrite has transformed to Figure 1. With limited ability to display compositions
austenite during solid state cooling, resulting in a nearly quantitatively in black and white, color scales were enhanced
100% austenitic microstructure. to dramatize the compositional variations as a function of
clearly show the variations in absolute concentrations in black position and to display the solidification segregation effects.
and white so the coloring was adjusted to enhance the changes leading to appreciable enrichment in the liquid, is shown. The
in concentration. The model results clearly show the results show the power and potential of the phase field
segregation behavior. The variation in carbon concentration modeling method in simulating the microstructural evolution
along the displayed line trace is shown more quantitatively in during welding.
Figure 11. Rejection of carbon by ferrite into the liquid,

integration with advanced heat and fluid flow models is still
needed. With these advanced models, real multicomponent
systems under realistic welding conditions can be addressed.
The use of modeling as applied to welding can be used to
identify optimum weld consumable alloys and welding
conditions that lead to microstructural control and improved
properties. While experimental research has achieved these
same goals in the past, the intensive use of models can
produce more optimum solutions in a fraction of the time and
at a fraction of the cost that has been the norm when using
experimental techniques. Combined modeling and judicial
experimentation is the ultimate means for achieving superior
weld performance.

This research was sponsored by the Division of Materials
Sciences and Engineering, U. S. Department of Energy, under
contract DE-AC05-00OR22725 with UT-Battelle, LLC.

1. S. A. David and J. M. Vitek, Correlation Between
Solidification Parameters and Weld Microstructures,
Inter. Mater. Rev. 34(5), 213-245 (1989)
2. See proceedings from earlier conferences in the
Trends in Welding series, e.g., Trends in Welding
Research, eds. S. A. David, T. DebRoy, J. C.
Lippold, H. B. Smartt and J. M. Vitek, ASM-
International, Materials Park, OH (2003) or Trends
in Welding Research, eds. J. M. Vitek, S. A. David,
T. DebRoy, J. A. Johnson, H. B. Smartt and T.
DebRoy, ASM-International, Materials Park, OH
3. See special issue of Calphad, 26(2), 143-312 (2002)
4. M.C. Flemings, Solidification Processing, p. 159,
Figure 11: (top) carbon concentration map at intermediate McGraw-Hill, New York, NY (1974)
stage of solidification (same as Figure 1 and Figure 10) and 5. W. Kurz and D.J. Fisher, Fundamentals of
(bottom) quantitative display of composition versus distance Solidification, p. 240, Trans Tech Publications Ltd.,
for the line scan shown by the grey bar in the top figure. Aedermannsdorf, Switzerland (1984)
6. J-O. Andersson, T. Helander, L. Höglund, P. Shi and
Summary B. Sundman, ThermoCalc and Dictra, Computational
Tools for Materials Science, Calphad 26(2), 273-312
Advances in solidification theory have led to the development (2002)
of computational models that can be used effectively to 7. O. J. Pereira and J. Beech, Factors Influencing the
describe the solidification behavior during welding and the Delta Ferrite Content of Cast Stainless Steels, p 315-
phase stability during subsequent cooling, heat treatment, and 321 in Solidification Technology in the Foundry and
service. The challenge that remains is to integrate existing Casthouse, Metals Society, London (1983)
models so as to better simulate the entire welding process and 8. S. Fukumoto and W. Kurz, Solidification Phase and
its inherent complexity. The examples that were presented Microstructure Selection Maps for Fe-Cr-Ni Alloys,
demonstrate the potential of model integration. Computational ISIJ International, 39, 1270-1279 (1999)
thermodynamics provides a solid framework for the 9. S. S. Babu, J. W. Elmer, J. M. Vitek and S. A. David,
application of kinetics and phase transformation models, for Time-Resolved X-Ray Diffraction Investigation of
solidification models that predict the solidification Primary Weld Solidification in Fe-C-Al-Mn Steel
morphology and competition among phases, and for Welds, Acta Mater., 50, 4763-4781 (2002)
microstructural models such as phase field models. Further

10. S. S. Babu, J. W. Elmer, S. A. David and M. A.
Quintana, In Situ Observations of Non-Equilibrium 18. M. Rappaz, S.A. David, J.M. Vitek, and L.A.
Austenite Formation during Weld Solidification of an Boatner, Development of Microstructures in Fe-
Fe-C-Al-Mn Low-Alloy Steel, Proc. R. Soc. Lond. A, 15Ni-15Cr Single Crystal Electron Beam Welds,
458, 811-821 (2002) Metall Trans A, 20A, 1125-1138 (1989)
11. T. Ganaha,, B. P. Pearce, and H. W. Kerr, Grain 19. M. Rappaz, S.A. David, J.M. Vitek, and L.A.
Structures in Aluminum Alloy GTA Welds, Metall Boatner, Analysis of Solidification Microstructures in
Trans A, 11A, 1351-1359 (1980) Fe-Ni-Cr Single Crystal Welds, Metall Trans A, 21A,
12. J.D. Hunt, Steady State Columnar and Equiaxed 1767-1782 (1990)
Growth of Dendrites and Eutectic, Mater Sci Engg, 20. M. Gäumann, S. Henry, F. Cléton, J.-D. Wagnière,
65, 75-83 (1984) and W. Kurz, Epitaxial Laser Metal Forming:
13. M. Gäumann, R. Trivedi, and W. Kurz, Nucleation Analysis of Microstructure Formation, Mater. Sci.
Ahead of the Advancing Interface in Directional Engg, A271, 232-241 (1999)
Solidification, Mater. Sci. Engg, A226-228, 763-769 21. J.M. Vitek, S.S. Babu, S.A. David, and J-W. Park,
(1997) Microstructure Development in Single Crystal Welds,
14. M. Gäumann, C. Bezençon, P. Canalis, and W. Kurz, Mater. Sci. Forum, 426-432, 4123-4128 (2003)
Single-Crystal Laser Deposition of Superalloys: 22. J.M. Vitek, S.S. Babu, J-W. Park, and S.A. David,
Processing-Microstructure Maps, Acta Mater., 49, Analysis of Stray Grain Formation in Single-Crystal
1051-1062 (2001) Nickel-Based Superalloy Welds, p 459-466 in
15. J.M. Vitek, S.A. David, and L.A. Boatner, Superalloys 2004, eds. K.A. Green, T.M. Pollock, H.
Microstructural Development in Single Crystal Harada, T.E. Howson, R.C. Reed, J.J. Schirra, and S.
Nickel-Base Superalloy Welds, Sci. Technol. Weld. Walston, TMS, Warrendale, PA (2004)
Joining, 2(3), 109-118 (1997) 23. J. M. Vitek, S. S. Babu and S. A. David, Welding of
16. J.-W. Park, S.S. Babu, J.M. Vitek, E.A. Kenik, and Single-Crystal Nickel-Based Superalloys: How to
S.A. David, Stray Grain Formation in Single Crystal Avoid Stray Grains and Why, in proceedings of this
Ni-Base Superalloy Welds, J. Appl. Phys, 94(6), conference
4203-4209 (2003) 24. MICRESS software, Access e.V., Aachen, Germany
17. J. M. Vitek, The Effect of Welding Conditions on
Stray Grain Formation in Single Crystal Welds -
Theoretical Analysis, Acta Mater., 53(1), 53-67

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Relationship Between the Formation of Hollow Bead Defects and Hydrogen

Assisted Cold Cracking
I. H. Brown, G. L. F. Powell, V. M. Linton
The University of Adelaide, Adelaide, South Australia

A. Kufner
F-H Konstanz, Konstanz, Germany

Hydrogen assisted cold cracking (HACC), also referred to as
The relationship between the occurrence of Hollow Bead delayed cracking or cold cracking, occurs in the weld metal of
Defects in root runs of pipeline welds, weld metal high strength welds. The factors leading to HACC are well
microsegregation, and hydrogen assisted cold cracking is known [3], namely the simultaneous presence of a critical
investigated. Hollow Bead Defects and hydrogen assisted cold concentration of diffusible hydrogen, a residual or applied
cracking are both found to occur in welds containing stress and a susceptible microstructure. In addition, the
segregated regions of manganese and silicon. An cracking normally occurs at a temperature below 200°C.
experimentally substantiated model for the formation of
Hollow Bead Defects is proposed. Since the major gas inside the Hollow Bead Defect is
hydrogen, it seems likely that there could be a relationship
Keywords: welding, hydrogen assisted cold cracking, hollow between the occurrence of Hollow Bead Defects and HACC.
In pipeline welding, Hollow Bead Defects are not seen as a
major defect if their length and size is within the limits
Introduction specified in the relevant Australian Standard [4]. This is a
consequence of the rounded nature of the pore. However these
Oil and gas pipelines are commonly welded using the stove- limits may need reassessment if it could be shown that the
pipe technique, a manual metal arc welding technique using conditions for the formation of Hollow Bead Defects are
cellulosic electrodes. Cellulosic electrodes provide good similar to those for the appearance of hydrogen assisted cold
penetration and high travel speeds and hence high cracking (HACC).
Hollow Bead Defect occurs in the root pass of pipeline welds
and is commonly described as an elongated linear porosity
Welded samples of two line-pipe steels, API 5L X70 (X70)
located in the root pass of a pipeline weld.
and API 5L X80 (X80) were examined. The samples
manufactured from 8.3mm thick API 5L X70 steel plate were
Cantin [1] and Barkow [2] both attempted to experimentally
supplied by Cantin. Further test plates were manufactured
determine the conditions under which Hollow Bead occurs.
from 9mm thick API 5L X80 steel plate under conditions
Cantin [1] found that the most important factor for the
likely to produce Hollow Bead. The composition of both steels
occurrence of Hollow Bead Defects was a fast weld travel
is given in Table 1.
speed combined with high welding current. The major gas in
the Hollow Bead pore is hydrogen, which has been rejected
Table 1: Chemical compositions of X70 [1] and X80 [5] plates
from the super-saturated liquid ahead of the solid-liquid
interface. On the basis of these findings Cantin postulated that
C Mn Si Ni Cr Mo Al Nb Ti
the hydrogen gas rejected by the saturated liquid ahead of the
X70 0.09 1.56 .33 .022 .017 .002 .03 .04 .012
solid-liquid interface accumulates in one single gas bubble,
which is enclosed by columnar grains growing parallel to the X80 0.09 1.7 .38 - - .035 .05 .08 0.25
welding direction. However no evidence was provided to
substantiate this mechanism.

Copyright © 2006 ASM International® 11

All welds were single pass root runs with no filling runs and When etched with LePera’s reagent, all of the transverse
were produced using cellulosic electrodes (AWS E6010/AS sections showed long columnar grains growing from the
E4110). These electrodes are commonly used for the field parent material towards the centre of the weld. However, in
welding of line pipe. Welds were produced in an automated the region of the Hollow Bead pore, the microstructure
manual metal arc welding machine designed to simulate the changed. There was a triangular region above the pore that
stovepipe welding technique using a travel speed of appeared to contain equiaxed grains. The apex of the triangle
500mm/min, a welding voltage of approximately 30V and a was on the centreline of the weld and the region was
welding current of 190A giving a heat input of 0.66kJ/mm. symmetrical either side of the pore. This feature was observed
Joints in the 200mm x 450mm steel plates were prepared to in the X70 and X80 samples. Figure 3 is an optical micrograph
the Australian Standard AS2885.2 – 2002 with a root face of where this feature has been highlighted.
1.6 – 2.1mm, a root gap of 1.3 – 1.6mm and a bevel angle of
30o. The welding was done at room temperature with no

After welding, the plates were x-rayed to locate Hollow Bead

Defects. Cross sections of those welds containing Hollow
Bead Defects were prepared using standard metallographic
techniques for examination using optical light microscopy.
The samples were examined after etching with 2% Nital, and
then repolished and etched in LePera’s reagent [5] Nital was
used to reveal the phases resulting from the solid-state
transformation, whereas LePera’s reagent was used to reveal
the microsegregation resulting from liquid to solid

The samples were further analyzed using a FEI/Phillips XL30

FESEM field emission scanning electron microscope on
etched and unetched samples, and a Cameca SX51 electron
microprobe was used on unetched samples for x-ray analysis
and x-ray mapping.
Figure 3: The microstructure at the weld centerline above a
Results Hollow Bead Pore in X70 steel: Segregation is indicated by
the darker etched regions of the cellular dendrite boundaries
Segregation at the Weld Centreline The triangular region referred to in the text is marked. The
Visibly sound welds produced using the welding parameters cellular dendrites are approximately 20μm in diameter.
described above were x-rayed, and Hollow Bead Defects were Etchant LePera’s reagent.
found in all x-rayed welds.

A transverse section of a Hollow Bead Defect is shown in A crack is evident along the centreline of the weld shown in
Figure 2. Although not discernable at this magnification a Figure 4. The crack follows a zigzag pattern along the
crack emanates from the defect towards the top surface of the boundaries of the cellular dendrites as indicated by the black
weld. arrows. The growth direction of the cellular dendrites at the
mid height of the weld is horizontal; in the upper region is
inclined slightly upwards towards the centre of the weld and in
the lower part of the weld is inclined downwards towards the
Hollow Bead pore. This indicates the change of the direction
of the heat extraction and hence the direction of solidification.

To identify the segregated elements in the weld centreline, x-

ray line scans were carried out across the crack along the
white line shown in Figure 4 using an x-ray analytical facility
attached to an SEM. The scans were made with a voltage of
20kV, a dwell time of 100 seconds per point and 128 points
per line. A typical line scan is shown in Figure 5.
Figure 2. Transverse section of Hollow Bead Defect in X70
steel with crack initiation point arrowed. Magnification x1

Manganese and silicon are clearly segregated at the weld
centreline, with manganese in particular showing a high
intensity peak. All samples which contained Hollow Bead
Defects showed similar characteristics:

Investigation of segregation surrounding a crack

Figure 4: Collage of optical micrographs extending from a

hollow bead pore to the weld surface showing cracking along
Figure: 6 SEM collage of a crack emanating from a Hollow
the segregation at the centerline of the weld. The position of
Bead Defect (bottom) and following a path through oxide
the x-ray line scan (Figure 5) is shown by the white arrowed
inclusions (arrowed). Etchant: LePera’s reagent.
line. Etchant: LePera’s reagent.
Figure 6 is a collage of micrographs taken using the back-
scattered electron mode of the SEM. The sample has been
Line 1
etched in LePera’s reagent. The figure shows that the crack
centreline has grown in a band where the material was smoother after
8 etching in LePera’s reagent than the surrounding areas. The
crack is tight, follows a staircase pattern and is branched. It
net intensity

can also be seen that the crack travels along a path running
between inclusions.
To identify the elements segregated around the crack, x-ray
maps of the crack shown in Figure 6 were collected using
0 20 40 60 80 100
electron probe microanalysis.
distance in um


Figure 5: Linescan across weld centerline showing increased

concentrations of manganese and silicon.

increased from approximately 1% to 5% with a subsequent
decrease in the iron content from approximately 99% to 95%.
Investigation of Hollow Bead Pore
Cantin [1] suggested that the growth direction around the
Hollow Bead Defects is parallel to the welding direction
(which is also the orientation of the Hollow Bead Pore). This
theory was supported by Powell et al. [6], whose research on
elongated gas pores in welds, observed cellular dendrites
growing parallel to the gas pores. To substantiate Cantin’s
hypothesis, it was considered important to further investigate
the structure surrounding the Hollow Bead pores.
Mn To investigate the solidification mode around the Hollow
Bead Defects, longitudinal sections of welds containing
Hollow Bead pores were cut along the centreline of the pores.
Figure 7: X-ray map of the cracked region using the The samples were polished and etched with LePera’s reagent
microprobe. The manganese segregation appears in the to reveal the microstructure as shown in Figure 9.
outlined area as the lighter region. (Unetched)


direction of

Black arrow indicates cellular

Hollow Bead Pore
growth direction

Si Figure 9: Microstructure of a longitudinal section of a hollow

bead pore. The pore is at the bottom left of the image and the
black arrow indicates the direction of welding. The lighter
etched regions indicate intercellular segregation. Etchant:
LePera’s reagent
Figure 8: X-ray map of the cracked region using the
microprobe. The silicon segregation appears as the lighter
region. (Unetched) The section shows that in the region to the left of the dotted
line in Figure 9 the cellular dendrites grow parallel to the
The maps (Figures 7 and 8) show that the crack follows a welding direction, which is the growth direction of the pore.
segregated region which contains considerably higher Further away i.e. to the right of the dotted line in Figure 9, the
manganese and considerably higher silicon contents than the columnar grains grow in a direction almost normal to the page
bulk of the weld metal. The segregated region corresponds to i.e. almost at 90 degrees to the growth direction of the pore.
the smooth region in Figure 6.
To substantiate these findings the inside surface of Hollow
The nominal concentrations of silicon, manganese and iron Bead pores were closely examined. Cross-section samples of
were confirmed using x-ray line scans across the segregated welds containing Hollow Bead Defects were
area containing the crack shown in Figure 6. These confirmed metallographically ground to the mid height of the pore, and
the results shown in the area scans Figures 7 and 8. The scans the inside surface of the pores examined in the SEM.
were collected at approximately the mid length of the crack. In
the segregated region the silicon content increased from On the internal surface of the pore, protrusions which were
approximately 0.2% to 0.5%. and the manganese content aligned in the longitudinal direction of the pore and the
welding direction, were observed. These protrusions were also

evident on the bottom surface of the pore. The protrusions do cellular dendrites growing in the direction of the Hollow Bead
not run absolutely parallel to the elongated direction of the pore
pore, but run slightly towards the centre of the pore. The
distance between the protrusions is approximately 20μm,
which corresponds to the diameter of the cellular dendrites as weld centre-line
shown in Figure 3.
Segregation surrounding a crack
A hydrogen assisted cold crack was found in the root pass of a
weld in X70 steel. The crack was initiated at a Hollow Bead
Defect. X-ray line and area scans revealed that the crack parent metal parent metal
occurred in a region where the manganese and silicon contents
were higher than in the surrounding weld metal. The localised
increase in manganese and silicon content was a result of
segregation at the cellular dendrite boundaries during the bottom
solidification of the weld metal [7]. This increase in the alloy
content of these two elements increased the hardenability of Hollow Bead Defect
these regions which increased their susceptibility to HACC

Scanning electron micrographs showed the crack path ran

between oxide inclusions. This is expected because these
inclusions would also be segregated to the cellular dendrite Figure 10: Schematic of a transverse cross section of a weld
boundaries during solidification. It has previously been containing Hollow Bead Defect in the root pass.
reported [8] that HACC follows these inclusions and that these
oxide inclusions not only act as stress raisers but also as
Examination of the inside of the Hollow Bead pores, revealed
hydrogen traps, i.e. the residence time of diffusible hydrogen
protrusions on the inside surface. The distance between the
is longer at the inclusion than in other parts of the
protrusions was approximately 20μm, which is approximately
microstructure. This leads to an increase in triaxial stresses
the diameter of the cellular dendrites. These protrusions were
around the inclusion which further increases the susceptibility
almost parallel to the welding direction. It was shown by
to cracking.
Powell [7] that these protrusions are regions of segregation at
the cellular dendrite boundaries as a consequence of the lower
Segregation associated with Hollow Bead Defect
liquidus temperature of the enriched liquid. The protrusions
With the chosen welding parameters, the production of
point towards the centre of the pore. A model for the
Hollow Bead Defects in the welds was very reproducible.
formation of Hollow Bead based on these findings, and
Every weld contained several Hollow Bead Defects, generally
supported by the hypothesis suggested by Cantin [1] is
in the bottom part of the weld
presented in Figure 11.
The occurrence of Hollow Bead Defects corresponded with
Cantin [1] showed that the gas inside Hollow Bead Defects is
with the weld growth pattern shown in Figure 3.
almost 100% hydrogen. The solutibility of hydrogen in the
liquid phase of steel is higher than in the solid phase.
The solidification mode of the welds is as described by Savage
Hydrogen was rejected from the supersaturated liquid phase
[9]. A transverse section of the solidification morphology is
ahead of the solid-liquid interface as the temperature
shown schematically in Figure 10. The cellular dendrites grow
decreased. The hydrogen accumulated in the form of hydrogen
from the parent metal towards the centre of the weld, where
bubbles. These bubbles were forced towards the outer surface
they meet. The growth direction at the mid-height position of
of the weld by the advancing solidification front, where they
the weld is horizontal, whereas in the top section of the weld it
would normally escape. However, due to the high weld travel
is slightly upwards, and in the bottom section of the weld it is
speeds and the high cooling rates involved in the present
slightly downwards. In the welds with Hollow Bead Defects, a
experiments, a layer of solid metal rapidly forms at the surface
triangular shaped area just above or below the Hollow Bead,
of the weld and the hydrogen cannot escape. The bubbles
and generally towards the middle of the weld, was observed.
become encapsulated in a thin layer of the last metal to
In this area, the grains appeared to be equiaxed, but it could be
solidify, metal that is enriched in alloying elements and at the
shown by taking a longitudinal section through the Hollow
centreline of the weld bead.
Bead and etching in LePera’s reagent, that they were in fact

solid-liquid interface growth direction of the cellular dendrites from perpendicular
to the welding direction towards the welding direction and
also produce a line of segregation from the bottom to the top
of the weld. In addition it was found that in the samples
welding direction containing the Hollow Bead Defect the cellular dendrites
surrounding the Hollow Bead Defect and the Hollow Bead
Defect itself grew parallel to the welding direction. When
welded at slower travel speeds and lower currents no change
in growth direction or Hollow Bead Defect were detected.
rejected hydrogen
top A crack initiated at the surface of a Hollow Bead Defect and
last region to solidify
travelling to the weld surface was investigated. The crack
followed the segregated regions at the centreline of the weld.
The segregated regions were higher in manganese and silicon
indicating that these regions would have higher hardenability.
film of solidified metal on weld surface Therefore all of the factors necessary for the formation of a
Hollow Bead Defect cold crack were present, hydrogen, residual stress due to
solidification and regions of increased hardenability. The
morphology of the crack reflected cold cracks previously
investigated and reported [7] in that it was tight, branched and
Figure 11: Transverse section of a schematic model for the
linked inclusions
formation of Hollow Bead Defects.
The Hollow Bead pore propagates in the direction of the 1. G. M. D. Cantin, An Investigation of the Formation
solidification front, i.e. the welding direction as more and of Hollow Bead Defects in Pipeline Field Welds, PhD
more diffusible hydrogen is rejected from the liquid metal. Thesis, 1998
2. A. G. Barkow, New Welding Problem for Pipeliners,
The cellular dendrites next to the Hollow Bead pore are The Oil and Gas Journal, 71 40-47 (1973)
observed to be relatively large compared to the epitaxial grains 3. P .H .M. Hart,., Resistance to Hydrogen Cracking in
growing from the parent metal. These grains were probably Steel Weld Metals. Welding Journal, 14-22 (1986)
nucleated at the skin around the gas pore ahead of the epitaxial 4. AS2885.2, Pipelines - Gas and liquid petroleum. Part
growth solidification front. They are located in the last metal 2, Welding. Standards Australia, 2002
to solidify at which stage the heat transfer is relatively slow, 5. F. S. LePera, J. Met., 32, 38-39 (1980)
giving the grains sufficient time to develop their size. 6. G. L. F. Powell, and P. G. Lloyd, Characterisation of
an Elongated Gas Pore in a Weld in Terms of
In welding tests, conducted with the low travel speed, cellular Solidification Mechanics. Prakt. Metallogr., 1995. 32:
dendrites growing parallel to the welding direction were not p. 25-31
found. This suggests that growth in the direction of welding 7. I. H. Brown, G. L. F. Powell, J. L. Davidson, V. M.
requires a high travel speed with a corresponding high cooling Linton, Cold Cracking and Segregation in Multi-pass
rate. None of the specimens taken from the low weld travel Welds of a Quenched and Tempered Steel, 6th Int.
speed welds (300mm/minute) showed either cellular dendrites Conf. on Trends in Welding Research, Georgia USA,
parallel to the welding direction or Hollow Bead Defects. The 2002.
cellular dendrites grew from the parent material towards the 8. Bhadeshia H., Microstructure modelling in weld
weld centreline. This was the case from the top of the weld to metal, Mathematical modelling of weld phenomena
the bottom of the weld in all of samples (Figure 12). 3, H. Cejrak ed., 650, 1997, 229 - 282.
9. Savage, W.F., 1980 Houdremont Lecture. Welding in
Conclusion the world, 18(5/6), 89-114, 1980

From the results of the present research it has been possible to

postulate a model for the formation of the Hollow Bead
Defect. Previous research on Hollow Bead Defects has
concentrated on the conditions for its formation rather than the
mechanism of formation [1][2]. In agreement with that
previous work, this work found that Hollow Bead Defects
formed at higher weld travel speeds and higher welding
currents. These welding conditions produce a change in the

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Achieving Grain Refinement through Weld Pool Oscillation

T.J. McInerney, R.B. Madigan
Montana Tech of the Univ. of Montana, Butte, Montana, USA

P. Xu, C.E. Cross

Federal Institute for Materials Research and Testing, Berlin, Germany

Abstract The intention of this study was to investigate different

combinations of frequency and amplitude using mechanical
Autogenous, bead-on-plate, gas tungsten arc welds made on oscillation, and to identify regimes capable of achieving weld
aluminum 2219 plate have been mechanically oscillated metal grain refinement through pool oscillation. Aluminum
during welding in order to achieve grain refinement in the alloy 2219 was used in this study because of its availability
weld metal. Various combinations of frequency and and similar composition to alloy 2519 used in the AAAV.
amplitude of oscillation have been examined to define the Also of interest was the identification of natural frequencies
conditions necessary for breaking up the columnar grain for aluminum welds, with the possibility of achieving grain
structure. A critical energy has been identified as being refinement through pool agitation at its natural frequency.
required for grain refinement, expressed in terms of the
product: frequency x amplitude (fa). The possibility for grain Background
refinement by exciting the weld pool at its natural frequency
has also been considered. Outlined below are several models for grain refinement
pertinent to this study, taken from the literature. Also
Introduction considered here are predictions for natural frequencies in
Weld metal grains are typically columnar, nucleating aluminum weld pools.
epitaxially on base metal grains, and growing in a continuous
manner normal to the weld pool interface. The presence of Grain Refinement
long, continuous grain boundaries is deleterious to both Grain refinement in solidification represents a columnar-to-
toughness and weldability. equiaxed transition (CET); i.e. an interuption in the continuous
advancement of a columnar grain by the nucleation of new
In welded aluminum alloys, grain refinement is normally equiaxed grains ahead of it. In order to effectively nucleate
controlled with TiAl3 grain refiner additions to the filler alloy new grains ahead of a columnar solid/liquid interface requires
[1]. However, in cases where impact toughness is of both undercooling and a suitable substrate. How this might be
particular importance and the presence of intermetallic achieved in weld solidification is discussed below.
compounds must be minimized, an alternative method for
grain refinement becomes desirable. Undercooling. Accounting for both partitioning and surface
tension effects, the undercooling ahead of a dendritic
One such application involves welds made on aluminum alloy solidification front can be shown to take the following form
2519 used in the construction of the US Navy’s Advanced [3], expressed in terms of growth rate R and temperature
Amphibious Assault Vehicle (AAAV), where weld ballistic gradient G :
properties have been identified as a potential problem [2]. GD
Greater resistance to cracking from ballistic impact could be ΔT = + AR1 / 2 (1)
achieved through improved weld metal toughness associated R
with grain refinement. Use of a low copper filler alloy would
also be desirable (i.e. fewer intermetallic compounds), in , where D is solute diffusivity in the liquid, and A is a material
which case grain refinement would be needed for weldability. constant. While the first term in this expression can usually be
neglected for most casting applications, weld solidification
poses a special boundary condition wherein the solidification

Copyright © 2006 ASM International® 17

rate approaches zero along the weld pool boundary, making Table 1: Comparison of physical properties for liquid metals
this term a major contributor to undercooling [4]. near their respective melting point taken from [17].
γ (N/m) ρ (x103Kg/m3) γ/ρ
Considering the second term in Equation 1, undercooling also Al 0.91 2.4 0.38
becomes high along weld centerline where R becomes large as Fe 1.87 7.0 0.27
it approaches the torch travel speed (Rmax). Thus, one might
normally expect to find grain refinement along the fusion line
and at the weld center, with columnar grains in between [5]. The natural oscillation frequency of a molten weld pool (f) has
In practice, equiaxed grains at the weld center are common [6, been considered to be similar to that of a molten droplet [14],
7], but they are seldom observed along the fusion line except which is proportional to mass (M), volume (V), and physical
in certain alloys (e.g. aluminum alloys containing zirconium properties surface tension (γ) and density (ρ) :
dispersoids, ref. 4).

γ γ
Applying this knowledge to an oscillated weld pool requires f =N =N (2)
an understanding of how the solidification front advances with M ρV
time. If you take a steadily advancing weld pool (i.e. constant
torch travel speed), and impose a periodic oscillation motion
, where N is a constant related to the mode of oscillation. Use
in the direction of travel, Rmax can be expected to vary between
is made of this relationship in order to predict the natural
two extreams: Rmax¯ and Rmax+. In the case where remelting
frequency of an aluminum weld, based upon published values
occurs, it follows that Rmax¯ is negative, and thus Rmax must at for low alloy steel welds. Using the properties given in Table
some point pass through zero. When this happens, the 1:
undercooling will momentarily become very high (from
Equation 1) giving the possibility for grain refinement. Such
f Al (γ / ρ ) Al 0.38
is the case with banding in weld metal, where the non-uniform = = = 1 .2 (3)
advancement of the weld pool has been shown to result in both f Fe (γ / ρ ) Fe 0.27
macro-segregation and grain refinement [8].
Thus, a comparable sized aluminum weld should have a
Substrate. The most effective substrate for the nucleation of a natural frequency 1.2 times larger than its steel counterpart
new grain is the grain material itself [9], fulfiling desirable when oscillating in the same mode. It is understood, however,
attributes of wettability and similar chemical and that these high temperature properties are not known to a high
crystallographic structure. Mechanisms that allow for this level of certainty, which leaves the exact value of this factor in
scenario include peritectic reactions, dendrite fragmentation, question.
and grain detachment [7, 9-11].
Natural frequencies reported for partial penetration mild steel
Of particular interest to this study is the possibility for welds are compared below, derived from data published for
dendrite fragmentation or bending, caused by shear forces specific modes [18], and expressed as a function of pool
associated with the periodic sloshing motion of the weld pool. diameter (D) :
A comprehensive review of this subject for castings by
Campbell [12] has demonstrated this mechanism to be a −3 / 2
tangible possiblity for grain refinement. In particular, Mode 1 (peak-valley mode): f = 2,030 D (4)
−3 / 2
Campbell examined the work involved in fragmentation or Mode 2 (slosh mode): f = 967 D (5)
bending, expressed in terms of the product: frequency x
amplitude (fa). The derivation for this relationship is
, where f is in herz and D is in mm. The D-3/2 dependence
presented in the Appendix.
follows from the volume term in Equation 2, assuming a
hemispherical shaped weld pool. Equations 4 and 5 represent
Natural Frequency linear regressions for data with considerable scatter (±10 Hz).
Use has been made of natural frequency measurements to Thus, a 6 mm diameter steel weld should oscillate at
control weld penetration in steel weldments [13-16]. Control approximately 138 herz for mode 1, and 66 herz for mode 2.
methodologies have involved sensing a change in natural
Based on Equation 3, corresponding frequencies for an
frequency, representing a change in oscillation mode in going aluminum weld would be 166 herz for mode 1, and 79 herz for
from partial to full penetration. The question arises, can use mode 2. Viewed from another approach, it is interesting to
also be made of this natural frequency to agitate the weld pool
note that a numerical analysis of slosh dynamics, for liquids in
into grain refinement. This appears to be a little studied topic a moving container, predicts a natural frequency of 52 Hz for
for welds. Also, most studies regarding weld pool natural
a 6 mm diameter, half-filled spherical vessel [19], which is
frequency characterization have been limited to steel welds.
somewhat close to the previous prediction.

Experimental directly at the arc, attaching leads to the electrode collet holder
and to the ground cable. Voltage signals were stepped down
Material and protected against welding current surges before being sent
The material used in this study was aluminum Alloy 2219- to an analog-digital converter and computer processing. A
T87, an aluminum-copper binary alloy with a nominal Fast Fourier Transform (FFT) program was used to identify
composition range: 5.8 to 6.8 wt.% Cu. Coupons (110 mm periodic fluctuations in the arc voltage.
wide x 125 mm long x 6 mm thick) were cut from 6 mm rolled
plate. Welds were performed in the rolling (long) direction, Metallography
with two welds per coupon. Surface preparation for welding The distinction between columnar and equiaxed grains in the
included oxide removal with a plastic abrasive pad, followed weld metal was determined by means of optical
by acetone degreasing. metallography. The complete elimination of any columnar
structure was taken to represent grain refinement.
Welding Process Metallographic specimens were sectioned both transverse and
An autogenous, bead-on-plate, gas tungsten arc weld was longitudinal to the welding direction and prepared by grinding
made on aluminum 2219 plate using a constant-current power and polishing to 1 μm alumina, followed by immersion in
supply with direct current and negative electrode polarity Keller’s double etch.
(DCEN). The welding torch was mounted to a motorized
carriage, permitting controlled advancement of the torch at a Natural Frequency Measurement
constant travel speed. Welding parameters, given in Table 2, Two different methods of pool excitation were examined:
were held constant, producing a partial penetration weld of hammer blow and current pulsation. In the first method, a
approximately 8 mm width and 4 mm depth. The weld shape single hammer blow was applied to a plate where a stationary
was found to vary somewhat with oscillating conditions. weld pool was being made. This was done on both aluminum
2219 and 2519 plates, utilizing the parameters in Table 2.
Table 2: Welding parameters held constant. Plate thickness was 38 mm for 2219, and 16 mm for 2519.
Welding Current 140 amps
Travel Speed 4.2 mm/s In the second method, a current pulse was periodically applied
(5 Hz) to a moving weld pool. Gas tungsten arc welding
Tungsten Electrode 2%La, 2.4 mm Ø, 30° tip
parameters were: 200 A (20 ms), 60 (180 ms), 2 mm arc gap,
Arc Gap 2 mm
4.2 mm/s, and helium shielding gas. This was done on 4 mm
Shielding Gas 28 L/min, Helium
thick aluminum 6060 plate to produce partial penetration
welds of approximate 6 mm diameter. The voltage signals
Mechanical Oscillation resulting from both methods of excitation were FFT processed
A mechanical oscillation test devise was constructed wherein to identify recurring frequencies.
the weld coupon was positioned between two opposing audio
speakers, each rated at 900 watts. Displacement of the coupon
was achieved through fixed contact with the speaker Results and Discussion
transducers. A sinusoidal wave was applied using a function
generator and dual channel amplifier, 600 watts per channel. Arc Gap-Arc Voltage Calibration
The calibration curve for arc voltage, expressed in terms of the
Frequency and amplitude could be varied independently over stepped down voltage signal, is shown in Figure 1.
a range of 25 to 2500 Hz with a maximum displacement of 11
mm. By phase shifting signals to each speaker, they could be
operated in a coordinated push-pull mode, with controlled
displacement of the coupon in the direction of welding. A
total of 5 different frequencies were examined: 25, 38, 50, 63,
and 75 Hz. The amplitude of oscillation was varied at each
frequency, in order to identify conditions over which grain
refinement occurs.

Oscillation Measurement
The frequency and amplitude of pool oscillation was
monitored by means of measuring small fluctuations in the arc
voltage. This possibility arises due to the changes in arc gap
associated with pool oscillation, together with a linear
relationship between arc gap and arc voltage at constant
current power. The correlation between arc gap and arc Figure 1: Calibration curve for voltage signal (proportional to arc
voltage) and arc gap.
voltage was accurately calibrated. Arc voltage was measured

It is apparent that a linear relation exists over the range of arc
voltage and arc gap examined. Use was made of this linear
relationship to calculate oscillation amplitudes in the
frequency-amplitude analysis that follows.

Frequency-Amplitude Analysis
A comparison is made in Figure 2 of the various frequencies
and amplitudes investigated. Each of the peaks located at 25,
38, 50, 63, and 75 Hz represents one test that was evaluated
for grain refinement using metallography. Some harmonic
peaks are observed at higher frequencies. Amplitude is
displayed in this figure in terms of both voltage signal and
calculated displacement (in mm), based upon Figure 1.

Figure 3: A log-log plot of frequency versus amplitude for aluminum

2219 weld pool oscillation demonstrating that the critical conditions
for grain refinement follow the relation: fa=constant, and fall within a
range of critical values observed for castings [12]. Shown are data
points for three different frequencies: 25, 50 and 75 Hz.
Combinations of frequency and amplitude below the line fa=19 Hz-
mm do not result in grain refinement.

Figure 2: Graphic comparison of oscillation conditions examined.

Columnar grain structure is absent in welds made above the
superimposed line.

A line has been superimposed on the peaks of Figure 2

demarking the boundary between those welds with a columnar
grain structure and those without. Welds made with an
amplitude above this line were observed to be grain refined. It
is clear that at lower frequencies, higher amplitudes are
required for grain refinement. Although the CET boundary
has not been defined precisely at any given frequency, a
narrow amplitude range has been defined (between peaks)
within which grain refinement is known to occur.

When this critical range for grain refinement is plotted on a

graph of log frequency versus log amplitude, as presented in
Figure 3, it is observed that the data falls on a line with a slope
of negative 1. This implies that the product fa is constant,
which suggests that a critical energy is required for grain
refinement (see Appendix). This energy could be taken to
represent the work required to either fragment or bend
dendrites in the sloshing motion of the weld pool. When the
oscillation parameters lie below this line, there is insufficient
energy to achieve grain refinement.

Also observed in Figure 3, the critical line for aluminum

welding (fa=19 Hz-mm) falls within the bounds reported by Figure 4: Arc voltage oscillation signals resulting from the hammer
blow method of pool excitation for two different aluminum plates: a)
Campbell [12] to represent conditions appropriate for grain
38 mm thick Alloy 2219 and b) 16 mm thick Alloy 2519.
refinement in the casting of numerous different materials,
based upon an extensive literature review.

Natural Frequency
Results from the hammer blow experiment are presented in
Figures 4 and 5, showing both voltage signals and FFT
analyses, respectively, for stationary 2219 and 2519 aluminum
welds. A natural frequency is indicated at approximately 83
Hz. The size of the weld pool was not accurately determined.
Part of the problem with this method is that the pool size
increases with time as the plate heats. Also, there is the
possibility that the results may be affected by the oscillation of
the welding table.

Figure 6: Arc voltage trace of moving weld pool with current

pulsation showing regions of pool excitation and FFT analysis.

Figure 5: FFT analysis of oscillation voltage signals generated

in hammer blow experiments presented in Figure 4. The peak
at approximately 83 Hz shows this to be a recurring frequency.

Results from the current pulsing experiment are given in

Figures 6 and 7 showing a typical voltage trace and FFT
analysis, respectively. A natural frequency is indicated at 76
Hz for a continuous weld made on 6060 aluminum,
approximately 6 mm in diameter. This corresponds closely
with predictions based upon data available for the slosh mode
in steel welds as discussed earlier. Examination of aluminum
pool movements using a high speed video camera also suggest Figure 7: FFT analysis of voltage oscillations in Figure 6 showing
a slosh mode of oscillation. strong recurring frequency at 76 Hz.


It has been demonstrated that grain refinement can be

achieved in aluminum 2219 alloy weld metal using
mechanical oscillation applied in the direction of welding.
Furthermore, it appears that grain refinement can be achieved
at any frequency (within the range examined), provided the The authors wish to acknowledge the support of the US Office
amplitude is sufficiently high, as defined by fa=19 Hz-mm. of Naval Research (ONR) and the Bundesanstalt für
This suggests that a critical energy is required for grain Materialforschung und –prüfung (BAM) for providing funding
refinement, giving credence to a dendrite fragmentation or for this research. The authors are also grateful to AMET Inc.,
bending mechanism associated with the sloshing motion of the Rexburg, Idaho, for providing equipment for voltage signal
weld pool. conditioning.

A natural frequency has been experimentally identified for

aluminum welds, occurring at approximately 80 Hz in the
slosh mode for a 6 mm diameter weld pool. This value
corresponds well with theoretical predictions.

References Pulsed GTA Welding”, Welding J., 77, 181s-187s
1. M.G. Mousavi, C.E. Cross and Ø. Grong, “Effect of 17. T. Iida and R.I.L. Guthrie, The Physical Properties of
Scandium and Titanium-Boron on Grain Refinement Liquid Metals, Clarendon Press, Oxford (1993)
and Hat Cracking of Aluminium Alloy 7108”, Sci. 18. Y.H. Xiao and G. den Ouden, “Weld Pool Oscillation
Tech. Weld. Join., (1999) During GTA Welding of Mild Steel”, Welding J., 72,
2. D.J. Gutscher, “Effects of Iron and Copper on the 428s-434s (1993)
Mirostructure and Ballistic Properties of Aluminum 19. F.T. Dodge, “Dynamic Behavior of Liquids in
Alloy 2519-T87 Weldments”, MSc Thesis, Montana Moving Containers”, Internal Report, Southwest
Tech, Butte, Montana (2003) Research Institute, San Antonio, p. 17 (2000)
3. M.H. Burden and J.D. Hunt, “Cellular and Dendritic
Growth II”, J. Crystal Growth, 22, 109-116 (1974)
4. C.E. Cross, Ø. Grong and M. Mousavi, “A Model for
Equiaxed Grain Formation Along the Weld Metal Appendix
Fusion Line”, Scripta Materialia, 40, 1139-1144
5. Ø. Grong and C.E. Cross, “A Model for Predicting The following derivation comes from Campbell [12], which
Weld Metal Grain Refinement in G-V Space”, Mat. shows the relationship between work W and the oscillation
Res. Symp. Proc., 578, 431-438 (2000) frequency f and amplitude a. For periodic, sinusoidal
6. T. Ganaha, B.P. Pearce and H.W. Kerr, “Grain oscillation of angular frequency ω, the acceleration s is given
Structures in Aluminum Alloy GTA Welds”, Met. by:
Trans., 11A, 1351-1359 (1980) s = −ω 2 y (6)
7. S. Kou and Y. Le, “Nucleation Mechanisms and
Grain Refining of Weld Metal”, Welding J., 65, 305s- , where y represents oscillation displacement, which varies
313s (1986) between ±a. Making use of the definition of angular
8. A.T. D’Annessa, “Characteristic Redistribution of frequency ω=2πf, it follows that force F can be expressed in
Solute in Fusion Welding”, Welding J., 45, 569s-576s terms of mass m and acceleration s:
9. L.F. Mondolfo, “Grain Refinement in the Casting of
Non-Ferrous Alloys”, in Grain Refinement in F = ms = −mω 2 y = − m(4π 2 f 2 ) y (7)
Castings and Welds, TMS-AIME, 3-50 (1982)
10. F.A. Crossley and L.F. Mondolfo, “Mechanism of Considering the work done in ¼ cycle (e.g. as the weld pool
Grain Refinement in Aluminum Alloys”, J. Metals, sloshes back towards the solidification front):
191, 1143-1148 (1951)
11. M.G. Mousavi, M.J.M. Hermans, I.M. Richardson 0 0

and G. den Ouden, “Grain Refinement due to Grain W = ∫ Fdy = −4π mf 2 2

∫ ydy = 2π
m( fa ) 2 (8)
Detachment in Electromagnetically Stirred AA7020 a a
Welds”, Sci. Tech. Weld. Join., 8, 309-312 (2003)
12. J. Campbell, “Effects of Vibration During Thus, it is observed that a constant value of fa represents a
Solidification”, Int. Met. Rev., 26, 71-108 (1981) constant amount of work performed. Equation 8 can be
13. R.B. Madigan, R.J. Renwick, D.F. Farson and R.W. applied to the line in Figure 3 (fa = 19 Hz-mm) to relate this to
Richardson, “Computer Based Control of Full a quantity of work:
Penetration GTA Welds Using Pool Oscillation
Sensing”, in Proc. Int. Conf. Computer Tech.
Welding, The Welding Institute, Cambridge, 165-174
W = 2π 2Vρ (19 x10 −3 m / s ) 2 ≅ 2.3x10−6 Nm (9)
14. G.E. Cook, R.J. Barnett, A.M. Strauss and K. , where V=1.34x10-7m3 is the volume of an 8 mm diameter
Andersen, “Penetration Control for Gas Tungsten Arc hemispherical weld pool of density ρ=2.4x103Κg/m3.
Welding”, in Modeling and Control of Joining
Processes, AWS, 19-26 (1993)
15. Q.L. Wang, C.L. Yang and Z. Geng, “Separately
Excited Resonance Phenomenon of the Weld Pool
and its Application”, Welding J., 72, 455s-462s
16. A.J.R. Aendenroomer and G. den Ouden, “Weld Pool
Oscillation as a Tool for Penetration Sensing During

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Modeling and Analysis of a New Solidification Cracking Test

L. Zhu, S. Bachani, R. Nordstrom, M. V. Li, and J. Devletian
Portland State University, Portland, Oregon, USA

Abstract increased cracking susceptibility in the low carbon range of

0.03-0.05%C and in a narrow range band close to 0.1%C. A
Ejig is a new solidification cracking test. It is similar to the new solidification cracking test was developed and used in
Sigmajig test in that the specimen is loaded in tension to a their study. The testing procedure has been described in detail
preset value before welding. The difference is that the by Bachani and Devletian.[5] The test is similar to the
displacement at the ends of the Ejig test specimen is fixed Sigmajig test originally developed by Goodwin of Oak Ridge
during welding whereas the stress in the Sigmajig test National Laboratory. In the Sigmajig test, a predetermined
specimen is maintained during welding. stress/force is applied to the weld coupon during welding. In
this new test, a predetermined strain/displacement is applied to
Ejig tests on AISI 1018 steel specimens are analyzed in this the weld coupon. Sigma is the Greek symbol of stress. This
study using a three-dimensional finite element model and a new test is called Ejig test where E denotes engineering strain.
fully coupled and integrated thermal-metallurgical-mechanical
analysis procedure. The objective of this study is to Like many other laboratory weldability tests, Ejig test is useful
understand the metallurgical and mechanical conditions of in studying the characteristics of solidification cracking and in
solidification cracking. With the proper consideration of evaluating compositional, metallurgical, and mechanical
melting, solidification, and property changes in the weld, this effects on solidification cracking. However, it is extremly
study proves it feasible to predict the thermal, metallurgical, difficult, if not impossible, to reliably apply the laboratory
and mechanical conditions for solidification cracking to occur. weldability test results to actual fabrication problems. The
The length of centerline solidification cracking is found to be difference primarily lies at the mechanical driving forces
proportional to the mechanical strain at the terminal stage of between laboratory test welds and actual field welds. There is
the weld metal solidifiction. a lack of quantitative understanding of the metallurgical and
mechanical interactions in the terminal stage of the weld metal
Introduction solidification.

Solidification cracking is one of the most detrimental defects This study analyzes the mechanical response of the weld metal
in welding manufacturing and fabrication. It occurs at the during the Ejig test using advanced finite element analysis
terminal stage of weld metal solidification. Metallurgically, techniques. The objectives of the study are to better
solidification cracking is attributed to solidification conditions understand the thermal-mechanical conditions associated with
and temperature range, solidification grain structure and sizes, solidification cracking and to aid the interpretation of
solute segregation, and presence of low melting films. experimental results.
Mechanically, solidification cracking is attributed to the stress
buildup due to metal shrinkage in the terminal stage of Materials and Experiments
solidification and slow recovery of ductility.
Steel plates in this study are AISI 1018. The chemistry of the
Susceptibility of metals to solidification cracking has been steel plates is presented in Table 1.
often determined from the “brittle temperature range” using
longitutinal varestraint and transverse varenstraint tests,[1] Table 1: Chemical composition of steel AISI 1018 (wt%).
Houldcroft test,[2] and Sigmajig test.[3] Recently, Shankar
and Devletian[4] studied the unique solidification cracking C Mn Si Ni Cr Cu S P
mechanism in low carbon steel weld metal. They observed 0.18 0.69 0.22 0.08 0.15 0.27 0.014 0.039

Copyright © 2006 ASM International® 23

Test specimens are of the dimension of 6.4×50×150 mm marked. The finite element model is shown in Figure 4. It
between two hardened round pins. Run-on and run-off tabs are consists of 8,248 linear brick elements and 10,314 nodes with
used. The test set-up is illustrated in Figure 1. Prior to both temperature and displacement degrees of freedom.
welding, a prescribed transverse stress is applied to the weld
coupon through two pins. Then the two pins are fixed to allow Table 2: Test conditions and crack lengths.
no displacement in the transverse direction. An autogenous
gas tungsten arc weld is made along the centerline of the Arc Travel Crack
Current Preload
specimen, i.e. the direction perpendicular to the preset load. A voltage speed length
(A) (kN)
transducer attached to the fixture is used to record the (V) (mm/s) (mm)
mechanical response (load variations) of the coupon to the Case I 25 213 2.5 36 0
welding. A photo of the actual experimental setup is shown in Case II 25 220 2.5 32 7.64
Figure 2. Case III 25 213 2.5 41 25.4

Figure 1: Illustration of Ejig test coupon and procedure.

Figure 3: Schematic illustration of the FEA model

Figure 2: Experimental setup of Ejig test.

To understand the mechanical response of the material during

the Ejig test, three cases are selected for modeling and 3

analysis. These three cases represent no crack, small crack, Figure 4: Finite element model of Ejig test
and large crack. The welding parameters and preload levels
for these three cases are in Table 2. Compared to the no-crack The hardened pins, run-on and run-off tabs are included in the
Case I, Case II has a lower preload but slightly higher heat model. A pin hole is created with its center at point ‘O’. Nodes
input which leads to a small crack; Case III has the same heat at the top and bottom surfaces of the test specimen around the
input but a higher preload, which leads to a large crack. hole are constrained in the Z (thickness) direction. A rigid
revolved surface is used to simulate the pin. The rigid surface
Modeling and Analysis is fixed in both X (welding direction) and Z (thickness)
directions. A contact pair between the rigid surface and the
Finite Element Model inside surface of the hole is created to simulate the surface
Commercial finite element code ABAQUS/Standard 6.4.1 is interaction. The rigid surface is prescribed with a displacement
used to simulate the Ejig test. This study utilizes a three- in the Y (transverse) direction to produce stress in the test
dimensional finite element model to represent the test coupon. specimen to a preset value. Computed reaction force at the
Due to the symmetry in both the shape of the specimen and reference point ‘O’ of the rigid surface during the welding
the loading condition with respect to the weld centerline, only simulation is saved in a result file. Variations of reaction force
half of the specimen is considered. Figure 3 illustrates the from finite element analyses are compared with the
geometric representation of the model with points of interest experimentally measured load trace using a transducer.

Modeling Procedure Thermal Physical Properties
A fully coupled thermal-metallurgical-mechanical analysis Transient heat transfer analysis requires the values of mass
procedure is applied to simulate the Ejig test. The coupled density, thermal conductivity, specific heat, and latent heat of
analysis procedure leverages the coupled temperature- fusion and solid state phase transformations. Density is needed
displacement analysis solver capability in finite element code in conjunction of specific heat to compute the internal energy.
ABAQUS/Standard. As illustrated in Figure 5, the coupling Since the computation of internal energy always refers to the
between heat transfer and metallurgical analyses is achieved original volume, room temperature value of the density is
with user subroutine UMATHT, which defines the thermal required.
constitutive behavior of the material. Microstructural
constituents are treated as solution dependent state variables Thermal conductivity of steels are dependent not only on
that are computed and updated at each time increment in the temperature but also on microstructure. Figure 6 presents the
UMATHT subroutine. Enthalpy change caused by phase reported thermal conductivity values of AISI 1018 steel[11].
transformations is considered in the thermal solutions. The During cooling, austenite may exist in a mestastable state until
metallurgical models are based on the previous work by decomposition reactions are completed. Note that the reported
Li[6,7]. thermal conductivity values at temperatures above the Ac3
temperature appear to be linear. Thus in this study, the thermal
conductivity value of austenite is linearly extrapolated to
lower temperature and linear rule of mixture is used to
obtained the thermal conductivity of a mixed microstructure
that may consist of austenite, ferrite, pearlite, bainite, and

Thermal Conductivity (W/m.°C) 140

Stable Microstructure
Unstable Austenite



Figure 5: Coupled Analysis of Welding.
Thermal strain and strain induced by the volumetric changes
due to phase transformations are non-mechanical strains and 0
are computed in user subroutine UEXPAN. Rate components 0 400 800 1200 1600 2000
of mechanical strains are determined in user subroutine
UMAT Temperature (°C)
ε& mech = ε& el + ε& pl + ε& tp (1)
Figure 6: Thermal conductivity of AISI 1018 steel.
where ε& mech
, ε& , ε& and ε& are the rates of mechanical
el pl tp

strain, elastic strain, plastic strain and transformation Temperature-dependent specific heat values are used in this
plasticity. The elasticity model is based on classic theory of study. The dependency of specific heat on microstructure is
isotropic elasticity. Plasticity model is based on the von Mises known to be weak when no phease transformation occurs.
yield criteria and isotropic hardening. Transformation Specific heat values for AISI 1018 steel are obtained from the
plasticity model was based on the work of Leblond[8,9] reported data in Ref. 12.
considering the Greenwood-Johnson mechanism[10].
Temperature dependent elastic modulus and Poisson’s ratio
Heat Source are used in this study. The temperature dependency of yield
A moving heat source with prescribed power density is used to strength of AISI 1018 is shown in Figure 7 based on the data
simulate the heat input from the gas tungsten arc. The model is published in Ref. 13.
in accordance with the double ellipsoidal heat source model
proposed by Goldak et al.[11] This heat source model is Melting and Solidification
implemented in the finite element analysis using user Upon heating, the melting is assumed to occur closely
subroutine DFLUX in ABAQUS/Standard. following the conditions of phase equilibrium. Phase changes
during heating normally occur much faster than those during

cooling due to higher diffusivity and the additional driving cracking during solidification of the remaining 10% liquid in
force provided by the reduction of interface energies. The the interdendrtic regions. This is considered in the user
solidification of weld metal is a non-equilibrium process. It material subroutine UMAT.
can be simulated using the Scheil equation coupled with
computational thermodynamics models and databases. Results and Discussion
Commerical software JMatPro is used to compute the enthalpy
change of AISI 1018 steel during heating with melting Mechanical load variations during Ejig tests were measured
included and during cooling with solidification included. The using a transducer attached to a hardened pin. The measured
computed results are shown in Figure 8. Latent heat of fusion load traces were compared with the reaction force predicted
and enthalpy change due to solid state phase transformations from finite element analyses. Figure 9 shows the comparison
are obtained from the calculated enthalpy results. for Case III in which cracking occurred. A rapid drop of the
mechanical load is observed in the experimental curve. In all
600 cases, predicted reaction forces match very well with the
experimental measurements until cracking occurs. Since the
Yield fracture constitutive model is not yet developed, the current
Tensile model is unable to predict the rapid change of reaction force
due to cracking. Nevertheless, it is useful to compare the
Strength (MPa)

mechanical stress and strain in the weld when cracking occurs
300 to understand the conditions of solidification cracking.

200 50
100 Simulation

Reaction Force (kN)

0 200 400 600 800 1000 30
Temperature (°C)

Figure 7: Mechanical properties of AISI 1018.


Heating 10
Cooling 0 5 10 15 20 25 30
1200 Time (sec)
Ehnthalpy (J/g)

1000 Figure 9: Experimental load variation and predicted reaction

force in Case III.
600 Welding in the Ejig test is conducted along the width direction
of the test specimen at a travel speed of 2.5 mm/s. The test
specimen is 50 mm wide. At 17 sec, the welding arc has just
200 passed the middle of the specimen. The predicted temperature
0 distribution for Case III is presented in Figure 10(a).
Temperatures in the red region exceed the liquidus of the
800 1000 1200 1400 1600
material. This region thus represents the weld pool. Transverse
Temperature (°C) stress (S22) at this moment is shown in Figure 10(b).
Transverse plastic strain (PE22) at this moment is shown in
Figure 8: Enthalpy change during heating and cooling.
Figure 10(c). As the weld cools down to room temperature,
Mechanically, all plastic strains accumulated in the material predicted transverse residual stress is shown in Figure 11,
which is consistent to existing knowledge on residual stress in
are annihilated once melting occurs. Upon melting, elastic
modulus is substantially reduced and the Poisson’s ratio
approaches the limiting value of 0.5. Computation of plastic
Martensitic transformation introduces increased yield strength,
strain resumes when 90% fraction of solid is reached. At this
volumetric expansion, and transformation plasticity, thus is
point, the primary dendrites are well developed that prevent
the liquid metal in the bulk to flow into the interdendritic expected to have marked effect on the residual stress and
strain in the weld. However, martensitic transformation should
regions. Material is capable to carry limited stress but has little
ductility, which gives rises to the propensity of solidification not affect the cracking in the Ejig test. When the welding arc

moves out of the test plate, the temperature at the starting edge propogate to the top surface. All cracks observed in the
of the test plate is still above the MS temperature. Moreover, experimental welds are centerline cracks. Metallographic
AISI 1018 is not very hardenable. Only a small amount of examinations revealed that cracks indeed initiated under the
martensite is formed in the weld HAZ. The effect of the surface. Some cracks did not propagate to the surface.
martensitic transformation on residual stress and strain is not
fully manifested. 400
30 sec
630 sec

Transverse Stress S22 (MPa)




(a) temperature
0 10 20 30 40 50
Welding distance X (mm)

Figure 11: Transverse stress along the weld centerline.

Transverse Plastic Strain (PE22) 0.035

(b) transverse stress 0.030



Case III
Case II
0.005 Case I

0 1 2 3 4 5 6
(c) transverse plastic strain Thickness (mm)

Figure 10: Temperature, transver stress and transverse plastic Figure 12: Transverse plastic strain through plate thickess
strain during welding (17 sec) (path C-J-F at 30 sec).

It is commonly agreed that it is the solidification cracking is At the time the welding arc extinguishes (30 sec), transverse
attributed to the mechanical strain at the terminal stage of the plastic strains along the welding centerline (path A-C-D-B in
solidification.[14] Mechanical strain consists of elastic strain, Figure 3) for three cases are presented in Figure 13. Top views
plastic strain, and transformation induced plastic strain. Note of the experimental welds are shown in Figure 14. It is clear
that the material trailing the weld pool picks up plastic strain that the maximum plastic strains occur in the first 10 to 20 mm
rapidly underneath the weld bead surface. of the weld length and underneath the surface. A wider
cracking openning is observed at this location. Crack lengths
During the Ejig test, the welding arc starts at the run-on tab, appear proportional to the magnitude of the transverse plastic
travels through the width of the test specimen, finishes on the strain.
run-off tab in about 30 seconds. Transverse plastic strains at
this time instance through the thickness along the path C-J-F Conclusions
in Figure 3 are presented in Figure 12 for all the three cases.
Note that maximum plastic strains occur underneath the plate A three-dimensional finite element model and a fully coupled
surface. The transverse plastic strain distribution profiles and integrated thermal-metallurgical-mechanical analysis
suggest solidification cracks initiate under the surface and procedure have been developed to simulate the Ejig test for

the evaluation of solidification cracking. Changes of material References
properties due to metallurgical changes, especially the melting
and solidification, are properly accounted for. The simulated 1. C. D. Lundin and W.F. Savage. The Varestraint Test.
results show a good agreement with the experiments, Welding Journal, 44:433–442, (1965)
indicating that solidification cracking initiated underneath the 2. N. Bailey and S. B. Jones, Solidification Cracking of
surface and the crack length is proportional to the transverse Ferritic Steel during Submerged Arc Welding,
plastic strain. Welding Journal, 57:217-231 (1978).
3. G. M. Goodwin, Development of a New Hot-
0.025 Cracking Test — the Sigmajig, Welding Journal, 66,
Case III 33–38 (1987)
Transverse Plastic Strain (PE22)

Case II 4. V. Shankar and J. H. Devletian, Solidification

Cracking in Low Alloy Steel Welds, Science and
Case I
Technology of Welding and Joining, 10:236-243
0.015 (2005).
5. S. Bachani and J. H. Devletian: Use of Acoustic
Emission to Detect Solidification Cracking in Steel,
46th Annual Meeting of International Acoustic
Emission Working Group, Portland Oregon,
0.005 August 5, 2003.
6. M. Li, Computational Modeling of Heat Transfer and
0.000 Microstructure Development in the Electroslag
0 10 20 30 40 50
Cladding Heat Affected Zone of Low Alloy Steel,
Weld Distance (mm)
Ph.D. Dissertation, Oregon Graduate Institute of
Science and Technology, 1996.
Figure 13: Transverse plastic strain along the weld centerline 7. M. V. Li, D. V. Niebuhr, L. L. Meekisho, and D. G.
(path A-C-D-B at 30 sec) Atteridge, A Computational Model for the Prediction
of Steel Hardenability, Metallurgical and Materials
Transactions B, 29B, 661-672, 1998.
8. J. B. Leblond, J. Devaux and J. C. Devaux.
Mathematical Modeling of Transformation Plasticity
in Steels I: Case of Ideal-Plastic Phases, International
Journal of Plasticity, 5:551–572, (1989).
9 J. Leblond, G. Mottet, J. Devaux and J. C. Devaux,
Mathematical Models of Anisothermal Phase
(a) Case I Transformations in Steels, and Predicted Plastic
Behaviour, Materials Science and Technology,
1:815-822 (1984).
10. G. W. Greenwood and R. W. Johnson, The
Deformation of Metals under Small Stresses during
Phase Transformations, Proceedings of the Royal
Society, A283: 403-421 (1965)
11. J. Goldak, A. Chakravarti, and M. Bibby, A New
Finite Element Model for Welding Heat Source,
Metallurgical Transactions B, 15B:299–305 (1984)
12. British Iron and Steel Research Association, Physical
(b) Case II Constants of Some Commercial Steels at Elevated
Temperatures, Butterworths Scientific Publications,
London, 1963.
13. F. J. Clauss, Engineer’s Guide to High-Temperature
Materials, Addison-Wesley, 1969.
14. Z. Feng. A computational analysis of thermal and
mechanical conditions for weld metal solidification
cracking. Welding in the world, 33:34–41, 1994.

(c) Case III

Figure 14: Topview of experimental welds.

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Proposal of Independent Two Phase Growth during Solidification

in Austenitic Stainless Steels
Steel Research Laboratories, Nippon Steel Corporation, Japan

Department of Materials Engineering, The University of Tokyo, Japan

Abstract Experimental procedures

Solidification morphologies of austenitic stainless steel weld The material used in the present study was an austenitic
metals that solidified with primary ferrite were investigated in stainless steel containing approximately 19wt%Cr and
terms of crystallography. At the fusion boundaries, austenite 11wt%Ni. The contents of the solute elements other than Cr and
grows first with plane-front morphology from the base metal Ni were reduced as low as possible.
austenite in an epitaxial manner. Then, ferrite forms on the In the present study, the welding process was adopted to
growing austenite by keeping the Kurdjumov-Sachs orientation examine solidification behaviors to confirm the change in
relationship with the austenite. The ferrite grows, as the primary crystallographic orientation relationships between the primary
phase, more rapidly with dendritic morphology than the planar phase and the secondary phase by changing the solidification
austenite. Though the phase diagram indicates that the growth direction determined by the movement of heat source.
formation of the austenite results from the eutectic reaction in Autogenous welding was performed using a gas tungsten arc
the primary ferrite solidification mode, no specific orientation (GTA) welding process at a current of 150A and a voltage of
relationship was confirmed by crystallographic studies between 12V with a travel speed of 1.67mm/sec. To examine the
the primary ferrite and the interdendritic austenite. The microstructural change during weld solidification, the liquid tin
austenite is found to grow independently, growing along <100> quenching method14) was used to quench the solidification front.
direction, even when the primary ferrite changes its growth The identification of phases and their crystallographic
direction. Consequently, it is suggested that the austenite in the orientations were performed using an SEM equipped with an
interdendritic regions is not crystallographically restricted by electron back scattering pattern (EBSP) analyzer15).
the primary ferrite during the growth. The growth manner of the
primary ferrite and secondary austenite is named as Results and discussion
“independent two-phase growth”, and was confirmed not only
in weld metals but in cast metals. Microstructural change during solidification
Figure 1. shows the microstructure around solidification front
Key Words in the weld metal obtained by liquid tin quenching method.
solidification; independent two phase growth; welding; casting; Ferrite dendrites can be distinctly observed within a region of
dendrite; nucleation; austenitic stainless steel; primary ferrite; around 50μm from solidification front. At the ferrite dendrite
crystallographic orientation relationship boundaries, austenite is solidified in succession to retained
liquid phase. This result indicates that the solidification mode
Introduction of the material used in the present study is FA mode.
Most austenitic stainless steel weld and cast metals are
designed to solidify to give primary ferrite and secondary
austenite to minimize the occurrence of hot cracks. This
solidification mode is known as ferritic-austenitic solidification
mode (FA mode)1-3). In the case of this solidification mode, the
phase stability and/or the phase selection were mainly
discussed4-10), but solidification morphologies of each phase is
still uncertain11-13). In the present study, solidification behaviors
of austenitic stainless steels that solidified with primary ferrite Fig.1 Microstructural change during solidification in the
were investigated from the viewpoint of the crystallography of weld metal obtained by liquid tin quenching method.
the ferrite and the austenite.
Figure 2 shows the front area of austenite, which is solidified
as the secondary phase in Fig.1. At the interdendritic region of

Copyright © 2006 ASM International® 29

the preceding ferrite dendrites, the cellular austenite is grows along the [100] γ preferential growth direction of
solidified and the tip of austenite overhangs forward the austenite. Therefore, the [100]δ direction of ferrite is found to
solid-liquid interface with curvature. Though the phase diagram be almost parallel to the [100] γ direction of austenite.
indicates that the formation of the austenite of the present steel Austenite, however, has two crystallographic orientations (γ1,
results from the eutectic reaction in the primary ferrite γ2) : these two orientations are obtained by rotating the crystal
solidification mode, neither lamellar nor rod-like eutectic of structure around the [100]γ direction. This result indicates that
ferrite and austenite is confirmed but the only cellular austenite no additional specific orientation relationships are identified at
single phase is observed. This result suggests that the austenite the interface between the primary ferrite and the interdendritic
is formed as divorced eutectic16). austenite.
In the case of this FA solidification mode, it was suspected
δ that the austenite at the dendrite boundaries of the preceding
ferrite dendrites was formed as divorced eutectic from the
viewpoint of the microstructural change during solidification
Liquid γ shown in Fig.2. It was reported that because the secondary
phase nucleates on the primary phase as divorced eutectic16),
the specific orientation relationship needs to be established
δ between the two phases17). However, there is no specific
orientation relationship between the ferrite and the austenite in
Liquid γ the present study.
Weld metal first forms from HAZ (Heat Affected Zone) of
base metals. Figure 4 shows the EBSP analysis results near the
Fig.2 Growth front of the interdendritic austenite
during primary ferrite solidification. fusion boundaries in the weld metal obtained by liquid tin
quenching method. The upper side in the micrograph is weld
Crystallographic orientation relationship between ferrite metal and the lower side is base metal (HAZ). The surrounded
and austenite during solidification region including ferrite by dotted line corresponds to the
Figure 3 shows the crystallographic orientation relationship primary ferrite formed during solidification. All of the austenite
between primary ferrite and interdendritic austenite (measured in both base metal (points 1-2) and weld metal (points 3-12) has
by EBSP) near the solidification front in the weld metal an identical crystallographic orientation, and therefore it is
quenched during solidification. The upward direction in the confirmed that the austenite formed at the fusion boundaries in
micrograph corresponds to the solidification growth direction. the weld metal results from epitaxial growth with plane-front
All the primary ferrites at dendrite cores (points 1-12) have an morphology from the austenite of base metal. Then, ferrite is
identical crystallographic orientation, showing growth along the formed on this planar austenite. Namely, the formation of ferrite
[100]δ direction, the preferential growth direction of ferrite. is somewhat delayed and occurs away from the fusion
The austenite located at dendrite boundaries (points 13-27) also boundaries. All of these ferrite (points 13-16), however, has
different crystallographic orientation, respectively, furthermore,
the parallel relationship between the <100> direction of ferrite
and the <100> direction of austenite is not established, unlike
the crystallographic orientation relationship near the
solidification front shown in Fig.3. However, between the
ferrite and the austenite, the Kurdjumov-Sachs (K-S)
orientation relationship18) is present (0-11)δ15//(-1-11)γ, [111]
δ 15//[011] γ :(1-10) δ 16//(1-11) γ ,[111] δ 16//[011] γ ) or close
packed planes19) are found to be parallel ((011)δ13//(-111)γ
,(-101) δ 14//(-1-11) γ ). These results indicate that the
crystallographic orientation relationship between the ferrite and
the austenite can not be determined even within one weld metal,
and therefore the relationship near the fusion boundaries is
different from that near the solidification front. The fact that the
K-S relationship or the parallel relationship between close
packed plane is established between the ferrite and the austenite
at fusion boundaries indicates that the ferrite is nucleated on the
planar austenite with coherency, and it is necessary for the
ferrite to keep the crystallographic correlation with the austenite
for its formation. On the other hand, the growth of austenite at
Fig.3 EBSP analysis results for the primary ferrite and the dendrite boundaries as a secondary phase is epitaxial from the
interdendritic austenite near the solidification front. base metal. It is suspected that austenite is not crystallo-
graphically restricted by the primary ferrite during the growth.

to this result, however, it must be kept in mind that the
crystallographic orientation of the austenite does not change
even if the crystallographic orientation of the ferrite changes. If
the secondary austenite in the interdendritic region is formed as
divorced eutectic, the specific orientation relationship is
established between the primary ferrite and the secondary
austenite, and therefore the crystallographic orientation of
austenite must be changed with the formation of the new ferrite.
However, the change in the crystallographic orientation of the
austenite can not be ascertained as to the result of Fig.5.

Fig.5 EBSP analysis results for ferrite and austenite.

Fig.4 EBSP analysis results for ferrite and austenite Note that the ferrite changes its growth direction
formed near the fusion boundary. in the middle of the area examined.

During formation of weld metals, the solidification growth Figure 6 shows the EBSP analysis results for the primary
direction is changed in turn by the movement of the weld heat ferrite and the interdendritic austenite at the region away from
source. Figure 5 shows the EBSP analysis results at the region the fusion boundaries in the weld metal cooled to room
where the ferrite changes its growth direction in the weld metal. temperature. The upward direction in the micrograph
Though Fig.5 is the analysis results of the weld metal cooled to corresponds to the heat flow direction. All the ferrite (points
room temperature, the crystallographic orientation relationship 1-4) has an identical crystallographic orientation, and the [100]
during solidification can be analyzed because the orientation δ direction of the ferrite corresponds to the upward direction of
relationship established during solidification is retained on the micrograph. On the other hand, the austenite (points 5-13)
cooling to room temperature12). The right-upward direction in has two different crystallographic orientations (γ1: points 5-9,
the micrograph corresponds to the heat flow direction. All the γ2: points 10-13), and the austenite grain boundary is seen in
austenite (points 1-18) has identical crystallographic orientation. the middle of the micrograph (indicated by dotted line). The
It is confirmed that this region is one austenite grain. Though <100>γ1 direction of the lower side austenite (γ1) does not
all the ferrites (points 19-28) are remaining within one austenite agree with the heat flow direction, but the <100>γ2 direction of
grain, two different crystallographic orientations (δ1: points the upper side austenite (γ2) approaches the upward direction
19-23, δ 2: points 24-28) corresponded to their growth of the micrograph and gets near to parallel with the [100]δ
directions are present. It is considered that the growth of the direction of the ferrite. It is considered that the lower side
prior ferrite ( δ 1) stopped because of the large deviation austenite (γ1) is weeded out by the upper side austenite (γ2),
between its preferential growth direction and the heat flow whose preferential growth direction is closer to the heat flow
direction determined by the movement of the heat source, and direction than that of γ1, because of the large deviation
then, the new ferrite (δ2) nucleated instead of the prior ferrite between the preferential growth direction of γ1 and the heat
(δ1), or the secondary or the tertiary dendrite arm of the flow direction. This result is the inverse of the result shown in
another ferrite (δ2) grew instead of the prior ferrite (δ1). As Fig.5 and indicates that the austenite is displaced regardless of

the growth of the primary ferrite. Therefore, the ferrite with manner, which two phases grow dependently while keeping a
identical crystallographic orientation grows across two more specific crystallographic orientation relationship. Consequently,
austenite grains with different crystallographic orientations. it is suggested that another solidification manner should exist.

Solidification manner in the austenitic stainless steel

solidified as FA mode
The solidification manner of the ferrite and the austenite in
the austenitic stainless steel solidified as FA mode can be
summarized in a schematic illustration shown in Figure 7.

Fig.6 EBSP analysis results for ferrite and austenite. Note

that austenite grain boundary is seen in the middle
of the area examined while ferrite goes through the
area without any change in growth direction.

The crystallographic characteristics in the austenitic stainless

steel weld metals solidified as FA solidification mode obtained
in the present study are summarized as follows:
a. At the solidification front, the <100> directions of primary
ferrite and interdendritic austenite are parallel along the
solidification growth direction, but the specific orientation
relationships do not exist at the interface between the
ferrite and the austenite.
b. At the fusion boundaries, the austenite in the weld metal
grows from the base metal austenite in an epitaxial
c. At the fusion boundaries, the parallel relationship between Fig.7 Schematic illustration on the formation and the
the <100> direction of the ferrite and that of the austenite growth of ferrite and austenite near the fusion
boundary of weld metal solidified in FA mode.
is rare, but the K-S relationship or the parallel relationship
between close packed planes is established between the When the base metal is fully austenite, the growth of
ferrite and the austenite. austenite is more favorable than the nucleation of ferrite
d. The ferrite with different crystallographic orientation is because of no nucleation barrier of austenite at the fusion
formed within one austenite grain with identical orienta- boundaries, and therefore austenite first grows epitaxially from
tion. the base metal austenite with plane-front morphology. During
e. The ferrite with identical crystallographic orientation this planar austenite solidification, Cr is rejected into the liquid,
grows across more than one austenite grains with different which increases the stability of ferrite and causes the nucleation
orientations. of ferrite on the growing planar austenite by keeping the
Eutectic reaction, peritectic reaction etc. are known as solidi- favorably coherent crystallographic orientation relationship
fication manner of most alloying metals. These reactions are with the austenite. Once the ferrite forms, it grows more rapidly
provided with specific crystallographic orientation relationships as the primary phase with dendritic morphology to dominate
between two phases20,21). In other words, two phases grow over the planar austenite growth. And the only ferrite whose
dependently each other not only as to composition but as to crystallographic preferential growth direction is aligned with
crystallographic orientation. It was considered that the austenite the heat flow direction continues to grow. When the preferential
as a secondary phase in the weld metal of the stainless steel growth direction of ferrite deviates significantly from the heat
used in the present study was formed as outwardly divorced flow direction, its growth stops, whereas new ferrites nucleate
eutectic. However, the above (a-e) crystallographic character- successively and the ferrite with the preferential growth
istics can not be explained by the conventional solidification direction nearly along the heat flow direction among them can

grow. On the other hand, in the interdendritic region of the intact columnar morphology even if titanium and nitrogen are
primary ferrite at the final stage of solidification, Ni is rejected added. The similar phenomenon was suggested in the weld
into the liquid, which increases the stability of austenite and metal of type 321 stainless steel26).
causes the formation of austenite. Austenite, however, grows If the secondary austenite forms in relation to the
more easily from the austenite which has already solidified than crystallographic orientation of the primary ferrite after the
the nucleation of austenite on the preceding ferrite or in the equiaxed solidification of the primary ferrite, each equiaxed
liquid, and therefore the formation of the austenite at the ferrite should be surrounded by the respective austenite.
dendrite boundaries is invariably epitaxial growth and fills the However, the equiaxed solidification of ferrite and the
interdendritic region of the primary ferrite. There is no columnar solidification of austenite occur simultaneously. Thus,
inevitability of keeping the favorably coherent crystallographic the phenomena, shown in Fig.8, can be explained only by the
orientation relationship between the austenite and the ferrite, “Independent Two-Phase Growth” manner.
and the interdendritic austenite is not crystallographically
restricted by the preceding ferrite during the growth.
Consequently, the austenite grows independently, growing
along the preferential growth direction, even when the primary
ferrite changes its growth direction. And the austenite whose
preferential growth direction is nearly aligned with the heat
flow direction can grow as forming columnar grain. As the
ferrite and the austenite independently repeat the competitive
growth respectively, the parallel relationship between the
<100> δ direction of ferrite and the <100> γ direction of
austenite along the heat flow direction is finally established at
the solidification front, but no specific orientation relationship
exists between ferrite and austenite during solidification.
The crystallographic characteristics obtained in the austenitic
stainless steel weld metals solidified as FA mode can be
explained by the following manner. At only the nucleation stage
of new ferrite, the specific crystallographic orientation
relationship is established between ferrite and austenite.
However, the following growth of austenite at the dendrite
boundaries of the primary ferrite is invariably epitaxial and is
not crystallographically restricted by the preceding ferrite
during the growth. This means that ferrite and austenite grow (a) No addition (b) Ti and N addition
independently, and therefore the authors name the growth
manner as “Independent Two-Phase Growth”22).
Fig.8 Microstructures and macrostructures of austenitic
stainless steel cast ingots,
Verification of “independent two-phase growth” manner
(a) no addition, (b) Ti and N addition.
To verify the “Independent Two-Phase growth” manner
proposed, the following experiment was performed. It is well
known that titanium nitride (TiN) is effective for the ferrite In the austenitic stainless steels with similar compositions to
nucleus and makes equiaxed solidification of ferrite promote in the present study, eutectic structures were confirmed at the
ferritic stainless steel23-25). Though the material used in the lower solidification velocity (10-6m/sec) by directionally
present study is an austenitic stainless steel, the primary solidification method and the K-S orientation relationship
solidification phase is the ferrite because of FA solidification between the eutectic two phases was confirmed9,10,13). On the
mode. Consequently, even in the austenitic stainless steel other hand, the solidification velocity in the present study is
solidified as FA mode, it is expected that the primary ferrite approximately from 10-4m/sec to 10-3m/sec. As the solidi-
solidifies as an equiaxed morphology by TiN. Titanium and fication velocity increase, the solidification morphologies of
nitrogen are added to the austenitic stainless steel with the same ferrite is shifted from eutectic growth to cellular or dendritic
compositions of the present study’s steel. Figure 8 shows the growth9,10), even if the chemical compositions are identical. It is
microstructures and macrostructures of this austenitic stainless considered that the “Independent Two-Phase Growth” manner
steel cast ingot. The fine ferrite is dispersed and TiN is proposed in this study becomes more feasible at the
observed at the center of the fine ferrite. The equiaxed solidification conditions of practical solidification processes,
solidification of the primary ferrite caused by TiN is confirmed. such as welding, continuous casting etc., in which the primary
However, in the macrostructure of the same cast ingot, a coarse phase grows as cellular or dendritic morphology.
grain; namely, the columnar grain of austenite is observed. It is Most previous investigations for solidification behaviors
found that the only primary ferrite is solidified as equiaxed were performed by directionally solidification method. Because
morphology but the secondary austenite is solidified as an the heat flow direction in the directionally solidification method

is not changed, a crystal continues to grow while maintaining independently, and this growth manner is named as
the crystallographic orientation relationship established at the “Independent Two-Phase Growth”.
stage of nucleation, and therefore the specific crystallographic (4) The coexistence of equiaxed solidification of ferrite and the
orientation relationship is more likely to exist. In other words, columnar solidification of austenite in TiN dispersed
the crystallographic orientation relationship obtained by the stainless steel can be explained only by the “Independent
directionally solidification method is the orientation relation- Two-Phase Growth” mechanism mentioned above.
ship at the nucleation stage, but does not always indicate the (5) The “Independent Two-Phase Growth” manner could be
orientation relationship during growth. On the other hand, the clarified by the solidification process that can change the
results obtained in the present study can indicate the orientation growth orientation of solid (i.e., the heat flow direction).
relationship between two phases during growth. This relation-
ship could be clarified by the examinations using the solidi- References
fication process which the crystallographic orientation (the heat
flow direction) changes. Furthermore, the results obtained in 1) N.Suutala, T.Takalo and T.Moisio:Metall.Trans.A,11A,718
the present study are important for materials engineering (1980)
because the crystallographic orientation (the heat flow 2) J.A.Brooks, J.C.Williams and A.W.Thompson: Metall.
direction) changes in many practical solidification processes. Trans.A,14A,1271 (1983)
3) S.A.David:Weld.J.,60, 63s (1981)
Conclusions 4) M.Bobadilla and G.Lesoult: Mem.Sci.Rev.Met., 78, 345
Solidification morphologies of austenitic stainless steels 5) Th.Schubert, W.Loser, S.Schinnerling and I.Bacher:Mater.
solidified as FA solidification mode were investigated in terms Sci.and Tech.,11, 181(1995)
of crystallography. The results obtained in the present study are 6) H.Mizukami, T.Suzuki, T.Umeda and W.Kurz: Mater.Sci.
as follows: Eng.,A173, 361 (1993)
(1) In the austenitic stainless steel weld metals solidified as FA 7) T,Koseki and M.C.Flemings: Metall.Trans.A, 27A, 3226
mode, the cellular austenite as a secondary phase is formed (1996)
at the dendrite boundaries of the preceding primary ferrite 8) T.Koseki and M.C.Flemings: Metall.Trans.A, 28A, 2385
during solidification. (1997)
(2) The following crystallographic characteristics (a-e) were 9) S.Fukumoto and W.Kurz: ISIJ Int.,39, 1270 (1999)
observed in the weld metals solidified as FA mode. 10) T.Okane and T.Umeda: ISIJ Int.,38, 454 (1998)
a. At the solidification front, the <100> directions of primary 11) H.Inoue, T.Koseki, S.Ohkita and M.Fuji:Quarterly J.Jpn.
ferrite and interdendritic austenite are parallel along the Weld.Soc.,15, 88 (1997)
solidification growth direction, but the specific orientation 12) H.Inoue, T.Koseki, S.Ohkita and M.Fuji: Sci .and Tech. of
relationships do not exist at the interface between the Weld. and Join., 5, 385 (2000)
ferrite and the austenite. 13) S.Fukumoto, T.Okane, T.Umeda and W.Kurz: ISIJ Int.,40,
b. At the fusion boundaries, the austenite in the weld metal 677 (2000)
grows from the base metal austenite in an epitaxial 14) H.Inoue and T.Ogawa: Quarterly J.Jpn.Weld.Soc.,9, 129
manner. (1991)
c. At the fusion boundaries, the parallel relationship between 15) D.J.Dingley and V.Randle: J.Mater.Sci.,27, 4545 (1992)
the <100> direction of the ferrite and that of the austenite 16) B.Chalmers: Principles of Solidification, John Wiley &
is rare, but the K-S relationship or the parallel relationship Sons Inc., New York,218 (1964)
between close packed planes is established between the 17) S.Wang, T.Akatsu, Y.Tanabe and E.Yasuda: J.Mater.Sci.,
ferrite and the austenite. 35, 2757 (2000)
d. The ferrites with different crystallographic orientations are 18) G.Kurdjumov and G.Sachs: Zeitschrift fur Physic.,64, 325
formed within one same austenite grain. (1930)
e. The ferrite with identical crystallographic orientation 19) K.Ameyama, T.Maki and H.Tamura: J.Jpn.Inst.Metals,50,
grows across more than one austenite grains with different 602 (1986)
orientations. 20) L.M.Hogan, R.W.Kraft and F.D.Lemkey: Advances in
(3) In austenitic stainless steel weld metals solidified as FA Materials Research vol.5,83, John Wiley & Sons Inc.,New
mode, ferrite and austenite do not grow dependently in York, 83 (1971)
terms of crystallographic orientation during the 21) R.W.Kraft: Trans.Met.Soc.AIME,224, 65 (1962)
solidification. Only at the nucleation stage of new ferrite 22) H.Inoue and T.Koseki: Tetsu-to-Hagane,87, 692 (2001)
on austenite, the specific crystallographic orientation 23) T.Koseki and H.Inoue: J.Jpn.Inst.Metals, 65, 644 (2001)
relationship is established between the ferrite and the 24) J.C.Villafuerte, E.Pardo and H.W.Kerr: Metall.Trans.A,
austenite. The following growth of the austenite in the 21A, 2009 (1990)
dendrite boundaries is invariably epitaxial and would not 25) A.Ostrowski and E.W.Langer: J.Metall,8, 153 (1979)
be restricted crystallographically by the preceding ferrite 26) J.C.Villafuerte and H.W.Kerr: Metall.Trans.A, 21A, 979
during the growth. Thus, the ferrite and the austenite grow (1990)

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

A Transport Phenomena Based Model to Prevent Liquation Cracking in

Aluminum Alloy Welds
S. Mishra, S. Chakraborty1 and T. DebRoy
The Pennsylvania State University, University Park, PA, USA
Currently with the Indian Institute of Technology, Kharagpur, India

Abstract Furthermore, both the thermodynamics and the kinetics of

solidification affect the solute partitioning during
A numerical model has been developed for non-equilibrium solidification. Both the partitioning of the solute and the
solidification in welds that considers momentum, heat and mixing of the filler metal with the base metal need to be
solute transport. The model uses an effective partition considered to understand the solidification process during
coefficient, which considers both the local interface velocity welding. However, many previous attempts to understand
and the undercooling for accurate prediction of solute weld pool solidification considered thermal field alone [5,6]
concentration in the mushy zone. The calculations show that and ignored the convective solute transport in the weld pool.
convection plays a dominant role in the solute transport Chakraborty and Dutta [7] developed a solidification model
inside the weld pool. The predicted weld metal solute for studying heat and mass transfer in a single-pass laser
content agreed well with the independent experimental surface alloying process. However, they [7] assumed
observations. The liquation cracking susceptibility in Al-Cu equilibrium at the solid-liquid interface that may not be
alloy weldments could be reliably predicted by the model, attained when the interface speed is comparable with or
based on the computed solidifying weld metal composition faster than the diffusion speed. Both the velocity of the
and the resulting solid fraction considering non-equilibrium solidification front and the undercooling must be considered
solidification. to accurately represent solidification during welding. The
complex coupling of momentum, heat and solute transport
Keywords: Liquation cracking, solidification, transport under non-equilibrium conditions during fusion welding
phenomena have not been investigated. Such an approach is desirable
for accurate prediction of the evolution of the solute
Introduction concentration and improved understanding of the weldment
solidification structure.
Liquation cracking, also known as edge-of-weld cracking,
base metal cracking, hot cracking and heat-affected zone The goal of the research presented here is to understand the
(HAZ) cracking, occurs in the HAZ when a low melting solute concentration field and the solidification process,
point region, i.e., partially melted zone (PMZ), is formed considering filler metal addition, fluid flow and solute
during welding. Cracks form when the PMZ cannot transport in the weld pool and non-equilibrium
withstand the tensile stresses generated during solidification solidification. The model predictions of liquation cracking
[1]. The occurrence of liquation cracking in aluminum susceptibility in various aluminum alloy weldments based
alloys have been confirmed by experiments [1-4]. Huang on Huang and Kou [1] criterion are compared with the
and Kou [1] found that the partially melted zone (PMZ) results of independent experimental studies to assess the
becomes prone to liquation cracking when the solid fraction capability of the model.
in the PMZ becomes lower than that in the mushy zone of
the aluminum alloy weld metal. They argued that a lower Mathematical Model
solid fraction in the PMZ makes this region weaker than the
weld metal mushy zone, making the PMZ vulnerable to The gas metal arc welding (GMAW) process is considered
liquation cracking. in the present study. This process involves a heat source
with a constant speed. The welding arc heats the work-piece
Composition of the weld metal results from the mixing of surface and contributes to the formation of a molten pool.
the base metal with the filler metal. Solutes are distributed The filler metal melts and mixes with the molten base metal
within the weld pool by convection and diffusion. by convection and diffusion. As the heat source moves

Copyright © 2006 ASM International® 35

away, the molten zone solidifies. A numerical model for density (Sv) and considers the interaction between the metal
non-equilibrium solidification in gas metal arc welds has droplets and the weld pool for various welding conditions
been developed that considers momentum, heat and solute [8,9]. The radius of the volumetric heat source is assumed to
transport. The model uses an effective partition coefficient, be 2.7 times the droplet radius [12]. The calculation of the
which considers both the local interface velocity and the volumetric heat source term has been described in the
undercooling for accurate prediction of solid fraction in the literature [8,9] and are not repeated here. Solute addition
weldment. The solution of the governing equations of from the filler metal is considered by incorporating a time-
conservation of mass, momentum, and energy has been well averaged volumetric mass source term, Sm, in the solute
explained in the literature [5,8-10] and are not repeated here. conservation equation, Eq. (2). The dimensions of the
Only the salient features of the model dealing with the volumetric mass source are assumed to be the same as the
solute transport, filler metal addition and non-equilibrium volumetric heat source. The net mass of solute from the
solidification are described. metal droplets, Qt, is given as:

Q t = ρπrw w f (C f − C)
Species conservation and solidification: The general form (3)
of the species conservation equation is given by:
where ρ is the density, rw is the radius of the wire, wf is the
∂ (ρC ) ∂ (ρu i C ) ∂ ⎛ ∂C ⎞ ∂C wire feeding rate, Cf is the concentration of solute in the
+ = ⎜ ρD ⎟⎟ − ρU (1)
∂t ∂x i ∂x i ⎜⎝ ∂x i ⎠ ∂x filler metal drops, and C is the local solute concentration.
The time-averaged volumetric mass source, Sm, for grid
where C is the solute concentration and D is the effective points within the mass source region is given by:
mass diffusivity of the solute. The variable C embodies
components from both solid and liquid phases. Voller et al. Qt
Sm = (4)
πD d d
[11] have shown that Eq. (1) may be rewritten in terms of
liquid phase concentration, Cl, and non-equilibrium partition
coefficient. Following their approach and neglecting where Dd is the droplet diameter.
diffusion in solid, Eq. (1) may be rewritten as [11]:
Thermo-solutal-flow coupling: The following iterative
∂ ∂(ρu i Cl ) ∂ ⎛ ∂Cl ⎞ ∂ enthalpy updating scheme proposed by Brent et al. [13] is
(ρ Cl ) + = ⎜⎜ ρf l Dl ⎟+ (ρ f s Cl )
∂t ∂xi ∂xi ⎝ ∂ x i ⎟⎠ ∂ t chosen in the present study for its adaptability in a fixed-
(2) grid enthalpy based framework:
∂ ∂C l
-k pCl (ρ f s ) − ρU −Sm
∂t ∂x
[ΔH P ]n+1 = [ΔH P ]n + a 0P λ [{h P }n − F −1{ΔH P }n ] (5)
where kp is the partition coefficient. In Eq. (2), Dl is an
effective diffusion coefficient to be described subsequently, where a P and a 0P are the coefficients of enthalpy for the
f denotes the appropriate phase fraction with subscripts l and
nodal point P in the discretized energy equation for the
s referring to liquid and solid phases, respectively, and Sm is
current and the previous time steps, respectively [14], ΔHP
a time-averaged volumetric mass source term to incorporate
is the latent heat content, hP is the sensible enthalpy
the filler metal addition.
associated with the nodal point P, n is the number of
Since the solute partitioning at the solid-liquid interface
may not reach thermodynamic equilibrium, calculations of iterations, λ is a relaxation factor, and F-1 is a suitable
fs and fl and the prescription of an appropriate partition function that relates local enthalpy with temperature and
coefficient in Eq. (2) requires a rigorous non-equilibrium liquid fraction. In order to include the appropriate thermo-
solidification model. Equation (2) also indicates a strong solutal effects [15] in F-1, the interfacial temperature, T, is
coupling between the thermal, solutal and velocity fields. A represented as a function of local liquidus composition, Cl,
key factor in this coupling is the appropriate modeling of the as [16]:
liquid fraction that affects the orientation and location of the
T = Tm + m L C l − V / β 0 − Γκ (6)
pool boundaries. The iterative updating of liquid fraction
(or, equivalently, nodal enthalpy) is done in the entire
concentration field, since the enthalpy and concentration where Tm is the melting point of the solvent, mL is the non-
fields are coupled through the non-equilibrium solidification equilibrium liquidus-line slope described by Eq. (7), V is
kinetics at the interface. the normal interface velocity, β0 is a kinetic coefficient of
interface motion, Γ is a capillary constant calculated by the
Filler metal addition: Filler metal addition to the weld pool expression (γ.Tm)/(L.ρ), γ is the surface tension, L is the
is considered by incorporating time-averaged volumetric latent heat of freezing, ρ is the density, and κ is the mean
heat and mass sources in the enthalpy and solute curvature of the solid-liquid interface. Equation (6)
conservation equations. The volumetric heat source is represents a deviation of the interfacial temperature from its
characterized by its radius (Rv), height (d) and power local equilibrium value due to motion of the interface, the

local interfacial curvature-undercooling effect and the non- (C l − C s )df s = (1 − f s )dC l (12)
equilibrium partitioning of the solute. The partitioning effect
is considered by relating the non-equilibrium liquidus-line where fs is the mass fraction of the solid and Cs is the solid
slope (mL) in Eq. (6) with its equilibrium value ( m*L ) as phase composition. Replacing fs by (1-fl) in Eq. (12), where
[16,17]: fl is the mass fraction of the liquid, and integrating, fl is
obtained as:
⎧⎪ ⎫⎪
m L = m *L ⎨1 +

[ ]
k *p − k P (1 − ln( k p / k *p )) ⎬ (7) ⎧⎪ C l dC l ⎫⎪
⎪⎩ ⎪⎭
1 k p f l = exp ⎨− ∫ ⎬ (13)
⎪⎩ C 0 C l (1 − k p ) ⎪⎭
where kP is the modified partition coefficient that can be
expressed in terms of the equilibrium partition coefficient, where kp, which replaces Cs/Cl, is a convection-corrected
k *P , as [16]: partition-coefficient representing the non-equilibrium
effects, given by Eq. (8). Equation (13) can be integrated if
the variation of kp with Cl is known. For the specific case of
⎧ k * [1 − ( V / VDL ) 2 ] + V / VDL ⎫
kP = ⎨ P ⎬ L for V < VD
a composition independent partition coefficient, integration
⎩ 1 − ( V / V L 2
) + V / V L
⎭ (8)
D D of Eq. (13) gives [18]:
= 1L for V ≥ VDL
k p −1
Cl = C0 f l (14)
In Eq. (8), V is the diffusive speed in the liquid, which can
be calculated as [16]: where C0 is the initial composition of the liquid. Although
Eq. (14) appears similar in form to the well-known Scheil’s
VDL = ( D *l / τ LD ) 0.5 (9) equation [18], a key difference lies in the fact that the
partition coefficient in Eq. (14) is a strong function of the
where τLD refers to the time of diffusional relaxation of interface growth rate governed by Eq. (10) and not a
collective atoms (molecules, particles) to their equilibrium
state in a local volume of alloy [16]. The diffusive speed at
With the aid of Eqs. (6) and (14), a final form of the
which a solute atom diffuses through the interfacial region
enthalpy updating function appearing in Eq. (5) can now be
can be approximated by the ratio of diffusivity of the solute
obtained as:
atoms at the interface and a length scale characterizing the
interface width. Furthermore, the interface velocity (V) in (k p −1)
Eq. (8) can be calculated in an iterative manner using the ⎡ ⎛ ΔH ⎞ ⎤
F −1 ( ΔH ) = c p ⎢Tm + m L C 0 ⎜ ⎟ − V / β 0 − Γκ ⎥ (15)
following equation [5]: ⎣⎢ ⎝ L ⎠ ⎦⎥

k SG S − k l G l
V= (10) Finally, the liquid fraction is calculated by using Eq. (5) as:
fl L
fl = (16)
where GS and Gl are the temperature gradients in solid and L
mushy zone at the mushy zone/solid interface, respectively,
kS and kl are thermal conductivities of the solid and liquid Possible unrealistic intermediate estimates predicted by Eq.
phases, respectively, fl is the liquid fraction, and L is the (16) during iterations can be avoided by imposing the
latent heat of freezing. For a known interface velocity, the following constraints:
diffusion coefficient appearing in Eq. (2) can be prescribed
as [16]: f l = 0 if f l < 0
= 1 if f l > 1
D l = D*l [1 − ( V / VDL ) 2 ]L for V < VDL
= 0 L for V ≥ VDL Boundary conditions: The temperature and velocity
boundary conditions are available in the literature [5] and
where D*l is the diffusion coefficient in the liquid under the are not repeated here. The boundary conditions for solute
conditions of interfacial equilibrium. transport at the solidification interface needs to consider
non-equilibrium partitioning of solute at the solidification
Equations (7) to (10) can be effectively used to complete the front:
iteration cycle involving updating of enthalpy using Eq. (5).
The calculations require an appropriate functional relation Cl = (18)
between liquid composition, Cl, and liquid fraction, fl, kP
consistent with the local solute balance:

where Cl is the local solute concentration in the liquid and from the base metal resulting in a weld metal composition
Cw is the solute concentration in the solidified weld metal. that lies between the filler metal and the base metal
Similarly, the boundary condition at the melting front can be compositions. At the solidification front, the solute is
written as: rejected from the solidified material into the molten pool.
As a result, high solute concentration is observed at the
Cl = Cb (19) solidification front in Fig. 2. Similarly, in the transverse
sections ahead of the heat source, the composition near the
where Cb is the concentration of the solute in the base metal. melting front is same as that of the base metal. However, in
transverse sections behind the heat source, segregation of
Results and Discussion the solute is observed near the solidification front. In the
middle of the weld pool, a large amount of filler metal is
GMA welding of 2219 aluminum-copper alloy containing added and the solute concentration is fairly close to the filler
6.3 wt% Cu was simulated. The filler metal compositions metal composition.
considered varied from 0.08 wt% Cu to 9.0 wt% Cu. The
data used in the calculations are summarized in Table 1.
Figure 1 shows the computed velocity and temperature
fields in the weld pool of 2219 alloy with the filler metal
composition of 0.08 wt% Cu. The weld pool is wide and
shallow because the aluminum alloy has a negative
temperature coefficient of surface tension (dγ/dT) which
causes the liquid metal to move from the middle to the
periphery on the weld pool surface. The Peclet number for
heat transfer is 12 while the Peclet number for mass transfer
is 2.9×105, which indicates that convection is the primary
mode of heat and mass transport in the weld pool.
Therefore, it is necessary to consider convective heat and
mass transport in order to accurately predict the temperature
and velocity fields as well as solute concentration Figure 1: Velocity and temperature fields in the weld pool
distribution in the weld pool. for the welding conditions indicated in Table 1. The filler
metal concentration was 0.08 wt% copper. All the
Table 1: Data used in the calculations. temperatures are in degree Kelvin.

Problem data/physical property Valuea

Arc current (amp) 140
Arc voltage (volt) 22
Welding speed (m.s-1) 4.2 × 10-3
Density (kg.m-3) 2400
Viscosity of liquid (kg.m-1.s-1) 0.3 × 10-3
Specific heat (J.kg-1.K-1) 1.06 × 103
Thermal conductivity (W.m-1.K-1) 192
Equilibrium partition coefficient 0.16
Equilibrium slope of liquidus line (K/wt%) -3.37
Solidus temperature of alloy 2219 (K) 821
Liquidus temperature of alloy 2219 (K) 911
Mass diffusivity of copper in liquid
aluminum (m2.s-1) under interfacial 3.0×10-9
equilibrium conditions
See Reference [19]. Figure 2: Concentration field (wt% Cu) in the weld pool.
The filler metal composition was 0.08 wt% Cu.

Figure 2 also shows that the high computed solute content at

Figure 2 shows the computed solute concentration
distribution within the weld pool for filler metal the solidification front does not have as much influence on
composition of 0.08 wt% Cu. It can be observed that the overall concentration distribution as the mixing of the
filler metal with the base metal. This behavior can be
convection plays a dominant role in solute distribution in
the weld pool causing efficient mixing of the base metal attributed to the very low mass diffusivity of copper in the
with the filler metal. At the melting front, in front of the alloy and very low liquid velocities in the two phase region
adjacent to the solidification front. The rejected solute is
pool, the base metal melts and forms a liquid of the same
composition. The filler metal then mixes with the liquid confined to a very small region and the low velocities in the

two phase region prevent rapid mixing of the rejected solute the equilibrium value, k *P , which is consistent with the
into the weld pool. shrinking of the two phase region at high solidification
rates. Once k *P and kp were obtained for each temperature,
Since the composition of a single-pass GMA aluminum
Eq. (7) was used to get the non-equilibrium liquidus line
weld is essentially uniform [1], the solidified weld metal
slope, ml, at these temperatures. Next, the modified liquidus
solute concentration was assumed to be equal to the average
composition, Cl, at each temperature was calculated using
concentration of the solute in the molten weld pool. Table 2
kp, ml, and the melting point of pure aluminum, 933 K. The
lists the solute content (wt% Cu) of the solidified weld
corresponding values of the modified solidus compositions,
metal and the solidifying weld metal (mushy zone at the
Cs, at each temperature could be obtained from the values of
trailing end), corresponding to the four different
kp and Cl at these temperatures. The modified Cs and Cl
compositions of the filler metal used in the present study.
were then used to calculate the non-equilibrium solid
Table 2 indicates that the concentration of the solute in the
fractions. The computed solid fraction for the 2219 alloy,
solidified weld metal and the solidifying weld metal
having 6.3 wt% Cu, is shown in Fig. 3. The solid line curve
strongly depends on the filler metal composition. The
in Fig. 3 was calculated from the equilibrium phase diagram
accuracy of the calculated solidified weld metal composition
using the equilibrium Cs and Cl values. The non-equilibrium
can be examined by comparing the computed solidified
solid fraction is lower than the corresponding equilibrium
weld metal composition for 0.08 wt% filler metal addition
value because of undercooling which prevents solidification
with the corresponding independent experimental result of
to occur at equilibrium temperature at high solidification
Huang and Kou [1]. For a GMA weld of 2219 alloy using a
filler metal containing 0.08 wt% copper, Huang and Kou [1]
measured the weld metal composition to be 3.43 wt%
copper. For the same welding conditions, the computed 1
solidified weld metal composition was equal to 3.17 wt%
copper as shown in Fig. 2 and Table 2, thus confirming the
accuracy of the calculations.
Table 2: Composition of the filler metal (wt% Cu) and the
Solid fraction (f s)

corresponding solute content (wt% Cu) of the solidified

weld metal and the solidifying weld metal (mushy zone at 0.7

the trailing end). Four different filler metal compositions

were used. Base metal contained 6.3 wt% Cu, and the rate 0.6 Equilibrium
of addition of the filler metal was 9.31 cm/s.
Filler metal Solidified weld Solidifying
metal weld metal 0.4
(wt% Cu) (wt% Cu) (wt% Cu)
0.08 3.17 0.31 0.3
823 833 843 853 863 873 883 893 903
2.0 4.13 2.16
Temperature (K)
4.0 5.14 4.09
9.0 7.67 8.91
Figure 3: Solid fraction versus temperature for alloy
composition of 6.3 wt% copper. The solid line was obtained
from the equilibrium phase diagram and the dotted line was
Gittos and Scott [20] proposed that liquation cracking
obtained using the modified non-equilibrium solidus and
occurs when the base metal solidus temperature is below the
liquidus lines.
weld-metal solidus temperature. In other words, if the base
metal solute content is higher than that of the weld metal,
The variation of non-equilibrium solid fraction with
then the PMZ is susceptible to liquation cracking. However,
temperature for three cases has been plotted in Fig. 4. One
Huang and Kou [1] argued that the cooling rate during
plot is for the base metal or the PMZ composition, i.e., 6.3
welding may be too high for equilibrium solidification to
wt% Cu, and the other two are for the solidifying weld
occur, and solidification can continue far below the
metal compositions in the mushy region computed from the
equilibrium solidus temperature. They [1] proposed that the
solute transport model for two welding conditions. The
partially molten zone (PMZ) becomes prone to liquation
lower plot represents welding of 2219 alloy using a filler
cracking when the solid fraction in the PMZ becomes lower
metal with 9.0 wt% Cu that leads to a mushy zone
than that of the mushy zone in the weld metal. A weaker
composition of 8.91 wt% Cu as listed in Table 2 for the
PMZ with lower solid fraction makes this region vulnerable
welding parameters indicated in Table 1. Similarly, the
to liquation cracking. Calculation of the solid fraction in the
upper plot indicates the use of a filler metal of 2.0 wt% Cu,
solidifying region requires the computed values of non-
for the same welding conditions, that leads to a mushy zone
equilibrium partition coefficient. The non-equilibrium
composition of 2.16 wt% Cu. The solid fraction in the PMZ
partition coefficient, kp, was found to be higher than that of

can be compared with that in the solidifying region, i.e., the In each case, the susceptibility of liquation cracking was
mushy zone. The lower graph representing 9.0 wt% Cu determined by Huang and Kou’s criteria, i.e., by comparing
containing filler metal has a lower solid fraction in the the solid fraction in the solidifying weld metal with the
mushy zone than that in the PMZ. Therefore, the solidifying corresponding value in the PMZ. The model predictions of
weld metal has a lower strength than the PMZ and the PMZ liquation cracking susceptibility in Al-Cu alloy weldments
is not susceptible to liquation cracking. In contrast, when were confirmed by independent experiments for various
the 2.0 wt% Cu containing filler metal is used, the solid filler metal compositions.
fraction in the solidifying metal is higher than that in the
PMZ. Consequently, the solidifying weld metal is stronger Acknowledgements
than the PMZ making the PMZ susceptible to liquation
cracking. Thus, the present calculations considering This research was supported by a grant from the U.S.
convective solute transport, non-equilibrium solidification Department of Energy, Office of Basic Energy Sciences,
and filler metal addition can be used to predict liquation Division of Materials Sciences, under grant number DE-
cracking susceptibility in aluminum alloy welds. FGO2-01ER45900.


1. C. Huang and S. Kou, Weld. J., 83, 50s (2004).
region 2. H. Zhao and T. DebRoy, Metall. Mater. Trans. B, 32,
Non-equilibrium solid fraction

0.8 163 (2001).

3. H. W. Kerr and M. Katoh, Weld. J., 66, 251s (1987).
4. C. Huang and S. Kou, Weld. J., 82, 184s (2003).
Safe region
5. W. Zhang, G. G. Roy, J. W. Elmer and T. DebRoy, J.
Appl. Phys., 93, 3022 (2003).
6. X. He, P. W. Fuerschbach and T. DebRoy, J. Phys. D:
0.4 2.16 wt% Cu Appl. Phys., 36, 1388 (2003).
6.30 wt% Cu
8.91 wt% Cu
7. S. Chakraborty and P. Dutta, Materials and
Manufacturing Processes, 17, 455 (2002).
0.2 8. W. Zhang, C.-H. Kim and T. DebRoy, J. Appl. Phys.,
95, 5210 (2004).
9. W. Zhang, C.-H. Kim and T. DebRoy, J. Appl. Phys.,
0 95, 5220 (2004).
823 833 843 853 863 873 883 893 903
10. K. Mundra, T. DebRoy and K. Kelkar, Numerical Heat
Temperature (K) Transfer A, 29, 115 (1996).
11. V. R. Voller, A. D. Brent and C. Prakash, Int. J. Heat
Figure 4: Non-equilibrium solid fraction versus temperature Mass Transfer, 32, 1719 (1989).
calculated using modified non-equilibrium solidus and 12. A. Kumar and T. DebRoy, Int. J. Heat Mass Transfer,
liquidus lines for different alloy compositions. The PMZ 47, 5793 (2004).
composition was 6.3 wt% Cu. The calculated mushy zone 13. A. D. Brent, V. R., Voller and K. J. Reid, Numerical
compositions, or the solidifying weld metal compositions, of Heat Transfer, 13, 297 (1988).
2.16 wt% Cu and 8.91 wt% Cu correspond to the filler 14. S. V. Patankar, Numerical Heat Transfer and Fluid
metal compositions of 2.0 wt% Cu and 9.0 wt% Cu, Flow, Hemisphere/McGraw-Hill, Washington DC
respectively. (1980).
15. S. Chakraborty and P. Dutta, Metall. Mater. Trans. B,
Conclusions 32, 562 (2001).
16. P. Galenko and S. Sobolev, Phys. Rev. E, 55, 343
A numerical model for non-equilibrium solidification in (1997).
welds has been developed that considers momentum, heat 17. W. Kurz and D. J. Fisher, Fundamentals of
and solute transport. The model uses an effective partition Solidification, 3rd Edition, Trans. Tech. Publications,
coefficient, which considers both the local interface velocity Switzerland (1992).
and the undercooling for accurate prediction of solid 18. M. C. Flemings, Solidification Processing, McGraw
fraction in the weldment. The solute concentration Hill, New York (1974).
distribution in the weld pool was effectively simulated. The 19. Q. Z. Diao and H. L. Tsai, Metall. Trans. A, 24, 963
predicted weld metal solute content agreed well with the (1993).
independent experimental observations. Using the computed 20. N. F. Gittos and M. H. Scott, Weld. J., 60, 95s (1981).
average composition in the two phase mushy region, the
solid fraction in the solidifying weld metal was compared
with that in the PMZ for various filler metal compositions.

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Prediction of Mechanical Properties of Steel Spot-Welds

M. Mimer, L-E Svensson
Volvo Truck Corporation, Gothenburg, Sweden

Abstract overall joint strength. The intense heat input associated with
the welding process has a radical effect on metallurgy and
Spot welding will continue to be the dominating process for strength of the spot-welds, which makes it necessary to
joining of thin steel sheets in the automotive industry for many perform different kinds of mechanical testing, such as static
years, due to the reliability and low cost of the process. The tensile testing, fatigue testing and even high strain rate testing.
overall strength of a component is determined by the strength With the introduction of high strength steels, established
of the individual spot welds, together with parameters like relations between base material strength, size of the weld and
spot interdistance, number of spot welds etc. Thus, the mechanical properties of the weld is changed (1). In addition,
mechanical properties of spot welds are essential to know. it has been found during testing that certain grades of high
From work based on mild strength steels, it is known that the strength steels are susceptible to less desired behaviours like
strength of a spot weld is closely related to the properties of interfacial failure and brittle fractures (2, 3). Thus, it might be
the base material and the size of the nugget in the spot weld. necessary to predict mechanical properties such as strength
However, with the introduction of higher strength steels more and hardness in order to avoid less desirable fracture
complex relationships may appear. For example, joints behaviours and provide sufficient information for further finite
between steels having large differences in strength are element analysis.
becoming much more common. Such joints may also be more A second benefit of an increased amount of simulation and
difficult to weld, due to the differences in steel characteristics. modelling is the possible reduction in number of prototypes
To reduce cost and increase speed of product development, and amount of time-consuming testing, which must be reduced
simulation is becoming more important. Simulation of the spot as the lead time for introduction of new products in the
welding process can now be made on standard PC’s using the automotive industry is continuously decreasing. Through a
simulation program SORPAS®. With SORPAS® the size of chain of simulation and modelling it is possible to predict the
the nugget in a spot weld can be estimated. For prediction of mechanical properties of any steel spot-weld and thereby
the strength and hardness of spot welds, two different avoid undesirable fractures as well as enable a reduction in
approaches have been used. A model was constructed using a amount of testing.
DOE program, relating spot weld strength to base metal
strength and spot-weld dimensions. As an alternative approach Work description
the hardness of a spot weld was estimated from the chemical
composition of the spot welds, following equations given by In the work of creating equations for prediction of mechanical
Blondeau et al (1). properties it is necessary to start with a process simulation in
In the paper, it is demonstrated how the chain of simulation order to get input to the creation of predictive formulas.
programs can be used to predict the mechanical properties of SORPAS® is an acronym for Simulation Of Resistance
spot-welds in a wide range of steels. Projection And Spot-welding. In this paper, the outputs used
from the software are nugget diameter and cooling rate.
Introduction The cooling rate obtained from the process simulation can
then be further used to predict the hardness of the spot welds.
Safety, weight reduction and cost efficiency have been and For this, it is necessary also to know the chemical composition
will continue to be the three main factors controlling the of the weld. If both these pieces of information are available, it
development of car bodies, the so called Body-In-White. The is possible to predict the weld nugget hardness using an
properties of a Body-In-White are determined by material equation suggested by Blondeau et al (4). The strength of the
choice and thickness, design solutions and the joining spot weld cannot, however, be predicted from the hardness. It
processes used. In the case of spot welding, which is the is anyway useful to know the hardness of the spot weld since
dominating joining process for thin sheet applications, the it gives an indication of the risk of for example cleavage
diameter of the spot-welds as well as the interdistance between fracture or interfacial failure (2, 3).
spots is commonly assumed to have the largest effect on

Copyright © 2006 ASM International® 41

The strength of spot-welds as function of variables like sheet • Welding current and electrode force as a function of
tensile strength, sheet thickness and spot-weld size can be time.
predicted using data from previous testing. In this paper, only • Electrical contact resistance and thermal conductance
predictions of static tensile strength for shear and cross tension at the interface together with friction conditions at the
are made. The data, which the prediction models are based surface. (8)
upon, comes from a number of spot-welding projects at the Outputs from the simulations are nugget diameter and
Swedish Institute of Metals Research (SIMR) (5, 6, 7). In electrode indentation as well as information of temperature
total, the prediction model for shear strength is based on 198 and resistance as a function of time. Figure 2 is showing an
tensile tested spot-welds and the model for cross tension example of how a simulated weld nugget compares to a real
strength is based on 428 tensile tested spot-welds. For all test nugget. In this case the simulated weld nugget diameter is
specimens, the material thickness, yield and tensile strength as slightly larger and the electrode indentation of the real weld is
well as plug size was measured. These parameters were used larger.
as factors, to which a model for strength was fitted using PLS
(Projection to Latent Structures). A software for statistical
modelling was used to obtain as good models as possible. This
paper only demonstrates prediction models where two
materials of similar grade and thickness are welded. However,
by adding more factors to the model, similar prediction
models for welds of dissimilar material and thickness can be
made. For future developments, the weld size can be
calculated using SORPAS®. The other factors are easily
Figure 2: Illustration of simulated and welded nuggets. The
Results material is a high strength steel with 7μm of zinc coating. (9)

Spot-weld simulation with SORPAS® From previous work, comparing simulated weld nuggets with
physical spot-welds (9) the accuracy of simulations made
With the software SORPAS® it is possible to simulate the spot
welding process on a standard PC. The simulation is based using SORPAS® has been determined. Just as shown in Fig. 2,
upon a finite element analysis where metallurgical, electrical, the results correlate well in many cases, but some areas of the
mechanical and thermal models work together in close program need improvements.
interaction; see Fig. 1 (8).
Prediction model for weld hardness
The calculations of weld nugget hardness are based on the
following equations (4)

HV (martensite) = 127 + 949C + 27Si + 11Mn + 8Ni

+ 16Cr + + 21×log V [1]

HV (bainite) = -323 + 185C + 330Si + 153Mn +

65Ni + 144Cr + 191Mo + (89 + 53C – 55Si – 22Mn
– 20Cr – 33Mo)×log V [2]

HV(ferrite/pearlite) = 42 + 223C + 30Mn + 12.6Ni +

7Cr + 19Mo + (10 - 19Si + 4Ni + 8Cr + 130V)×log V

Here, C, Mn etc stand for the concentration of the elements in

wt% and V is the cooling rate in °C/hour.

From SORPAS® simulations it can be seen that the cooling

® rate of a spot-weld is initially, i.e. at the moment when current
Figure 1: The algorithm used by SORPAS includes four
models, which all are closely interrelated. is turned off, as high as 2000 °C/s, but it gradually decreases
as the temperature drops. In equations 1-3 above the cooling
The input data for the software includes: rate comes in as a logarithmic term, which means that a large
• Material properties, including electrical resistivity, variation in cooling rate does not make a large difference in
thermal conductivity, heat capacity, material strength final weld hardness. For the calculations made here, a cooling
of the work pieces and electrodes. rate of 1 000 000 °C/hour is assumed.

The chemical composition of the weld nugget was estimated
by assuming that each sheet contributed to composition in 30
proportion to its thickness. This is of course a relatively coarse 28
assumption. It is possible that a future development of 26
SORPAS® may provide a more accurate calculation of how 24

much from each sheet is melted into the nugget. 20
By comparing the calculated hardness in equations 1-3 above 18
to measured weld hardness of 24 spot-welds it was noted that 16
equation [1] provides the best results for the calculated weld 14
hardness for all welds except when the welded materials are 10
very lean alloyed mild steels, where equation [2] provide 8
better results. Fig. 3 shows a plot of how the predicted 6
hardness relates to measured hardness for the 24 welds. The 6 8 10 12 14 16 18 20 22 24 26 28 30
24 welds were made on two different types of test specimen, Predicted
here called coupon and H-specimen. The hardness tests were Figure 4: Measured and predicted strength values for shear
made using a Vickers diamond and a load of 1kg. specimens. The values are given in kN.
As seen in Fig. 3, the fit between calculated and measured
values is not perfect, but taking the scatter in measured The suggested predictive equation for shear strength is:
hardness and the rough assumption regarding chemical
composition of the weld nugget into account the results must
FSS = -8,01976 – 14,1373×t + 0,0185996×Rm +
be seen as encouraging.
3,24976×d + 11,2113×t2 - 1,07619e-5×Rm2 –
0,224022×d2 + 0,0061255×(t×Rm) [4]
Here, t stands for material thickness, Rm is the tensile strength
of the material and d is the plug diameter of the weld.
450 Coupons

Mild steel coupons (Mart)

If the coefficients for each terms in equation 4 is scaled and
Predicted hardness

Mild steel coupons (Bainite)
centered it becomes clear that the material thickness followed
350 by base material strength has the largest individual influence
on spot-weld strength. Such a plot is shown in Fig. 5. The
scatter bars indicate the 95% confidence level of each


150 200 250 300 350 400 450 500
Measured hardness 3
Figure 3: Comparison between measured and calculated
hardness for the 24 welds. The error bars show the minimum 2

and maximum hardness in each nugget. 1

Prediction models for weld strength 0

As described above, the prediction models for strength are
based on a large amount of tensile tests in shear mode as well
as in cross tension mode. The fracture force was recorded -2





during the tests and was used as a response in these models.

As the model for shear strength was fitted using PLS, a very
good model was obtained. In Fig. 4, the observed values for Figure 5: Scaled and centered coefficients for spot-weld shear
shear strength are plotted against the values predicted by the strength. Large bars indicate large effect on strength.
model. Once again, the results must be seen as encouraging.
The relation between predicted and measured values is In graphical illustrations, shown in Fig. 6, it once again
particularly good in the lower load range, whereas some becomes clear that material thickness has the largest
scatter in results are seen in the higher load range. individual influence on strength. One can also see that
material strength has a positive effect on joint strength,
especially for thicker sheets. The image to the right, showing

the effect of plug diameter, indicates minor influence
compared to material thickness and strength.


1 2 3 4 5 6 7 8 9 10 11 12 13
Figure 7: Observed versus predicted plot for cross tension
strength. The values are given in kN.

The predictive equation for cross tension strength is:

FCT = 6,55921 - 23,1511×t - 0,00152876×Rm +

2,51058×d + 13,3216×t2 - 0,160177×d2 [5]

By looking at the plot of the scaled and centered coefficients

for the terms in equation [5] (Fig. 8), it becomes very clear the
thickness has the largest individual influence on weld strength.
It can also be seen that the material tensile strength has a
negative influence on joint strength, i.e. by increasing the
tensile strength of the material, the cross tension strength is

Figure 6: Surface plots, showing the predicted shear strength.
To the left the plug diameter is fixed at 5,96mm and to the 1,0
right the material strength is fixed at 965MPa.

Just as the model for shear strength, the fitted model for cross
tension strength reflects the measured values very good.
Figure 7 is showing the observed versus predicted plot for
cross tension. It can be seen that the predictions are especially
good in the lower load range, whereas some deviations is seen -0,5
for specimens which fractured with higher loads.



Figure 8: Scaled and centered coefficients for cross tension

strength. The scatter bars indicate 95% confidence level.

The graphical illustration (Fig. 9) for cross tension strength

shows that the material strength has minor influence on
strength and the thickness and plug size have larger and
positive influence on joint strength.

observed and predicted values is most likely the rough
assumption of the chemical composition in the weld nugget.
However, by combining the results from SORPAS®
simulations with the predictive models for strength and
hardness, the possibility to predict the properties of single
spot-welds from scratch is very promising.

It is worth noting that equations for calculating spot-weld

strength in shear and cross tension have been published by
several other (10, 11) authors. Although somewhat different
approaches have been chosen, the results seem to be fairly
similar. When using the equations mentioned above, some
material constants have to be used. Due to these constants,
which might not be available for all materials of interest,
equations [4] and [5] presented here may be favourable. It
must also be stated that the predicting equations in this paper
as well as in the papers mentioned above are valid for full plug
failures. In the case of partial plug or interfacial failures the
fracture forces are normally lower.

Statistical modelling of data from mechanical testing of welds

is a good way of making predictions. This paper has been
limited to static testing only and also to sheet materials of the
same grade and thickness. However, similar work can easily
be done regarding predictions of other mechanical properties
such as elongation during static testing, fatigue life of spot
welds or high strain rate testing. It is also possible to make the
predictive equations valid for just any combination of steel
grade and thickness simply by introducing factors for each
sheet into the equations.

Figure 9: Graphical illustration of cross tension strength. To
the left, the plug size is held constant at 5,75mm and to the The main conclusions of this study can be summarized as
right material tensile strength is constant at 965MPa. follows:
• The thickness of the sheets being welded has the
largest individual influence on spot-weld strength. In
Discussion addition, larger nugget (or plug) diameter increases
the strength. The tensile strength of the base material
By using a chain of simulation and modelling software, as mainly affects the strength in shear mode.
demonstrated here, not only the mechanical properties of each • Simulation of the spot welding process using
individual weld is possible to predict, but also the weldability
SORPAS® provide results, which correlate well with
for any combination of materials, which are to be welded.
physical spot-welds in most situations. In certain
areas does the program need improvement in order to
It has been shown in this paper that the mechanical properties
reflect process properly.
of a steel spot weld can be predicted with reasonable accuracy.
• The equation for prediction of martensite hardness,
From a statistical point of view, the models reflect the
suggested by Blondeau et al. is valid for all steel
measured values to approximately 90%, which must be seen
spot-welds except very lean alloyed mild steels where
as very good at this stage. The large amount of data, which the
the equation for bainite provide more accurate
models are based upon provide a solid basis for the modelling.
results. Some further work needs to be done
However, in order to come up with even better models, the
regarding the chemical composition of the weld
data needs to be designed in such a way that a large volume of
nugget when to sheets of dissimilar material grade
the input factors are being tested. That has not been the case
and thickness are being welded.
for the data in this paper. Another limiting factor for achieving
better models, is the natural scatter in spot weld strength, • Predicting spot-weld strength by statistical modelling
which is obtained by testing. When it comes to weld hardness of historical data provide a solid basis for further
predictions the limiting factor for better correlation between modelling and analysis.

• A combination of simulation using SORPAS® with
the other predictive equations will further enhance
the possibility for predicting the properties of single
spot welds from scratch.

This paper is published by the permission of Volvo AB.
Discussion with Dr Wenqi Zhang, Swantec A/S is gratefully

1. L-E Svensson, Prediction of hardness of spot welds
in steels, Submitted for publication in Welding in the
2. M. Eliasson, L-E Svensson, R. Johansson and J.K.
Larsson, Doc IV-850-03 Improvement of mechanical
properties of laser welded TRIP 700 steel, 56th
Annual Assembly of the International Institute of
Welding (IIW) 2003
3. M. Mimer, L-E Svensson and R. Johansson, Process
Adjustments to Improve Fracture Behaviour in
Resistance Spot Welds of EHSS and UHSS, Welding
in the World, 2004, Vol. 48, no. ¾, 14 – 18.
4. R Blondeau, P Maynier, J Dollet and B Vieillard-
Baron, Mathematical model for the calculation of
mechanical properties of low alloy steels
metallurgical products: a few examples of its
applications, Proc Int Conf Heat Treatments’76,
1976, Metals Society, London
5. J Hedegård, J Andersson, E Tolf, Resistance spot
welding of ultra high strength C-Mn and stainless
steels – enclosures, IM-2002-561:2
6. J Hedegård, J Andersson: Resistance spot welding of
some EHS and UHS steels, Part 1, C-Mn steels, IM-
7. E Tolf, Possibilities with Advanced Control of the
Resistance Spot Welding Process, Swedish Institute
for Metals Research 2004/05
8. SORPAS® User Manual version 5.0 Professional,
Swantec software and engineering Aps. 2004
9. D. Axelsson, Optimization of the Resistance Spot
Welding Process and Evaluation of the Simulation
Software SORPAS®, Diploma workNo.56/2004,
ISSN 1651-0003, Chalmers University of
10. S. Ferrasse, P.Verrier, F. Meesmaecker, Resistance
spot weldability of high strength steels for use in car
industry, Welding in the World 1998 Vol 41 p.177-
11. T. Nilsson et al., Fogningshandboken SSAB Tunnplåt
AB 2004, p3.15 p3.22-24

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Predictions of TIG weld depth from a unified electrode-arc-workpiece treatment.

John J Lowke
CSIRO Industrial Physics Sydney, NSW, Australia

Manabu Tanaka and Masao Ushio

Joining and Welding Research Institute, Osaka, Japan

Abstract specific heat, viscosity, thermal conductivity and electrical

conductivity are required as a function of temperature for the
Two dimensional predictions have been made of temperature arc plasma, the electrode, the solid work piece and also for the
profiles of the arc, electrode and work-piece for stationary liquid weld pool. For the arc plasma, values are obtained
TIG arcs in a unified treatment. Increases in weld depth of assuming “local thermodynamic equilibrium” of the plasma.
over a factor of two are predicted for different surface tension
properties of the weld pool or if an insulating layer exists on Solution of the three conservation equations of mass,
the weld pool. The calculations indicate possible explanations momentum and energy follows conventional heat transfer
of weld depth variability in practical welding and also possible theory for heat transfer in fluids, and we have followed the
explanations for the increase in weld depths obtained in TIG numerical methods of Patankar [4]. But there are two
welding activated by a flux (ATIG). important extensions to this treatment.

Introduction. Firstly, account must be taken of the magnetic pinch forces, or

“j×B” forces, in the momentum transfer equation; j is the
Theoretical predictions of weld depth depend on the whole current density and B the magnetic field. These forces
electrode-arc-workpiece system. For TIG welding, the heat dominate the convective flow for arcs, in that the current
input to the workpiece depends on the current density density is highest at the cathode tip, resulting in an increased
distribution at the surface of the workpiece, which depends on pressure near the cathode tip due to the magnetic field, which
the arc. The current density distribution in the arc depends, in drives plasma flow from the electrode to the workpiece.
turn, on the shape and sharpness of the cathode. It follows that Similarly magnetic pinch forces can be dominant in
for predictions of weld pool properties, the whole arc- determining the convective flow in the weld-pool.
electrode system needs to be included in the calculation.
Analyses can then be made of the dependence of quantities, Secondly, the electrical conductivity of the arc plasma poses a
such as weld depth, on particular properties such as the surface special problem because in the regions adjacent to the
tension of the weld pool, arc gas properties or electrode electrodes, where the temperatures are very low compared
properties. with usual plasma temperatures, the equilibrium electrical
conductivity is near zero. Electrical conduction through these
Calculations of temperature profiles [1],[2],[3], and thus also electrode regions is accounted for by considering ambipolar
weld depth, have been made in two dimensions for an diffusion of electrons from the plasma to the electrodes. The
idealised TIG system, consisting of a stationary and vertical electron continuity equation is solved for the arc region to
tungsten cathode producing an arc in argon of current 150-200 determine the non-equilibrium electron density accounting for
A to a stainless steel electrode, the bottom surface of which is ambipolar diffusion. The current continuity equation is also
maintained at room temperature, i.e. 300 K. Solutions are solved to obtain distributions of the electric potential and thus
obtained of the coupled equations for the conservation of the electric field. But solution of the current continuity
mass, energy and momentum for the whole electrode, arc, equation differs from conventional solutions in that (1) we use
workpiece and weld pool region. These equations define the calculated non-equilibrium electrical conductivity obtained
temperature, velocity and pressure, for this region including from the electron density and (2) electron diffusion current is
the weld pool. Values of transport coefficients such as density,

Copyright © 2006 ASM International® 47

included in addition to the usual electric field component of the surface of the weldpool is flat. No consideration is made of
current. space charge sheaths at the electrodes.

Physical Model A significant heating mechanism at the surface of the cathode

is due to ion currents, contributing jiVi, per unit area of the
(1) Magnetic Forces. surface, if the derived total current density is greater than the
A term, jz×Bθ is introduced into the axial momentum equation theoretical thermionic emission current; the ion current ji is
[2] to represent the magnetic pinch forces. The azimuthal then assumed to be equal to the difference between the derived
magnetic field Bθ induced by the arc current is evaluated by total current and the theoretical thermionic emission current;
Vi is the ionization potential.
Maxwell’s equation,
A boundary condition was imposed at the surface of the weld
1 ∂
(rB ) = μ0 jz pool setting the axial velocity of the liquid to zero. This
r ∂r θ boundary condition effectively uncouples the pressure of the
arc region from that of the weld pool region, so that no
where r is the radial coordinate, μ0 is the permeability of free
account was taken of the arc pressure in depressing the surface
of the weld pool. Convective velocities were also set to zero at
space and jz is the axial component of the current density. the base of the weld pool. The position of this boundary was
re-determined after each numerical iteration according to the
(2) Electrical Conductivity. temperature of each cell. If the temperature of a cell was
Instead of the usual representation of the current density, j, as above the melting point, the cell was regarded as being liquid,
dependent only on the electric field E, by Ohm’s law, otherwise solid. We have a non-uniform grid with a total of 95
j = σ E , where σ is the electrical conductivity, we also points in the axial direction and 75 points in the radial
include a term to account for diffusion current from electrons. direction. These points are needed to span the cathode,
This term overcomes the problem that the equilibrium cathode tip, the arc and the weld pool, but we were still able to
electrical conductivity is effectively zero in the plasma close have 20-23 points in the axial direction for the weld pool and
to the electrodes owing to the low plasma temperature. Thus, 41-43 points in the radial direction at the top of the weld pool.
Mesh sizes varied from 0.0001 cm at the surface of the weld
∂V ∂n pool to 0.5 cm at the base of the cathode.
jr = −σ + eDe e
∂r ∂r
and Specific calculations.
∂V ∂n
jz = −σ + eDe e
∂z ∂z (1) 150 A, Stainless steel.
Figure 1 shows calculated results of temperature profiles for a
where De is the electron diffusion coefficient, e is the 150 A arc in argon with a 1 cm thick workpiece of stainless
steel. The contour for a temperature of 1750 K in the
electron charge and ne is the electron number density. The
workpiece gives the limit of the liquid weld pool, as 1750 K is
electron continuity equation in terms of ambipolar diffusion is the melting point of steel. Principal features of this calculation
are in good agreement with experimental measurements i.e.
1 ∂ ∂n ∂ ∂n the central temperature of the arc, the temperature of the tip of
(rDamb e ) + (Damb e )
r∂r ∂r ∂z ∂z the cathode, and the depth and extent of the weld pool. Fig. 2
shows calculated velocity profiles for both the arc and the
+ γ [K eq (T )n e na − ne3 ] = 0 weld-pool for the same conditions as Fig. 1. Contours are also
obtained for current density, potential and pressure, as
components of the total calculation.
where Damb is the ambipolar coefficient, K eq (T) is the Saha
function, γ is the three-body recombination coefficient, T is (2) Influence of surface tension on weld pool depth.
the temperature and na is the neutral number density. The It is found that the predicted depth of the weld-pool depends
non-equilibrium electrical conductivity is given by σ = neμ critically on the direction of the convective flow in the weld
where ne is obtained from the solution of the electron pool. The upward flow at the weld pool center, shown in Fig.
2, is normal for cases where the surface tension gets smaller
continuity equation and μ is the electron mobility [2].
with increasing in temperature. However, for high sulfur
steels, or if a surface flux dissolving in the weld pool changes
(3) Boundary Conditions.
the temperature dependence of the surface tension to increase
The calculations assume laminar flow in the arc plasma and
with temperature, the central flow in the weld pool can be
the weldpool, neglect ion currents in the arc and assume that

downwards and the predicted weld depth then can increase by Figure 2: Calculated velocity profiles for the conditions of
a factor of two or more. Fig. 1; from [1].

0 2 4 6 8 10 Fig. 3 shows calculated temperature profiles in the work piece

for a 200 A arc where the gradient in the surface tension
2000 K Ar, 150 A coefficient with temperature, dγ/dT, was taken to be constant
-6 τ = 20s with temperature at a value of -0.6 dyne/(cm K). In this case
2500 K
-4 the convective flow in the weld pool, indicated by the arrows
3000 K
of Fig. 3, is radially outward at the surface of the weld pool.
3500 K
-2 The calculated shape of the weld pool is then very shallow,
1000 K,
Interval 2000 K
being less than 1 mm in depth. The calculated extent of the
Axial distance (mm)

0 17000 K
weld pool is the region having a temperature above 1750 K of
Fig. 1, as the melting point of stainless steel was taken to be
2 1750 K. Arrows indicate the direction of convective flow
within the weld pool.
Fig. 4 shows the calculated temperature contours, again for a
6 current of 200 A, but for the case of the gradient of the surface
1750 K
tension coefficient dγ/dT being 0.6 dyne/(cm K). In this case
8 1500 K the convective flow is radially inwards at the top of the weld
pool, again shown by arrows in Fig. 4, and there is a strong
2000 K
1000 K flow downwards at the centre of the weld pool. This flow
carries heat downwards with the result that the calculated weld
500 K
SUS 304 (LS) depth is significantly larger than that of Fig. 3, i.e. 3.2 mm,
compared with 0.6 mm for dγ/dT = -0.6 dynes/(cm K).
0 2 4 6 8 10
Radial distance (mm)

1750 K
Figure 1: Calculated temperature profiles for a 150 A arc in
argon with a stainless steel workpiece 1 cm thick and an 0.2
electrode-workpiece spacing of 5 mm; from [1]. 1500 K

0.4 1250 K
0 2 4 6 8 10

150 A 1000 K
-6 τ = 20 s
750 K
Max. 201 m/s 0.8

-2 500 K
Axial distance (mm)

0 1.00
0.2 0.4 0.6
Radius; cm
Max. 54 cm/s

Fig. 3 Calculated temperature contours in the work-piece for
200 A, for dγ/dT = -0.6 dyne/(cm K). Arrows show the
6 direction of convective flow in the weld-pool; from [3].


SUS 304

0 2 4 6 8 10
Radial distance (mm)

The calculated voltage for Fig. 5 is 15 V compared with
voltage calculations of 9 V for Figs. 3 and 4. Of course the
diameter that is taken for the extent of the insulating surface
2000 K introduced by the flux simulation of Fig. 5 is just an estimate.
For a larger diameter of the central conducting region, the
1750 K convective flow effects that increase the weld depth would be
1500 K weaker and the increase in the calculated arc voltage due to
the increased current density would also be smaller. In Fig. 6,
1250 K we show calculated temperature profiles as in the case of Fig.
1000 K
5, but for the arc and the tungsten cathode as well. It is noted
from the 16000 K contour of Fig. 6, that there is an arc spot at
750K the centre of the anode, as well as at the high temperature
region in front of the tungsten cathode. This anode arc spot
500 K does not exist in the calculations of Figs. 3 and 4. The spot
1.0 0
from Fig. 6 is introduced because of the high current density
0.2 0.4 0.6
Radius; cm
region forced to be at the centre of the anode by the insulating
layer imposed on the weld pool surface. Such an anode spot is
Fig. 4 Calculated temperature contours in the work-piece for seen experimentally in investigations of ATIG arcs [5], but is
200A, for dγ/dT = 0.6 dynes/cm K. Arrows show the direction not seen in normal TIG arcs.
of convective flow in the weld pool; from [3].
Insulating Layer
(3) Influence of a surface insulating layer. 0

If the surface of the work piece is covered with an insulating

layer, for example an oxide layer or a flux, the arc diameter at 0.2
1750 K

the surface of the work piece is reduced, magnetic pinch 1500 K

forces in the weld pool are increased at the weld pool surface
and the resulting increased pressure can drive convective flow 0.4
downwards leading to a markedly increased weld depth. 1250 K
2000 K
Figure 5 shows the calculated temperature contours where we
1000 K
have simulated the effect of flux placed on the surface of the
metal, by introducing an insulating layer on the surface of the 0.8 750 K
weld pool and metal, except for a circular region of diameter 4
mm at the centre of the weld pool. It is assumed that in the 500 K
high temperature region at the centre of the weld pool, the 1.0
0 0.2 0.4 0.6
insulating flux is evaporated to allow arc current to enter the Radius; cm
weld pool. There is then a much higher current density at the
Fig. 5. Calculations of temperature contours in the weld pool
centre of the weld pool, as the radius of the current region is
where the effect of a flux is simulated by an insulating layer on
now only 2 mm, compared with about 5mm for the weld pool
the outer surface of the weld pool, beyond a radius of 2 mm.
radius in Figs. 3 and 4. This increased current density leads to
Arrows show the direction of convective flow in the weldpool.
a significant j×B force, both in the plasma and in the molten
metal at the surface of the weld pool. Thus there is an
increased pressure at the arc centre, which forces a strong Discussion
downward convective flow in the metal, as indicated by the
arrows of Fig. 5. This downward convective flow carries heat The calculations of weld depth indicate that weld depth can
energy downwards and leads to a significantly increased weld vary by a factor of two or more depending on the surface
pool depth. The weld pool depth of Fig. 5 is 7 mm, compared tension properties of the weld pool surface, or if an insulating
with a weld pool depth of 0.6 mm of Fig. 3 and 3 mm in Fig. layer exists on the liquid surface. These calculations give a
4. possible explanation of the large variability experienced in
weld depth experienced in practical welding. Impurities of
The value of dγ/dT for Fig. 5 was taken to be 0.6 dynes/(cm oxygen or sulfur of as little as 0.01 percent can change the
K), as in Fig. 4. If we set dγ/dT to be -0.6 dynes/(cm K), as in surface tension to be an increasing instead of a decreasing
Fig. 3, the calculated weld depth corresponding to Fig. 5 is 5 function of temperature [6]. Such a change in temperature
mm, still larger than for Fig. 4. There is then a very thin dependence can result in the convective flow in the weld pool
convective circulation of metal outward at the surface of the changing from being radially outward to radially inward,
weld pool, on top of the major circulative flow as in Fig. 5. resulting in large changes in weld depth. Effects of an oxide

layer on the metal being welded could also produce changes in
convective flow and thus a change in weld depth.

200 A; Argon

500 K

1750 K
10000 K
5000 K

16000 K
Insulating Layer

1750 K 500 K

0 0.5 1.0 1.5
Radius; cm

Fig. 6. Predicted temperatures of cathode, arc, weld pool and

workpiece, for the conditions of Fig. 5. An arc spot is evident
at the weld-pool surface.

An explanation can also be given for the large changes in weld

depth observed when the metal surface is first coated with an
activating flux. The flux could dissolve in the molten weld
metal producing a change in surface tension properties and
thus a change in weld depth, as in the previous paragraph. Or
the effect of the insulating layer of the flux can produce an
increase in current density at the centre of the weld and thus
downward convective flow from the magnetic forces and thus
a similar increase in weld depth.


[1] M. Ushio, M. Tanaka and J.J. Lowke, Anode melting from

free burning argon arcs, IEEE Trans. Plasma Science
32,108-117 (2004).
[2] L. Sansonnens, J. Haidar and J.J. Lowke, Prediction of
properties of free burning argon arcs including effects of
ambipolar diffusion, J Phys D: Appl. Phys. 33, 148-157
[3] J.J. Lowke, M. Tanaka and M. Ushio, Arc modeling for
prediction of weld depth variability in TIG welding,
Australasian Welding Journal 47, 2, 33-37 ( 2002).
[4] S.V. Patankar, Numerical heat transfer and fluid flow,
McGraw Hill, New York (1980).
[5] M. Tanaka, T. Shimzu, H. Teresaki, M. Ushio, F. Koshi-
ishi and C.L. Yang, Effects of activating flux on arc
phenomena in gas tungsten arc welding, Sc. & Tech.
Welding and Joining, 5, 6, 397 (2000).
[6] B.J. Keene, Review of data for surface tension of iron and
its binary alloys, Int. Materials Reviews 33, 1-37, (1988).

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Algorithms for Inverse Analysis of Welding Processes

1 2 1
S. G. Lambrakos, J.O. Milewski, D.W. Moon
U.S. Naval Research Laboratory, Washington, DC, USA
Los Alamos National Laboratory, Los Alamos, NM, USA

Abstract calculated by inverse methods represent a mapping from data

space into parameter space. It is therefore preferable to
adopt a parametric function representation whose form tends
An examination of the structure of algorithms for inverse
to minimize any bias resulting from its mathematical form.
analysis of welding and heat deposition processes is
presented. The mathematical foundation of these algorithms Second, a set of parameters associated with a physically
consistent representation can in principle be used to extract
is that a very large class of temperature fields, associated
relationships between parameters, which can provide further
with different types of welding processes, can be represented
insight related to physical characteristics. Third, control and
parametrically by a relatively small set of functions that
optimization of heat deposition processes associated with a
includes linear combinations of solutions to the heat
specific application requires a quantitative assessment of
conduction equation. Issues concerning the use of algorithms
process characteristics over a wide range of values of process
based on different types of parametric representations related
to different types of welding processes are discussed. Our parameters, e.g., beam current, accelerating voltage and
chemical composition of the interacting environment.
emphasis is on the characteristics of algorithms for inverse
System identification for purposes of process control and
analysis as well as general aspects of the inverse problem
optimization is only realizable by specification of a
approach. A prototype analysis is presented that serves to
parametric representation, which establishes a
demonstrate many of the characteristics of these algorithms
correspondence between model and process parameters over
that are significant for their practical application.
a sufficiently wide range of values. And fourth, a
sufficiently general parametric representation can be adjusted
Introduction to include influences due to incomplete information
concerning the system.
According to the inverse-problem approach, a system is
represented by a model and associated set of adjustable In the case of heat deposition processes, such as welding,
parameters (1-5). The particular choice of a model (or parameterizations that are both physically consistent and
equivalently, set of model parameters) is termed a sufficiently general in terms of their mathematical
“parameterization” of the system. The choice of a representation can be defined in terms of a relatively small
parameterization to be used to describe a system, however, is set of basis functions. These basis functions are the modified
in general not unique. In order to address the property of Beer-Lambert law and Gaussian function, for the general
non-uniqueness of system parameterization, inverse problem representation of heat sources associated with deposition-
theory has adopted the concept of “model space,” where each type processes, and solutions to the heat conduction equation
point of this space represents a “conceivable” corresponding to a finite set of boundary conditions.
parameterization of the system (3). Given the model space
of a specific system, quantitative inverse analysis of the
system is further enhanced by isolating the regions of model
Formulation of Inverse-Problem Approach
space that correspond to parameterizations that are physically
consistent, sufficiently general in terms of mathematical Representations of Temperature Field: Presented in this
representation, and optimal in the sense that the number of section is a representation of the temperature field, which
model parameters is the least required for representing the provides a general foundation of inverse methods and of the
system. A physically consistent and sufficiently general concept of a model space for welding processes. A
parameterization of heat deposition processes is significant parametric representation of the temperature field T(x,y,z)
for the following reasons. First, temperature distributions within the workpiece is given by

Copyright © 2006 ASM International® 53

N input (e.g., energy per distance); and (h) information based
T(x, y, z) =  wi fi (x, y, z) , (1) on physical model representations of aspects of heat
i =1 deposition process. The functions fi(x,y,z) provide for the
inclusion of the following types of information: (a) boundary
and T(x k ,y k ,z k ) = C k (2) conditions on the workpiece; (b) estimates of the temperature
field for subsequent optimization based on constraints; and
where the functions fi(x,y,z) are assumed to have trends (c), explicit physical model representation of the process.
consistent with that of the temperature field associated with Selection of a well posed set of fi(x,y,z) is based on
the specific welding or heat deposition process. The determining the characteristic model space of welding
constraint conditions are represented by the quantities Ck. processes and subsequently isolating those regions of the
model space corresponding to parameterizations that are both
The formal procedure underlying the inverse method physically consistent and sufficiently general in terms of
considered here entails the adjustment of a temperature field their mathematical representation. It follows that for heat
T(x,y,z) defined over the entire spatial region of the deposition processes f(x,y,z) is given by
workpiece. This approach defines an optimization procedure
where a temperature field spanning the spatial region of the f(x,y,z) = C1 (x,y,z) C 2 (x,y)TB (x,y,z) (4)
workpiece is adopted as the quantity to be optimized.
Depending on the characteristics of the spatial distribution of where the functional forms of C1, C2 and TB are the Beer-
constraint information, the temperature field is represented Lambert law, Gaussian function, and solutions to the heat
parametrically by Eq.(1). Accordingly, an optimization conduction equation for a given set of boundary conditions,
criterion may be defined by minimizing the value of an respectively. This follows for C1(x,y,z) since deposition-type
objective function of the form processes,(e.g., transmission of electrons or photons) can be
represented with respect to penetration by the modified Beer-
N1 Lambert law, which given in Napierian form is
Z =  w n ( T(x ,y ,z )  T
n )
n =1 I 4

N2 2 C1 (x,y,L) = -ln  = μCLf1 + f 2 =  i Li (5)

m =1
+  w m max{T(x,y cm ,z cm )}  Tmc ) (3) i= 0

where I0 is the incident intensity at the origin o, I is the

intensity at a distance L from the origin, μ is the extinction
where wn and wm are weighting coefficients and the coefficient, C is the concentration of the ambient medium, f1
quantities having superscript “c” designate constraints on is a pathlength factor, which accounts for increases in
temperatures or positions within the workpiece. The pathlength caused by scattering within the material, and f2 is
mathematical structure underlying the general formulation a geometry factor, which accounts for instrument geometry,
defined by Eqs.(1)-(3) is that of a parametric function- e.g., shape or spatial profile of beam source. The general
approximation, using a linear combination of nonlinear functional form of the Beer-Lambert law (see Fig.1) is that of
functions. This structure establishes an equivalence between a monotonically decreasing function that is characteristically
represented by five parameters (e.g., right side of Eq.(5)).
inverse methods and methods of constrained parameter
Similarly, a sufficiently general representation of the
optimization. Reference 5 examines the formal equivalence
transverse character of heat sources C2(x,y), is that of a
of inverse methods for heat conduction problems and the Gaussian function, i.e.,
least-squares method. Reference 6 examines formulations of
the inverse problems in terms of optimization. The
mathematical structure of the formulation defined by eqns C 2 (x,y) = Aexp[-1 (x - x 0 ) 2   2 (y  y 0 ) 2 ] . (6)
(1)-(3) provides for the inclusion of information either
through specification of constraints Ck or the functions

The constraints Ck provide for the inclusion of the following


types of information: (a) solidification cross-sections (e.g.,

transverse, longitudinal and top surface cross-sections); (b)
spatial character of energy source (e.g., position of maximum
temperature, shape and relative location of keyhole in deep-
penetration welds); (c) geometric information (e.g., shape
features of workpiece and top surface of weld); (d) boundary
r0 r1 r2 r3 r4
conditions on workpiece; (e) information related to
temperature history (e.g., microstructure correlation with
temperature); (f) thermocouple measurements; (g) energy

Fig. 1 General form of basis function C1(x,y,r) defined in where
Eq.(5). C(xk ,y k ,z k )  (x - x )2 
Tk (x k ,y k ,z k ,nt) = TA + exp - k 
Although heat deposition processes may be characterized by nt 
 4 1 (nt) 
complex coupling between the heat source and workpiece, as     m 2  2 (nt)  my  myk 
well as complex geometries associated with either the  1 + 2 exp  - 2 2
 cos   cos 
workpiece or deposition process, (e.g., reinforcement  m=1

structures), in terms of inverse analysis, the general
functional form of the temperature fields associated with all
such processes is within a restricted set of functions.     m 2  2 (nt)  mz  mz 
Accordingly, a sufficient approximation of this restricted set  1 + 2 exp  - 3 2
 cos  cos k




of functions are the solutions to heat conduction equation
TB(x,y,z) for the set of boundary conditions given in Fig. 2.
A parameterization based on this set is both sufficiently
general and convenient relative to optimization. where TA is the ambient temperature of the wire structure
and the time t = N t t ( N t = 1, 2, 3, …) is expressed as N t
multiples of discrete time steps t . The coefficients
C(xk,yk,zk), which specify the spatial distribution of discrete
surface heat elements, and the diffusivities  1 ,  2 and  3
are adjustable parameters of the inverse model defined by
Eq.(7). The quantities a and l are the lengths of the sides of
the rectangular cross section of the wire structure. In general
the wire structure is characterized by an anisotropic thermal
diffusivity such that the “through-thickness” diffusivities,
i.e.,  2 and  3 , are significantly less than the “in-plane”
diffusivity  1 . The spatial coordinates (xk,yk,zk) and (x, y, z)
are the discrete locations of the heat elements on the surface
of the model wire structure and positions within the wire
structure at which the calculated temperature is defined,
respectively. The solution to the heat conduction equation
given by Eq.(7) represents a modification of that given in
reference 7.
Fig. 2 Boundary conditions for basis functions TB(x,y,z)
defined in Eq.(4). The wire structure considered for the present analysis was a
silicone rubber coated carbon fiber having a rectangular cross
Case Study Analysis section, a = 0.125 in and l = 0.25 in. Surface heating of this
wire structure was achieved using an electric hot air source
For the present case study we consider an analysis of a whose maximum air temperature was 600 oC. The heating
prototype wire structure of finite cross section whose surface pattern, which appeared white, is correlated with the
is heated over a given period of time by a heat source whose formation of a silica coating resulting from decomposition of
spatial distribution is that of a Gaussian function. Although the silicone rubber coating of the wire structure. This pattern
we adopt as input data for this analysis thermocouple was adopted for assigning the spatial distribution of surface
measurements of temperature histories for a grapite-fiber heat elements, i.e., the values of coefficients C and set of
wire structure, the prototype wire structure modeled is positions (xk,yk,zk) defined in Eq.(7). For the purpose of our
representative of a wide range of different types of material prototype analysis two thermocouple measurements were
response characteristics that can be associated with heat made during the heating period. One thermocouple was
deposition processes in general. In particular, the analysis attached within the center of the wire structure, below the
presented can be applied to processes involving spot welding point of maximum heating, while the other was attached on
and heat treatment of complex structures. the surface of the structure at the point of maximum heating
from the hot air source. Shown in Fig. (3) are temperature
Heating of a Wire Structure: For unsteady heat deposition histories corresponding to thermocouple measurements at
within a wire structure of finite cross section a consistent these two positions, on and within the graphite-fiber
parametric representation of the time dependent temperature prototype wire structure.
field is
The inverse model defined by Eq.(7) is function of the
Nk Nt
adjustable parameters C(xk,yk,zk),  1 ,  2 ,  3 and t . For
T(x, y,z, t) =   T (x ,y
k k k ,z k , nt) (7) the purpose of our prototype analysis we consider only
k=1 n=1
adjustment of the parameters C(xk,yk,zk) and  1 . This is

consistent with our use of two thermocouple measurements assigned according to their relationship to  1 .
of temperature histories. The values of  2 and  3 are
Fig. 3 Bottom and top “coarse” curves are for experimentally measured temperatures at center and surface, respectively. Bottom




and top “smooth” curves are for simulated temperatures at center and surface, respectively. Parameter values are given in Fig.4.

Fig. 4 Simulations of heat transfer within the prototype wire structure corresponding to the three different modes of heat
2 -3 2
conduction. (a)  1 = 7.7 x 10-4 m /s,  2 =  1 ,  3 =  1 , adjusted according to Fig.3 (top); (b)  1 = 3.0 x10 m /s,  2 = 0.27  1 ,
 3 =  2 , adjusted according to Fig.3 (middle); and (c)  1 = 3.7 x10 -3 m 2 /s,  2 = 0.0176  1 , adjusted according to Fig.3 (bottom).

This relationship is specified according to the mode of heat presented an examination of the foundation of algorithms for
conduction occurring within the wire structure. We consider inverse analysis of heat deposition processes such as
three modes of heat conduction, purely isotropic, moderately welding. These algorithms can be based, in principle, on a
anisotropic, and anisotropic heat conduction (typical of relatively small set of mathematical functions, or
laminate wire structures). A typical value of  2 /  1 for a equivalently, parameterizations. From the perspective of
graphite/epoxy laminate is 0.01757, which we have adopted inverse problem theory, the fact that a given process can in
for our prototype analysis. The discrete time step t and general be represented by a relatively small set of functions
that is both physical consistent and convenient with respect
total number of time steps N t were 0.4865 s and 1000,
to parameter optimization is based on the concept of a
respectively. “model space” for a given system. According to this concept,
although parameterizations for inverse analysis are not in
Shown in Fig.(3) are calculated temperature histories, at the general unique, there does exist for any given system an
locations of thermocouple measurement, corresponding to optimal inverse model representation based on its physical
three different modes of heat conduction within the prototype characteristics. This concept is discussed further in reference
wire structure. For each case the values of C(xk,yk,zk) and 3. The case study presented demonstrates the use of a
 1 have been adjusted so as to achieve the best relatively optimal parameterization for inverse analysis of the
correspondence of the calculated and measured temperature heating of complex wire structures and further demonstrates
histories over the entire time interval extending over 500 s. the general flexibility and convenience of the inverse-
The spatial distribution of C(xk,yk,zk) was that of the problem approach for analysis of complex processes
decomposition pattern such that the discrete locations involving heat deposition.
(xk,yk,zk) were at a grid spacing l = 1.27 x 10 m. Shown
in Fig.4 are simulations of heat transfer within the prototype Acknowledgements
wire structure over a period of time prior to the assumed
point of thermal breakdown corresponding to the three The authors would like to thank the Office of Naval
different modes of heat conduction: Research, Arlington, VA for support of this research.
 1 = 7.7 x 10-4 m /s,  2 =  1 ,  3 =  1 ; References
-3 2
 1 = 3.0 x 10 m /s,  2 = 0.27  1 ,  3 =  2 ; 1. S.G. Lambrakos and J.O. Milewski, “Analysis of
and Processes Involving Heat Deposition using Constrained
-3 2
 1 = 3.7 x 10 m /s,  2 = 0.0176  1 ,  3 =  2 . Optimization,” Science and Technology of welding and
Joining, 7 (3) 137, 2002.
The time dependent temperature fields shown in Fig.4 are for
the xz-plane at midpoint of the y axis. The three simulations 2. S.G. Lambrakos and J.O. Milewski, “Analysis of
shown in Fig.4 are for the same spatial distribution of heat Welding and Heat Deposition Processes Using an
Inverse-Problem Approach,” to appear in Mathematical
sources. The values of the parameters C(xk,yk,zk),  1 ,  2 ,
Modeling of Weld Phenomena 7 (ed H. Cerjak).
 3 and t have been adjusted according to the inverse
model defined by Eq.(7) with respect to this specific heat 3. A. Tarantola: “Inverse Problem Theory and Methods for
source. These parameters will, in principle, provide an Model Parameter Estimation,” SIAM, Philadelphia, PA,
estimate of the response of the wire structure to heat sources 2005.
having different spatial and temporal characteristics. The
level of anisotropy is specified according to the ratio  1 /  2 4. K.A. Woodbury, editor: “Inverse Engineering
(where  3 =  2 ). In practice, this quantity may be assigned Handbook,” CRC Press, New York, 2003.
according to the temperature history measured by an 5. M. Rappaz, M. Bellet and M. Deville: “Numerical
additional thermocouple measurement at a different location Modeling in Materials Science and Engineering,”
along the x-coordinate. One can, in principle, adjust both Springer-Verlag, Berlin, 2003, pp. 448-475.
 1 /  2 and  1 /  3 according temperature histories
measured by two additional thermocouples. In the case of 6. C.R. Vogel, “Computational Methods for Inverse
more than three measured temperature histories, the model Problems,” SIAM, Philadelphia, PA, 2002.
parameters should be optimized following a least-squares
procedure (see references 5 and 6). 7. H.S. Carslaw and J.C. Jaegar: “Conduction of Heat in
Solids,” Clarendon Press, Oxford, 2nd edn: 1959, p. 374.
Summary and Conclusions
General Aspects of the inverse problem approach for a range
of different welding and heat deposition processes have been
presented in more detail elsewhere (1,2). Here, we have

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Improving the Understanding of Laser Deposition Processes

Through Process Simulation
R. P. Martukanitz1, A. C. Naber2, R. M. Melnychuk3, and R. W. McVey3
Applied Research Laboratory, 2Department of Mathematics, and 3Department of Engineering Science and Mechanics
Pennsylvania State University, University Park, PA 16803

As stated, laser surface modification is a versatile process
The laser surface modification process, as described using pre- through which one can modify the surface properties of metals
placed powder, is typified by a relatively diffuse energy input and ceramics to obtain improved component performance.
from a laser beam to metal powder residing on the surface of a This is accomplished through modifying the material’s
metal substrate. Based on this description, a model properties by localized melting and solidification on or near
representing the laser deposition process using pre-placed the surface of the substrate. Many of the laser surface
powder has been constructed. In the computation, special modification processes utilize powders as well as solid
attention is given to the interaction of laser radiation with the material in the form of wire for depositing material. When
powder bed. Initial results of simulations under a relatively high deposition rates are required, the powdered materials are
wide variety of conditions indicate the utility of the model for typically fed into the interaction area directly ahead of a
improving the basic understanding of the laser deposition diffuse laser beam, and may be described as pre-placing the
process. powder onto the substrate. This process is illustrated below in
Figure 1.

Introduction Diffuse Beam

There is a continued need for new and improved coatings that

provide increased life for critical mechanical components.
This is particularly true for heavy industries such as mining
Powder Feed
and mineral processing, primary material production and Nozzle
fabrication, paper and pulp, and agriculture. Typically,
coating technology utilized for these industries requires
metallurgical bonding for durability, high deposition rates
affordable restoration, and hard surfaces for maintaining high
wear resistance. Laser surface modifications, in the form of Deposited Layer
cladding and alloying are gaining widespread use because of
its ability to provide high deposition rates, low thermal
distortion, low base-metal dilution, low metallurgical Substrate
degradation of the base material, and refined microstructures
in the deposition due to high solidification rates. A variety of Figure 1. Schematic of laser deposition process using pre-
applications have been developed for the laser surface placed powder.
modification process, and the most common uses are for
improving resistance to corrosion, abrasion, erosion,
oxidation, and wear. Since the properties of the modified
surface depend on the microstructure that evolves during Model of the Laser Deposition Process
melting and cooling, an accurate depiction of the thermal
history associated with the process can play an important role The laser surface modification process, as described using pre-
in developing improved coatings and advanced processes. placed powder, is typified by a relatively diffuse energy input

Copyright © 2006 ASM International® 59

from a laser beam to the metal powder residing on the surface These relationships were approximated through the explicit
of a metal substrate. The powder is melted along with a small finite difference method involving a three-dimensional
amount of substrate material, metallurgically bonding the computational space. A schematic of two dimensions of the
powder layer to the substrate, thus creating a new surface. space, along with a Gaussian energy input, is shown in Figure
The energy balance for such a system may be represented as: 2. Also illustrated in the figure is the variable grid spacing
that was employed to represent depth and width and the
∂H powder layer, shown as the shaded portion of the
= ∇ ⋅ (k∇T ) + Q (1) computational space. A fine node spacing of 0.1 mm, which
∂t was used for computation within the vicinity of the powder
layer that was exposed to the beam and then melted, was
where Q represents energy introduced to the powder and required to accurately depict the energy attenuation within the
substrate by the laser and surface heat losses, while enthalpy is powder layer.
utilized to account for latent heat and may directly describe
temperature within the computation space through:

H = ρ (C P T + L) (2)

Although standard boundary conditions are applied, such as

surface losses due to radiation and convection, unique
properties of the computation involve the spatial distribution
of energy and absorption associated with the laser source and
the attenuation due to scattering to finite depths within the
powder layer. For a heat source having some intensity (I)
distribution, which may or may not be Gaussian:

dΔI - γ 2 I = 0 in Ω (3)

where Ω represents the powder domain γ represents the

absorption coefficient due to attenuation within the powder Figure 2. Schematic of computational space in the x and y
layer. Letting L(x) be the spatial distribution of the laser plane for the finite differencing model.
energy, within the powder domain that is addressed by the
beam, I(x) is equal to L(x) ; whereas, within the powder
domain that is outside the area of laser interaction, I(x) is Model Parameters
equal to 0 . Hence, γ if is known, the intensity within the
Several parameters that are critical to the accuracy of the
powder layer may be described as:
model must be discussed. This includes the variation of
thermal conductivity and density of material in the powder,
∫ I(x)dx = Qin (4) liquid, and solid state, and absorption of laser energy within
Ω the porous powder layer. As described earlier, the dramatic
difference in thermal conductivity between a material in
where Qin is the total laser energy provided to the substrate. powder and solid state, which can be two orders of magnitude,
coupled with the decreased density of powder, which may be
Another important factor specific to the description of the reasonably approximated at 60-percent of the bulk material,
laser deposition process involves the large difference in results in a significant variation of thermal diffusivity between
thermal diffusivity associated with the porous powder layer, the powder, melted or consolidated powder, and the substrate.
the melted powder, and the metallic substrate. For positions
on the boundary that is separating materials reflecting this Various relationships were evaluated for describing the
difference, i.e. the interface between the powder and substrate, effective conductivity of the powder in terms of the
conservation of energy requires1: conductivity of the bulk material comprising the powder;
however, the relationship by Batchelor, et al. was found to
provide values within the range of other relationships, while
k1 (∇T)i,1 = k1 (∇T)i,2 (5)
also providing the most accurate simulation results. This
relationship determines the effective conductivity of the
powder based on the conductivity of the bulk material

comprising the powder and the thermal conductivity Results of Simulations and Discussion
representing the interstitial voids, assumed to be air2:
Various simulations were conducted to ascertain the
2kair ln(k powder /kair ) sensitivity of the model to processing conditions. Shown in
k= [ − 1] (6) Figure 3 are cross-sectional isotherms for laser irradiation of
1 − (kair /k powder ) 1 − (kair /k powder ) iron powder with 3.0 kW of power using a Nd:YAG and CO2
laser. The circular heat source has a Gaussian distribution, is
The amount of energy absorbed during laser irradiation of 5.0 mm in diameter, and is moving at a rate of 8.47 mm/s over
powder layers is extremely important in accurately a 2.0 mm thick powder layer pre-placed on an iron substrate.
representing the process in numerical simulations. When laser The isotherms represent the mid-plane of the powder layer and
energy is directed onto a powder layer, the reflectance of the substrate with the center of the beam slightly ahead of the
beam is confined to the particles near the top surface and a mid-plane. Temperatures above the melting point of the iron,
fraction of the laser radiation propagates through the powder 1808K, are shown in white.
layers by way of the void spaces between particles. The
radiation is then thought to scatter (produce secondary
radiation) through the thickness. This results in significant
“penetration” into the powder and an increased absorption,
when compared to solid materials, and although it has long
been recognized that the use of powder significantly improves
the absorption of the laser beam3-6, data that accurately
describes this phenomena is virtually nonexistent.

Experimental data was recently obtained through closely

controlled experiments designed to measure the amount of
reflectance, transmission, and absorption of CO2 (wavelength
of 10,600 nm) and Nd:YAG (wavelength of 1,064 nm)
irradiation. These experiments utilized an integrating sphere
to measure the optical response during irradiation of pure iron
or pure copper powder at varying bed thickness, as well as
various powder size distributions. In all cases, the degree of Figure 3. Isotherms at the mid-plane during laser deposition
absorption within the powder followed an exponential decay, of a 2.0 mm thick pre-place iron powder layer using 3.0 kW of
which could then be used to estimate the absorption power with Nd:YAG (top) and CO2 (bottom) lasers.
coefficient. The estimated absorption coefficients (γ ) for the
pure iron and pure copper powders from these experiments are
shown in Table 1. It should be noted that the values shown in As shown in Figure 3, the energy density associated with the
Table 1 are up to 4 orders of magnitude larger than those 3.0 kW over the 5 mm diameter spot is not sufficient to cause
obtained for solid materials. melting into the substrate when using the CO2 laser; whereas,
the Nd:YAG laser was easily able to achieve full melting of
the substrate. This large difference in thermal input is chiefly
due to the absorption of the iron powder at the CO2 and
Table 1: Estimated absorption coefficients for three size Nd:YAG laser wavelengths. The absorption of iron with
distribution of pure iron and pure copper powder for Nd:YAG Nd:YAG radiation is nearly twice that of CO2 radiation6.
Shown in Figures 3 and 4 are the time dependent temperatures
at the top surface of the powder and the original interface
Powder Fe Fe Fe Cu Cu Cu
Material 44-149 > 106 < 45 44 -149 >106 < 45
position between the powder layer and the substrate for the
and Size μm μm μm μm μm μm Nd:YAG and CO2 laser irradiation, respectively.

Nd:YAG Shown in Figure 5 are results for laser deposition of a cobalt-

0.7802 0.5443 0.0365 0.0204 0.8803 0.1086
γ value based powder using 4.0 kW of power distributed over a 12
mm by 1 mm area and assuming a wavelength similar to a
0.0193 0.0130 0.0258 0.0158 0.0092 0.0253 Nd:YAG laser. The long axis of the heat source is
γ value perpendicular to the traverse direction. In this simulation, an
absorption of 0.6 was utilized for the powder and an
absorption of 0.45 was employed to represent the consolidated

431SS Powder on 4140 with 3kW Nd:YAG
Because of the attempt to accurately represent the interaction
of the laser energy with the powder bed, the model has also
shown the ability to illustrate the dynamic nature of the laser
6000 deposition process. Shown in Figure 6 is a top view of a
Powder Surface
Substrate Surface simulated 1.5 kW Nd:YAG beam irradiating a metal powder
bed. Based on experimental measurements, the powder bed is
Temperature (K)

4000 assumed to have an absorption value of 0.55 and the

consolidated or melted powder represents an absorption of
0.486. Upon scrutinizing the resultant temperatures, it is
2000 theorized that an area of high absorption near the leading edge
of the beam may be the controlling factor in powder melting
and deposition, and this phenomena is driven by the
0 interaction of the laser irradiation with the virgin powder.
0 1 2 3 4 5 6
Time (s)

Figure 3. Time dependent temperatures at the top surface of Powder and two
the powder and the original interface position between the phase region
powder layer and the substrate for surface deposition using a under beam
Nd:YAG laser.

431SS Powder on 4140 with 3kW CO 2

Area under
the beam
Powder Surface
Substrate Surface
Temperature (K)


Figure 6. Top view of simulation of laser deposition of a
2000 nickel-based powder bed with a 1.5 kW Nd:YAG laser.

0 1 2 3 4 5 6
Time (s)

A model representing the laser deposition process using pre-

Figure 4. Time dependent temperatures at the top surface of
placed powder has been constructed. In the computation,
the powder and the original interface position between the
special attention is given to the interaction of laser radiation
powder layer and the substrate for surface deposition using a
with the powder bed. Initial results of simulations under a
CO2 laser.
relatively wide variety of conditions indicate the utility of the
model for improving the basic understanding of the laser
deposition process.


Research sponsored by the Department of Energy’s

Industrial Materials for the Future Program under
contract DE-FC07-02ID14247. The authors would also
like to acknowledge the contribution of Mr. Jay Tressler
in supporting various experimental tasks described in
Figure 4. Isotherms at the mid-plane during laser deposition this work.
of a 2.0 mm thick pre-place iron powder layer using 3.0 kW of
power with Nd:YAG (top) and CO2 (bottom) lasers.


1. Barbašin, E. A. Approximation of the Solution of

Partial Differential Equations. American p. 463,
Mathematics Society, Providence, RI (1963).

2. Ba tchelor, G.K., O’Brien, R.W., Thermal or

Electrical Conduction Through a Granular Boundary,
Proceedings of the Royal Society of London. Series
A, Mathematical and Physical Sciences, 335, 313-
333 (1977).

3. Haag, M., Hugel, H, Albright, C.E., and Ramasamy,

S., CO2 Laser Light Absorption Characteristics of
Metal Powders, J. Applied Physics, 79 (1995).

4. Bohren, C. F., and Huffman, D. R., Absorption and

Scattering of Light by Small Particles, John Wiley
and Sons, Inc. (1983).

5. Tolochko, N.K., T. Laoui, Y.V. Khlopkov, S.E.

Mozzharov, V.I. Titov, and M.B. Ignatiev,
Absorptance of powder materials suitable for laser
sintering. Rapid Prototyping Journal., Vol. 6, 217-
223 (2000).

6. Martukanitz, R.P., Melnychuk, R.M., and

Copley, S.M., Dynamic Absorption of a Powder
Layer, Proceedings of the 23rd International
Congress on Applications of Lasers & Electro-
Optics, Laser Institute of America, 97, 1404-
1409 (2004).

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

A Microstructure Model for Laser Processing of Ti-6Al-4V

S.M. Kelly
Applied Research Laboratory, The Pennsylvania State University, State College, Pennsylvania, USA

S.S. Babu, S.A. David, T. Zacharia

Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA

S.L. Kampe
Materials Science and Engineering Department, Virginia Tech, Blacksburg, Virginia, USA

Abstract Figure 1. The layer band consists of a colony-α morphology,

while the material between the layer band is a distinctly
The current challenge in laser processing titanium alloys using different morphology consisting of basketweave-α. A region
methods such as Laser Metal Deposition (LMD) is in of material containing the colony-α layer band and adjacent
understanding the complex microstructure evolution during scale-graded basketweave-α comprise a fully developed
multiple passes of the laser. The microstructure is affected by characteristic layer. The characteristic layer is found in each
the repeated thermal cycling that occurs during the deposition deposited layer except for the last three to be deposited. The
process. The current work focuses on the thermal and last three layers of the deposit are in an undeveloped state
microstructural modeling of multilayered Ti-6Al-4V deposits. indicating that the characteristic layer forms in a layer n after
Prior work with LMD-Ti-6Al-4V has shown that a complex the addition of three layers of material (n+3th layer
microstructure evolves consisting of a two-phase α+β deposition).
structure that is measurably different across the deposit. A
microstructure model has been developed to predict the
evolution of the alpha fraction during thermal cycling. Alpha
dissolution and growth rates were obtained using
computational thermodynamics and diffusional phase
transformation software as well as available TTT diagrams.
The results indicate that during the n+3 layer addition, the
material in layer n will experience the greatest change in
evolution path. The results of the microstructure model will
be discussed in relation to the as-deposited microstructure.

One of the current challenges in laser processing titanium
alloys (Ti-6Al-4V) using methods such as Laser Metal
Deposition (LMD) is in understanding the complex
microstructure evolution during multiple passes of the laser. It
has been shown that the microstructure evolution is affected
by the repeated thermal cycling that occurs as multiple layers
of material are deposited.1-3 The periodic movement of the
heat affected zone (defined by the depth of the beta transus)
results in a similar variation in the two phase α+β phase
fraction and morphology. An example of the variation of Figure 1: Macrograph of a laser deposited Ti-6Al-4V cross-
microstructure within the deposit is shown in Figure 1. section showing the last 6 layers of an 18 layer deposit. Layer
bands (“LB”) are highlighted. Layer bands are periodic
Previous work has focused on the characterization2 of the regions of colony-α and are present for every layer except for
periodic microstructure that evolves in laser deposited Ti-6Al- the last three layers. The material between layer bands is
4V. Particular emphasis was placed on understanding the basketweave-α . In this cross section, x is the direction of
origin of the banded structure seen in the macrograph of laser travel, and z is the direction of layer addition.

Copyright © 2006 ASM International® 65

The focus of this paper will be to model the variation of alpha
phase fraction during the deposition of multiple layers of Ti-
6Al-4V using a commercial layer deposition process.4 In
addition, existing theories2,5-7 as to the thermal cycle
responsible for the for the formation of variation in
microstructure will be reinforced with the microstructure
model results.

The microstructure model seeks to describe the evolution of
alpha phase fraction with temperature. Ti-6Al-4V is a two
phase, α (HCP) + β (BCC), at room temperature. Above
approximately 1273K (beta transus tempertaure) the beta
phase is stable. The model will consider the nucleation and
growth and dissolution of the alpha phase seperately. Figure 2: Modeled (single transformation curve) and JMatPro
calculated TTT diagram for grain boundary and intragranular
Alpha nucleation and growth nucleation and growth.
During cooling through the beta transus, alpha nucleates first
at the beta grain boundary, typically as allotriomorphic
sideplates (colony-α). At further undercooling, the alpha Alpha dissolution
phase appears to nucleate intragranularly as basketweave-α. Alpha completely dissolves to beta under slow heating
The alpha phase that froms during cooling through the beta conditions as the temperature rises above TBETA = 1273K. The
transus is designated primary-alpha (αP). The isothermal alpha dissolution reaction occurs by movement of the alpha
beta interface, which is caused by the transport of beta
nucleation and growth of the α phase can be described by
stabilizing elements across the interface.17,18 In this sense
classical Johnson-Mehl-Avrami-Kolmorgorov8-10 (JMAK)
alpha dissolution can be thought of as diffusion controlled beta
kinetic equation:
growth. The dissolution of alpha / growth of beta is modeled
assuming a parabolic beta growth rate (α*(T)) determined
ζ = 1 − exp(−k (T )t n ) , (1) from isothermal diffusion controlled beta growth simulations
as shown in Eq. 2.
where the function k(T) describes the kinetics (nucleation and
growth rates) and is strongly dependent on temperature, while ζ = α * (T )t1 2 (2)
n describes the nucleation/growth mechanism of
transformation taking place and is independent of temperature.
Kinetic parameters for alpha dissolution in Ti-6Al-4V have
JMAK kinetic equations have been derived for Ti-6Al-4V by been derived, based on JMAK kinetics, by Elmer, et al19 from
Malinov, et al11; however, the differential scanning time-resolved x-ray diffraction experiments; however, when
calorimetry experiments used to calculate k(T) were performed applying the resulting kinetic parameters to continuous heating
at cooling rates much slower than those expected in laser models, it is predicted that melting will occur before complete
deposition processes and are not applicable here. Thus, k(T) is dissolution of the alpha phase for heating rates greater than
calculated from existing time-temperature transformation 100K/s. Hence the validity of this data is questioned.
diagrams. The material property software, JMatPro12-14 is
used to calculate the TTT diagram for an alloy composition Briefly, α*(T) is calculated using diffusion controlled
Ti-6.00Al – 3.99V – 0.25Fe – 0.20O – 0.02C – 0.01N (wt%) transformation (DiCTra) software20 and ThermoCalc solid
using the titanium alloy database (Ti-DATA).15 The JMatPro solution (SSOL2)21 and mobility (MOB2)22 databases. A
calculated TTT diagram contains transformation curves for diffusion cell is set-up to represent the room temperature alpha
grain boundary and intragranularly nucleated alpha, e.g. two and beta phase composition and fraction that is typically found
equations (k(T), n) for each morphology.13,16 In the current in Ti-6Al-4V (91%- α, 9%-β). The beta thickness is
model k(T) is determined for a single start curve, assuming n = approximately 100μm, after TEM observations of the as-
5/2 as detailed in Ref [1]. It should be noted that k(T) is a deposited material.1 The diffusion cell is instantaneosly
nonlinear function with temperature and is incorporated into heated to a temperature, Ti, below the beta transus and the
the alpha phase fraction calculation using numerical alpha phase is allowed to dissolve. α*(Ti) is the slope of the
interpolation techniques. The resulting TTT diagram showing thickness versus time1/2 plot for the isothermal temperature.
the JMatPro calculated and modeled transformation start and The best-fit equation describing the parabolic thickening rate
finish curves is shown in Figure 2. as a function of isothermal temperature found to be:

1/ n
⎛ ⎛ ⎞⎞
α D* (T ) = 2.2 * 10 −31
T 9.89
(3) ⎜ ln⎜1 − fα ,i −1 ⎟ ⎟
⎜ ⎜ fα ,eq (Ti ) ⎟⎠ ⎟
tG* ,i = ⎜ ⎝ ⎟ (10)
Further details of this calculation may be found in Ref [1]. ⎜ − k (Ti ) ⎟
⎜ ⎟
⎜ ⎟
Application to non-isothermal heating and cooling ⎝ ⎠
The aforementioned methods for describing alpha nucleation fα ,i = fα ,eq (Ti )ζ α ,i (11)
and growth and alpha dissolution are developed for isothermal
heat treatments. In other words, the methods are theoretical The subscript i refers to the current time step, i-1 refers to the
tools for predicting time-temperature transformation (TTT) previous time step, D refers to dissolution, and G refers to
diagrams. In order to apply the above methods to non-
growth. The variable ζ refers to the extent of reaction. The
isothermal treatments, such as the LMD thermal cycles, one
flow of the phase fraction model calculations begins by
must integrate the kinetic equations over the thermal cycle.
checking the alpha fraction against the equilibrium fraction for
The principle of additivity23-26 is applied to accomplish this
the current (ith) temperature and time step. Based on whether
task: to predict microstructure evolution for non-isothermal
dissolution or growth is occurring (Eq. 4), the appropriate
heat treatments based on isothermal experiments.
dissolution (Eqs. 5-8) or growth (Eqs. 9-11) equations are
evaluated for the current temperature and time step. The
Now consider a thermal cycle where alpha dissolution and
alpha growth occur during heating and cooling, respectively. phase fraction of alpha (fα,i) is stored for each time step. This
The thermal cycle has been discretized into many small time process is repeated each time step in the entire thermal cycle.
steps, defined by Ti and ti. The equilibrium phase diagram Further details of the phase fraction model may be found in
fraction of alpha determines the appropriate sub model Ref[1].
(dissolution or growth) that is used for the ith time step. The
model will utilize the guidelines in Eq. 4, where fα,eq and fα,i
are the equilibrium and current alpha fractions. The
equilibrium alpha fraction is calculated using ThermoCalc Results and discussion
v.P27 and the Ti-DATA thermodynamic database. The
appropriate sub-model is chosen based on the equilibrium and Review of thermal model results
current α fractions for the current temperature and time step
(i). The results presented herein are based on calculated thermal
cycles for the laser deposition of Ti-6Al-4V; however, the
fα ,i > fα ,eq Alpha Dissolution above equations may be utilized in any non-isothermal heating
and cooling sequence. The thermal model utilizes implicit
fα ,i < fα ,eq Alpha Growth (4) finite difference techniques to solve two-dimensional heat
fα ,i = fα ,eq Equilibrium conduction equations, assuming a volumetric distribution of
heat input caused by the deposition of each layer. Further
Alpha dissolution is described by Eqs. 5-8: details of the thermal model may be found in Refs [1,3,28].
Figure 3 shows the calculated maximum temperature during
( )
⎧⎪α * (T ) Δt 1/ 2 the eighth layer addition in the modeled deposit in the form of
(i ) −(i −1) + t D ,i 0 < Δt (i )−(i −1) + t D* ,i ≤ t D,i ,crit
ζ β ,i = ⎨ D i (5) a contour map. Also shown is a cross section of the deposit,
⎪⎩ 1 Δt (i )−(i −1) + t D* ,i > t D ,i ,crit with similar regions of the thermal modeled and as-deposited
t D , i , crit = α D* (Ti ) )−2
microstructure highlighted. In the thermal model results,
discussed in detail elsewhere1, it was observed that after the
2 third layer is deposited a heat affected zone develops that
⎛ f β ,i −1 ⎞
t D* , i = ⎜ ⎟ (7) extends approximately 4 layers into the deposit. With each
⎜ f β , eq (Ti )α D (Ti ) ⎟
additional layer deposition the heat affected zone increments 1
⎝ ⎠
layer in the positive z direction, which in theory should result
fα ,i = 1 − f β ,i = 1 − f β , eq (Ti )ζ β ,i
(8) in an evolution of microstructure that is periodic, as observed
in the as-deposited material. The periodic characteristic layer
Alpha growth is described by Eqs. 9-11: containing the graded basketweave and layer-band (colony α)
is designated layer n and forms during the deposition of layer

ζ α ,i = 1 − exp ⎡⎢− k (Ti ) Δt(i ) − (i −1) + tG* , i ⎤⎥
)⎦ (9)
n+3. The last three layers (n+1 through n+3) consist of a
microstructure that is uniquely different from n, n-1,...n-j and
define the heat-affected-zone microstructure in the deposit.
Layers below n, e.g. n-1 are unaffected by thermal cycles after
layer n+2.

The results of the microstructure model will focus on the
phase fraction and morphological evolution in the last 5 layers
of material, which will provide evidence that the characteristic
microstructure in layer n develops during the deposition of
layer n+3 and remains unaltered by the deposition of layer
n+4 as originally hypothesized.2

Phase fraction model results

The discussion of the phase fraction evolution begins with the

last five layers to be added in the modeled deposit. These are
layers 4 through 8, where layer 4 can be thought of as layer
“n” and layer eight would be layer “n+4”. In Figure 4 the
alpha phase fraction evolution with time is shown at two z-
positions along the centerline (y=0) of the deposit, within
layer 4 (or layer n). The thermal cycles these two points
experience are also shown. The periodic increases in
temperature are caused by the addition of molten layers of
Figure 3: Comparison of modeled maximum temperature with material to the deposit, which eventually cool to near room
the as-deposited macrostructure. The modeled data is taken temperature before the next layer is added. A phase fraction
after the eighth layer addition. The as-deposited amount of zero corresponds to 100% β, except when β
macrostructure consists of the last 6 layers added to the transforms to liquid, which is indicated in Figure 4 by a gap in
deposit, with the layer n representing the characteristic, the alpha fraction curve (This occurs near 300 and 400
repeated layer. Layers n+1 through n+3 represent a transient seconds). In all instances, alpha transforms completely to beta
layer. The locations of the characteristic and transient layers before melting.
are indicated by boxes on the macrograph and modeled data.
During the first three thermal cycles (n through n+2),
temperatures above the beta transus are experienced by both z-
positions. The alpha phase fraction evolution path is
essentially equivalent at both positions during the first 3
thermal cycles. During heating, alpha dissolves rapidly to beta
before melting due to the high heating rates. Upon cooling,
equivalent amounts of alpha form. The final amount of alpha
after the deposition of each layer n through n+2 is the room
temperature fraction (91% α, 9% β).

During the thermal cycle caused by the deposition of n+3, the

greater z-position experiences a temperature above the beta
transus and alpha dissolves completely. The lower z-position
is heated high into the two phase (α+β) field and as a result,
only 20% of the alpha dissolves during the deposition of layer
n+3. As result, a significant change in the evolution path is
observed, though the final amount of alpha is the same for
both positions: 91%-α, 9%-β. Further thermal cycles, i.e., the
deposition of n+4, will be insufficient to produce changes in
the alpha fraction with time since the temperatures will be
below the dissolution start temperature (TDISS = 981K).
Figure 4: Temperature and alpha phase fraction evolution
with time at two z positions along the centerline (y=0) of the The effect of the position sensitive variation in phase fraction
deposit. The positions lie within a layer n and experience an evolution path as shown in Figure 4 is further illustrated if the
additional three thermal cycles (n+3), with the n+3 cycle variation of alpha morphology is considered. Morphology
causing the most variation in the phase fraction evolution being defined here as colony-α and basketweave-α. The
path. Gaps in the phase fraction data near 400 and 500 morphology model assumes that any alpha to form above
seconds distinguish between 100%-β and 100%-liquid. 1100K does so as colony-α, while below 1100K both colony-
α and basketweave-α may form. This is appropriate
considering the JMatPro calculated TTT diagram shown in

Figure 2, where at low undercooling (above 1100K) alpha In addition, Figure 5 shows that the fourth most recent layer
forms at the grain boundary as Widmanstatten alpha (36 < Z <41mm) is the last layer to contain the periodic
sideplates. With further undercooling, the driving force for variation in colony-alpha, thereby representing a characteristic
nucleation increases and intragranular nucleation dominates. layer that is periodically repeated through the deposit. In the
Further details of the morphology model may be found in Ref. as-deposited structure, the characteristic layer contained a
[1]. layer band consisting of colony alpha and a region below the
layer band consisting of a basketweave morphology with a
Figure 5 clearly shows a periodic variation in the amount of variation in the individual alpha lath size. In Figure 5, the
colony-alpha present in the final deposit. This variation is regions of higher colony alpha correspond to the colony
present in each layer except the last three layers deposited. In morphology observed in the as-deposited layer band. This
these last layers, the alpha phase morphology would be near conclusion is drawn based on the assumption that the kinetics
90% basketweave alpha (fα =0.91, fC-α ≈0, fBW-α ≈0.91). Both for grain boundary nucleated colony-alpha formation is
of these results are in good agreement with the as-deposited equivalent to the kinetics for intragranular colony alpha.
results, where the 3 most recent layers to be deposited Further experimental work must be performed in order to
consisted primarily of basketweave and fine-colony alpha. develop better kinetic parameters to model the dissolution and
growth of different alpha morphologies, especially for thermal
cycles where incomplete dissolution occurs prior to alpha

Summary and Conclusions

The thermal model results for LMD of Ti-6Al-4V were used
as input to an alpha phase fraction model, resulting in a
predicted microstructure that is qualitatively similar to the as-
deposited microstructure. The phase fraction model predicts a
near equilibrium evolution with the greatest spatial (z)
variation in the maximum alpha fraction during the n+3
thermal cycle, corresponding to heating into the α+β phase
field. The phase fraction evolution paths for the deposition of
layers n through n+2 are similar, while the deposition of n+4
leads to no change in the phase fraction. The phase fraction
results reinforced the theory that the as-deposited
characteristic layer forms after the deposition of layer n+3
onto layer n.

A microstructure model has been developed to describe the

evolution of both alpha phase fraction and morphology
assuming accurate thermal cycles are known. As such, this
model may be applied to any type of thermal processing of Ti-


Research was sponsored by an appointment through the Joint

Figure 5: Calculated amounts of colony-α after the Institute for Computational Sciences at the Oak Ridge
deposition of eight layers of material using a morphology National Laboratory. The Division of Materials Sciences and
model and calculated thermal cycles. White indicates a high Engineering at the Oak Ridge National Laboratory sponsored
amount of colony-α while black indicates basketweave-α. For a portion of this research. The Oak Ridge National Laboratory
reference, only half of the deposit is shown (due to symmetry). is operated by UT-Battelle, LLC for the U.S. Department of
The substrate is 16 mm thick and each layer is 5 mm thick. Energy under contract number DE-AC05-00OR22725. A
Layers are added in the +z direction and the laser travels in portion of the research was sponsored through the Office of
the +x direction. Naval Research under contract number N00014-98-3-0022
with Virginia Polytechnic Institute and State University.

References TCS Alloys Mobility Database (MOB2), v. 2.0.
(ThermoCalc Software, 1999.
S. M. Kelly, Dissertation, Virginia J. W. Cahn, Acta Metallurgica 4, 449-459 (1956).
Tech,(http://scholar.lib.vt.edu/theses/available/etd- J. W. Cahn, Acta Metallurgica 4, 572-575 (1956).
11242004-211009/) 2004. J. W. Christian, Theory of phase transformations in
S. M. Kelly and S. L. Kampe, Metall Mater Trans A metals and alloys - Part I (Pergamon Press, Oxford,
35A, 1861-1867 (2004). UK, 1975).
S. M. Kelly and S. L. Kampe, Metall Mater Trans A O. Grong and H. R. Shercliff, Progress in Materials
35A, 1869-1879 (2004). Science 47, 163-282 (2002).
F. G. Arcella and F. H. Froes, JOM 52, 28-30 (2000). Thermo-Calc Classic, v. P. (Thermo-Calc Software,
P. S. Goodwin, C. Mitchell, J. Liang, J. Mei, and X. Stockholm, Sweden) 2002.
Wu, in Proc. 2002 Int. Conf. on Metal Powder S. M. Kelly, S. S. Babu, S. A. David, T. Zacharia,
Deposition for Rapid Manufacturing, edited by D. M. and S. L. Kampe, in Cost-Affordable Titanium
Keicher, J. W. Sears, and J. E. Smugeresky (Metal Symposium in Honor of Professor Harvey Flower,
Powder Industries Federation, Princeton, N.J., 2002), edited by F. H. Froes (TMS, Warrendale, 2004).
p. 87-95.
P. A. Kobryn, E. H. Moore, and S. L. Semiatin,
Scripta Materialia 43, 299-305 (2000).
P. A. Kobryn and S. L. Semiatin, JOM 53, 40-42
M. Avrami, Journal of Chemical Physics 8, 212-224
W. A. Johnson and R. F. Mehl, Transactions of the
AIME 135, 416 (1939).
A. N. Kolmorgorov, Izv. Akad. Nauk. SSSR. 3, 355
S. Malinov, Z. Guo, W. Sha, and A. Wilson,
Metallurgical and Materials Transactions A 32, 879-
887 (2001).
X. Li, A. P. Miodownik, and N. Saunders, Materials
Science and Technology 18, 861-868 (2002).
N. Saunders, Z. Guo, X. Li, A. P. Miodownik, and J.
P. Schille, JOM - Journal of the Minerals Metals &
Materials Society 55, 60-65 (2003).
JMatPro, v. 3.0. 2004.
N. Saunders, in Titanium '95: Proceedings of the
Eighth World Conference on Titanium; Vol. 3, edited
by W. J. Blenkinsop, W. J. Evans, and H. M. Flower
(IOM, London, 1996), p. 2167-2177.
N. Saunders, X. Li, A. P. Miodownik, and J. P.
Schille, in Ti-2003: Proc 10th World Conference on
Titanium, edited by G. Luetjering and J. Albrecht
(Wiley, Hamburg, Germany, 2003).
O. M. Ivasishin and R. V. Teliovich, Materials
Science and Engineering A 263, 142-154 (1999).
W. Szkliniarz and G. Smolka, Journal of Materials
Processing Technology 53, 413-422 (1995).
J. W. Elmer, T. A. Palmer, S. S. Babu, W. Zhang,
and T. DebRoy, Journal of Applied Physics 95, 8327-
39 (2004).
J. O. Andersson, L. Hoglund, B. Jonsson, and J.
Agren, in Fundamentals and Applications of Ternary
Diffusion, edited by G. R. Purdy (Pergammon, New
York, 1990), p. 153-163.
SSOL2 - SGTE Solutions Database, v. 2.0.
(ThermoCalc Software, 2002.

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Simulation of Effect of Weld Variables on Thermal Cycles during Twin Wire Welding

Abhay Sharma, Navneet Arora, S. R. Gupta.

Mechanical and Industrial Engineering Department,

Indian Institute of Technology, Roorkee, India, 247 667.

Abstract the help of a double set of drive rolls and guides. As

shown in Figure 1, both the wires are fed at the
The weld thermal cycles during twin wire welding same feed rate and current is supplied from a single
has been simulated with three dimensional heat power source. The current splits between the two
transfers modeling. An appropriate model has been wires and results in lower penetration and higher
developed for twin wire welding process which has deposition rate. The technique of twin wire welding
been validated with experiments. The model has has been known for a long time [1-5] and it is
been considered with two different heat sources for practiced where higher deposit rate without deep
leading and trailing wires and it gives better results penetration is advantageous. Although the twin arc
in comparison to the conventional double ellipsoidal welding is used in practice and has numerous
heat source. advantages over single-wire welding, it has almost
not been investigated [6].
Due to consideration of different heat sources at two
wires it has become possible to better model the
process in which leading and trailing wires have
Power source
been subjected to different surroundings. The
consequence of this difference has been quantified Contact tube
in terms of energy share of particular arc to melt the
base metal and contribution in producing

Effects of different weld variables like speed,

voltage and current on thermal cycles, peak
temperature and cooling time have been simulated
at different points including HAZ and points Figure 1: Twin wire welding system.
situated at surface. These effects have been
quantified and plotted which can be useful to In the present state of welding research the main
visualize the interrelationship between weld emphasis has been on modeling of physical process
parameters and thermal responses. and simulating outcome of the process with help of
Keywords: Twin wire welding, Thermal cycle, developed model. This exercise has helped
Heat transfer, Modeling, HAZ researchers to make process mechanism more
transparent as well as time and cost involved in
Introduction experiments has been avoided. Not only the
Twin wire welding is a variant of conventional techniques under development and evaluation but
single wire welding in which two wires are fed with

Copyright © 2006 ASM International® 71

also the existing techniques can be better b) If above doesn’t work, whether two
understood with help of modeling and simulation identical double ellipsoidal heat sources at a
exercise. given distance (distance between wires)
would work.
As welding is mainly governed by pattern of heat c) Else if two different double ellipsoidal heat
transfer between heat source and work piece, heat sources (Figure 2b) would be needed to
transfer study is an essential part of any welding simulate different surroundings of leading
process. The fundamental requirement for heat and trailing wires, as the leading wire has
transfer modeling has been to define heat source in been subjected to fresh metal while the later
terms of appropriate mathematical function, which is subjected to molten metal generated by
can correspond with the work piece in the same the former.
manner as the actual heat source does. Initial studies
in this area had been made with use of a point heat
source [7] and with due course of time 2-D and 3-D Qg
distributed sources came into picture. In case of arc x
welding process double ellipsoidal heat source a
given by Goldak [8] has been considered one of the z
authentic and versatile heat source models. Various c
br bf
heat sources for different process can be modeled
by adapting the above stated model. The present
work has been involved in same line of action and
an appropriate heat source model for twin wire (a)
welding has been developed.
Qrr Qfr
Development of heat source model x
a a ξ
In present investigation heat source has been z
modeled with adaptation in the double ellipsoidal cr cf
brr brf bfr bff
heat source model. This heat source model
corresponds to the heat generation term (Qg) in the
fundamental heat condition equation (Eq.1).

∇.(K∇T) + Qg = ρ.C p = (1)
Dt Dt
Where K is the temperature dependent thermal Figure 2: (a) Double ellipsoidal heat distribution
conductivity of the material, T is temperature, Qg is and (b) Combination of two double ellipsoidal heat
volumetric heat generation rate (heat source), ρ is distributions.
density, Cp is specific heat, H is enthalpy and t is
time. Mathematical formulation for double ellipsoidal
heat source is given as follows:
Following approaches has been considered for
⎛ 6 3Q ⎞ ⎛ ⎛ ⎞
model development: ⎜ p ⎟ ⎜ − 3x 2 ⎞⎟ ⎜ − 3(ξ)2 ⎟ ⎛⎜ − 3 z 2 ⎞⎟
Qg = f j ⎜ ⎟exp exp⎜ ⎟exp (2)
a) Whether the double ellipsoidal heat source ⎜ π π a.b j.c ⎟ ⎜⎝ a 2 ⎟⎠ ⎜ b2 ⎟ ⎜⎝ c2 ⎟⎠
⎝ ⎠ ⎝ j ⎠
(Figure 2a) with change in its geometric
parameters would work and if an equivalent Where Qp is amount of energy transferred to work
heat source would effectively represent the piece from the arc which is equal to ηVI, V is
heat pattern generated by combination of voltage, I is current and η is arc efficiency and j = f,
two wires. r stands for front and rear of the source respectively.

In the front half of the source, ff and bf refer to heat both the source has been considered as half of the
input fraction and semi axis along longitudinal weld width. Semi axis of trailing source in depth
direction respectively. Where as ‘a’ and ‘c’ direction has been considered equal to depth of
represents semi axis in transverse and depth penetration. Remaining parameters have been
directions. The sum of the fractions, fj, between the varied and the combination, which can reproduce
heat deposited in the front and rear must equal two, experimental result, has been considered.
i.e., ff + fr = 2. ξ is moving co-ordinate fixed with
the arc. To get continuity in Eq.2 when ξ =0 the
following condition must yield,

2b f
ff = (3a ) C
bf + b r
2b r
fr = (3b)
bf + br A

In case of combination of two double ellipsoidal

heat sources, new geometric parameters and heat
input fractions have been incorporated. Figure 3: Positions of points for temperature
measurements: A – at bottom of weld pool, B – at
Mathematical formulation considering two double surface (8 mm from center line) and C – at surface
ellipsoidal heat sources has been given as follows: (9 mm from center line.

⎛ 6 3Q ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ It has been observed that approach (a) and (b)

⎜ p ⎟ ⎜ − 3x 2 ⎟ ⎜ − 3(ξ) 2 ⎟ ⎜ − 3 z 2 ⎟
Qij = fijdi ⎜ ⎟exp⎜ ⎟ exp⎜ ⎟exp⎜ ⎟ (4)
⎜ π π a i bij.ci ⎟ ⎜ a 2 ⎟ ⎜ b 2 ⎟ ⎜ c 2 ⎟ produced error in computed temperature while
⎝ ⎠ ⎝ i ⎠ ⎝ ij ⎠ ⎝ i ⎠
approach (c) i.e., two dissimilar heat sources
produced computed results in good agreement with
Where i = f, r for leading and trailing heat source experimental results. Numerical values of df, dr, cf,
respectively and j = f, r for front and rear half bff, brr equal to 0.25, 0.75, 0.65, a/4, a/2
of particular source and di represents power respectively, where ‘a’ is half of weld width, has
density factor been found most appropriate (Appendix).
For determination of above stated parameters, As the leading wire is subjected to fresh metal and
inverse modeling has been applied. At different colder surroundings, spread of arc at this wire is
values of these parameters heat conduction equation leaser in comparison with the trailing wire. Due to
has been solved and peak temperature at three smaller spread, heat density has been more, which
different points situate on surface and within the results in higher proportion (65 %) of penetration.
body of work piece have been compared with On the other hand due to application of heat to melt
experimental results. These three points are shown flux, proportion of heat supplied for conduction has
in Figure 3. been lesser (25 %).
The conduction equation has been solved with finite This outcome has been further justified with
difference method with enthalpy method and semi- comparison of computed and measured weld
discrete technique [9] coded in C language while dimensions as shown in Table 1. Computed depth
experimental measurements have been made on of fusion zone and heat affected zone has been
bead on plate welds produced with submerged arc found in good agreement with experimental results.
welding. The semi axis in transverse direction for

Table 1: Comparison of FZ and HAZ depth temperatures at different speeds have not been
differing. It has been due to the reason that this
Measured Calculated point has been directly in contact with the arc,
Depth of FZ (mm) 11.750 11.035 hence time lag between heat generation and
Depth of HAZ (mm) 13.580 14.033 conduction has not been present. As soon arc has
passed from this point, temperature has reached to
peak point without any time lag.
Simulation of effect of weld variables
Simulation of effect of speed and position on 20 cm/min
thermal cycle 30 cm/min
Thermal cycle at different points situated at the 1400

Temperature in °C
40 cm/min
surface and within the body of the work piece has 1200
been simulated at different speeds with the help of 1000
developed model. Figures 4 to 6 show the same. It
is evident from Figure 4 that speed considerably
effect the thermal cycle. 600
At lower speed, higher peak temperature has been 200
observed. This is due to the reason that heat
available per unit length is more at lower speed as 0 6 12 18 24 30 36 42 48 54 60
well as sufficient time has been available for heat to
time in sec.
conduct up to bottom of the weld pool. On the
contrary, responses at points situated at surface are Figure 5: Effect of welding speed on the thermal
considerably different from the point situated at the cycle of point situated at surface (8 mm form center
bottom of the weld pool. line), current 950 Amp and voltage 33V.

1800 1800
1600 20 cm/min 1600 20 cm/min
30 cm/min
1400 1400 30 cm/min
Temperature in °C

40 cm/min
Temperature in °C

40 cm/min
1200 1200
1000 1000
800 800
600 600
400 400
200 200
0 0
0 6 12 18 24 30 36 42 48 54 60 0 6 12 18 24 30 36 42 48 54 60
time in sec. time in sec.

Figure 4: Effect of welding speed on the thermal Figure 6: Effect of welding speed on the thermal
cycle of point situated at the bottom of the weld cycle of point situated at surface (9 mm form center
pool, current 950 Amp and voltage 33V. line), current 950 Amp and voltage 33V.

It can be seen in the Figure 5 that incase of point The point situated far from weld centerline at
situated at surface (8 mm form center line), peak surface of the work piece has also shown difference

in peak temperatures (Figure 6) at different welding due to increase in current, heat input has increased
speeds but it has not been as dominant as in the case in lesser amount yet both the changes in current and
of the point situated at the bottom of the weld. On speed have resulted in almost same amount of
carefully studying, there has been one more change in peak temperature. Current has been more
difference in the responses. Thermal cycles of dominant in producing the peak temperature
surface points have shown two maxima points because of its role in joule heating and melting of
where as same has not been observed in the other wire.
case of weld bottom. It seems that effect of two
wires on thermal response has been prevailing at the 1600
surface as well as it has been more dominant in case
of higher welding speed. 850 amp

1200 950 Amp

Temperature in °C
Simulation of effect of current and position on
thermal cycle
With increase in current, peak temperatures of all 800
the points of consideration have increased. It has 600
been evident from Figure 7 to 9 that change in peak
temperature due to change in current has almost 400

been independent of position of the point of 200

0 6 12 18 24 30 36 42 48 54 60
1600 time in sec.
1400 850 amp
Figure 8: Effect of current on the thermal cycle of
1200 950 Amp
Temperature in °C

point situated at surface (8 mm form center line)

1000 speed) 30 cm/min and voltage 33V.

600 1400 850 amp

1200 950 Amp

Temperature in °C


200 1000

0 800
0 6 12 18 24 30 36 42 48 54 60
time in sec.
Figure 7: Effect of welding current on the thermal 200
cycle of point situated at the bottom of the weld
pool, speed 30 cm/min and voltage 33V. 0
0 6 12 18 24 30 36 42 48 54 60
time in sec.
The heat input per unit length of the work piece can
Figure 9: Effect of current on the thermal cycle of
be changed by change in current and speed both. It
point situated at surface (9 mm form center line,
can be seen that change in current from 850 Amp to
speed 30 cm/min and voltage 33V.
950 Amp has increased the heat input
approximately 1.1 times and on the other hand
In cases of all the three points peak temperature has
reduction in speed from 40 cm/min to 30 cm/min
increased in proportional manner, irrespective to the
has increased heat input by 1.33 times. Although

position of the point of consideration. This has been Conclusions
due to the fact that generation of peak temperature
is a momentary event which has occurred just at the Heat source model for twin wire welding has been
moment when arc has been passing through the developed by adapting the conventional double
plane of particular point of consideration. ellipsoidal model. With help of this model twin wire
welding can be better represented. Thermal
Simulation of effect of variables on cooling time responses have been examined under the effect of
Increment in speed has resulted in reduction in different set of welding variables. Simulation of
cooling time (T800-500). It has been observed that thermal cycle with developed heat source model is
cooling time has also been a function of the position capable to produce results which can be justified
of the point of consideration as shown in Figure 10. with help of physics of the process. Role of twin
wires in producing the thermal response at different
12 point of body of the work piece has also been
evaluated and it has been observed that twin wires
850 Amp, 33 V, Point A has significant effect on the thermal response of the
10 950 Amp, 33 V, Point A points situated near the surface of work piece.
9 850 Amp, 33 V, Point B
cooling time in sec

950 Amp, 33 V, Point B Further research will be carried out to extend the
present work on twin wire with consideration of
7 different wire combinations and thermo-mechanical
6 aspect of the process.
4 References
1. T. Asthon, Twin arc submerged arc welding.
20 30 40
Welding J ,33(4),350-54(1954)
2. D.E. Knight, Multiple electrode welding by
Speed in cm/min
Unionmelt process, Welding J, 33(4), 303-
Figure 10: Effect of welding speed on cooling time 3. S.L. Mandel and V.E. Lopta, The high speed
T 800-500 at different points. twin SAW Process, Welding Production
February No-2, 25-28(1966)
In case of point situated at bottom of the weld, 4. J.E. Hinkel and F.W. Forsthoefl, High
cooling time has been inversely proportional to current density SAW with twin electrode,
speed and it is almost independent of change in Welding J 55(3),175-180(1976)
current. While on the other points which have been 5. P.J. Konkol and G.F. Koons, Optimization
situated at surface, rate of decrement in cooling of parameters for Two-Wire AC-AC- SAW,
time has been higher then the earlier case as well as Welding J, 57(12),367s-74s,(1978)
change in current also affect the cooling time. This 6. Janez Tuesk, Mathematical modelling of
has been due to the reason that the surface is melting rate in twin-wire welding. J of
subjected to continuous heat loss due to convection. Material Processing Technology,100,250-
At higher speed, heat input per unit remains lesser 56(2000)
but loss due to convection is almost same which 7. D. Rosenthal, The theory of moving source
causes faster cooling. of heat and its application to metal
treatments. Transaction of ASME, 849-

8. J .Goldak, A. Chakravarti and Bibby, A new dimenssional transient heat transfer
finite-element model for welding heat- computations of autogeneous arc welding,
sources. Metall Trans B—Process Metall., Metallurgical Transactions, 21B, 1033-
15,299–305(1984) 47(1990)
9. Robert L. Ule, Yogendra Joshi and Eugene
B. Sedy, A new technique for 3-

df dr cf* bfr brf bff ( bf) brr(br) Computed peak Temperature in RMSE
1 - - - - - a** 4a 1122 923 619 307.59
2 - - - - - a 3a 1248 908 656 283.01
3 - - - - - a/2 3a 1243 911 655 282.35
4 - - - - - a/2 2a 1254 1002 715 219.25
5 - - - - - a/2 3a/2 1261 1068 758 173.83
6 - - - - - a/2 a 1276 1166 822 106.06
7 - - - - - a/2 3a/4 1272 1219 867 68.52
8 - - - - - a/2 3a/5 1275 1267 908 36.63
9 - - - - - a/4 3a 1239 912 655 282.25
10 - - - - - a/4 2a 1249 1005 716 218.04
11 0.5 0.5 1 p p*** a/2 a/2 1466 1164 852 127.48
12 0.5 0.5 0.9 p p a/4 a/2 1427 1166 851 113.02
13 0.5 0.5 0.8 p p a/4 a/2 1347 1166 851 95.36
14 0.5 0.5 0.7 p p a/4 a/2 1250 1164 840 105.73
15 0.4 0.6 1 p p a/4 a/2 1465 1211 879 105.88
16 0.4 0.6 0.9 p p a/4 a/2 1425 1212 887 86.60
17 0.4 0.6 0.8 p p a/4 a/2 1379 1213 887 69.72
18 0.4 0.6 0.7 p p a/4 a/2 1292 1212 876 65.41
19 0.3 0.7 0.75 p p a/4 a/2 1356 1246 913 42.92
20 0.3 0.7 0.7 p p a/4 a/2 1327 1247 913 36.71
21 0.3 0.7 0.6 p p a/4 a/2 1264 1245 910 49.14
22 0.25 0.75 0.7 p/2 p/2 a/4 a/2 1338 1263 926 28.12
23 0.25 0.75 0.6 p/2 p/2 a/4 a/2 1285 1262 924 32.55
24 0.25 0.75 0.5 p/2 p/2 a/4 a/2 1261 1262 923 42.25
25 0.25 0.75 0.65 p/2 p a/4 a/2 1327 1253 917 32.77
26 0.25 0.75 0.65 p/2 p/2 a/4 a/2 1323 1264 926 25.23
Experimental 1318 1307 932
*** cf = fraction of semi axis along depth direction for trailing arc i.e. cf = 0.5 means half of cr ,**2a= weld
width,***p = distance between wires

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Three-dimensional modeling of transient heat transfer and fluid flow during

orbital gas tungsten arc welding of pipes
W. Zhang, C. Conrardy and I. Harris
Edison Welding Institute Inc., Columbus, Ohio, USA

Abstract model to calculate thermal cycles and fusion boundary during

orbital GTA welding of aluminum alloy pipes. In their model,
The evolution of temperature and velocity fields during orbital the heat source is stationary in space and the material is
gas tungsten arc (GTA) welding is studied using a transient rotating at a constant angular velocity. Lho et al. [2] studied
thermo-fluid model. The model solves the equations of the transient heat flow considering a moving heat source.
conservation of mass, momentum and energy using the Finite Mesh with variable spacing was regenerated at each time step
Volume Method (FVM) in an unstructured mesh. The driving to effectively resolve the high spatial gradients of
forces for melt convection include the electromagnetic, temperatures in the vicinity of the heat source. Matsutani et
gravitational and surface tension gradient forces. An adaptive al. [3] developed a heat conduction model to calculate the
meshing technique is used to maximize the resolution of transient temperature distribution and molten pool size. The
solution variables while minimizing the computational cost. computed molten pool boundary was then used in a set of
The thermo-fluid model is applied to calculate the fusion zone equations, which were based on the minimization of total
geometry and thermal cycles during orbital GTA welding of surface energy consisting of surface tension and gravity
aluminum 6061 alloy pipes. The calculated weld bead potential energies, to determine the shapes of the top and
geometries at different locations along the circumferential bottom surfaces of weld pool. Although these heat conduction
joint were compared with those measured experimentally, and models have provide many useful insights into the orbital
the model predictions were in fair agreement with the GTA welding process, they did not take into account an
experimental results. The transient model provides a baseline important physic phenomenon in the weld pool, i.e., liquid
for future work in optimizing welding conditions to achieve metal convection, which is dominant in dissipating the heat
uniform root beads on full penetration butt joints, minimize energy in the pool and largely determines the pool shape [4,5].
procedure development cost, and avoid weld defects such as Comprehensive modeling of fluid flow and heat transfer is
porosity. required for a more fundamental understanding of the welding
Keyword: weld pool modeling, orbital GTA welding,
aluminum pipe In the present study, a three-dimensional transient heat transfer
and fluid flow model is developed to calculate the liquid
Introduction convection in the weld pool, the temperature distribution in
the pipe, and the shape and size of the weld bead. The
During orbital gas tungsten arc (GTA) welding of pipes, it is numerical model solves the equations of conservation of mass,
important to obtain uniform bead shape, size and penetration momentum and energy using the Finite Volume Method
over the full length of the joint. For pipes of small diameters (FVM) in an unstructured mesh. The electromagnetic force,
and high thermal diffusivities, the bead width becomes wider gravitational force, and Marangoni shear stress are considered
as the welding proceeds, if the welding conditions are kept for driving melt flow in the weld pool. An adaptive meshing
constant. In traditional applications, multiple welding technique generates a box of dense mesh in the vicinity of the
schedules using decreasing heat input are empirically moving welding arc. Such meshing technique maximizes the
developed to compensate for the increasing preheat. Excessive resolution of variables while minimizing the computational
experimental costs are typical to achieve satisfactory weld cost. The transient model is used to compute the weld bead
results. size and thermal cycles during orbital GTA welding of
aluminum 6061 alloy pipes. The computed results are verified
Since the past two decades, various computational models with the available experimental data.
have been developed to simulate the orbital GTA welding
process. Kou et al. [1] developed an unsteady heat conduction

Copyright © 2006 ASM International® 79

Mathematical Formulation
where φ is the electric scalar potential, σ is the electric
In the present work, the welding arc is not simulated and the
conductivity assumed to be constant, B is the magnetic field,
interactions between the arc and the test piece material are r
taken into account by using the appropriate boundary μm is the magnetic permeability, and J is the electric current
conditions. The mathematical equations representing the heat density vector which can be obtained using the following
transfer and fluid flow in the weld and their numerical relationship.
solutions are described as follows.
Governing equations The basic assumptions of
J = −σ∇φ (6)
the transient thermo-fluid model are the following.
Solving Eqs. (4) to (6), the electric current density vector and
(I) The thermo-physical properties of the material are magnetic field can be obtained. The electromagnetic force is
assumed to be constant. The model is capable of then calculated by the following equation.
easily incorporating temperature dependent thermo-
r r r
physical properties. F = J ×B (7)
(II) An incompressible, laminar and Newtonian molten
metal flow is assumed in the weld pool.
(III) The effect of turbulence is considered using the Boundary conditions As shown in Fig. 1, only a
effective thermal conductivity and viscosity for the half of the pipe is considered since the weld is symmetric
molten metal. about the z = 0 plane. For the molten metal convection, the
(IV) The weld pool surface is assumed to be not boundary conditions at the outer and inner surfaces are the
deformable. following [7].
r r r r
With the above assumptions, the heat transfer and molten μ∇(u ⋅ t ) ⋅ n = γ T ∇T ⋅ t (8a)
r r
metal flow can be described by the following equations. u ⋅n = 0 (8b)
∂ui r
=0 where γT is the temperature coefficient of surface tension, t is
∂xi (1) r
a surface unit tangential vector, and n is the surface unit
∂u j ∂ (ui u j )⎛ ∂u j ⎞∂ normal vector. Eq. (8a) represents the Marangoni shear stress
ρ +ρ ⎜⎜ μ
= ⎟⎟ + F j + S j
∂t ∂xi ∂xi
⎝ ∂xi ⎠
at the surface, and Eq. (8b) indicates that the normal velocity
of molten metal is zero. At the end surface, u = 0 .
∂h ∂ (ui h ) ∂ ⎛ ∂h ⎞
ρ +ρ = ⎜α ⎟ + Sh
∂t ∂xi ∂xi ⎜⎝ ∂xi ⎟⎠ End surface

Equations (1), (2) and (3) represent the conservation of mass, Travel direction
momentum, and energy, respectively. In these equations, x is
the distance, u is the velocity of molten metal flow, t is the
time, µ is the viscosity, h is the enthalpy, and α is the thermal Welding arc
diffusivity. Details of these equations as well as the
momentum source term (Sj) and the enthalpy source term (Sh)
are available elsewhere in the literature [5] and are therefore
not repeated here. The term Fj in Eq. (2) corresponds to the
electromagnetic force and its calculation is described in the
following section.
Calculation of electromagnetic force The
electromagnetic force is obtained by solving the following two Outer surface
steady-state Maxwell equations [6]. z
Symmetric surface
x Inner surface
∂ ⎛ ∂φ ⎞
⎜⎜ σ ⎟⎟ = 0
∂xi ⎝ ∂xi ⎠ (4) Figure 1: Schematic plot showing the solution domain used in
r r
∇ × B = μm J the calculations.

For the heat transfer, the boundary condition at the outer
surface is the following.

3IVη ⎛ 3d 2 ⎞ Welding arc

qn = exp⎜⎜ − 2 ⎟⎟
πrb2 ⎝ rb ⎠ (9a)

where qn is the normal heat flux, I is the welding current, V is

the welding voltage, η is the arc efficiency, rb is the arc radius,
and d is the distance to the welding arc. At the inner surface,
the heat loss (qc) due to the convective heat transfer between
the workpiece and surrounding gas is given as:

qc = − hc (T − Ta ) (9b)
where hc is the heat transfer coefficient, T is the workpiece
surface temperature, and Ta is the ambient gas temperature.
T = Ta
At the end surface, .

For the electrical scalar potential equation, the boundary

condition at the outer surface is:

Welding arc
3I ⎛ 3d 2 ⎞
Jn = exp ⎜⎜ − 2 ⎟⎟ (10)
πrb2 ⎝ rb ⎠

where Jn is the normal electric current density. The inner

surface is assumed to be an insulated surface. Therefore, the
∂φ r = 0 . The
current density flux is equal to zero, i.e.,
end surface is assumed to be grounded, i.e., φ = 0 . (b)
Adaptive meshing A fine mesh is often
required near the welding arc to adequately capture the
vigorous molten metal convection and high spatial gradient of Figure 2: Adaptive meshing as the welding arc travels.
temperature. Since the welding arc was traveling along the
circumferential direction of the pipe, a very large number of Results and Discussion
cells would have been needed along this direction if no special
meshing technique had been used. In this study, an adaptive In the present work, the experimental data reported by Kou
meshing technique was utilized. As shown in Fig. 2(a), a box and Le [1] is used to validate the transient thermo-fluid model.
of dense mesh was placed near the welding arc. The dense Table 1 summarizes the welding conditions used in the orbital
mesh box followed the motion of the welding arc, as shown in GTA welding experiments conducted by Kou et al. The
Fig. 2(b). Such adaptive meshing technique maximized the thermo-physical property data of the aluminum 6061 alloy and
resolution of solution variables while minimizing the other parameters used in the calculations are summarized in
computational cost. Table 2.

Numerical solution The governing Electromagnetic force Figure 3(a) plots the
conservation equations are solved using the commercial distribution of the magnetic field at the outer surface of the
Computational Fluid Dynamics (CFD) code, FLUENT ®. pipe. Since the electrode was positive, the electric current
FLUENT discretizes the governing equations using FVM in flied into the outer surface. As a result, a clockwise magnetic
an unstructured mesh [8]. User-defined scalar transport field was generated. The electromagnetic force at the
equations are solved to calculate the electromagnetic force. symmetric surface is plotted in Fig. 3(b). In the absence of
other driving forces, the electromagnetic force will cause a
downward flow.

It should be noted that the electromagnetic force was
calculated by numerically solving the two steady-state Max = 8.5×10-3 Tesla (a)
Maxwell equations. Unlike the widely-used analytical
equation for the calculation of electromagnetic force in the
weld pool [5], the present method has the ability to take into
account a curved weld pool surface.

Welding arc
Table 1: Welding conditions during orbital GTA welding of
aluminum 6061 alloy [1].

Welding condition Value

Power supply DCEP
Welding current, I, (A) 110
Welding voltge, V, (V) 10
Travel speed (rpm) 2.12
Outer diameter of pipe (mm) 38.1 5 mm
Wall thickness of pipe (mm) 3.2
Shielding gas Argon
* DCEP – Direct current electrode positive

Welding arc (b)

Table 2: Physical property of aluminum 6061 and other
parameters used in the calculations.

Property Value
Density, ρ, (kg m-3) 2700
Thermal conductivity of solid (J m-1⋅s-1⋅K-1) 168
Effective thermal conductivity of liquid 420
(J m-1⋅s-1⋅K-1) 2 mm
Specific heat (J kg-1⋅K-1) 1066
Effective viscosity, μ, (kg m-1 s-1) 0.02
Solidus temperature (K) 855
Max = 2.4×105 N m-3
Liquidus temperature (K) 925
Temperature coefficient of surface tension, -3.5×10-4
γT, (N m-1⋅K-1)
Arc efficiency, η 82% Figure 3: Computed distribution of (a) magnetic field, B, at
Arc radius (mm) 3.0 the weld outer surface, and (b) electromagnetic force at the
symmetric surface.
Magnetic permeability, μm, (N A-2) 1.26×10-6
Ambient temperature, Ta, (K) 298 K

Heat transfer and melt convection in weld pool

Figure 4(a) shows the weld pool at a time equal to 0.4 s. At
this moment, the pipe is partially penetrated. Two distinct
flow patters can be observed in Fig. 4(a). At the top surface,
the liquid metal is pulled from center to periphery of the weld
pool by the surface tension. In the middle of the weld pool,
Lorenz force is dominant in driving the liquid flow downward.
Figure 4(b) shows the weld pool at a time equal to 2.2 s when
the pipe is fully penetrated. Flow patters similar to those
shown in Fig. 4(a) are observed in the fully penetrated weld

Figures 5(a) and 5(b) compare the calculated weld pool

geometry at two instants. As shown in these figures, the
model predictions agree well with the experimental results.

Temperature (K)

Temperature (K)
(a) Time = 0.4 s (a) Time = 0.4 s
Temperature (K)

Temperature (K)

(b) Time = 2.2 s

(b) Time = 2.2 s
Figure 5: Comparison between calculated and measured weld
Figure 4: Evolution of the weld pool during orbital GTA pool shape at two instances. Experimentally measured weld
welding of aluminum alloy pipe. The maximum velocity in (a) pool boundaries are represented by dashed lines. The
is 0.31 m/s while that in (b) is 0.33 m/s. maximum velocity in (a) is 0.31 m/s while that in (b) is 0.33

Thermal cycles Figure 6 shows the calculated

thermal cycles at different locations in the weldment. As Conclusions
shown in this figure, the temperatures start to increase again
15 s after the weld starts. This is because the heat continues to The evolution of temperature and velocity fields during orbital
build up during welding and the areas yet to be welded are GTA welding of aluminum 6061 alloy pipe was studied using
preheated. a three-dimensional transient thermo-fluid model. The
following conclusions can be made from the investigation.
The geometry of the FZ predicted from the model was in good
agreement with the corresponding experimental results. The
computed thermal cycles indicate that the build-up of the
preheat causes significant rises of temperatures in the

The transient model provides a baseline for future work in

optimizing welding conditions to achieve uniform root beads
on full-penetration butt joints, minimize procedure
development cost, and avoid weld defects such as porosity.


Temperature (K)

700 B B




0 5 10 15 20 25
Time (s)

Figure 6: Calculated thermal cycles at different locations

during orbital GTA welding of aluminum alloy pipe.

The authors would like to thank Drs. Wangen Lin, Robert
Yancey, and Dennis Harwig at Edison Welding Institute, Inc.
for his valuable suggestions.

1. S. Kou and Y. Le, Heat flow during the autogenous
GTA welding of pipes, Metall. Trans. A, 15A, 1165-
71 (1984)
2. T. J. Lho and S. J. Na, A study on three-dimensional
transient heat flow in circumferential GTA welding of
pipes using periodicity conditions, Proc. Instn. Mech.
Engrs., 205, 271-8 (1991).
3. T. Matsutani, F. Miyasaka, T. Oji and Y. Hirata,
Mathematical modeling of GTA girth welding of
pipes, Weld. Int., 11, 616-20 (1997).
4. T. DebRoy and S. A. David, Physical processes in
fusion welding, Rev. Mod. Phys., 67, 85-112 (1995).
5. W. Zhang, G. G. Roy, J. W. Elmer and T. DebRoy,
Modeling of heat transfer and fluid flow during gas
tungsten arc spot welding of low carbon steel, J.
Appl. Phys., 93, 3022-33 (2003)
6. R. Moreau, Magnetohydrodynamics, p. 30, Kluwer
Academic Publishers, Boston (1990)
7. W. Zhang, C. H. Kim and T. DebRoy, Heat and fluid
flow in complex joints during gas metal arc welding -
Part I: Numerical model of fillet welding, J. Appl.
Phys., 95, 5210-9 (2004)
8. FLUENT, Inc., FLUENT 6.2 User's Guide, Lebanon,
NH (2005)

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Reliable Modeling of Heat and Fluid Flow in Gas-Metal-Arc Fillet Welds through
Optimization of Uncertain Variables
A. Kumar and T. DebRoy
The Pennsylvania State University, University Park, PA

Abstract strong convection is considered in the phenomenological

models by enhancing the values of thermal conductivity and
Although numerical heat transfer and fluid flow models viscosity to account for the enhanced heat and mass transfer.
have provided significant insight about fusion welding Unlike the molecular values of the thermal conductivity and
processes and welded materials in recent years, several model viscosity, the enhanced values of these “transport properties”
input parameters cannot be easily prescribed from are not physical properties of the fluid and, as a result, their
fundamental principles. As a result, the model predictions do values cannot be obtained from the standard compilations of
not always agree with the experimental results. In order to thermophysical properties. In contemporary transport
address this problem, the approach adapted here is to develop phenomena, the enhanced values of transport properties are
and test a model that embodies a heat transfer and fluid flow calculated using an appropriate turbulence model.
sub-model and an algorithm for optimizing and learning the The momentum transport rates in the weld pool owing to
values of uncertain process variables from a limited volume of the strong recirculating velocities with fluctuating components
experimental data. The heat transfer and fluid flow sub-model is often taken into account [1-5] by simple ‘zero-equation’ or
numerically calculates three-dimensional temperature and algebraic turbulence models. In these calculations, the effects
velocity fields and the weld geometry during gas metal arc of turbulence are simulated by arbitrarily enhancing the
(GMA) welding of fillet joints. The proposed model could molecular values of thermal conductivity and viscosity by 10
estimate the unknown values of arc efficiency, effective to 100 times. The values of the enhanced viscosity and thermal
thermal conductivity and effective viscosity as a function of conductivity are properties of the specific welding system and
welding conditions based on only a few experimental cannot be easily assigned from fundamental principles [4-7].
measurements. A vorticity-based mixing length hypothesis Although the established turbulence models often serve as a
was also used to independently calculate the values of the basis for the estimation of enhanced transport properties, the
effective viscosity and effective thermal conductivity. Good empirical constants in these models have been determined
agreement between the experimental and the predicted weld using experimental data from large scale parabolic flows. In
geometry showed that this approach was useful in improving contrast, fluid flow in the weld pool is strongly recirculatory
reliability of heat transfer and fluid flow calculations. or elliptic in nature and the size of the weld pool is rather
small. Currently there is no unified basis to accurately
Keywords prescribe the values of effective thermal conductivity,
Gas-Metal-Arc fillet welding, Heat transfer and fluid flow, effective viscosity and arc efficiency. Values of these
Reliability, Optimization, Vorticity, Mixing length. parameters significantly affect the results of numerical heat
transfer and fluid flow calculations.
Introduction The present work seeks to enhance the reliability of the
In recent years, phenomenological models of fusion heat and mass transfer calculations in the weld pool by
welding have provided important understanding of welding determining how the uncertain input parameters, i.e., the arc
processes [1-7] and welded materials. Although these models efficiency, effective thermal conductivity and the effective
use time-tested fundamental equations of conservation of viscosity vary with heat input. The values of these parameters
mass, momentum and energy with appropriate boundary are determined from a limited volume of experimentally
conditions, their predictions are affected by the uncertainty in measured weld pool penetration, throat and the leg-length data
the values of various input parameters used in the modeling using a combination of an optimization algorithm and a
[4-7]. For example, the reported values of arc efficiency vary numerical heat transfer and fluid flow model. The
significantly for apparently similar welding conditions optimization algorithm minimizes the error between the
reflecting the complexity of the GMA welding process. predicted and the experimentally observed penetration, throat
Furthermore, strong convection currents are present in the and the leg-length during the GMA welding process by
weld pool during GMA welding process. Generally, this considering the sensitivity of these geometric parameters to

Copyright © 2006 ASM International® 85

each of the uncertain parameters. The Levenberg-Marquardt where η is the arc efficiency, ke is the effective thermal
(LM) [4-7] and two versions of conjugate gradient method conductivity, kL is the conductivity of the liquid material, μe is
(CG) i.e., Fletcher-Reeves and Polak-Ribiere of non-linear the effective viscosity, μL is the viscosity of the liquid
parameter optimization [4-6] are used to estimate these material, I is the current, V is voltage, rw is wire radius, wf is
uncertain parameters with a well-tested three-dimensional the wire feeding rate, ρ is the density, cp is the specific heat,
numerical heat transfer and fluid flow model. The obtained TL is the liquidus temperature, Ta is the ambient temperature,
values of effective thermal conductivity and viscosity are also L is the latent heat of the alloy, rb is the arc radius, Uw is the
validated by calculating their values using a vorticity based welding speed, and f1, f2, f3, f4, f5 and f6 are constants. In the
mixing-length turbulence model [8,9]. literature, it has been shown that arc efficiency varies linearly
with heat input per unit length [11] but the slope of this
Mathematical model variation depends on the welding conditions and the technique
[11]. Recent work on butt welding showed that effective
thermal conductivity and effective viscosity is also a linear
Modeling of Heat Transfer and Fluid Flow during Gas
function of heat input per unit length [7]. Also, due to narrow
Metal Arc Fillet Welding
range of experimental data used in the present work, it is
The heat transfer and fluid flow model takes into account
justifiable to use the linear variation of the arc efficiency,
the liquid metal convection in the weld pool, the complex fillet
effective thermal conductivity and effective viscosity with
joint geometry, the deformation of the weld pool top surface,
input power.
additions of the filler metal, and the heat transfer by metal
In order to calculate the values of arc efficiency,
droplets. The output from the model includes temperature and
effective thermal conductivity and effective viscosity, we
velocity fields, thermal cycles, fusion zone geometry and the
require values of constants i.e. f1, f2, f3, f4, f5 and f6 in
solidified geometry of the weld reinforcement. Since the
equations (1) to (3). To find the values of these terms, an
numerical model of heat transfer and fluid flow has been
objective function is minimized that depicts the difference
described in the literature [2-5], only its salient features are
between the computed and measured values of the weld
described here. By using a coordinate system attached to the
dimensions. For example, if the penetration, throat and the
heat source, the welding process can be treated as a steady
leg-length of the fusion zone are of interest, an objective
state problem [1,2]. For fillet welding, accurate solution of
function, O(f), can be defined as follows:
heat transfer and fluid flow with a deformable weld pool
(m ) (m ) (m )
M 2 M 2 M 2
surface and complex joint geometry requires the use of non- O (f ) = ∑ pe − pc + ∑ te − tc + ∑ le − lc (5)
m m m
orthogonal deformable curvilinear grid system [1, 2]. m =1 m =1 m =1
Therefore, the governing equations are transformed from the where pcm , t cm and lcm are the computed penetration, actual
Cartesian to curvilinear coordinate system [1-3]. The throat and the leg length of the weld bead, respectively and
transformed governing equations are discretized using the
control volume method, where the computational domain is pem , t em and lem are the corresponding experimentally
divided into many small rectangular control volumes. determined values of these three variables. The subscript m in
Discretized equations for a variable are formulated by equation (5) corresponds to a specific weld in a series of M
integrating the corresponding governing equation over the number of total welds. In equation (5), f refers to a set of six
control volumes in the computational domain. A power-law uncertain non-dimensional parameters, f1, f2, f3, f4, f5 and f6
based scheme is used to describe the convective flux at the that are constants in the assumed linear functions of
control volume faces [10]. A modified Semi-Implicit efficiency, η, effective thermal conductivity, ke, and effective
Algorithm for Pressure Linked Equations (SIMPLE) is used to viscosity, μe, expressed by equations (1) to (3). For the
solve the discretized equations. In the present work, the estimation of these uncertain variables, Levenberg-Marquardt
droplet heat transfer in the spray mode is effectively simulated (LM) method and two modifications of the conjugate gradient
by incorporating a time-averaged volumetric heat source term method suggested by Fletcher-Reeves and Polak-Ribiere are
(Sv) in the energy conservation equation [11, 12]. used in the present study. The mathematical descriptions of
Optimization of Uncertain Variables these techniques are available in literature [4, 6].
The goal of the optimization problem is to determine
how the uncertain parameters, e.g., arc efficiency, effective Results and discussion
thermal conductivity and effective viscosity vary with heat
input per unit length. For simplicity we assume the following The evolution of weld pool involves complex interaction
linear relations between these variables and heat input. of physical processes such as application of welding arc, metal
η = f1 + f 2 ⋅ Pe* (1) droplet transfer, heat transfer through conduction and
convection, free surface deformation and the fluid flow inside
k e = f3 ⋅ k L + f 4 ⋅ k L ⋅ Pi* (2) the weld pool. To simulate these simultaneous processes
μe = f5 ⋅μL + f6 ⋅μL ⋅ Pi* (3) correctly in the numerical heat transfer and fluid flow analysis,
the accurate values of arc efficiency (η), effective thermal
IV /(πrw2 w f ) IV /(πrb2 U w ) conductivity (ke) and effective viscosity (μe) are needed. The
Pe* = and Pi* = (4)
[ρc p (TL − Ta ) + ρL] [ρc p (TL − Ta ) + ρL] following section shows how the values of arc efficiency,

effective thermal conductivity and effective viscosity are 6

Objective Function (x10 m )

CG (Fletcher-Reeves)

Table 1: Welding conditions used in the experiments. CG (Polak-Ribiere)
CTWD feeding Voltage
No. speed current
(mm) rate (V)
(mm/s) (A) 2
1 22.2 169.3 4.2 31 312.0
2 22.2 211.7 6.4 31 362.0
3 22.2 169.3 6.4 33 312.0 0
0 5 10 15 20
4 22.2 211.7 4.2 33 362.0 Iteration Number
5 28.6 169.3 6.4 31 286.8
Figure 1: Computed values of the objective function, O(f),
6 28.6 169.3 4.2 33 286.8 using LM method and the two versions of CG method as a
7 28.6 211.7 4.2 31 331.4 function of iteration number.
8 28.6 211.7 6.4 33 331.4
1.0 1.5

Non-dimensional Weld Geometry

9 25.4 190.5 5.3 29.6 322.6 Arc efficiency

10 25.4 190.5 5.3 34.4 322.6

Actual throat
Penetration 1.3

Arc Efficiency
11 25.4 190.5 7.8 32.0 322.6 0.8
12 25.4 240.8 5.3 32.0 375.6

Optimized values of effective thermal conductivity, 0.6

viscosity and arc efficiency 0.9
Figure 1 depicts the variation in the objective function
(i.e. O(f)) with number of iterations. The objective function
decays rapidly in the first 4 iterations in the Levenberg- 0.4 0.7
4 9 14
Marquardt (LM) method and both versions of the conjugate Iteration Number
gradient (CG) method. After that, the objective function
becomes almost constant for several iterations and then starts (b)
fluctuating. Figure 1 shows that the Fletcher and Reeves’s CG 5
Enhancement Factor

method gives somewhat better convergence of the objective

function compared to the other two methods. In the Fletcher 4
and Reeves’s CG method, the minimum value of the objective
function obtained is 0.22 after 13 iterations while LM and
Polak-Ribiere’s CG method produced the value as 0.27 and 3
0.26 in 13 and 14 iterations, respectively. Therefore, the final Conductivity
optimized values of arc efficiency, effective thermal 2 Viscosity
conductivity and effective viscosity are calculated using
Fletcher and Reeves’s CG method. 1
Figures 2(a) and 2(b) show the variation in the values of 2 4 6 8 10 12 14
Iteration Number
non-dimensional weld dimensions, arc efficiency,
k Figure 2: Optimized values of (a) arc efficiency and non-
enhancement factor for thermal conductivity, f k = 1 + T ,
kL dimensional weld dimensions (b) enhancement factor of
μT thermal conductivity and viscosity by using Fletcher-Reeves
and viscosity, fμe = 1 + with iterations for case #2 in Table CG method for case #2 of Table 1.
1. Figure 2(a) shows that non-dimensional weld geometry Figures 2(a) and 2(b) show that enhancement factor for
parameters are initially very large due to the large value of the thermal conductivity and viscosity increases as the calculation
assumed arc efficiency. However, as the calculation progresses and the computed weld pool dimensions tend to
progresses, the weld dimensions decrease and tend to attain agree progressively better with the corresponding
the target value of one. The decreasing trend of the values of experimental values. The optimal value of these unknown
the weld dimensions is somewhat similar to that of arc parameters can be expressed as:
efficiency. These trends are consistent with the fact that arc
efficiency has a major impact on weld pool dimensions.

IV between the two variables is governed by the turbulent Prandtl
η = 0.31 + 4.65 × 10−6 (6) number (Pr) which is defined as:
μ c
IV Pr = T P (9)
k e = 41.80 + 3.17 × 10−5 (W/m-K) (7) kT
where μe = μ L + μT , k e = k L + k T , μT and kT are the turbulent
μe = 0.016 + 1.05 × 10 (kg/m-s) (8) viscosity and turbulent conductivity to account for the
Uw fluctuating fluid movement and resulting enhanced transport
where I is the current (A), V is voltage (V), wf is the wire of heat and mass within the weld pool. The optimized values
feeding speed (m/s) and Uw is the welding speed (m/s). The of μT and kT obtained from Fletcher and Reeves CG method
values of η, ke and μe calculated from equations (6), (7) and results in Prandtl numbers between 0.2 to 0.3. These values of
(8) can be used for the experimental conditions given in Table the Prandtl number lie between laminar and fully turbulent
1 for the GMA welding in spray mode. Equations (7) and (8) flow which suggests that the flow in GMA fillet weld is
also show that the effective value of thermal conductivity and neither laminar nor fully turbulent in traditional sense for the
viscosity is higher than the corresponding molocular values welding conditions given in Table 1. The structure of the flow
even at low heat input per unit length. It is due to high velocity in the weld pool is consistent with the need for enhanced
droplets impact on the weld pool surface which leads to values of transport properties for the heat transfer and fluid
turbulence in the weld pool during GMA fillet welding in flow calculations.
spray mode.
Table 2 show that the arc efficiency increases slightly as Table 2: Optimized values of arc efficiency, η, effective
the input power increases and as the wire feeding rate thermal conductivity, ke, and effective viscosity, μe, obtained
decreases (case #3 and case #8 of Table 1). Approximately, 8 using Fletcher and Reeves CG method for the first eight welds
% increase in the value of arc efficiency is observed with listed Table 1.
decrease in the value of wire feed rate from 211.7 mm/s to
169.3 mm/s for almost same heat input/length (case #3 and No. Heat input/ length η ke μe Pr
case #8 of Table 2). This behavior is consistent with the fact
(KJ/m) (J/m-s-K) (kg/m-s)
that with decrease in wire feed rate, less power is consumed in
melting the wire and more heat is available for the workpiece 1 2302.8 0.58 112.9 0.04 0.29
for the same heat input rate. Table 2 also shows that there can
be 50% variation in the value of the effective thermal 2 1753.4 0.56 96.1 0.03 0.24
conductivity depending on heat input rate. Equations (7) and 3 1608.7 0.59 92.0 0.03 0.23
(8) show that effective thermal conductivity and effective
viscosity increase with increase in the heat input per unit 4 2844.3 0.57 133.8 0.04 0.24
length. The increase in heat input rate enhances mixing in the 5 1389.2 0.55 87. 8 0.03 0.28
weld pool and increases the effective thermal conductivity and
viscosity. The optimized values indicate enhancement factors 6 2253.4 0.57 112.9 0.04 0.29
for thermal conductivity and viscosity to be in the range of 5 7 2446.0 0.54 121.2 0.04 0.26
to 9. This behavior is consistent with the presence of turbulent
flow in the weld pool during GMA welding as reported in the 8 1708. 8 0.55 96.1 0.03 0.24
literature [2, 4,13-16]. Hong and Weckman [13, 14] suggested
an enhancement factor between 12 and 15 for thermal Calculation of effective viscosity and thermal conductivity
conductivity and a factor more than 6 for the viscosity for using mixing-length hypothesis
GTA welding using 150 A current and 25 V based on peak A vorticity based mixing length turbulence model has
temperature analysis in the weld pool and k-ε turbulence been used for the calculation of effective viscosity and
model calculations. Choo and Szekely [16] suggested an effective thermal conductivity. In this model, the
enhancement factor of 8 for thermal conductivity and a factor computational effort is significantly less compared to the k-ε
of 30 for the viscosity at a current of 100 A by matching the turbulence model, since it is algebraic in nature and does not
calculated weld pool geometry with the experimentally require solution of any additional partial differential equations.
determined geometry. They also verified the weld pool shape Hong et al. [15] implemented a vorticity based turbulence
and values of enhancement factors using the k-ε turbulence model in their thermo-fluid calculation in the weld pool using
model. The values available in the literature [3, 7, 13-16] are a constant value of Prandtl mixing length that was calculated
specific to the welding procedure and the specific welding by taking the ratio of the weld pool volume to its interfacial
conditions. Because of the scarcity of data, the available area. The constant mixing length model cannot be applied to
literature cannot be used as a basis for the selection of finger-type penetration characteristic of the GMA fillet
enhanced transport parameters for any specific welding welding process where the flow is constrained by the weld
conditions. boundary which varies with the location. Therefore, mixing
The computed values of ke and μe for various heat inputs length was calculated in the present work using Van Driest
indicate that the rates of transport of momentum and heat are model [8, 9] which can accommodate local variation of
considerably higher than that for laminar flow. The relation mixing length in a weld pool of irregular geometry containing

finger penetration. According to this model, mixing length at 4
distance y from the weld pool boundary is given by: (a)
/ Ao+
lmix = κy[1 − e− y

Penetration (mm)
] (10) 3
The values of the constants used in equation (10) are 0.41 and
26.0, respectively [8, 9], whereas the non-dimensional
distance, y+, from the weld pool boundary is calculated as 2 wwwwwwwwwwwwwwwwww
31V, 4.2mm/s, 169.3mm/s
31V, 6.4mm/s, 169.3mm/s
follows [9]: 33V, 4.2mm/s, 211.7mm/s
33V, 6.4mm/s, 211.7mm/s
ρ ⎛ ∂u ⎞ 1 31V, 4.2mm/s, 169.3mm/s
y+ = y ⎜ ⎟ (11) 31V, 6.4mm/s, 169.3mm/s
μ ⎝ ∂y ⎠ w 33V, 4.2mm/s,
33V, 6.4mm/s,
⎛ ∂u ⎞ 275 300 325 350
The term ⎜ ⎟ in equation (11) represents the velocity Arc current (Amp)
⎝ ∂y ⎠ w 14
gradient at the weld pool boundary. For the three-dimensional (b)
flow in the weld pool, the turbulent viscosity is calculated 12
using Badwin-Lomax model [8, 9] as follows:

Throat (mm)
μT = ρl2mix ω (12) 10

where, ω is the magnitude of the vorticity vector given by: wwwwwwwwwwwwwwwwww

31V, 4.2mm/s, 169.3mm/s
31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
1/ 2
⎡⎛ ∂v ∂u ⎞ ⎛ ∂w ∂v ⎞ ⎛ ∂u ∂w ⎞2 ⎤ 2 2 33V, 6.4mm/s, 211.7mm/s

ω = ⎢⎜ − ⎟ + ⎜ ⎟ ⎥
31V, 4.2mm/s, 169.3mm/s
− ⎟ +⎜ − (13) 6 31V, 6.4mm/s, 169.3mm/s
⎢⎝ ∂x ∂y ⎠ ⎝ ∂y ∂z ⎠ ⎝ ∂z ∂x ⎠ ⎥ 33V, 4.2mm/s, 211.7mm/s
⎣ ⎦ 33V, 6.4mm/s, 211.7mm/s
275 300 325 350
Viscosity (Vorticity) (a) Arc current (Amp)
Enhancement Factor

17 Viscosity (Proposed)
14 (c)
Leg-length (mm)


9 10

31V, 4.2mm/s, 169.3mm/s

8 31V, 6.4mm/s, 169.3mm/s
5 33V, 4.2mm/s, 211.7mm/s
33V, 6.4mm/s, 211.7mm/s
6 31V, 4.2mm/s, 169.3mm/s
1 31V, 6.4mm/s, 169.3mm/s
1.5 2 2.5 3 4
33V, 4.2mm/s, 211.7mm/s
Heat Input (KJ/ mm) 33V, 6.4mm/s, 211.7mm/s

275 300 325 350

Arc current (Amp)
Conductivity (Vorticity)
Conductivity (Proposed)
Enhancement Factor

17 Figure 4: Weld bead geometric parameters as a function of

welding current: (a) penetration, (b) actual throat, and (c)
13 leg-length obtained using the optimized values of arc
efficiency, effective thermal conductivity and effective
viscosity. The open symbols represent the experimental results
9 while the filled symbols indicate the corresponding calculated
result. The standard deviations of the experimental data are
5 shown by error bars. The values indicated in the legends
represent the arc voltage, welding speed and wire feed rate,
1.5 2 2.5 3
Heat Input (KJ/ mm) The calculated average enhancement factor of viscosity
by using equation (12) for welding conditions listed in Table 1
Figure 3: Computed values of enhancement factor for (a) is shown in fig. 3(a). This shows a reasonably good agreement
viscosity, and (b) thermal conductivity using estimated values between the values of enhancement factor in viscosity by
of unknown parameters by using proposed model and vorticity using vorticity based mixing length turbulence model and the
based turbulence model for different welding conditions listed proposed equation (8). Using the value of effective viscosity
in Table 1. and Pr, the enhancement factor for thermal conductivity was

calculated. Figure 3 shows the calculated enhancement factors References
for thermal conductivity and viscosity as a function of heat
input by using vorticity based turbulence model and the 1. C. H. Kim, W. Zhang, T. DebRoy, Modeling of
proposed equations (7) and (8). The calculated enhancement temperature field and solidified surface profile
factors for thermal conductivity and viscosity by using during gas metal arc fillet welding, J. Appl. Phys. 94,
vorticity based turbulence also increase with increase in the 2667-2679 (2003).
heat input per unit length as obtained by equations (7) and (8). 2. W. Zhang, C. H. Kim, T. DebRoy, Heat and fluid
The reasons for slightly lower values of the average flow in complex joints during gas metal arc welding -
enhancement factors for thermal conductivity and viscosity by part I: numerical model of fillet welding, J. Appl.
the vorticity based turbulence model compared to equations Phys. 95, 5210-5219 (2004).
(7) and (8) are not known. However, the fact that the values 3. W. Zhang, C. H. Kim, T. DebRoy, Heat and fluid
of constants κ and A o in equation (10) were obtained from flow in complex joints during gas metal arc welding -
experiments in parabolic flows may be a contributing factor. part II: application to fillet welding of mild steel, J.
Figures 4(a), 4(b) and 4(c) depict comparison between Appl. Phys. 95, 5220-5229 (2004).
the computed and the experimentally obtained weld pool 4. A. Kumar, W. Zhang, and T. DebRoy, Improving
dimensions using the optimized values of arc efficiency, reliability of modelling heat and fluid flow in
effective thermal conductivity and effective viscosity obtained complex gas metal arcfillet welds—part I: an
from equations (6) to (8) for the welding conditions listed in engineering physics model, J. Phys.D: Appl. Phys.
Table 1. These figures show satisfactory agreement between 38, 119-126 (2005).
the computed and the experimentally obtained weld geometry 5. A. Kumar and T. DebRoy, Improving reliability of
for various welding conditions. The reliability of numerical modelling heat and fluid flow in complex gas metal
heat transfer and fluid flow calculations can be significantly arcfillet welds—part II: application to welding of
enhanced by using the optimized values of uncertain welding steel, J. Phys.D: Appl. Phys. 38, 127-134 (2005).
parameters from a limited volume of measured weld 6. A. Kumar and T. DebRoy, Guaranteed Fillet Weld
dimensions. Geometry from Heat Transfer Model and
Multivariable Optimization, Int. J. Heat Mass Tran.
Summary and conclusions 47, 5793-5806 (2004).
7. A. De and T. DebRoy, Probing unknown welding
parameters from convective heat transfer calculation
To improve the reliability of numerical heat transfer and
and multivariable optimization, J. Phys. D: Appl.
fluid flow calculations, a comprehensive model has been
Phys. 37, 140-150 (2004).
developed and tested which embodies a heat transfer and fluid
8. D. C. Wilcox, Turbulence Modeling for CFD, DCW
flow sub-model and an algorithm for searching and optimizing
Industries, California (1993).
the values of uncertain process variables from a limited
9. B. E. Launder and D. B. Spalding, Lectures in
volume of experimental data. This model was able to estimate
Mathematical Models of Turbulence, Academic
the unknown values of arc efficiency, effective thermal
Press, London (1972).
conductivity and effective viscosity based on only a few
10. S.V. Patankar, Numerical Heat Transfer and Fluid
experimental measurements. The optimized values of arc
Flow, Hemisphere, NY (1980).
efficiency, effective thermal conductivity and effective
11. J. F. Lancaster, The Physics of the Welding (2nd Edn),
viscosity were found to depend on the welding conditions. The
Pergamon, Oxford, (1986).
enhancement factors for thermal conductivity and viscosity
12. S. Kumar and S. C. Bhaduri, Three-dimensional finite
were in the range of 5 to 9 for the welding conditions used in
element modeling of gas metal arc welding, Metall.
this study. The average values of the enhancement factors for
Trans. B 25, 435- 441 (1994).
thermal conductivity and viscosity using vorticity based
13. K. Hong, D. C. Weckmann, A. B. Strong and W.
mixing length turbulence model agreed well with the values
Zheng, Vorticity based turbulence model for
predicted by the proposed model. The reliability of the
thermofluids modeling of welds, Sci. Technol. Weld.
numerical heat transfer and fluid flow calculations in the weld
Joining, 8, 313-324 (2003).
pool can be significantly improved by including a suitable
14. K. Hong, D.C. Weckmann, A.B. Strong, and W.
optimization model to determine the uncertain welding
Zheng, Modelling turbulent thermofluid flow in
parameters from a limited volume of experimental data on
stationary gas tungsten arc weld pools, Sci. Technol.
weld dimensions.
Weld. Joining 7, 125-136 (2002).
15. K. Hong, D.C. Weckman, A. B. Strong and E. Pardo,
Acknowledgements in Proc. First Int. Conf. on Transport Phenomena in
Processing, Ed. Secuk I. Guceri, Technomic Pub.,
This research was supported by a grant from the U.S. 626-635 (1992).
Department of Energy, Office of Basic Energy Sciences, 16. R. T. C. Choo and J. Szekely, The possible role of
Division of Materials Sciences, under grant number DE- turbulence in GTA weld pool behaviour, Weld. J. 73,
FGO2-01ER45900. Mr. Amit Kumar gratefully acknowledges 25-31 (1994).
award of a Fellowship from the American Welding Society.

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Transport Phenomena and Genetic Algorithm based Window of Welding

Variables to Achieve a Target Gas Metal Arc Fillet Weld Geometry
A. Kumar and T. DebRoy
The Pennsylvania State University, University Park, PA, USA

Abstract sets to achieve a target weld geometry, based on scientific

principles, would be an important step to achieve this goal.
Current transport phenomena based welding models are What is needed, and not currently available, is the capability
designed to calculate temperature and velocity fields and other to systematically determine multiple paths to tailor weld
weld attributes such as the weld geometry. However, in many geometry and assess robustness of each individual solution to
instances, the desired attribute such as the geometry is known achieve safe welds. Here we show that a genetic algorithm
and the correct set of welding parameters need to be assisted numerical model of heat transfer and fluid flow in
determined. The mismatch between the practical need and the fusion welding can systematically determine multiple paths for
capability of the current models has restricted the use of these tailoring weld attributes and help in the production of reliable
powerful models. This paper shows that by combining a welds.
numerical thermo-fluid model with an appropriate genetic The welding variables can be linked with the weldment
algorithm based optimization scheme, many possible sets of attributes by means of statistical regression analysis [2, 3],
welding variables that are capable of producing a given weld artificial neural network [3, 4] or phenomenological modeling
geometry can be determined. A three-dimensional numerical [5-10]. The use of regression and neural network based
heat transfer and fluid flow model for the gas metal arc models require large volume of experiments that are time
(GMA) welding of fillet joints is combined with a genetic consuming and expensive [8, 9]. The phenomenological
algorithm (GA) based optimization scheme to obtain a models simulate the essential physical processes in welding
window of welding variables. To reduce the computation time, through the equations of conservation of mass, momentum and
the model is parallelized to run on multiple processors energy. These models are unidirectional in nature because
simultaneously. The approach outlined in this paper they are designed to function with known welding variables as
completely restructures numerical heat transfer and fluid flow input [8, 9]. The outputs from these models include
calculations and empowers users to determine a window of temperature and velocity fields, thermal cycles, and the weld
input variables consisting of several sets of welding variables shape and size. The available phenomenological models
all of which would lead to a target weld geometry. cannot currently predict multiple combinations of welding
variables that can result in a target weld attribute.
Introduction There are three main attributes of a model that are required for
systematic tailoring of a weld attribute such as weld geometry.
The welding related expenditure in the U.S. in the year First, the procedure should embody an adequate
2000 was approximately $34 billion, equivalent to about $325 phenomenological description of the complex physical
per household [1]. The large economic impact of welding processes in welding. Second, the model must have bi-
originates from its extensive use in industries that support the directional capability. In other words, in addition to the
nation’s defense, infrastructure, and standard of living. capability of the traditional forward (unidirectional) models to
Systematic tailoring of weld attributes can potentially save life compute the weld shape and size from a given set of welding
and property by preventing catastrophic failures of large variables, it should also be capable of reverse modeling, i.e., it
welded structures. Since there are a large number of process should be able to systematically search and optimize the
variables in welding, the desired weld attributes such as the variables needed to produce the desired weld geometry.
weld geometry and structure are commonly produced by Finally, the methodology must be able to determine a
empirically adjusting the welding variables. However, this population of solutions, i.e., various welding variable sets to
approach does not always produce optimum welds and attain a target weld attribute. The gas-metal-arc (GMA) fillet
inappropriate choice of variables can lead to poor welds. welding process involves nonlinearities and complex
Systematic tailoring of weld attributes based on scientific interactions among various welding variables [5-10]. As a
principles still remains an important milestone in changing result, a given weld attribute such as the fusion zone geometry
welding from an empirical art to a mainstream science-based is attainable using multiple combinations of welding variables.
technology. The ability to determine multiple welding variable Since multiple paths can lead to a target weld geometry, the

Copyright © 2006 ASM International® 91

classical gradient-based search and optimization methods that algorithm [16]. The three dimensional temperature field in the
produce a single optimum solution cannot be used. GMA weld, computed from the model, is used to calculate the
Furthermore, most of the classical optimization algorithms are weld dimensions, i.e., penetration, throat and leg-length. An
designed to solve a specific type of problems. For example, effective search for multiple welding variable sets to attain a
the conjugate directions or conjugate gradient methods have given geometry requires many runs of the heat transfer and
better convergence for solving quadratic objective functions fluid flow model. Only because of the recent advances in
having one optimal solution. The Frank-Wolfe’s successive computational hardware and software can such a
linear programming method works efficiently on linear computationally intensive task for tailoring weld attributes be
functions and constraints. Its performance largely depends on undertaken. More details of the numerical model of heat
the chosen initial conditions for solving non-linear problems. transfer and fluid flow of gas metal arc fillet welding are
Thus, many of the classical methods of optimization are available in the literature [6, 9].
suitable for finding a single value of a parameter as these The genetic algorithm based search for multiple sets of
methods use a point-by-point approach [11-15], where one welding variables to achieve a target weld geometry starts
relatively imperfect solution in each iteration is modified to a with many initial sets of randomly chosen values of the four
different more appropriate solution. Therefore, a combination most important welding variables, i.e., current, voltage,
of one of these classical optimization methods with a welding speed and the wire feed rate. A systematic global
phenomenological heat and fluid flow model can provide only search is next undertaken to find multiple sets of values of
a single local optimum solution where multiple solutions exist. these four welding variables that lead to least error between
In contrast, genetic algorithms (GA) mimic nature’s the calculated and the target weld dimensions, i.e., penetration,
evolutionary principles to derive its search towards a throat and the leg-length. The phenomenological model of
population of optimal solutions [11-15]. In the context of heat transfer and fluid flow calculates the values of these weld
welding, GA can systematically search for multiple dimensions for each set of input welding variables. The
combinations of welding variable sets that comply with the chosen values of welding variables do not always produce the
phenomenological laws of welding physics and improve with desired weld dimensions and the resulting mismatch between
iterations following certain rules of evolution. Therefore, the the computed and the desired weld dimensions is expressed by
GA, when combined with a transport phenomena based the following objective function, O(f):
model, has the potential to produce multiple sets of welding ⎛ pc ⎞
⎛ tc ⎞
⎛ lc ⎞

variables that can result in a particular weld attribute. We O (f ) = ⎜ − 1⎟ + ⎜ − 1⎟ + ⎜ − 1⎟ (1)

⎜ pt ⎟ ⎜ tt ⎟ ⎜ lt ⎟
show here that multiple sets of welding variables that are ⎝ ⎠ ⎝ ⎠ ⎝ ⎠
capable of producing a target weld geometry can be calculated c c c
where p , t and l are the computed penetration, throat and
by combining a numerical heat transfer and fluid flow model
for GMA fillet welding with an appropriate GA. the leg-length of the weld bead, respectively and p t , t t and
l t are the corresponding target or desired values of these three
Mathematical model parameters. The objective function, O(f), depends on four
main welding variables, i.e., current, I, voltage, V, welding
The main computational engine used here is a well- speed, U, and the wire feed rate, wf.
tested heat transfer and fluid flow model [5-10] that solves the ⎛ I V U wf ⎞
O(f) = O ( f1 , f 2 , f 3 , f 4 ) = O ⎜ , , , ⎟ (2)
equations of conservation of mass, momentum and energy in ⎝ I r Vr U r (w f ) r ⎠
three dimensions. This forward model takes into account the In Eq. 2, the reference values, Ir, Vr, Ur and (wf)r
complex fillet joint shape, the deformation of the weld pool represent the order of magnitude of the welding variables.
top surface, heat transfer by the hot metal droplets and the Note that Eq. 2 is made non-dimensional to preserve the
addition of the filler metal from the consumable electrode. The importance of all four welding variables by making their non-
convection in the weld pool is driven mainly by the spatial dimensional values comparable in magnitude. The GA
gradient of interfacial tension and the electromagnetic force produces new individuals, or sets of welding conditions, with
field and to a much lesser extent by the buoyancy force. The iterations based on the evolutionary principles as explained in
complex joint shape and the severely deformed weld pool the literature [14, 15].
surface in fillet welding require the use of deformable The genetic algorithm used in the present study is a
curvilinear grid system for accurate calculation of heat transfer parent centric recombination (PCX) operator based
and fluid flow. Therefore, the governing equations are generalized generation gap (G3) model [12, 14, 15]. To reduce
transformed from the Cartesian to curvilinear coordinate the computation time, the model is parallelized to run on
system [5,6]. The transformed governing equations are multiple processors simultaneously. The multiple island (or
discretized using the control volume method [16]. The deme) based parallelized GA model [12, 17] is used to achieve
discretized equations for enthalpy, three components of rapid search for the multiple sets of welding variables by
velocity and pressure are formulated by integrating the dividing the search space into multiple islands. Rapid
corresponding governing equation over all the interior control convergence is achieved when each island is assigned to a
volumes in the computational domain. The equations, separate processor because the size of each island is smaller
typically several hundreds of thousands of them, are then than the total population. After each iteration, a few solutions,
solved simultaneously at every iteration by a Gaussian selected based on their fitness or objective function value, are
elimination technique known as the tri-diagonal matrix

allowed to migrate to other islands [17]. The migration of the 4
best individuals to replace the worst individuals in the (a)
neighboring processors (islands) enhances the speed of

Penetration (mm)
convergence [17]. Figure 1 illustrates the working of the 3
model to find the window of welding parameters to achieve a
target weld geometry.
2 wwwwwwwwwwwwwwwwww
31V, 4.2mm/s, 169.3mm/s
31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
33V, 6.4mm/s, 211.7mm/s
1 31V, 4.2mm/s, 169.3mm/s
31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
33V, 6.4mm/s, 211.7mm/s
275 300 325 350
Arc current (Amp)


Throat (mm)

31V, 4.2mm/s, 169.3mm/s
8 31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
33V, 6.4mm/s, 211.7mm/s
31V, 4.2mm/s, 169.3mm/s
6 31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
33V, 6.4mm/s, 211.7mm/s
275 300 325 350
Arc current (Amp)
Figure 1: Flow chart of the parallelized generalized
generation gap (G3) model
14 (c)
Results and discussion
Leg-length (mm)

A reliable forward model provides a useful link between
the welding variables and the weld attributes. This link reveals
significant insight not just about the effect of welding 31V, 4.2mm/s, 169.3mm/s
variables but the very strategy of tailoring weld attributes 8 31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
based on scientific principles. For example, fig. 2 indicates the 33V, 6.4mm/s, 211.7mm/s
computed values of penetration, throat and leg-length for 6 31V, 4.2mm/s, 169.3mm/s
various welding conditions. These results demonstrate two 31V, 6.4mm/s, 169.3mm/s
33V, 4.2mm/s, 211.7mm/s
special features. First, a comparison of the computed and the 4 33V, 6.4mm/s, 211.7mm/s
experimental weld dimensions indicate that the forward model 275 300 325 350
provides both the expected trends and the correct values for Arc current (Amp)
various combinations of welding variables I, V, U and wf.
The agreement provides confidence about the adequacy of Figure 2: Weld bead geometric parameters as a function of
quantitative representation of the essential physical processes welding current: (a) penetration, (b) throat, and (c) leg-
in the phenomenological model and demonstrates its promise length. The open symbols represent the experimental results
to serve as the main computational engine in the overall goal while the filled symbols indicate the corresponding calculated
of systematic tailoring of weld attributes. Second, and more result. The standard deviations of the experimental data are
interesting, the results show that a given weld dimension can shown by error bars. The values indicated in the legends
be obtained using various sets of welding variable values. represent the arc voltage, welding speed and wire feed rate,
Since GA can provide a population of solutions, the heat respectively.
transfer and fluid flow model must be combined with an Let us consider an example to illustrate how the overall
appropriate GA to tailor weld attributes from scientific model to tailor weld geometry from scientific principles
principles. works. The task involves three steps. First, a target weld

geometry is selected by specifying one set of values of processors with I and V plotted as their product for
penetration, throat and leg-length. Second, the model is run to accommodating all variables in three dimensional space.
obtain multiple combinations of welding variable sets to Values of the variables I, V, U and wf were chosen in different
achieve this target geometry. Third and final, the results ranges on the processors to maintain the diversity and explore
obtained from the model are adequately verified. Let us now the solution space. To analyze the effect of island model on
examine these three steps in detail. the solution, two different schemes were used for initializing
The first step, i.e., the specification of the target the initial population on each processor. In the first scheme,
geometry, in principle, involves stating any plausible only the values of U were chosen in different range on
combinations of the three weld dimensions, i.e., penetration, different processors while I, U and wf were selected randomly
throat and leg-length. However, it is useful to specify the between 250.0 to 400.0 A, 25.0 to 40.0 V and 100.0 to 250.0
weld dimensions from an actual welding experiment as a mm/s, respectively (fig. 3(a)). In the other scheme (fig. 3(b)),
target geometry. Such a choice provides a useful advantage the diagonal of the solution hyperspace was divided into parts
because one of the solutions, i.e., the welding variable set used depending on the number of processors to select the values of
to produce this target geometry is known. If the model works the variables randomly.
correctly, the solutions obtained from the model must include
a set of welding variables that are fairly close to a set used in 1.6
the experiment. It should be noted that the ability of the (a) Processor 1
Processor 2

Objective function
model to produce this solution is only a necessary, but not Processor 3
sufficient, component of the model verification. Since the 1.2
model produces multiple solutions, other solutions obtained 1
from the model also need to be verified.
(a) 0.6
3 4 5 6 7 8
Welding velocity (mm/s)
0.6 Processor 1
(b) Processor 2
Processor 3
Objective function

(b) 0.4


3 4 5 6 7 8
Welding velocity (mm/s)

Figure 4: The value of the objective function corresponding to

welding speed variable of all the individuals in the initial
Figure 3: Initial values of individual welding variable sets (a)
population generated using (a) scheme 1 and (b) scheme 2.
Scheme 1-The initial ranges for these variables are as follows:
I (250.0 to 400.0 A), V (25.0 to 40.0 V), wf (100.0 to 250.0
Figures 4(a) and (b) show the initial values of the
mm/s) while U is in the range of 2.25 to 4.5 cm/s, 4.25 to 6.5
objective function for various values of welding speed on all
cm/s and 5.5 to 8.0 cm/ s, respectively on the three processors.
three processors for both the schemes. An individual with a
(b) Scheme 2-The initial ranges for these variables are defined
low objective function indicates correct values of current,
along the diagonal of the search space.
voltage, welding speed, and wire feed rate that result in low
error between the computed and the target weld geometry.
The second step, i.e., the operation of the model to
For example, fig. 4 shows the computed values of the
tailor weld geometry starts with specifying a large population
objective functions for all the individuals depicted in fig. 3.
of potential solutions. Each individual in the population
The values of objective function of the randomly created
represents a set of randomly chosen values of welding
individuals is lower for scheme 2 than that of scheme 1. It
variables. Figure 3 depicts the initialization of individuals, i.e.,
shows that solution may exist in the region of hyperspace
sets of values of I, V, U and wf of each population on different

diagonal where the individuals in scheme 2 are selected. (a)
Figure 4 also shows that for many sets of welding variables,
the values of the objective function, O(f), are fairly low
indicating that each of these welding variable sets can produce
a weld geometry that is close to the target geometry.
To prevent premature loss of useful solutions, different
values of GA parameters (i.e. the random seed to initialize the
population) were set on each processors so that some
populations converge slower than the others. This procedure
maintains the diversity for longer duration in the population.
These welding variable sets are then improved iteratively
using a combination of GA and the forward phenomenological (b)
model as shown in fig. 1. The progress of the iterations is
monitored by calculating the objective function values for
each set of welding variables (individuals) after each iteration.
Figure 5 shows that with increase in the number of iterations,
the objective function decays rapidly for the best individuals
whereas the average value of the objective function of the
whole population decreases at a relatively slower pace for both
the schemes. It is because the GA tries to explore the solution
space which can produce sets of welding parameters that may
have higher O(f). The diversity of population decreases with
the increase in the number of iterations and the solution starts
crowding in some specific regions providing the window of Figure 6: The plots show the welding variable sets that
welding parameters. produced low values of the objective function, O(f) with
iterations. (a) individuals after five iterations with O(f) less
O(f) evaluations than 8.0×10-3, and (b) individuals after ten iterations with O(f)
100 150 200 less than 4.0×10-3.
Scheme 1 (Processor 1)
Scheme 1 (Processor 2) The third and the last step involves verification of the
Scheme 1 (Processor 3)
Scheme 2 (Processor 1) computed solutions. Since the target weld geometry was
Average O(f)

Scheme 2 (Processor 2) produced by an experiment, a preliminary test is to check if

0.2 Scheme 2 (Processor 3)
the population of solutions produced by the model include an
individual set of welding variables that is very close to, if not
the same as, that used to produce the weld. These
0.1 experimental welding variables are presented in Table 1
together with the computed optimum welding variable sets.
The table shows that the computed values of current, voltage,
welding speed and the wire feed rate in the individual solution
(b) in the table all lie within less than 1% of the corresponding
20 40 60 80 experimental welding variables. However, the model also
Generation predicted other individual solutions (a) through (f) listed in the
Fig. 5: Variation of average objective function of all the table. The accuracy of these individual solutions was
individuals in the population with number of generations. examined by calculating weld geometry for each welding
variable sets (a) and (c) through (f) in Table 1 and comparing
Figures 6(a) and (b) indicate several clusters of the computed weld dimensions with the target weld geometry.
welding variable combinations that have objective function The values of the four welding variables listed in the table
values lower than 0.008 and 0.004, corresponding to the 5th differ from each other considerably. For example, the voltage,
generation and 10th generation of individuals using scheme 2. current, welding speed and wire feed rate values vary between
What is of special interest in fig. 6 is that sets of welding solutions by 38%, 12%, 47% and 50%, respectively. In
variables are distributed throughout the welding variable space solutions (a) and (f), the powers required to achieve the target
signifying the existence of multiple paths to attain the target geometry were 9.7 and 14.2 kW, respectively. Furthermore,
geometry. Improved solutions are obtained with iterations (or other welding parameters such as the welding speed and the
generations) as observed from the reduction of the values of wire feed rate were also significantly different in the two
the objective function of the best individuals. When the values solutions. All these differences in the important welding
of the objective functions are low and they do not decrease variables indicate significant diversity in the paths, all of
further with iterations, the computed welding variable sets which lead to the same set of target weld dimensions.
constitute the final solutions and these are presented in Table

Table 1: Optimized window of welding parameters, i.e., arc 3. I. S. Kim, S. H. Lee, P. K. D. V. Yarlagadda, Comparison
current (I), arc voltage (V), welding speed (U) and wire feed of multiple regression and back propagation neural
rate (wf) to achieve the following target weld dimensions: network approaches in modeling top bead height of
penetration = 3.8 mm, throat = 11.0 mm, and leg-length = multipass gas metal arc welds, Sci. Technol. Weld Joining,
12.5 mm. The target weld geometry was obtained 8, 347-352 (2003).
experimentally using the following welding variables: I = 4. H. B. Smartt, J. A. Johnson, C. J. Einerson, and G. A.
331.4 A, V= 31 V, U = 4.2 mm/s and wf = 211.7 mm/s. Cordes, Development of a Connectionist Fuzzy Logic
System for Control of Gas Metal Arc Welding, in
No. I (Amp) V (Volt) U (mm/s) wf (mm/s) Proceedings of the Artificial Neural Networks in
Engineering (ANNIE’91), Ed. C. H. Dagli, S. R. T.
(a) 315.0 30.7 4.1 239.3 Kumara, Y. C. Shin, Eds. ASME, NY, 711-716 (1991).
(b) 330.0 31.2 4.2 212.1 5. C. H. Kim, W. Zhang, T. DebRoy, Modeling of
temperature field and solidified surface profile during gas
(c) 331.5 30.2 4.0 228.4 metal arc fillet welding, J. Appl. Phys. 94, 2667-2679
(d) 321.5 30.7 4.1 239.2 (2003).
6. W. Zhang, C. H. Kim, T. DebRoy, Heat and fluid flow in
(e) 354.0 31.4 4.1 230.4 complex joints during gas metal arc welding -part I:
(f) 338.0 41.9 5.9 319.2 numerical model of fillet welding, J. Appl. Phys. 95, 5210-
5219 (2004).
7. W. Zhang, C. H. Kim, T. DebRoy, Heat and fluid flow in
Concluding remarks complex joints during gas metal arc welding -part II:
application to fillet welding of mild steel, J. Appl. Phys.
The ability of the model to correctly predict multiple 95, 5220-5229 (2004).
welding variable sets that can lead to the target weld 8. A. Kumar and T. DebRoy, Guaranteed Fillet Weld
dimensions proves that by combing the principles of Geometry from Heat Transfer Model and Multivariable
evolutionary biology with welding physics, a useful Optimization, Int. J. Heat Mass Tran. 47, 5793-5806
phenomenological framework can be created to systematically (2004).
tailor a weld attribute via multiple paths. Although the work 9. A. Kumar, W. Zhang, and T. DebRoy, Improving
reported here focuses on tailoring of weld geometry, these reliability of modelling heat and fluid flow in complex gas
results provide hope that with the development of new metal arcfillet welds—part I: an engineering physics
methodologies at the crossroads of basic and applied sciences, model, J. Phys.D: Appl. Phys. 38, 119-126 (2005).
the promise of science based tailoring of structure and 10. A. Kumar and T. DebRoy, Improving reliability of
properties of weldments may also become attainable in the modelling heat and fluid flow in complex gas metal arc
future. fillet welds—part II: application to welding of steel, J.
Phys.D: Appl. Phys. 38, 127-134 (2005).
Acknowledgement 11. D.E. Goldberg, Genetic Algorithm in Search, Optimization
and Machine Learning, Addison-Wesley, MA (1989).
This research was supported by a grant from the U.S. 12. T. Back, D.B. Fogel, Z. Michalewicz, Eds., Handbook of
Department of Energy, Office of Basic Energy Sciences, Evolutionary Computations, IOP Publishing Ltd., Oxford
Division of Materials Sciences, under grant number DE- University Press (2000).
FGO2-01ER45900. Mr. Amit Kumar gratefully acknowledges 13. K. Deb, Multi-objective optimization using evolutionary
award of a Fellowship from the American Welding Society. algorithms, Wiley, New York (2001).
We have received helpful comments from Professors L.Q. 14. A. Kumar, S. Mishra, J. W. Elmer and T. DebRoy,
Chen, W. M. Small and R. Roy of Penn State, and Dr. S.A. Optimization of Johnson-Mehl-Avarami Equation
David of Oak Ridge National Laboratory during preparation Parameters for α-Ferrite to γ-Austenite Transformation in
of this manuscript. Steel Welds using a Genetic Algorithm, Metall. Mat.
Trans. A. 36, 15-22 (2005).
References 15. K. Deb, A. Anand and D. Joshi, A computationally
efficient evolutionary algorithm for real parameter
1. American Welding Society, Welding Related optimization, Evolutionary Computation, 10, 371-395
Expenditures, Investments, and Productivity Measurement (2002).
in U.S. Manufacturing, Construction, and Mining 16. S.V. Patankar, Numerical Heat Transfer and Fluid Flow,
Industries, AWS, Miami, Fl. (May 2002). Hemisphere, NY (1980).
2. P.W. Fuerschbach, G.R. Eisler, and R.J. Steele, Weld 17. E. Cantú-Paz, Efficient and Parallel Genetic Algorithms,
Procedure Development with OSLW- Optimization Kluwer Academic Publishers, Boston (2000).
Software for Laser Welding, in Proc. of the Trends in
Welding Research, Pine Mountain, Georgia, ASM
International, OH (1998).

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Quantitative Observations of Surface Flow and

Solidification on Autogenous GTA Weld Pools
D. DeLapp, G. Cook, A. Strauss, W. Hofmeister
Vanderbilt Uuniversity, School of Engineering, Nashville, Tennessee, USA

Abstract metal is opaque so subsurface observation and measurements

are not presently possible. Temperature distribution and fluid
Autogenous GTA welds were performed on 3mm carbon steel flow patterns are measurements that would prove to be quite
plates. A macro camera and high speed frame grabber camera useful in understanding the melting and solidification of
system was employed to make direct observations and welding processes. Numerical modeling of the weld pool is
measurements of weld pool solidification and surface flows. the current method used to learn about the volume behavior of
Observations of the solidification process revealed fluctuating the weld pool. This method relies on an incomplete knowledge
rates of solidification, including rapidly occurring remelt of the thermo-physical properties of the liquid metal and how
events. Observation of the motion of normally occurring they vary with temperature, particularly for alloy systems [1].
particles floating on the surface of the weld pool revealed
rotational flow patterns. Various researchers have observed weld pool fluid behavior
and solidification using high speed imagery. Woods [2]
Solid growth often occurred in a direction tangent to the solid observed flow patterns on the bottom surface of full
liquid interface. Weld ripple formation and subsequent remelt penetration, stationary and moving GTA welds. Flow patterns
was also observed. Remelting of the solid when it occurred were made visible by the presence of normally occurring
was rapid, on the order of 300 mm/s, proceeding in the same oxide particles. Observed motion was correlated to mixing
counterclockwise direction as the weld pool circulation. patterns of dissimilar materials added to the weld pool
followed immediately by a rapid quench. Ercer [3] made
Subsequent particle motion measurements revealed significant observations of floating oxide particles on pulsed GTA welds
fluctuations in surface flow velocity. Assuming the flow was made on Fe-26Ni. Three types of pool motion were noted.
turbulent or at least in transition to turbulence based on a non- Radial flow, dominant in the early stages of the high and low
zero vorticity evaluation, the turbulent intensity was found to pulses, was attributed to sudden changes in arc pressure.
be 22%. The flow had a Reynolds number of 107, based on an Faster flow velocities corresponded to a larger time rate of
average velocity of 25.4 mm/s, and bead width of 2.5mm. change in current. Two symmetrical rotational flows appeared
during periods of constant current, were attributed to
Keywords: GTAW, convection, turbulence, vorticity, flow electromagnetic affects. Up and down pool surface oscillations
visualization. were attributed to changes in arc pressure. DeLapp [4]
observed the formation of surface ripples in welds made on
Introduction mild steel and stainless steel. Ripples were found to have a
fluctuating rate of formation and growth sometimes proceeded
Assessing weld bead characteristics (post weld) is readily in a direction tangential to the edge of the weld pool. Hall [5],
done using the wide variety of destructive and nondestructive [6], observed melting and solidification on the leading and
tests presently available. Assessing the weld pool behavior trailing edges, respectively, of moving GTA welds in stainless
(post weld) is an area of welding research which has been steel. Images of solidification revealed dendritic growth,
done primarily by metallographic examination of sectioned melting and ripple formation. Analysis of the images showed
and polished specimens, then extrapolating back to the weld that the solid-liquid interface advanced at a fluctuating rate.
pool behavior that must have been present to create the
observed microstructures. Actual measurements on the liquid The nature of the fluid flow within the weld pool is extremely
weld pool are more difficult to obtain. important, as the flow dictates the distribution of momentum,
energy (heat) and species (alloy concentration). In a laminar
The only part of a weld pool that is observable is the free flow regime, fluid particles move in a predictable manner,
surface although it is masked by the bright arc light. Liquid following well defined paths, with uniform velocity profiles.

Copyright © 2006 ASM International® 97

Fluid motion is predictable and in general is readily modeled.
Turbulent flow is markedly more complex than laminar flow.
There is no one single precise definition of turbulence
applicable to all turbulent flows [7-9].

Weld pool flow models initially assumed fluid flow to be

laminar. Better agreement between numerical model and
experimental result occurred when turbulent flow models were

The hypothesis of turbulent weld pool flow behavior has been

tested experimentally on a limited basis. Malinowski-
Brodnicka [10] applied an axial magnetic field to welds made
on austenitic stainless steel. An annular flow velocity of
0.5m/s was reported. A Reynolds number of 3000 was Figure 1: Welding torch end of the macro view camera.
reported and the flow was considered most probably turbulent. Orientation is 45° to the horizontal and 90° to the direction of
A critical Reynolds number associated with weld pool flow travel. A 105mm lens gave adequate magnification, depth of
was not cited. field and sufficient working distance to protect the lens from
Direct measurement of surface velocity has been done and the Experiments were run on 3.2mm x 102mm x 457mm 1018
corresponding Reynolds number calculated. Turbulent flow is cold rolled mild steel plate. All weld specimens were prepared
assumed based on the magnitude of the Reynolds number. by first cleaning off the oil coating with mineral spirits
Hong [11] stated that the Reynolds number of 3000 reported followed by a final cleaning with denatured alcohol. All welds
by Malinowski-Brodnicka was larger than the classical critical were non-reinforced autogenous. To avoid the heat sink
Reynolds number for transition from laminar to turbulent flow effects of direct contact between the 25.4mm thick steel
in pipes, Re=2100, and concluded that the flow was probably worktable and the thin 3.2mm workpiece, the weld specimens
turbulent. were placed on 3.2mm shims then clamped in place. Clamp
bolts were lightly torqued to provide uniform clamping force
This work will present an experimental means to quantify the and minimize heat transfer to the table.
vorticity vector and show that the weld pool flow is rotational
and is therefore not laminar.

Experimental Apparatus
Welding experiments were carried out using a semi-automated
GTAW apparatus. The experimental apparatus is composed of
the welding power supply, traversing system, macro camera,
and digital frame grabber camera. Welding power was
supplied by a TMC 500 welding power supply to a water
cooled torch assembly mounted to the weld traversing system.
On this apparatus the torch is held stationary and the work
Figure 2: Screen capture of the image playback controls of the
piece is traversed. The torch is mounted on slides which have
viewer program The yellow, red and green cursers can be
three degrees of freedom allowing the weld pool to be
moved to any pixel in the image. The x-y coordinate pixel
properly positioned in the ground glass view screen of the
position of each curser is given in the second and third
macro camera, Figure 1. A more complete description of this
columns from the left side of the dialog box shown in the lower
apparatus can be found in reference [12].
right of the figure. These coordinates multiplied by the
appropriate scale factor give spatial resolution on the order of
Digital images were recorded with a computerized digital
frame grabber system and later played back on a viewer 15-20mm.
program. The camera interface and viewer programs were Weld parameters were chosen to optimize the formation of
created at Vanderbilt University using National Instruments surface ripples while not overwhelming the macro camera
LabVIEW® software and analyzed. The viewer software was with excessive arc light. The most uniform ripples were
originally designed to study rapid solidification and was obtained when welding current was near 75-95A, voltage at
adapted for use on the macro camera to study weld pool 8.5-10V and travel speed in the range of 1.9-4.4mm/s. These
behavior. parameters resulted in partial penetration welds. These
parameters provide an upper and lower bound to what can be
imaged using the macro camera without the use of any light

The horizontal and vertical pixel resolution was used to
calculate the amount of material that remelted. Over the frame
interval 401-402, (1.15ms) a 0.304mm length of solid as
measured along the solidification front, melts back into the
weld pool at an average velocity of 264mm/s. The melting
velocity calculated for frame interval 402-403 and 403-404 is
302mm/s and 160mm/s, respectively. Subsequent work,
discussed below, has shown the average convection velocity to
be on the order of 25mm/s. The melting occurs at a rate
Figure 3: Typical autogenous weld investigated in this Ripples approximately one order of magnitude higher than the liquid
are uniform in appearance and spacing. Weld pool is round to convection velocity.
slightly elliptical. Material used was 3.2mm 1020 CRS. Weld
parameters which enhance weld ripple formation typically
range 8.0-9.5 V, 75-90A and travel speed of 2-3 mm/sec.

Ripple Formation
As reported in a previous work [4] when primary weld
parameters (arc voltage, current and travel speed) were varied
individually around an optimum range for ripple formation,
ripples were observed to have a fluctuating rate of formation.
An additional velocity component was sometimes found when
the high-speed images of the weld pool solidification were
played back at slow speed. Weld ripples were observed to not
always form in a direction normal to local isotherms. Ripple
formation was found to proceed in directions tangential to the
edge of the weld pool.

This can be seen in figure 2, where a weld image is shown as

displayed in the viewer program. A ripple is shown during
formation; growth proceeded along the circumference of the
weld pool, growing simultaneously towards the weld edge and
Figure 4: Composite image illustrating a remelting event in
Figure 3 shows a ripple pattern typical when weld parameters frames 401-404. Weld parameters: 85A, 9.5V, travel speed
are set within an optimal range. Near the weld centerline the 2.41mm/s and frame rate 870F/s.
ripple spacing is rather uniform. Near the edge the ripple lines
become less uniform. Some ripple lines extend from edge to
Figure 5 is a composite image showing repeated solidification
edge, but many do not. In this work the solid-liquid interface
and melting during the formation of a single weld ripple. The
is studied with attention given to solidification as influenced
line in each image represents the position of the solid-liquid
by surface fluid flow.
interface. The line in each left hand image was carefully
projected onto the right hand image maintaining its position
Remelting of Solid
with respect to a stationary coordinate frame so that the degree
to which solid material is melted back into the weld pool is
Figure 4 is a sequence of four consecutive frames that shows
readily seen.
the rapidity at which remelting of solid can occur. Frame 401
shows solid material that has been growing in a continuous
When comparing frame 421 to frame 425, it is evident that the
manner up to this point in time. Frame 402 shows the loss of
darker colored area, which represents solid material, has
solid material between cursor 1 and cursor 2. Frame 403
moved to the left. Since the travel direction was right to left,
shows the loss of solid material between cursor 2 and cursor 3.
with respect to a stationary reference frame attached to the
Frame 404 shows the loss of solid material between cursor 3
camera, the solid-liquid interface moved from right to left. The
and cursor 4. Following frame 404 during a time period of
light colored region above and to the left of the line indicates
33.4 ms, no apparent growth takes place. Ripple formation
the loss of solid material. A similar occurrence happens
resumes after this interval and continues for 411 ms when a
between frames 459 and 462. The melting occurred over most
similar remelting event occurs. The image exposure is such
of the solid-liquid interface shown. Only a small portion on
that the liquid region is too light to observe any floating
the lower left end of the line appears unchanged. In frame 497
particles which might give insight about the fluid flow when
the melting appears to be uniformly distributed along the
the melting occurred

the weld pool. When travel was resumed, the previous particle
The three frame pairs show three distinct re-melting events flow patterns were reestablished
that occurred during the formation of a single ripple. These
three events happened in a time period of 87ms, yielding a The melting behavior illustrated in figures four and five show
frequency of 34.3 remelts per second. Each time the remelt that a fluctuation in local temperature had occurred. This
event occurred it was observed to proceed from an upper right could be due to rapid release of latent heat of fusion [13],
position to a lower left position across each respective frame. Solidification mode instability [14-17], or liquid temperature
In the context of a circular weld pool moving from left to fluctuations due to turbulence. The first two possibilities lie
right, the melting event proceeded in a counter clockwise outside the capabilities of this experimental apparatus to
direction. measure, but the last possibility can be examined by tracking
the motion of oxide particles to ascertain the surface flow
Surface Flow
The frame grabber camera is unable to record images over the
entire weld pool surface due to the bright light emitted by the
arc plasma. The weld pool is observable only near the solid-
liquid interface,

Frame grabber images have a low pixel count, and a grainy

image resolution. Often a particle could be identified and its
position tracked over a sequence of frames. Yet when
individual frames from the same sequence were examined,
that particle was difficult to identify. The method used to
identify the position of an oxide particle in an individual frame
was to repeatedly play a short series of frames just prior to the
frame of interest and place a cursor at the estimated centroid
position of the particle.

The images in figure 6A and 6B show the flow path of

particles floating on weld pool surfaces using different weld
parameters. Lines have been added to help define the
liquid/solid interface. Weld ripples have formed to the left of
the lines and liquid material lies to the right. The flow paths
were counterclockwise with respect to the weld pool.

Weld parameters for figure 6A are 80A, 9.5V, and travel

speed of 2.96mm/s. Average velocity of the particle over this
interval was 25.0mm/s with a max/min variation of 34.6/20.4
mm/s. The Reynolds number for the flow in figure 6A was
107. Pool diameter was used as the characteristic length and
the average particle velocity for the characteristic velocity.
Figure 5: A composite image showing repeated solidification Turbulent intensity for this flow was 22%.
and melting during the formation of a single weld ripple. The
line in each left hand image is carefully projected onto the
right hand image maintaining its position with respect to the
stationary camera frame. Weld Parameters: 9.5V, 80A, travel
speed 2.96mm/s and frame rate of 862F/s

During the course of recording the various data files,

pool circulation patterns were sometimes revealed by oxide
particles floating on the weld pool surface. Generally, the
individual particle velocities are too fast to track by eye.
However, a "circular" (not geometrically round, but
recirculating ) flow pattern was noted. Sometimes a pair of Figure 6 A &B: The motion of an oxide particle was tracked
counter rotating eddies were seen on the weld pool surface. over a series of frames. Weld parameters, 6A: 80A, 9.5V
When travel was halted, the particles were seen to stop the 2.96mm/s 6B 90A, 9.5V 2.45mm/. Frame rate was 869F/s.
rotational motion and move quickly to the outer perimeter of

The image of figure 6B shows the path taken by a particle on a tracked over a short time interval and their relative motion was
weld using different parameters: 90A, 9.5V, and travel speed observed. Assuming that the particles are sufficiently close
of 2.45mm/s. More energy has been put into the weld pool together such that their motion approximates the theoretical
relative to figure 6A and the particle motion has become more motion of adjacent fluid particles and the time interval is
erratic. The average velocity was 9.73mm/s and the max/min sufficiently short such that the velocity field can be assumed
variation was 23.8/0.593mm/s. constant over this interval, the effect of the spatial derivatives
on the particle motion, the vorticity, can be measured.
Figure 7 is a graphical evaluation of vorticity. The particles
Turbulent flow possess several characteristics the absence or were tracked over 5 frames and their position normalized with
presence of which does not establish or eliminate the presence respect to particle A. Change in angle ABC was 0.29 r over
of turbulent flow [9]. One criterion however does provide a 11.7ms for an angular velocity of 4.96 r/s. The angular
necessary, but not sufficient, condition that is the presence of velocity is the magnitude of the vorticity vector, since it is non
vorticity. The presence of vorticity, ω ≠ 0 , in a fluid flow zero, the flow is rotational.
means that the flow is rotational. All turbulent flows are
rotational, but a rotational flow is not necessarily turbulent. A The diameter of the oxide particles was on the order of 50μm
rotational flow is definitely not a laminar flow. length scale between oxide particles was on the order of
500μm compared to a pool diameter on the order of 3000μm.
The vorticity vector represents rotation of adjacent fluid The approximation improves with smaller size scale. If a
particles relative to a moving fluid particle. If the origin of a camera with higher pixel count were used it is possible to view
set of coordinate axes were attached to a moving fluid particle smaller oxide particles that are closer together and
and the axes extended through each adjacent particle, then approximation of vorticity is more accurate.
each term of the vorticity vector represents an average angular
velocity about each coordinate axis. Equation (1) is the
mathematical expression for the vorticity vector in Cartesian

v 1 ⎡⎛ ∂Vz ∂V y ⎞ ˆ ⎛ ∂V x ∂Vz ⎞ ˆ ⎛ ∂V y ∂Vx ⎞ ˆ ⎤

ω = ⎢⎜⎜ − ⎟i + ⎜ − ⎟ j +⎜ − ⎟k ⎥ (1)
2 ⎣⎝ ∂y ∂z ⎟⎠ ⎝ ∂z ∂x ⎠ ⎜⎝ ∂x ∂y ⎟⎠ ⎦

The two spatial velocity derivatives inside each set of brackets

represent the instantaneous average angular velocity of two
fluid particles located on two orthogonal; axes relative to the
unit vector axis passing through the third particle. The
vorticity vector is the vector sum of the 3 average angular
velocity components about each coordinate axis.

Vorticity is a 3-dimensional phenomenon that if present would

exist through the entire volume of the weld pool. Experimental Figure 7: Graphical solution to the vorticity vector. Physical
determination of the velocity field in a weld pool is not significance of vorticity is the relative rotation of two fluid
feasible at the present time. Assume that the free surface particles B and C relative to a particle A. Finite space
serves as an observable 2-dimensional boundary for vorticity approximation to theory derived using infinitesimal space.
present in the flow. Applying the assumptions: Vz = 0 and Camera resolution, particle size and population density limit
the fineness of the approximation. Smaller distance between

= 0 at the free surface (z=0) the equation (1) reduces to: particles ABC would have given give better accuracy.
v 1 ⎛ ∂V ∂V ⎞ Summary
ω = ⎜⎜ y − x ⎟⎟kˆ (2)
2 ⎝ ∂x ∂y ⎠
High speed digital images of weld pool solidification and
The vorticity vector is evaluated experimentally by an surface flow were recorded and analyzed. Solidification was
approximation. Evaluation of the spatial derivatives was not found to occur at a fluctuating rate. Newly solidified material
feasible, but the effect the velocity field had on the motion of was observed to melt back into the weld pool before final
oxide particles was. solidification occurred.

On the weld pool surface three oxide particles were found Normally occurring oxide particles floating on the weld pool
oriented such that the lay approximately on an orthogonal x-y were used for flow visualization. Particle position was tracked
axis, shown in the schematic diagram of Figure 7. They are over a sequence of frames and average and frame to frame

velocity was calculated. Reynolds number was 107, and 11 K. Hong, D. C. Weckman, A. B. Weckman and A.B.
turbulent intensity was 22%. Strong, The Predicted Influence of Turbulence in
Stationary Gas Tungsten Arc Welds, Trends in
A graphical method which utilized the velocity data from Welding Research, ASM International Conference
floating oxide particles was presented for evaluating the Proceedings, pp 399-404, (1995).
vorticity vector. Results showed that the magnitude of the 12 D.R. DeLapp, D.A. Hartman, G.E. Cook, R.J.
vorticity vector was nonzero indicating that the flow was Barnett, and A.M. Strauss, The Development of a
rotational. GTAW Observation System, 5th International
Conference on Trends in Welding Research, ASM
Rotational flow is a necessary condition for turbulent flow or International Conference Proceedings pp 405-409,
flow in transition to turbulent flow. Either type of flow can (1998)
possess velocity or temperature fluctuations which may 13 D. L. Cheever and D. G. Howden, Technical Note:
account for the observed remelting behavior during Nature of Weld Surface Ripples, Welding J, 48 (4),
solidification. pp 179s-180s, (1969)
14 R. Travedi and W. Kurz, Morphological Stability of a
References Planar Interface Under Rapid Solidification
Conditions, Acta Metall. 34(8), 1663-1670 (1986)
1 S.A. David and J.M. Vitek, Principles of Weld 15 W. Kurz, B. Giovanola and R. Travedi, Theory of
Solidification and Microstructures,3rdInternational Microstructural Development During Rapid
Conference on Trends in Welding Research, ASM solidification, Acta Metall. 34(5) pp. 823-830 (1986)
International Conference Proceedings, p 147-156, 16 W. Kurz, Dendrite Growth in Welding, Mathematical
(1992) Modeling of Weld Phenomena 2 The Institute of
2 R.A. Woods and D.R. Milner, Motion in the Weld Materials, pp. 40-53, (1996)
Pool in Arc Welding, Welding J.,50(4),163s-173s, 17 W. Kurz and R. Travedi, Mat. Sci. Eng., A179/180,
(1971) pp. 46-51 (1994)
3 G.M. Ercer, A. Tzavaras, A Gokhale, and H.D.
Brody, Weld Fluid Motion and Ripple Formation in
Pulsed Current GTAW, Trends in Welding Research,
ASM International Conference Proceedings, pp 419-
442, (1982)
4 D.R. DeLapp, D.A. Hartman, W.H. Hofmeister, G.E.
Cook, and A.M. Strauss, An Investigation into the
Local Solidification Rate of the GTA Weld Pool, 5th
International Conference on Trends in Welding
Research, ASM International Conference
Proceedings, pp 400-404, (1998)
5 A.C. Hall, G.A. Knorovsky, C.V. Robino, J. Brooks,
D.O. Maccallum, M. Reece, and G. Poulter,
Characterizing the Microstructure of a GTA Weld In-
Process using High-speed, High Magnification
Digital Imaging 11th Int. Conf. on Computer. Tech. in
Welding, Columbus OH, pp 117-124, (2001).
6 A.C. Hall, C.V. Robino, , Microstructural Features
and Rippling Phenomenon in 304 stainless Steel Gas
Tungsten Arc Welds, Science and Technology of
Welding and Joining, pp 103-108, (2004)
7 F.P. Incropera and D.P. DeWitt, Fundamentals of
Heat And Mass Transfer, 2nded., John Wiley and
Sons, pp 293-296 (1985)
8 H. Tennekes, and J.L. Lumley, A First Course In
Turbulence, MIT Press Cambridge MA, pp 1-4
9 R. L. Panton, Incompressible Flow, John Wiley and
Sons, New York, pp704-709 (1984)
10 M. Malinowski-Brodnicka, G. den Ouden and W. J.
P. Vink, Effect of Electromagnetic Stirring on GTA
Welds in Austenitic Stainless Steel, Welding Journal
69(2), pp 52s-59s,(1990)

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Scaling Laws in Welding Modeling

P. F. Mendez
Colorado School of Mines, Golden, Colorado

This “handbook formulas” approach would be very useful for
There are few simple formulas to predict the results of a welding, but there are very few universally accepted such
welding process. Such formulas would be of enormous help formulas in this field. Very relevant welding characteristics
in the design of welding processes, and they are ubiquitous in such as penetration, arc size, current in GMAW, voltage in
other engineering disciplines such as strength of materials, GTAW lack such a general expression. For welders in the
fluid mechanics, and heat transfer, typically in the form of field, who must have an approximate estimation of the right
scaling laws. Scaling laws provide accurate approximations welding parameters before they weld, there are rough aids
and display clearly the trends in a problem. In this paper I such as the “Miller Calculators” [8], which provide ballpark
review the reasons why scaling laws are so scarce in welding, values for some particular welding activities. Sophisticated
highlight previous and current efforts to develop scaling laws software is often difficult to operate, time consuming, and not
for welding, and present two complimentary approaches of widely available to welders or welding engineers.
great potential to develop scaling laws specific for welding.
Scaling laws are of enormous utility during the early stages of
welding design, when the configuration of a system is still
uncertain. In this case, scaling laws could provide quick
estimations of the feasibility of a proposed process, help
The design of a new welding process involves so many
determine costs, and contribute to decisions about
parameters that it can seldom be predicted reliably; therefore,
configuration and materials. Scaling laws can be calculated in
extensive experimentation must take place in order to
negligible time; therefore, they can also be useful for robotic
determine an ideal process set-up. A set of simple and
welding systems predicting the behavior of a system in real
intuitive design laws based only on the most relevant
time, much faster than computationally intensive models such
parameters would be of enormous help in this case. Scaling
as finite element analysis, or computational fluid mechanics.
laws are particularly well suited for this purpose.
When experimental databases or numerical models exist,
Scaling laws appear in several disciplines such as physics,
scaling laws can be used to generalize and extrapolate the
biology[1, 2], geophysical [3, 4], Internet traffic[5], and even
results obtained. For existing processes, scaling laws are
economic systems[6]. A broad sample of problems that can be
useful for set-up and tuning operations and to compare
described with such scaling laws is presented in [7]. Scaling
different welding alternatives.
laws are ubiquitous in engineering; some of the reasons for
this are: 1) the combination of units has the form of a power
When modeling a welding process, it is convenient to divide it
law, 2) the expressions of many physical phenomena have the
into submodels such as the weld pool, arc, electrode, etc.
form of power laws as noted above, and 3) many empirical
Power laws are useful to link the submodels together. For
regressions of engineering data in log-log plots tend to give a
example weld pool depression depends on arc pressure.
straight line, which corresponds to a power law.
Power law solutions for the arc can relate the arc pressure to
the controllable parameters of the arc, thus embedding the arc
In engineering in particular, scaling laws constitute the
into the weld pool model. Power laws do not present
backbone of handbooks, together with the tables of values for
convergence problems of the type of numerical solutions.
the parameters involved. For example, the maximum
deflection of a cantilever beam is universally presented as a
simple power law involving the properties of the cross section
of the beam and the modulus of the beam material. This Scaling Laws, Dimensional Analysis, and
simple formula is accompanied by tables of properties of cross Similarity
sections and modulus of different materials. This power law is
valid for steel beams of a bridge, wooden beams of a house, or Scaling laws give us an accurate estimation a magnitude (e.g.
silicon beams in MEMS. penetration) as a function of the welding parameters. Welding

Copyright © 2006 ASM International® 103

parameters are not only the obvious current, voltage, wire feed straight line y’=ax’+b in a log-log plot corresponds to the
speed, etc. They also include thermophysical properties of power law y=10bxa, where y’=log y, and x’=log x.
materials involved such as heat conductivity, density, and heat
capacity for heat transfer in the solid, and the same plus Power laws are indeed very powerful. They clearly indicate
viscosity and surface tension for heat transfer in the weld pool. trends and can yield accurate predictions over several orders
Even more parameters are necessary to include a of magnitude. They also convey much intuitive meaning: the
characterization of the arc, electrode, laser beam or electron sensitivity of a power law to a given welding parameter is
beam. Some of these parameters can vary within the domain directly proportional to the exponent of the parameter. Using
of the problem, for example the surface tension depends on Equation (1) as an example, we see that penetration will
temperature, which varies spatially. increase approximately 13% if we increase the current by
Scaling laws have the functional form of a power law of the
relevant problem parameters such as the following proposed Dimensionless Groups
scaling law for penetration in SMAW [9] Dimensionless groups are a particular type of power law.
These groups have no units; therefore, their value is
P = G I4/3V-1/3E-2/3 ................................................................(1) independent of the unit system used, conferring them
generality. Well known dimensionless groups include the
Where G is a constant, I is the current, V is the travel speed, Reynolds number and the Peclet number.
and E is the voltage. We observe in this power law that the
parameters are raised to a constant power. Expressions where Dimensionless groups are at the core of the technique of
a constant is raised to a variable power are not power laws in Dimensional Analysis, in which a problem is represented in a
this context; thus, if a is a constant, and P, is a parameter, Pa is simpler way without any loss of generality. A cornerstone of
a power law, but aP is not. As power laws are aimed to Dimensional Analysis is Buckingham’s Π theorem[12], which
compare different welding processes, they are based on the roughly states that a problem involving n parameters and k
parameters, and not on the problem variables (space, time). units can be represented in a simpler form by m dimensionless
Thus if L is a length in the x direction, La is a useful power groups, where m=n-k. The technique of Dimensional
law to compare welding alternatives, while xa is not. Analysis can greatly simplify a problem, and has been of
substantial help to the fields of Aerodynamics and Fluid
Since power laws do not depend on the space and time Mechanics.
coordinates of the problem, a representative value must be
chosen for those parameters that vary with the coordinates. A fundamental requirement for proper dimensional analysis is
For example, a representative value of the density of the that all relevant parameters must be included. For simpler
molten metal is chosen for scaling laws of the weld-pool. problems this is not too difficult; for example, in
Occasionally, spatial variations are relevant and they generate Aerodynamics all geometrically similar problems can be
driving forces, for example variations in density cause completely described by the properties of the fluid (density ρ
buoyancy forces, and variations in surface tension cause and viscosity μ), a characteristic length L, and a velocity V.
Marangoni forces. In these cases, the varying magnitude can Such a complete description is practically impossible in
be expressed as two parameters, one capturing a characteristic welding because of the large number of parameters required.
value, and another capturing the variation. For density we The obvious current, voltage, wire feed speed, etc., are only a
thus have the Boussinesq approximation: ρ=ρ0+β(T-T0), small fraction of all the parameters needed. In fact, relevant
where β captures the variation of density with temperature. parameters have been overlooked at times; for example, before
For Marangoni flows we use σ=σ0+σT(T-T0). the early 1980’s[13], the effect of surface tension variation on
Marangoni forces had not been considered yet[14], and
Power laws are a natural consequence of dimensional welding engineers could not explain large penetration
homogeneity[10], which states that all terms in an equation variations in apparently identical welding conditions.
must have the same dimensions, and dimensions can only be
formed as power laws of basic units such as m, kg, s. Power The large amount of parameters to be considered makes
laws often appear naturally as the exact solution of asymptotic welding modeling especially challenging. While not all
cases. As a welding example, temperature distribution in parameters are simultaneously critical, we often do not know
Rosenthal’s solution[11], is expressed as a power law in units which ones to neglect before modeling. And models
of temperature, multiplied by a dimensionless function: considering all imaginable parameters can become intractable.

T-T0 = q/(4πkR)exp[-λν(x+R)]...........................................(2) Dimensional Analysis is less powerful for problems with

several parameters such as welding; while the number of
The fundamental nature of power laws is also evident in the parameters easily exceeds 10, the number of units is fixed to
engineering wisdom that “everything is a straight line when approximately 5. Buckingham’s theorem indicates that the
plotted in a log-log graph.” In mathematical terms, the problem still requires 5 or more dimensionless groups. In

comparison, in Aerodynamics, the four parameters mentioned 20 Lo w ke 1997
above (ρ, μ, L, V) involve three units (m, kg, s), and a single

Choo, 1990

M c K e llig e t 1 9 8 6

K im 1 9 9 7
dimensionless parameter (typically the Reynolds number) is

L ee 1 9 9 6
enough to characterize all problems. In this case, tabulation

R a m a k r is h n a n 1 9 7 8

H su 1 9 8 3
and understanding of the problem are relatively simple, while 14

G lic k s te in 1 9 7 9
n u m b e r o f d im e n sio n le ss g ro u p s
the same is virtually impossible for welding.

M a e c k e r 1 9 5 5 ( a p p r o x im a te )

S h e r c lif f 1 9 6 9 ( a n a ly tic a l)

S q u ir e 1 9 5 1 ( a n a ly tic a l)
Complete and Incomplete Similarity
When two problems can be described using the same set of
dimensionless groups, the two problems will be considered 4

“similar” when the corresponding groups have the same value 2

for both problems. For similar problems, the behavior of one 0
can be deduced from the behavior of the other through the use 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000
of scaling laws. For example, a model airplane in a wind ye a r o f p u b lic a tio n
tunnel and a real airplane in flight can be considered similar if
both have the same Reynolds number, and measurements
made in the wind tunnel can translate directly to the actual Figure 1: Evolution of arc modeling complexity [15-26]
plane with the use of the appropriate scaling laws.
I propose that the reason why there are relatively few general
When only some dimensionless groups have the same value scaling laws for welding is because useful models involve an
on two systems, these models have “incomplete similarity.” excessive number of dimensionless groups, and there are no
Incomplete similarity is especially important in welding, since established techniques that can help the engineer to form and
it is not possible to control all the dimensionless groups select the most useful dimensionless groups on which to
associated with the large number of parameters. If the establish similarity between model and reality, and upon
dimensionless groups with the same value capture the which to develop scaling laws. Especial tools are necessary to
dominant parameters, and the groups with different values accomplish this, and will be reviewed in a later section. In the
represent secondary phenomena, we can still obtain useful and next section we will discuss past and current applications of
representative results. Using weld pool flows as an example, scaling laws to welding.
buoyancy is a secondary force when Marangoni forces are
dominant. In this case, a model and experiment with a Scaling and Dimensional Analysis in Welding
different Grashoff number (which captures buoyancy) will still
show essentially the same behavior. Possibly the oldest scaling law developed for welding is the
estimation of penetration presented in Equation (1) and its
The choice of which dimensionless groups to preserve and successors[27]. This law was developed empirically for
which ones can be discarded is critical, and far from trivial. manual SMAW of ½” steel plate. We see that the constant G
Dimensional Analysis is useful to generate dimensionless needs units for this equation to be homogeneous, thus it
groups, but does not provide guidelines for which ones to captures some physics such as the thermophysical properties
keep. In addition, there are several sets of dimensionless of the material being welded, not considered by the explicit
groups that can represent a given problem; thus, there is the parameters. This formula is seldom seen today, possibly
additional challenge of selecting the best set of dimensionless because of its lack of generality.
groups that would provide for the minimum influence of the
discarded groups. The next improvement for scaling laws in welding was the
scaling of analytical solutions for heat transfer in the solid,
The problem of selecting the most representative neglecting the effects of the fluid flow in the weld pool.
dimensionless groups in partially similar problems has become
increasingly relevant since the introduction of powerful Scaling of analytical solutions for heat transfer in the solid
personal computers in the 1970’s. At that time there was a Rosenthal’s analytic solution for a point heat source in a semi-
surge of mathematical models of increasing complexity, as infinite solid[11] was generalized with great success by
illustrated in Figure 1. Christensen, and it is still widely used today. This problem is
formulated in terms of the thermophysical properties of the
Figure 1 uses the modeling of a welding arc as an example. In solid: heat conductivity k and heat diffusivity α, process
it, we appreciate the significant jump in the number of parameters: heat input q and traveling velocity V, and a
dimensionless groups involved in the models after 1970. characteristic temperature jump ΔT. These parameters involve
Similar trends are also evident for other aspects of welding the units m, kg, s, K, and thus the whole formulation of this
modeling such as the weld pool. problem could be characterized by a single dimensionless

group. This is actually the case, and Christensen used the groups such as the Reynolds number, Peclet number, and
“operating parameter” n magnetic Reynolds number.

n = qV /(4πkαΔT) ...............................................................(3) The increase in the number of dimensionless groups observed

in Figure 1 is due to the addition of thermal effects of the arc
All unknowns of this problem can be obtained from scaling and variations in geometry of the electrode or weld pool
laws involving only the four parameters listed above and a surface.
dimensionless function of the operating parameter.
Dimensional analysis of the whole welding process
This solution is restricted to thick plates and temperatures There have been attempts to capture the whole welding
significantly lower than the melting temperature of the base process with dimensionless groups. All of these efforts are
metal, thus it does not provide accurate information about based on heat transfer in the solid, and are aimed at
weld pool shape and HAZ thermal history. One possible generalizing numerical solutions or experiments.
reason for the success of Christensen’s generalization despite Krivosheya[37] presented an analysis of SAW butt joints
its limitations is that it is derived from first principles; based on seven parameters, thus obtaining three dimensionless
therefore, its generality is assured as long as the starting groups. Kou[38, 39] developed a dimensionless formulation
hypotheses are valid. that considers the Stefan problem of heat of melting. A
relevant aspect of his normalizations is the use of the same
There is a family of solutions following a similar approach to characteristic length for all three spatial directions.
Rosenthal and Christensen for thin plates, line heat sources Fuerschbach proposed a dimensionless parameter model for
and more. Comprehensive lists of solutions can be found in arc welding[40], and Fuerschbach and Knorovsky proposed a
the compilations by Grong [28, 29]. These last references also single dimensionless group to characterize the melting
list solutions for point heat sources on plates of intermediate efficiency of melting in PAW and GTAW[41]. A scaling
thickness. In this case, the plate thickness becomes an analysis for the butt welding of thermoplastics is developed
additional parameter, and the problem needs two in[42].
dimensionless parameters to be completely described.
Scaling laws and dimensional analysis of the weld pool
There are also solutions for multiple point heat sources, and In the early 1980’s, a simultaneous development of weld pool
for distributed heat sources. Eagar and Tsai presented understanding and the availability of computer power
solutions for a circular gaussian heat source on a thick prompted the introduction of fluid flow and heat transfer in the
plate.[30, 31] In this case, in addition to the operating models of the weld pool. Dimensional Analysis is a standard
parameter n, there is an additional dimensionless number u, tool of fluid mechanics, and soon models incorporated the
the “distribution parameter.” Further generalization to a traditional Peclet number, Reynolds number, and Marangoni
gaussian heat source on a plate of intermediate thickness were number.
presented by Manca et al.[32]. As expected, their solutions
depend on three dimensionless groups. Wang and Kou[43] extended Kou’s dimensionless formulation
of the welding problem to include convection in the weld pool.
Similarly to the earliest work on arc illustrated in Figure 1, Landmark studies of convection in the weld pool during spot
these analytical solutions trace their roots to the times in welding were developed by Oreper, Szekely, et al.[44-47]. In
which computers were not readily available. The welding them, the governing equations are normalized, generating the
problems were then just as complex as today, but their physics Reynolds, Prandtl, Grashoff, Stefan, Rayleigh, surface tension,
needed to be simplified until a tractable problem was obtained. and Marangoni numbers. An analysis of the relevance of the
driving forces is performed using these numbers, and scaling
Scaling laws and dimensional analysis of the welding arc laws are proposed for the characteristic velocity of the weld
For the arc, two of the earliest scaling laws are those for pool when the driving forces are Marangoni, electromagnetic,
pressure and velocity by Maecker[21]. These laws involve or buoyancy. In [48], Szekely also proposes scaling laws for
two dimensionless groups at the most, and they have little time constants in melting and solidification during the
influence on the resulting scaling laws. Thorough dimensional transient regime of a spot weld pool. This approach of
analysis of an electric arc was performed by Yas’ko and determining the dominant forces using known dimensionless
Shaskov[33, 34]. In their work they identified seven or more groups, then using the appropriate scaling laws is being used
dimensionless groups, proposed no particular solution today by DebRoy et al. for weld pools in GMAW[49], and
employing them, and indicated which groups were more spot[50] and traveling GTAW[51]. A well explained scaling
relevant for similarity of 10 different types of arc. Scaling analysis of fluid flow is presented by Rivas and Ostrach[52].
laws for pressure, velocity and temperatures in a laminar In this work they considered three different regimes of
welding arc are presented in[35, 36]. Most current work on thermocapillary driven flows: I) when there is no surface
arc modeling will include a standard set of dimensionless boundary layer, II) when there is a viscous surface boundary
layer, and III) when there is a thermal surface boundary layer.

Another systematic scaling of fuid flow in the weld pool is normalization of the differential expressions. It is called
presented by[53]. These last two references differ from the Order of Magnitude Scaling, and it is described in[65-70]. A
analysis by Kou in that they use intrinsic scales for lengths in similar approach to modeling was developed by Yip[71] using
different directions. This enables the estimation of differential concepts of Artificial Intelligence.
expressions without the need to solve the corresponding
differential equations. For weld pools under high currents, Another alternative for generating scaling laws is to analyze
dimensional analysis and scaling laws are presented in[54]. sensitivity data from experiments or numerical simulations. In
this case scaling laws can be generated by minimizing fitting
Dimensional analysis of LBW error. The field of artificial intelligence has been active in this
For weld penetration in LBW, early correlations based on task, although the models generated are difficult to extend to
three dimensionless groups were presented by Lubin[55]. An welding. An algorithm called SLAW[72, 73] and associated
empirical study of energy transfer efficiency based on software were developed to generate scaling laws and
dimensionless groups has been presented by Fuerschbach[56], dimensionless groups ranked by relevance. This algorithm
and dimensionless maps for laser processing of materials correctly reproduced the scaling laws previously developed for
based on dimensionless groups have been developed by Ion, ceramic to metal joining[74].
Shercliff and Ashby[57]. An extension to LBW of the GTAW
weld pool scaling laws and dimensional analysis is presented Summary
in [58, 59].
Summarizing, scaling laws are desirable in welding. They
Dimensional analysis of GMAW would help integrate models, speed-up weld process design,
Several correlations for metal transfer in GMAW based on and would facilitate costing estimations. However, scaling
dimensionless groups are presented by Murray in[60]. In[61], laws are much less developed in the welding field than in
he presents dimensionless correlations for penetration, and in other disciplines such as fluid mechanics. The reason for this
[62] there is a dimensionless analysis of droplet detachment. disparity is that welding involves a much larger number of
An scaling analysis for flows in the weld pool during fillet parameters than other disciplines. It is difficult to decide
welding is presented in [49], and a thorough dimensional before modeling which parameters to neglect, and it is
analysis of weld pool phenomena involving eight parameters difficult to model without neglecting some parameters.
is presented in[63]. Traditional tools such as Dimensional Analysis or dominant
balance are not powerful enough for the complexity of
Tools for Generation of Scaling Laws welding. Current scaling laws were generated by solving
analytically very simplified systems, or by inspectional
Welding modeling has a difficulty that other disciplines do not analysis of the normalized governing equations. Two methods
have: the number and parameters and dimensionless groups is are proposed to systematically generate scaling laws and
relatively large. Partial similarity is key to simplifying the associated dimensionless groups. The first, Order of
models and generating useful scaling laws, and to accomplish Magnitude Scaling, is proposed for analyzing systems of
it effectively, two tasks must be completed; first, an coupled differential equations. The second, SLAW is
appropriate set of dimensionless groups must be generated; proposed to analyze experimental or numerical sensitivity
second, these dimensionless groups must be ranked by studies.
relevance. References
Dimensional Analysis provides no guidance to accomplish 1. R. K. Azad, P. Bernaola-Galvan, R. Ramaswamy, J. S. Rao,
these tasks. Applied mathematics tools such as dominant Physical Review E, 65, art. no.-051909 (2002).
balance[64] are limited to single equations, and are not 2. V. B. Kokshenev, Physica a-Statistical Mechanics and Its
effective for the systems of equations often found in welding Applications, 322, 491-505 (2003).
modeling. 3. J. Y. N. Cho, R. E. Newell, G. W. Sachse, Geophysical Research
Letters, (1999).
Most scaling laws for welding were generated either from 4. K. R. Housen, R. M. Schmidt, K. A. Holsapple, Journal of
Geophysical Research, 88, 2485-99 (1983).
exact solutions of simplified problems, or from intuitive 5. J. M. Carlson, J. Doyle, Physical Review Letters, 84, 56-9 (2000).
normalization of governing equations. The intuitive approach 6. F. J. d. Jong, W. Quade, Dimensional analysis for economists,
is not necessarily systematic, and requires a great deal of North Holland Pub. Co., Amsterdam, (1967).
experience to determine the best normalization scheme. A 7. T. Szirtes, P. Rózsa, Applied Dimensional Analysis and Modeling,
systematic approach would involve trying all possible McGraw Hill, New York, (1997).
normalization schemes, and checking for consistency of the 8. Miller, (1999).
assumptions made. Such an approach was followed by Rivas 9. R. Gunnert, Welding Journal, 542- (1948).
and Ostrach for a simplified weld pool with no 10. P. W. Bridgman, Dimensional Analysis, Yale University Press,
electromagnetic or buoyancy forces[52]. This approach has New Haven, (1922).
11. D. Rosenthal, Transactions of the A.S.M.E., 849-66 (1946).
been generalized using linear algebra, and improved 12. E. Buckingham, Physics Review, 4, 345-76 (1914).

13. C. R. Heiple, J. R. Roper, Welding Journal, 97s-102s (1982). 47. M. E. Thompson, J. Szekely, Int. J. Heat Mass Transfer, 32,
14. S. S. Glickstein, W. Yeniscavich, Welding Research Council 1007-19 (1989).
Bulletin, Welding Research Council, Report # 226 (1977). 48. J. Szekely, in Proceedings of International Conference on Trends
15. R. T. C. Choo, J. Szekely, R. C. Westhoff, Welding Journal, 69, in Welding Research, ASM International, 3-14 (1986).
346s-61s (1990). 49. W. Zhang, C. H. Kim, T. DebRoy, Journal Of Applied Physics,
16. S. S. Glickstein, in Proceedings of Arc Physics and Weld Pool 95, 5220-9 (2004).
Behaviour, London, UK, 1-16 (1979). 50. W. Zhang, G. G. Roy, J. W. Elmer, T. DebRoy, Journal Of
17. K. C. Hsu, K. Etemadi, E. Pfender, Journal of Applied Physics, Applied Physics, 93, 3022-33 (2003).
54, 1293-301 (1983). 51. T. Debroy, S. A. David, Reviews of Modern Physics, 67, 85-112
18. W.-H. Kim, H. G. Fan, S.-J. Na, Metallurgical Transactions B, (1995).
28B, 679-86 (1997). 52. D. Rivas, S. Ostrach, International Journal of Heat and Mass
19. S.-Y. Lee, S.-J. Na, Welding Journal, 269s-79s (1996). Transfer, 35, 1469-79 (1992).
20. J. J. Lowke, J. Phys IV France, 7, C4-283 - C4-94 (1997). 53. S. Chakraborty, S. Sarkar, P. Dutta, Science and Technology of
21. H. Maecker, Zeitschrift fur Physik, 141, 198-216 (1955). Welding and Joining, 7, 88-94 (2001).
22. J. McKelliget, J. Szekely, Metallurgical Transactions A, 17A, 54. P. F. Mendez, T. W. Eagar, Estimation of the Characteristic
1139-47 (1986). Properties of the Weld Pool During High Productivity Arc
23. F. Qian, B. Farouk, R. Mutharasan, in Proceedings of Welding, in: Mathematical Modelling of Weld Phenomena 5, H.
Fundamentals of Heat Transfer in Electromagnetic, Electrostatic, Cerjak, H. K. D. H. Bhadeshia (Ed.), Institute of Materials,
and Acustic fields, 53-63 (1993). London, UK, (2001).
24. S. Ramakrishnan, A. D. Stokes, J. J. Lowke, Journal of Physics 55. B. T. Lubin, Welding Journal, 140s-4s (1968).
D: Applied Physics, 11, 2267-80 (1978). 56. P. W. Fuerschbach, Welding Journal, 24s-34s (1996).
25. J. A. Shercliff, Journal of Fluid Mechanics, 40, 241-50 (1970). 57. J. C. Ion, H. R. Shercliff, M. F. Ashby, Acta Metallurgica Et
26. H. B. Squire, Quarterly Journal of Mechanics and Applied Materialia, 40, 1539-51 (1992).
Mathematics, IV, 321-9 (1951). 58. X. He, P. W. Fuerschbach, T. DebRoy, Journal Of Physics D-
27. C. E. Jackson, A. E. Shrubsall, Welding Journal, 172s-8s (1953). Applied Physics, 36, 1388-98 (2003).
28. O. Grong, Metallurgical Modelling of Welding, The Institute of 59. A. Robert, T. Debroy, Metallurgical and Materials Transactions
Materials, Cambridge, Great Britain, (1994). B-Process Metallurgy and Materials Processing Science, 32, 941-
29. O. R. Myhr, O. Grong, Acta Metallurgica Et Materialia, 38, 449- 7 (2001).
60 (1990). 60. P. E. Murray, Welding Journal, 81, 125S-31S (2002).
30. T. W. Eagar, N.-S. Tsai, Welding Journal, 346s-55s (1983). 61. P. E. Murray, A. Scotti, Science And Technology Of Welding
31. N.-S. Tsai, T. W. Eagar, in Proceedings of Modeling of Casting And Joining, 4, 112-7 (1999).
and Welding Processes II, New England College, Henniker, NH, 62. P. E. Murray, Science And Technology Of Welding And Joining,
317-28 (1983). 5, 221-6 (2000).
32. O. Manca, B. Morrone, V. Naso, 38, 1305-15 (1995). 63. S. K. Choi, Y. S. Kim, C. D. Yoo, Journal of Physics D-Applied
33. A. G. Shashkov, O. I. Yas'ko, IEEE Trans. Plasma Sci., PS-1, 21- Physics, 32, 326-34 (1999).
37 (1973). 64. C. M. Bender, S. A. Orszag, Advanced Mathematical Methods
34. O. I. Yas'ko, British Journal of Applied Physics (Journal of for Scientists and Engineers, McGraw-Hill, New York, (1978).
Physics D: Applied Physics), 2, 733-51 (1969). 65. P. F. Mendez, T. W. Eagar, in Proceedings of Seventeenth
35. P. F. Mendez, M. A. Ramirez, G. Trapaga, T. W. Eagar, Scaling Symposium on Energy Engineering Sciences, Argonne, IL, 106-
Laws in the Welding Arc, in: Mathematical Modelling of Weld 13 (1999).
Phenomena 6, (Ed.), Institute of Materials, London, UK, (2002). 66. P. F. Mendez, T. W. Eagar, Journal of Materials Processing
36. P. F. Mendez, M. A. Ramirez, G. Trapaga, T. W. Eagar, Technology, 117, CD-ROM, Section B9: Modeling (2001).
Metallurgical Transactions B, 32B, 547-54 (2001). 67. P. F. Mendez, T. W. Eagar, in Proceedings of Fourth International
37. V. E. Krivosheya, Automatic Welding Ussr, 31, 5-8 (1978). Workshop on Similarity Methods, Stuttgart, Germany, 51-67
38. S. Kou, Journal Of Metals, 31, 111- (1979). (2001).
39. S. Kou, Metallurgical Transactions A-Physical Metallurgy And 68. P. F. Mendez, T. W. Eagar, in Proceedings of International
Materials Science, 13, 363-71 (1982). Conference on Modeling and Simulation of Metal Technologies,
40. P. W. Fuerschbach, in Proceedings of 4th International MMT-2000, Ariel, Israel, 636-45 (2000).
Conference on Trends in Welding Research, Gatlinburg, TN, 69. P. F. Mendez, S. B. Brown, T. W. Eagar, Mechanical
493-7 (1995). Engineering, 126, Design Supplement, 16-20 (2004).
41. P. W. Fuerschbach, G. A. Knorovsky, Welding Journal, 70, S287- 70. P. F. Mendez, Order of Magnitude Scaling of Complex
S97 (1991). Engineering Problems, and its Application to High Productivity
42. H. Potente, M. Heil, in Proceedings of Advances in Joining Arc Welding, Doctor of Philosophy Department of Materials
Newer Structural Materials, Montreal, Canada, 341-6 (1990). Science and Engineering, Massachusetts Institute of Technology,
43. Y. H. Wang, S. Kou, in Proceedings of Modeling and Control of (1999).
Casting and Welding Processes, Santa Barbara, CA, 197-209 71. K. M. K. Yip, Artificial Intelligence, 80, 309-48 (1996).
(1986). 72. P. F. Mendez, F. Ordonez, Journal of Applied Mechanics,
44. G. M. Oreper, T. W. Eagar, J. Szekely, Welding Journal, 307s- Accepted for Publication, (2004).
12s (1983). 73. P. F. Mendez, F. Ordonez, in Proceedings of Fifth International
45. G. M. Oreper, J. Szekely, Journal of Fluid Mechanics, 147, 53-79 Workshop on Similarity Methods, Stuttgart, Germany, 21-31
(1984). (2002).
46. G. M. Oreper, J. Szekely, Metallurgical Transactions A, 18A, 74. J.-W. Park, P. F. Mendez, T. W. Eagar, Acta Materialia, 50, 883-
1325-32 (1987). 99 (2002).

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Alloying Element Vaporization and Liquid Metal Expulsion during Laser

Microjoining of Stainless Steel with Short Pulse
X. He, J. T. Norris*, P. W. Fuerschbach*, T. DebRoy
Department of Materials Science and Engineering
The Pennsylvania State University, Pennsylvania, USA
*Joining and Coating Department
Sandia National Laboratories, New Mexico, USA

Abstract found that the hardness of the weld metal was lower than that
of the base metal due to the magnesium vaporization. The loss
Vaporization and liquid metal expulsion are the two main of hardness was attributed to a reduction in the solid solution
mechanisms of material loss during laser microjoining. strengthening effect as a result of lower magnesium
Various factors that affect alloying element vaporization and concentration. In the electronics industry, where components
liquid metal expulsion during microjoining of 304 stainless are often processed in a clean room environment, discharge of
steel with short Nd-yttrium aluminum garnet (YAG) laser metal vapors is not acceptable. During laser assisted joining of
pulses were investigated experimentally and theoretically. The components, loss of alloying elements needs to be minimized.
temperature fields used to simulate the loss of materials were Therefore, quantitative understanding of the evaporation of
obtained from a well-tested comprehensive three-dimensional alloying elements and liquid metal expulsion is important in
transient heat transfer and fluid flow model. The calculated the welding of engineering alloys.
fusion zone geometry and composition change produced by Vaporization of alloying elements during laser spot
laser microjoining under various welding conditions showed welding is different from that during linear welding in several
fair agreement with the corresponding experimental results. ways. First, the evaporation rate is strongly time dependent,
The conditions necessary for the initiation of liquid metal i.e., the rate is negligible at the initiation of the pulse and
expulsion were determined by balancing the vapor recoil force gradually increases owing to increase in temperature. Second,
with the surface tension force at the periphery of the liquid because of the short duration of the laser pulse, experimental
pool. The laser power density and pulse duration are important determination of temperature and velocity fields is difficult
parameters for liquid metal expulsion. Higher power density and remains both an important goal and a major challenge in
and longer pulse duration increases the tendency of liquid the field. Third, although both surface area and the volume of
metal expulsion during laser microjoining. the weld pool are small, they change significantly with time.
Introduction As a result of these difficulties, very little information is
available in the literature about measurements of important
Laser spot welding is characterized by its small length variables such as the temperature field during microjoining.
scale, fairly short pulse duration, highly transient nature and In order to have a quantitative understanding of
very high heating and cooling rates. Because of the high vaporization and liquid metal expulsion during microjoining, a
power density used, the temperatures of weld metal often comprehensive model is needed. Aden et al 8 investigated the
exceed the boiling points of materials. In such situations, the laser induced vaporization from steel and aluminum surfaces
equilibrium pressure on the weld pool surface becomes higher as a function of laser intensity and material properties. A
than the atmospheric pressure, and significant vaporization material-dependent minimum laser intensity above which no
takes place from weld pool surface.1-5 If the weld pool further expansion of the metal vapor occurs was discussed.
temperature is very high, the escaping vapor exerts a large Anisimov9 and Knight10 derived expressions for the vapor
recoil force on the weld pool surface.6 As a consequence, the temperature, density, velocity and the extent of condensation
molten metal may be expelled from the weld pool. by solving the equations of conservation of mass, momentum
Vaporization and liquid metal expulsion are the two main and energy in a thin layer adjacent to the liquid-vapor
mechanisms of material loss during laser microjoining. The interface, known as the Knudsen layer. Their approach has
loss of alloying elements can result in significant changes in been incorporated into vaporization models1,2 to calculate the
the microstructure and degradation of mechanical properties of laser-induced vaporization rate. von Allmen11 indicated that
weldments.3-7 Moon and Metzbower7 investigated the change vapor pressure acts like a piston that exerts a pressure onto the
of properties of aluminum alloy before and after welding using melt, squirting it out of the hole radially. He also developed a
a CO2 laser with He gas shield. They found that the tensile theoretical model to calculate drilling velocity and drilling
properties of the welds were inferior to the base metal, mainly efficiency as a function of absorbed intensity.12 Chan and
because of magnesium depletion, loss of strain hardened Mazumder13 developed a one-dimensional steady state model
structure, and porosity. Cieslak and Fuerschbach3 investigated to describe the laser induced damage caused by materials
the property change of aluminum alloys 5456 and 5086. They removal through vaporization and liquid metal expulsion.

Copyright © 2006 ASM International® 109

Depending on the materials and beam power density, either Spectroscopy) analysis were taken for the interior surface of
vaporization or liquid expulsion was thought to be the the quartz tube for every experiment.
dominant mechanism of material removal. Basu and DebRoy6
examined the conditions for the initiation of liquid metal Mathematical modeling
expulsion during laser irradiation both experimentally and Transient temperature profiles. The temperature profile
theoretically. They proposed that liquid metal expulsion takes was calculated using a well tested three-dimensional
place when the vapor recoil force exceeds the surface tension comprehensive transient heat transfer and fluid flow model.
force of the liquid metal at the periphery of the weld pool. The weld metal was assumed to be an incompressible,
In this paper, the temperature fields used to simulate the Newtonian fluid. Constant thermophysical properties, such as
loss of materials were obtained from a well-tested viscosity, thermal conductivity and specific heat, were used
comprehensive three-dimensional transient heat transfer and for simplicity. The variation of absorption coefficient of the
fluid flow model. Using the computed temperature fields, laser energy by the stainless steel at different temperatures was
vaporization rate, composition change and total mass loss due ignored. The driving forces for the liquid flow in the weld
to vaporization of various alloying elements resulting from pool considered in the present model were the Marangoni
both concentration and pressure driven transport were force and the buoyancy force. The boundary condition, model
calculated. The calculated fusion zone geometry and framework and the solution procedure have been described in
composition change produced by laser microjoining of details elsewhere.15,16 The computed temperature fields as a
stainless steel under various welding conditions were function of time were then used to calculate the vapor pressure
compared with the corresponding experimental results. The and vaporization rates of alloying elements. The data used for
conditions necessary for the initiation of liquid metal calculations14,17-20 are presented in Table I.
expulsion were determined by balancing the vapor recoil force
with the surface tension force at the periphery of the liquid Table I Data used for calculations.14,17-20
pool. In addition, the influences of laser power density and Property/Parameter Value
pulse duration on liquid metal expulsion were analyzed. Density of liquid metal (kg/m3) 7.2 × 103
Absorption coefficient 0.27
Experimental procedure
Effective viscosity (kg/m sec) 0.1
Several 304 stainless steel welds were fabricated at the Solidus temperature (K) 1697
Sandia National Laboratories. The steel had the following Liquidus temperature (K) 1727
composition: 1 wt% Mn, 18.1 wt% Cr, 8.6 wt% Ni, 0.69 wt%
Enthalpy of solid at melting point (J/kg) 1.20 × 106
Si, 0.046 wt% C, 0.012 wt% P, 0.003 wt% S, and balance Fe.
A Raytheon SS 525 pulsed Nd:YAG laser was used for laser Enthalpy of liquid at melting point (J/kg) 1.26 × 106
spot welding with pulse energies of 2.1 J, 3.2 J and 5.9 J and Specific heat of solid (J/kg K) 711.8
pulse durations of 4.0 ms and 3.0 ms. Individual spot welds Specific heat of liquid (J/kg K) 837.4
from a pulsed laser beam were made on 3 by 10 by 17 mm Thermal conductivity of solid (J/m sec K) 19.26
EDM wire cut samples. Up to 15 individual spot welds were Effective thermal conductivity of liquid (J/m sec K) 209.3
made on each of the samples, using a single pulse for each Temperature coefficient of surface tension (N/m K) -0.43 × 10-3
individual spot weld. Coefficient of thermal expansion 1.96 × 10-5
Laser spot size was measured with 50 μm Kapton film Surface tension coefficient (N/m) 1.872
using the method described elsewhere.14 Longitudinal
metallographic cross-section measurements through several Vaporization due to concentration gradient. At the
collinear welds for each plate were averaged to determine weld pool surface, the concentrations of the alloying elements
weld pool size. The mass loss was experimentally determined in the vapor are higher than those in the bulk shielding gas.
by weighing each specimen before and after welding with a The vaporization flux of element i, Jci, can be defined as:
Metler MT5 micro-balance. To increase the accuracy of the
weight loss measurements, the reported weight loss per pulse (
J ci = K gi M i ai Pi 0 RTl − Cib ) (1)
is the average of the fifteen spot welds made on each sample. where Kgi is the mass transfer coefficient of element i, Mi is
During welding experiments, an open ended quartz tube, the molecular weight of the element i, ai is the activity of
25 mm in length, having a 6 mm inner diameter and a 1 mm 0
element i in the liquid metal, Pi is the equilibrium vapor
wall thickness, was placed co-axial to the laser beam and right
above the 304 stainless steel samples. A portion of the pressure of element i over its pure liquid, R is the gas constant,
vaporized elements and ejected metal droplets were collected Tl is the temperature at the weld pool surface, Cib is the
on the interior surface of the tube. The deposit and the concentration of element i in the shielding gas which is
particles were examined from each experiment. The SEM significantly lower than the concentration at the weld pool
micrographs and EDS (Energy-Dispersive x-ray surface. The mass transfer coefficient between the weld pool
surface and the shielding gas is calculated from the graphical

results of Schlunder and Gniclinski21 for a jet impinging on a temperature, the distribution patterns of vapor fluxes are
flat surface and is given by: similar to the surface temperature profiles. From Fig. 3(a), the
( (
K gi = 2 Pr 0.42 Re 0.5 (Di d ) 1 + Re 0.55 200 )) 0. 5
peak temperature near the weld center of the beam-workpiece
interaction zone exceeds the boiling point of the alloy. As a
[0.483 − 0.108(r d ) + 7.71 × 10 −3
(r d ) 2
] (2) result, the vaporization is predominantly driven by the
pressure gradient. Most of the vaporization from the weld pool
where Pr is Prandtl number, Re is the Reynolds number at the surface occurs from this active region. The diameter of this
nozzle exit, Di is the average diffusivity of element i in the region is approximately 0.2 mm as can be observed from Fig.
shielding gas at average temperature Tav, d is the diameter of 3(b). This dimension is comparable but somewhat smaller than
the nozzle, and r is the radial distance on weld pool surface. the laser beam diameter of 0.52 mm. The vaporization flux
Vaporization due to pressure gradient. During laser outside this region is much lower. The vaporization flux is
microjoining, the peak temperature reached on the weld pool driven mainly by the concentration gradient.
surface often exceeds the boiling point of the alloy. As a
result, the vapor pressure at the weld pool surface can be
higher than the ambient pressure, and the excess pressure
provides a driving force for the vapor to move away from the
surface. Therefore, the convective flux of the vaporized
elements, driven by the excess pressure is an important
contributor to the overall vaporization flux.
The velocity distribution functions of the vapor
molecules, f1, f2, and f3, escaping from the weld pool surface
at various locations are shown schematically in Fig. 1. On the
weld pool surface, the molecules cannot travel in the negative Fig. 1 A schematic diagram of the velocity distribution
direction, and as a consequence, the velocity distribution is functions in the Knudsen layer and in adjacent regions.
half-Maxwellian. Close to the weld pool, there exists a space
-0.5 0 0.5
of several mean free paths length, known as the Knudsen 1.5 1.5
layer, at the outer edge of which the velocity distribution just 1.4 1.4
z (mm)

reaches the equilibrium distribution. Considering the velocity K

1.3 97 1.3
distribution functions, the rate of vaporization and 16
condensation were calculated based on the works of 1.2 600 mm/s 1.2
Anisimov9 and Knight10 by solving the equations of 1.1 1.1
conservation of mass, momentum and kinetic energy. The -0.5 0 0.5
detailed procedure for the calculation of vaporization flux due x (mm)
to pressure gradient, Jpi, is available in a recent paper22 and is Fig. 2 Experimental and calculated weld pool cross section.
not presented here. Laser power: 1507 W, pulse duration: 4 ms, and spot radius:
0.389 mm.
Results and discussion 0.2
Temperature in K (a)
Validity of heat transfer and fluid flow model. The 0.15
experimentally determined weld pool cross sections are
y (mm)

0.1 3000
compared with the corresponding computed values in Fig. 2. It 3150
is observed that the calculated weld pool geometry and 0.05 3250
dimensions agree well with the experimental results. Since the
temperature coefficient of surface tension is negative, the -0.2 -0.1 0 0.1 0.2
molten metal on the surface flows from the center to the x (mm)
periphery of the pool. As a result, the convection in the weld Total vapor flux in kg/m2 -s (b)
pool aids in the transport of heat from the middle to the 0.15
y (mm)

periphery. The fair agreement indicates validity of the

0.1 10
transient heat transfer and fluid flow model. The temperatures 50
calculated from this model are reliable and can be used in the 0.05 100
calculation of vaporization and liquid metal expulsion. 0
-0.2 -0.1 0 0.1 0.2
Vaporization rate and composition change. Fig. 3 x (mm)
shows the computed temperature distribution and total vapor Fig. 3 Calculated distribution of temperature and the total
fluxes on weld pool surface after 3.0 ms. Because vapor vapor flux on the weld pool surface after 3 ms. Laser power:
pressures of all the alloying elements are strong functions of 1067 W, pulse duration: 3 ms, and spot radius: 0.26 mm.

For the laser spot welding, vaporization is time- properties of alloys, the successful prediction of composition
dependent. Fig. 4 shows the variation of calculated change by the model is helpful to understand how these
vaporization rates with time. It can be seen that the properties are affected by laser microjoining.
vaporization rates of the constituent alloying elements increase

Concentration Change (wt %)

with time. At the end of the pulse cycle, vaporization rates
Calculated value
decrease suddenly and the vaporization of alloying elements 0.4
Experimental value
stops. It can be also seen that iron is the main vaporizing 0.2
species, followed by chromium and manganese. Although
manganese has the highest vapor pressure over its pure 0

liquid,23 its low equilibrium vapor pressure over the alloy -0.2
results in the lower vaporization rate than iron and chromium.
As a result of vaporization of constituent alloying
elements, the concentrations of alloying elements in the weld -0.6
Fe Mn Cr Ni
metal will change after laser microjoining. The fair agreement
between experimental and computed concentration changes of Fig. 5 Comparison between the calculated and the
four alloying elements is shown in Fig. 5. It is shown that the experimentally determined composition change of 304
concentrations of manganese and chromium decrease while stainless steel. Laser power: 1067 W, pulse duration: 3 ms,
those of iron and nickel increase after laser microjoining. and spot radius: 0.225 mm.
Although the total mass of iron and nickel in the weld pool is

Composition Change of Mn (wt %)

lower than that before the welding, the total mass of the weld 0

pool has decreased because of the loss of manganese,

chromium, iron and nickel. As a result, the concentrations of
iron and nickel in the fusion zone are higher than those in the -0.4
base metal because of the pronounced loss of manganese and
chromium. -0.6
Experimental values
2 -0.8
Calculated values
Vaporization Rate (mg/s)

total 2000 3000 4000 5000 6000 7000
Power Density (W/mm2)

Fig. 6 Experimental and calculated concentration change of

1 Fe Mn as a function of power density. Pulse duration: 3 ms.
Liquid metal expulsion. The total weight loss of 304
0.5 Cr stainless steel after laser microjoining at various laser power
densities, obtained from the calculated and experimental
Mn results, are shown in Fig. 7. As the laser power density
0 Ni increases, the temperature at the weld pool surface exceeds the
0 1 2 3 4
Time (ms) 50
Fig. 4 Change of vaporization rates of the four alloying
Total Weight Loss ( μ g)

elements with time. Laser power: 1067 W, pulse duration: 3 40 Calculated value
ms, and spot radius: 0.26 mm. Experimental value
The fair agreement between the experimental and the
computed concentration change of manganese as a function of 20
power density can be seen in Fig. 6. In experiments, several
electron microprobe traces were made for every sample. The 10
final concentration is effected by two factors: volume of weld
pool and total weight loss. As laser power density increases, 0
both the volume and total weight loss increase. As a result, the 1000 2000 3000 4000 5000 6000 7000
change of concentration with laser power density is not Power Density (W/mm2)
monotonous. Depending on how the rates of volume and total Fig. 7 The calculated vaporization loss is compared with
weight loss change with power density, the concentration measured mass loss for different power densities. Laser
change either increases or decreases. Because the composition power: 1067 W, pulse duration: 3.0 ms.
change of alloys has effects on the mechanical or corrosion

boiling point of the steel. As a result, the total vaporization occurrence of intermittent or heavy expulsion. If laser power
loss increases significantly due to pressure driven density keeps constant, for example, keeping it at 7 kW/mm2,
vaporization. However, it is observed that the experimental no vapor deposit were observed on the inner surface of quartz
weight loss is always higher than the computed weight loss tube under the pulse duration of 2 ms. When the pulse
due to vaporization. A possible reason for the observed duration increases to 3 ms, metal vapor was found on the inner
discrepancy between the experimental weight loss and the wall of quartz tube. When the pulse duration is equal or higher
calculated vaporization loss is that only a portion of the weight than 4 ms, it is possible for intermittent even heavy expulsion
loss occurs due to vaporization and the remainder of the loss to take place. It can be expected that longer pulse duration
must be attributed to some other mechanism. Therefore, the results in a lower critical laser power density for the
possibility of ejection of the tiny metal droplets from the weld occurrence of liquid metal expulsion, which was also
pool owing to the recoil force exerted by the metal vapors was successfully predicted from calculation results, as shown by
examined both experimentally and theoretically. the solid line in the figure.
Expulsion of metal drops takes place when the vapor 600
recoil force exceeds the surface tension force of the liquid (a)
metal at the periphery of the weld pool.6. The vapor recoil

Force (gm-cm/sec2)
force, Fr , and the surface tension force at the periphery, Fs , 400
Surface tension
can be expressed by:
Fr = 2π
∫ rΔP(r )dr
Recoil force

FS = 2πr0σ (4) 0
0 0.5 1 1.5 2 2.5 3
where rB is the radial distance at which the surface temperature Time (ms)
reaches the boiling point, ΔP(r) is the difference between the
local equilibrium vapor pressure and the atmosphere pressure
and is the function of radial distance from the beam axis, r0 is
the radial distance at which the temperature is equal to the
solidus temperature of the alloy and σ is the surface tension
coefficient at that temperature. Fig. 8(a) shows the computed
values of these two forces during laser microjoining. At the
start of the pulse, the surface tension force is higher than the (b) 0.5 mm
recoil force. As the temperature increases with time, both the
surface tension force and recoil force increase. However, the Fig. 8 (a) Calculated recoil force and surface tension force at
recoil force increases faster than the surface tension force. At the periphery of the liquid pool as a function of time. (b)
about 2.6 ms after the start of the pulse, the two forces are Vaporized elements and tiny droplets ejected from the weld
roughly equal. Further heating results in the recoil force pool of 304 stainless steel were captured on the inner surface
surpassing the surface tension force, at which point, ejection of a both end open quartz tube placed co-axial with the laser
of metal droplets is anticipated. The liquid metal expulsion has beam during laser microjoining. Laser power: 1067 W, pulse
been verified during experiments, as shown in Fig. 8(b). duration: 3 ms, spot diameter: 0.405 mm.
Clearly, mass loss is contributed by both vaporization of
alloying elements as well as the liquid metal expulsion.
The combinations of laser power and spot diameter that
lead to liquid metal expulsion are shown in Fig. 9. The points
Spot Diameter (mm)

on the same curve have same laser power density, defined by 3.0 kW/mm2
laser power per unit area. It is observed that the liquid
expulsion occurs above a critical laser power density, which is
about 7 kW/mm2 in Fig. 9. The recoil and surface tension 0.4
forces were compared to predict the critical beam diameter Heavy expulsion
0.2 Intermittent expulsion
under different laser powers, as shown by the solid line. It is Vapor
close to that obtained from experiments. No vapor
Beside laser power density, pulse duration also has 0 0.5 1 1.5 2

important effects on liquid metal expulsion. Fig. 10 shows the Laser Power (kW)
extent of liquid metal expulsion under different laser power Fig. 9 Liquid metal expulsion data under different laser power
densities and pulse durations. At constant pulse duration, densities for laser microjoining of 304 stainless steel. Pulse
higher laser power density increases the tendency of the duration: 3 ms.

20 YAG laser welds in aluminum-alloys 6061, 5456, and
Heavy expulsion
Intermittent expulsion 5086, Metall. Trans. B, 19B, 319-29 (1988)
Power Density (kW/mm2)

16 Vapor 4. A. Block-bolten and T. W. Eager, Metal vaporization from

No vapor
weld pools, Metall. Trans. B, 15B, 461-69 (1984)
5. T. A. Palmer and T. DebRoy, Numerical modeling of
enhanced nitrogen dissolution during gas tungsten arc
welding. Metall. Trans. B, 31B, 1371-85 (2000)
6. S. Basu and T. DebRoy, Liquid-metal expulsion during
laser irradiation, J. Appl. Phys., 72, 3317-22 (1992)
7. D. W. Moon and E. A. Metzbower, Laser Beam Welding
1 3 5 7 9 of Aluminum Alloy 5456, Welding J., 62, 53s-58s (1983)
Pulse Duration (ms) 8. M. Aden, E. Beyer, G. Herziger and H. Kunze, Laser-
Fig. 10 Analysis of liquid metal expulsion under different induce vaporization of a metal-surface, J. of Phys. D:
laser power densities and pulse durations for laser spot Appl. Phys., 25, 57-65 (1992)
welding of 304 stainless steel. 9. S. I. Anisimov and A. Kh Rakhmatulina, Soviet Physics –
JETP, 37, 441-44 (1973)
Summary and conclusions 10. C. J. Knight, Theoretical modeling of rapid surface
vaporization with back pressure, AIAA J. 17, 519-23
A comprehensive model to calculate temperature,
vaporization rates of alloying elements and weld metal
11. M. von Allmen and A. Blatter, Laser-Beam Interactions
composition change during laser microjoining of 304 stainless
with Materials, Springer-Verlag, New York, (1995)
steel, taking into account both vaporization and condensation,
12. M. von Allmen, Laser drilling velocity in metals, J. Appl.
was developed. The calculated fusion zone geometry and
Phys., 47, 5460-63 (1976)
composition change produced by laser microjoining under
13 C. L. Chan and J. Mazumder, One-dimensional steady-
various welding conditions showed fair agreement with the
state model for damage by vaporization and liquid
corresponding experimental results. The vaporization rate
expulsion due to laser-material interaction, J. Appl. Phys.,
increased with time. The expulsion of metal droplets was
62, 4579-86 (1987)
predicted by computations and verified by experiments. The
14. P. W. Fuerschbach and G. R. Eisler, Effect of laser spot
conditions necessary for the initiation of liquid metal
weld energy and duration on melting and absorption, Sci.
expulsion were determined by balancing the vapor recoil force
Technol. Weld. Joining, 7, 241-46 (2002)
with the surface tension force at the periphery of the liquid
15. X. He, P. W. Fuerschbach and T. DebRoy, Heat transfer
pool. Higher power density and longer pulse duration
and fluid flow during laser spot welding of 304 stainless
increased the tendency of liquid metal expulsion during laser
steel, J. Phy. D: Appl. Phys., 36, 1388-98 (2003)
16. W. Zhang, G. Roy, J. Elmer and T. DebRoy, Modeling of
Acknowledgements heat transfer and fluid flow during gas tungsten arc spot
welding of low carbon steel, J. Appl. Phys., 93, 3022-33
The work was supported by a grant from the U.S. (2003)
Department of Energy, Office of Basic Energy Sciences, 17. D. Peckner and I. M. Bernstein, Handbook of Stainless
Division of Materials Sciences, under grant number DE- Steels, McGraw-Hill, New York (1977)
FGO2-01ER45900. Portions of this work, performed at Sandia 18. J. R. Davis, Metals Handbook, ASM International,
National Laboratories, were supported by United States Materials Park, OH (1998)
Department of Energy’s National Nuclear Security 19. J. R. Davis, ASM Specialty Handbook. Stainless Steel,
Administration under contract DE-AC04-94AL85000. The ASM International, Materials Park, OH (1994)
authors would like to thank Paul Hlava for his help with the 20. ASM International Handbook Committee, Metals
experimental measurements. handbook. Volume 1. Properties and Selection: Ion, steels,
References and high-performance alloys, ASM International,
Materials Park, OH (1990)
1. H. Zhao and T. DebRoy, Weld metal composition change 21. E. U. Schlunder and V. Gniclinski, Chem. Ing. Technol.,
during conduction mode laser welding of aluminum alloy 39, 578-84 (1967)
5182, Metall. Trans. B, 32B 163-72 (2001) 22. X. He, T. DebRoy and P. W. Fuerschbach, Composition
2. K. Mundra and T. DebRoy, Calculation of weld metal change of stainless steel during microjoining with short
composition change in high-power conduction mode laser pulse, J. Appl. Phys., 96, 4547-55 (2004)
carbon dioxide laser-welded stainless steel, Metall. Trans. 23. X. He, T. DebRoy and P. W. Fuerschbach, Probing
B, 24B, 145-55 (1993) temperature during laser spot welding from vapor
3. M. J. Cieslak and P. W. Fuerschbach, On the weldability, composition and modeling, 94, 6949-58 (2004)
composition, and hardness of pulsed and continuous Nd-

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Numerical Analysis for Optimization of Aluminum Tube Welding

J. Menke, D. F. Farson, M.H. Cho
Ohio State University, Columbus OH

B. Green, L. Brown
EWI, Columbus OH

presented and the effects of the heat source shape in the

Abstract modeling of laser welds as well as the Marangoni effect
(magnetohydrodynamics) were discussed. Jönsson, et.al.2
A computer model of the GTAW process is presented. The presented a very nice comprehensive survey of research on
model includes a moving, gaussian heat source, deposited transport phenomena associated with arc welding. It is, of
filler wire and temperature dependent material properties. The course, valid only up until the time it was written in 1993.
arc was represented by heat input and pressure distributions However it does provide an illustration of the history of 1-, 2-
imposed on the metal surface. The model was used to predict and 3-dimensional weld modeling, both coupled and
the location of liquidus and solidus isotherms and the weld uncoupled. For more than 20 years, the most extensive and
face and root surface reinforcement height and width around influential work on welding process has been done by a group
the circumference of a weld made with the pipe in a fixed at PSU; a few of the most recent relevant works are noted.3,4,5
(5G) position. When the arc heat input was calibrated from The primary contribution of the present work in comparison is
calorimetric measurements, the model roughly predicted the to the body of prior research is the use of a simulation
weld pool solidified face and root widths (maximum error was technique that allows prediction of the time-varying shape of
about 21%) and the temperature in the material away from the the weld pool surface and the solidified weld bead around the
weld pool. circumference of a fixed aluminum pipe in the horizontal
In construction of certain chemical plants, thousands of Modeling
aluminum pipe welds are required and welding, inspection and
repair costs comprise a significant cost of construction. The simulation deals with Gas Tungsten Arc Welding
A key factor in welding costs is meeting the stringent (GTAW) of 6061-T6 aluminum pipe with 4043 aluminum
requirements on the weld smoothness on the inner diameter of filler wire. In the orbital welding process considered in this
the header piping. simulation, the pipe axis is horizontal and the butt-weld joint
The objectives for this project are as follows: is vertically-oriented It is also important to model the weld as
1. To develop an accurate numerical process model of it completes a 360° rotation about the cylinder in order to
GTAW that incorporates the most important physical measure the temperature effects from the beginning of the
phenomena. weld.
2. Experimentally measured process heat inputs for
incorporation in the model. A key desired model predictions were the width and height of
3. Use model predictions of process parameters that the solidified weld root penetration. The idea is to make welds
optimize heat input around the weld for consistent where the extension into the pipe interior of the weld bead
penetration. does not exceed a maximum value and is relatively consistent
around the entire weld circumference. Other features that were
Background desired to predict accurately were weld bead front width and
Taylor, et al.1 developed a unified model that was primarily
concerned with the prediction of distortion, residual stress and The following affects were considered to be important to
the size of the heat affected zone. Their paper presented the evaluate for inclusion in a numerical model for accurate
computational modeling of welding phenomena within a prediction of the weld pool flow and solidified shape 6: surface
framework embracing models from both the fields of tension gradient (Marangoni affect), buoyancy (densitometric
Computational Fluid Dynamics (CFD) and Computational gradient), magnetohydrodynamics (MHD) , gravity force,
Solid Mechanics (CSM). With regard to the CFD modeling of viscosity, convection heat loss and filler wire addition.
the weld pool fluid dynamics, heat transfer and phase change,
cell-centered Finite Volume (FV) methods were employed. The arc is represented by heat input and pressure boundary
Vertex-based FV methods were employed with regard to the conditions on the fluid rather than being simulated from first
elasto-plastic deformation associated with the CSM. principals. Based on previous research7, the arc heat input
PHYSICA software was employed. Velocity profiles were distribution is assumed to be Gaussian. The magnitude of the
heat input is calculated as the product of arc current and

Copyright © 2006 ASM International® 115

voltage, multiplied by an empirical arc efficiency term
(measured by calorimetry). The radius of the arc heat input
distribution at the surface is taken from previous
measurements published in the literature.
3Q  − 3r 2 
q (r ) = exp 2  ,
πa 2  a 
where q(r) = heat flux at the material surface, r = radial
distance from center of arc,
Q = heat input magnitude and a = effective arc heat input

Despite that fact a cylindrical geometry would seem to be Figure 2. Simulation domain dimensions
necessary to simulate welding of pipe, it happens that a
simulation based on a flat or “linear” material geometry can
also be made to accurately represent pipe welding and actually
has some advantages over the more obvious choice. By
“unwinding” the pipe, starting at top dead center, a ‘linear’
model with a length equal to the circumference of the pipe
could be used to simulate the process. This transformation is
shown in Figure 1. The use of the linear model meant that
certain physical effects, such as the direction of the gravity
vector and convection to the surroundings, would have to be
varied with position along the length of the material. The
simulation results also require a certain amount of visual
interpretation to match the weld arc travel along the linear part
to the circumferential travel of a weld arc around a pipe. In the Figure 3. Simulation boundary conditions
linear model the x-axis was chosen as the direction of arc
travel along the weld, the y-axis was chosen as the direction To transform the model from the round pipe to the linear
normal to the weld and the z-axis was the direction through geometry, the gravity vector and convection heat transfer
the thickness of the material, with dimensions shown in Figure needed to be varied sinusoidally as a function of distance
2. along the weld. This model uses a single heat source that
begins centered over the beginning face of the material at x=0,
y=0. In this simulation the beginning and end faces of the
domain at x = 0 and x = 2πr are defined to have periodic
boundary conditions (Fig. 3). In this model, the heat source is
initially “split” by the beginning boundary at the weld start.
Therefore, one half of the Gaussian heat input distribution is
located on the right side of the domain.

The input power endpoint and slope values were obtained by

inserting the data from the weld process controller into an
Excel spreadsheet and applying a linear fit to the segments of
the weld cycle. This is shown graphically in Figure 4 below.

Figure 1. Conversion of cylindrical to linear geometry

y = -16.667(t) + 1242
1400 70%

1200 60%
Power, watts


Arc Efficiency


600 y = 134.244(t) + 771.181








y = -58.317(t) + 2286.7

200 20%

0 10%
RP 50 Amp
0 10 20 30 40
Time, seconds 0%

Figure 4. Pulsed power input and straight line fits used for Figure 5. Arc efficiencies measured by calorimetry.

Simulations done using the Volume of Fluid (VOF) numerical Model Results and Discussion
technique (implemented in commercial Flow-3D software8)
and were run in on a dual Xeon processor computer with 2 GB The temperature distribution sequence depicted in Figures 6(a-
of RAM and a 32 GB SCSI hard drive, using the Windows XP c) show progression of the top surface temperature of a weld
operating system. The time required for simulation of a weld from beginning to end. Figure 6(a) shows the temperature
around a 1”-diameter pipe was approximately 12 hours. distribution on the pipe surface shortly after the weld has
begun. The solidus temperature of aluminum is taken as 875 K
Arc Efficiency Measurements and is represented as pure red. Therefore, the boundaries of
the weld pool are clearly visible. The temperature distribution
The arc efficiency of the welding process to be modeled must shown in Figure 6(b) represents the weld as the heat source
be determined in order to know what percent of the power nears the right side of the simulation domain. All material
input to the arc is transferred as heat to the part. The arc within the simulation domain has reached a temperature above
efficiency is defined as the ratio of the energy transferred to a 400 K (about 260°F). The temperature distribution shown in
sample part from a weld arc to the total energy input to the arc Figure 6(c) represents the weld after the heat source has made
(integral of current x voltage). The procedure used was liquid one complete revolution around the pipe and is in the
nitrogen calorimetry9 whereby a heated sample part is “overlap” region near the end of the weld where the heat input
immersed into a bath of liquid nitrogen and the amount of is being decreased and the weld pool size is decreasing. All
gaseous nitrogen released is measured as the part is cooled to material within the simulation domain has then reached a
the equilibrium temperature of -196° C (77 K). By multiplying temperature above 500K (about 440°).
the mass of nitrogen released by the heat of transformation
one can calculate the amount of heat energy removed from the Figures 6(a-c) also provide a clear illustration of the affect of
welded sample. By measuring the heat energy in a room the periodic boundary condition applied on the left and right
temperature part and subtracting that from the heat energy of a sides of the material. When using this boundary condition, the
welded part you get the heat energy transferred to the part heat source is bisected by the left and right boundaries at the
from the weld arc. This is then compared to the energy input beginning of the simulation time, so heat is deposited on both
to the weld head to compute the arc efficiency. the far left and far right of the material region. As the heat
source travels to the right, the heat input distribution becomes
The results of the arc efficiency testing are summarized in entirely contained within the left side of the material region,
Figure 5. The measured efficiencies are comparable but although the heat that it deposited on the far right side of the
somewhat less that similar values in the literature.7-1010111213 region is still visible. This residual heat captures the
The arc efficiency of the reverse polarity process is much less “preheating affect” that causes the weld size to increase as the
than straight polarity, which is to be expected. The variable heat source approaches the right side of the material region.
polarity process yielded results that were essentially the same Thus, the periodic boundary condition is seen to one key to
as the straight polarity process which indicates that it is not realistically simulating a cylindrical geometry within a
simply a linear combination of the proportion of time spent on rectangular simulation domain.
each cycle of the waveform times the efficiency for that cycle.
Further work would be required to see if the efficiency of the
reverse polarity process is increased once a molten weld pool
is established.


Figure 6. Temperature distributions a) top, 3.6 s; b) bottom 18

s; c) bottom 32 s.

Another view of the temperature distribution along the weld

centerline predicted by the model is shown in Figure 7. In this
“cut-away view”, the isotherm colors have been adjusted to
reveal the locations of the liquidus and solidus isotherms, so
the weld pool fusion boundary shape is clearly visible. The
length of the weld pool on the top and bottom surfaces of the
weld can be easily estimated from these figures.


Figure 7. Cut-away view of temperature distribution along the

weld centerline showing weld pool fusion boundary shape.

Figure 8 shows the temperature history at a single distance of

3.5 in. along the weld seam from the start point and various
positions from the weld centerline on the top face of the pipe.
Similarly, Figure 9 shows the predicted temperature history at
a distance of 5 mm from the weld centerline but at various
locations along the weld joint. In both plots, the overall
heating of the pipe is evidenced by the general upward trend
of the temperature histories.
Temperature vs Time, x= 5.3 cm

(b) y=0.0 cm
1200 y=0.5 cm
Temperature, K

1000 y=0.25
cm y=1.0 cm



200 y=2.5 cm
y=5.0 cm
0 10 20 30 40
Time, sec

Figure 8. Top face temperature vs. time at a fixed location

along the weld seam and at various distances from the

3. Another cause may be in the way that the wire feed is
Temperature vs. Time, y= 0.5 cm
modeled as an integral part of the solid base material. This
900 technique entails simply apply a rectangular block of
800 aluminum material equal in swept volume to the wire feed
700 speed times cross-sectional area. Although this makes more
efficient use of computer resources, the arc heat input is
Temperature, K

applied to a surface that is a slightly above the material
x=0.0 cm

surface. This may contribute to the lack of penetration seen,
x=2.0 cm
particularly at the beginning of the weld cycle when the
300 x=4.0 cm
molten pool is just getting established.
200 x=6.0 cm

100 x=8.0 cm
4. Another issue is with the measurement of the welds,
0 5 10 15 20 25 30 35 40
themselves. Weld bead widths were estimated by visual
Time, seconds inspection of the face and root surfaces under a low
magnification binocular microscope. Because the distance
Figure 9. Top face temperature vs. time at a fixed distances between the liquidus and solidus isotherms is relatively wide,
from the centerline and at various locations along the weld the weld bead width measurements varied somewhat because
seam. of the difficulty to identify precisely where the fusion
boundary in the weld zone. and part of this is variation due to
the fact that different observations measuring the weld widths
As can be seen from the weld pool width results displayed in of the same welds determined values. Because This due to the.
Tables 1 and 2, the model predicted a greater outside fusion
zone width and also less root width than that which was seen 5. Identical parts welded with the same process and variable
on welded parts. Probable causes of this are as follows: settings have varying values for weld width and
reinforcement. This variation makes up most of the standard
1. The model and parts used for this comparison were two- deviation magnitude seen in the measurements.
inch diameter pieces. This required a very long material
domain. Issues of runtime vs. model accuracy were Table 1. Root Width, mm
encountered due to the high aspect ratio of length to material
thickness seen in these models. Flow-3D requires that the Measurement Prediction,
computational cells have aspect ratios no greater than 1.5, mm mm Error, %
which means that there are limits to how many cells there can
Pos deg 90 270 90 270 90 270
be through the thickness of the part. Increasing the number of
average 5.0 4.6 4.0 4.5 -20% -2%
cells in the z-direction (material thickness) from seven to, say,
ten requires a proportional increase in the number of cells in std dev 1.37 1.47
the x- and y-directions. This results in 1.33 = 220% increase in
model size. This can seriously slow the pace of development Table 2. Outside width
from a 4 hour model to an 18 hour model runtime. Also,
increasing the number of cells to ten is not really a substantial Measurement Prediction,
increase. Twenty cells are required before noticeable increases mm mm Error, %
in model accuracy seen. All this boils down to is that the lack Pos deg 90 270 90 270 90 270
of penetration seen in the computer model may be due to not average 6.9 6.7 8.0 8.1 16% 21%
enough computational divisions in the z-direction through the std dev 1.19 1.43
material thickness. Extreme gradients in temperature are not
handled well, in general, by digital simulations. The usual fix
is to increase the number of computational cells. Model-based Weld Optimization
2. It was discovered late in the model development that two The goal of this work was to produce a model that can be used
complete and empty computational cells are required above to assist in optimizing the welding of aluminum pipes. One
and below the material for proper solving of the model. The possible useful function would be to modify a baseline
cells above the material have met this condition throughout the welding procedure to have a faster or slower travel speed
development of the model due to the increased height of the while adjusting arc power to maintain a uniform back bead
wire on the top surface. This was not always the case below width.
the material. This may have affected the heat transfer to the
void below z = 0, causing an artificial cooling affect on the
bottom of the part and reducing the penetration and weld pool
width seen there.

1800 the semi-empirical simulation approach illustrated in this
1600 work, Gaussian heat input parameters can be calibrated to
1.2x TRAVEL SPEED determine the effects of welding energy, heat flow and gravity
1200 on weld bead shape around the entire weld joint, including the
Power, watts

overlap area at the end of the weld. The model approximately
predicts of weld pool shape as it varies around the pipe
(DATA FROM AMET) circumference. The model can be executed to provide
400 0.8x TRAVEL SPEED numerical modeling data to help develop and optimize
welding procedures.
0 100 200 300 400 500 600 References
Rotational Position, degrees
Gareth A. Taylor, Michael Hughes, Nadia Strusevich, and
Koulis Pericleous, “Finite Volume Methods Applied To The
Figure 10. Arc Power input profiles that produce a constant Computational Modeling Of Welding Phenomena” in
weld back bead width along the entire weld, plotted for 5 Proceedings of the Second International Conference on CFD
different travel speeds. in the Minerals and Process Industries, Melbourne,
Australia, 6-8 December, 1999. (CSIRO, 1999)
The results plotted in Figure 10 show arc power profiles 2
P. G. Jönsson, J. Szekely, R. T. C. Choo, and T. P. Quinn.
corresponding to various travel speeds that all are predicted to
“Mathematical Models Of Transport Phenomena Associated
yield the same back bead penetration. The curves are based on
With Arc Welding Processes: A Survey.” Modeling and
an initial heat input profile that was taken from a optimized
Simulation in Materials Science and Engineering, 2(5)
process (labeled as 1.0x travel speed). The travel speed was
1994: 995-1016.
increased and decreased from this initial value and the heat 3
input required to produce consistent back bead width was S Mishra, T DebRoy, A heat-transfer and fluid-flow-based
determined by iteratively executing the simulation to arrive at model to obtain a specific weld geometry using various
the correct heat input. The same data is re-plotted in Figure 11 combinations of welding variables, J. Appl. Phys. 98(4):
to show the relationship in a different way. Generally, the Art. No. 044902 (2005).
increase in travel speed required to maintain a constant weld TA Palmer, T DebRoy, Numerical modelling of enhanced
back bead width increased with travel speed with a rate that nitrogen dissolution during gas tungsten arc welding, Met
was higher-order than linear. Mat. Trans B 31(6):1371-1385 (2000).
W Zhang, GG Roy, JW Elmer, T DebRoy, Modeling of heat
transfer and fluid flow during gas tungsten arc spot welding
Heat Input vs Travel Speed of low carbon steel, J Appl Phys 93(5):3022-3033 (2003).
Sindo Kou 1996. Transport Phenomena And Material
Processing, 1st Edition. New York: Wiley-Interscience.
160% Sindo Kou 2003. Welding Metallurgy, 2nd Edition. New
Heat Input Change, %

140% York: Wiley-Interscience.

Flow-3D, Flow Science Inc, 683 Harkle Rd Suite A, Santa
Fe, NM 87505. 2004.
100% Joseph, D. Harwig, D. Farson, R. Richardson 2003
Measurement and Calculation of GMAW-P Arc Power and
Heat Transfer Efficiency, Science and Technology of
60% Welding and Joining, 8 (6): 400-406.
40% 60% 80% 100% 120% 140% 160% 10
Travel Speed Change, %
J. N. DuPont, and A. R. Marder 1995. “Thermal Efficiency
of Arc Welding Processes”, Welding Journal, 74 (12): 406-s
to 416-s.
Figure 11. General trend of heat input variation that was J. B. Wilkinson, and D. R. Milner 1960. “Heat Transfer
required for a fixed weld back bead width at various travel From Arcs”, British Welding Journal, 7 (2): 115 to 128.
speeds. R. W. Niles, and C. E. Jackson 1975. Weld Thermal
Efficiency of the GTAW Process, Welding Journal, 54
(1): 25-s to 32-s.
Conclusions 13
W. H. Giedt, L. N. Tallerico, and P. W. Fuerschback 1989.
“GTA Welding Efficiency: Calorimetric and Temperature
Field Measurements”, Welding Journal, 68 (1): 28-s to 32-s
This work demonstrated that a modified commercial VOF
code can, with some degree of accuracy, predict the weld pool
shape with input data taken from actual welds. By adopting

Proceedings of the 7th International Conference on Trends in Welding Research, May 16–20, 2005, Callaway Gardens Resort, Pine Mountain, Georgia, USA

Laser Plasma Powder Hybrid Welding

in Vertical-Up and Vertical-Down Positions
K. Stelling, Th. Boellinghaus, M. Lammers, H. Schobbert
Federal Institute for Materials Research and Testing, Berlin, Germany

Abstract improved the gap bridging and misalignment leveling

capacities. Such improvements drastically reduced the weld
Laser arc hybrid welding processes are increasingly applied to preparation costs and, together with the achievement of deeper
industrial fabrication welding, but are still restricted to penetration, made such hybrid welding techniques most
horizontal welding positions associated with complicated attractive for ship building and pipe manufacturing at even
component and assembly part movement. Out-of-position somewhat higher investment costs and lower welding speeds
hybrid welding has not yet been reported in the open literature. as compared to sole laser processes.
Based on a preliminary study with the austenitic stainless steel Laser arc hybrid techniques can generally be categorized into
1.4565 (AISI S34565) to adopt the LPPAW (Laser Plasma process coupling the laser to GMAW (Gas Metal Arc
Powder Arc Welding) to the PF and PG position, similar Welding) and those coupling the laser to PAW (Plasma Arc
experiments have been carried out by variation of the joint Welding). During the recent years, there is a tendency of
inclination angle. In order to widen also the material LGMAW (Laser Gas Metal Arc Welding) towards better gap
application range, a Nickel base alloy and an austenitic bridging capacities as compared to LPAW (Laser Plasma Arc
stainless steel with a remarkably lower content of the alloying Welding). But, although meanwhile improved by pulsing and
elements Cr und Ni have been investigated. With the present further arc control, LGMAW appears as a less stable process,
contribution, the process proved to be generally applicable to in particular, if the droplet transfer is extinguishing part time
vertical up and down welding of of such CRAs (Corrosion the laser keyhole. In such cases, decoupling the energy input
Resistant Alloys) and thus, the perspectives for out-of- from the filler material transfer appears as very effective for a
position laser arc hybrid welding have evidently become better weld process control and particularly LPPAW (Laser
wider. As a particular item with respect to corrosion resistance Plasma Powder Arc Welding) has been proven as a very stable
and safer operation of welded stainless steel components, hybrid welding procedure [1] and utilizes applications needing
more tightly focusing of the powder feeding in the plasma the fine dispersed filler material transferred deeply into the
torch avoids powder deposits alongside the weld seem and weld root to avoid hot cracking by alloying additions, for
improves the shielding of the process zone during welding. instance. A comparison of such multi parameter hybrid
techniques appears to be difficult and is further complicated
by the fact that both arc welding types have meanwhile been
Introduction coupled to various laser types, in particular to Nd:YAG as
well as to CO2 lasers. It can only be emphasized that
Considering fabrication welding efficiency as welded LGMAW (Laser Gas Metal Arc Welding) and LPAW (Laser
longitudinal section per unit time, modern laser welding Plasma Arc Welding) should not be regarded as concurrent
techniques have become increasingly attractive in nearly all processes, but increase the general application range of laser
industrial fabrication branches over the last two decades. In arc hybrid welding by their specialties. However, up to the
most cases, the high investment and operation costs have present laser arc hybrid welding has predominantly been
successfully been counterbalanced by production factors carried out in position PA [2], in order to avoid weld metal
related to the high welding speeds. fall-through and lack of fusion. Circumferential pipeline
In the last decade, further laser application limits were hybrid welding is performed in the PA or PC position [2],
overcome by the introduction of hybrid laser arc welding moving the tubular underneath the static hybrid welding
techniques. Such coupling of the laser to a respective arc system [3]. To avoid such complicated and cost extensive
welding process in a common process zone particularly fabrication procedures and to overcome the important

Copyright © 2006 ASM International® 121

industrial application restriction to in-position (PA) welding, Table 4: Chemical composition of the austenitic powder filler
laser arc hybrid techniques have to be utilized for out-of- metal 316 L, wt.%, bal. Fe
position welding. As a first step towards this direction, it has C Cr Ni Mo Ti Nb
been shown in a previous preliminary study [4] that LPPAW 0.014 16.3 10.6 2.15 n.s. n.s.
of a high nitrogen austenitic stainless steel 1.4565 (AISI Si Mn P S N Cu
S34565) is feasible in the vertical-up and vertical-down 0.52 1.18 <.01 0.004 n.s. n.s.
welding position, PF and PG [2], respectively. In order to n.s. not specified
widen the application range first towards other CRAs
(Corrosion Resistant Alloys) with respectively high surface Welding Process
quality requirements, similar welding experiments have been For the welding experiments, a diode pumped Nd:YAG-laser
carried out with an austenitic stainless steel lower alloyed in with a maximum power of 4.4 kW, a focal length of 200 mm
Nickel and with a Nickel base alloy. and a beam parameter product of 23.4 mm⋅mrad was used.
The laser beam was transmitted to the work station by an
Experimental optical fiber with a diameter of 600 µm. The laser focusing
optics together with the plasma powder torch were mounted
Materials on the arm of a six-axis robot.
The alloying composition of the Nickel base alloy 2.4463 In all experiments, the plasma powder process was coupled
(Nicrofer 5520Co) and the austenitic stainless steel 1.4828 into a common process zone behind the leading laser keyhole
(comparable to AISI 309) are listed in Table 1 and Table 2, in a trailing position, because previous studies showed that the
respectively. hybrid process can be controlled in such configuration in a
wider range of parameters. For the investigated CRAs it
Table 1: Chemical composition of the Ni-base alloy 2.4463 turned out that an additional protective gas trailer preserved
(Nicrofer 5520Co), wt.% the hot weld seam from oxidization.
C Cr Ni Mo Ti Al The plasma powder process was carried out in a non-keyhole
0.050 22.10 54.90 8.70 0.390 1.170 mode, due to high welding speeds and undesirable interaction
Si Mn P S Co Fe with the laser keyhole. Fig. 1 shows the construction principle
0.010 0.080 0.003 0.002 11.50 0.910 of the plasma powder torch. Welding was started by a pilot arc
between electrode and plasma nozzle which was ignited by a
Table 2: Chemical composition of the austenitic stainless steel high-frequency.
1.4828 (AISI 309), wt.%, bal. Fe
C Cr Ni Mo Ti Nb
0.043 19.2 10.88 0.170 0.019 -
Si Mn P S N Cu
1.957 1.058 0.022 0.003 0.053 0.099

The Nickel base alloy was welded with a powder filler metal
of the same composition (Table 3).

Table 3: Chemical composition of the nickel-based filler, wt.%

(nominal values)
C Cr Ni Mo Ti Al
0.05-0.1 20-23 bal. 8.0-10.0 0.20- 0.60-
Figure 1: Schematic illustration of the plasma powder hybrid
0.60 1.50
torch design (Type I)
Si Mn P S Co Fe
up to up to 0.012 0.008 10.0- k.A.
The powder filler metal was added into the weld pool from
0.70 0.70 13.0
behind the plasma arc.
Three different torch types were designed with various powder
The steel 1.4828 was welded with a powder of an alloying feeding nozzles
content similar to the steel AISI 316L. The particle size of the
vacuum evaporated powder types was approximately between Type I:
40 and 80 µm. The powder feeding is arranged via a nozzle, which forms a
concentric gap around the plasma nozzle. Focusing of the
The plate thickness of the 2.4663 plates was 5 mm and the powder into the weld pool is achieved by an additional
1.4828 steel plates had 6 mm thickness. The edge preparation focusing gas stream (Fig. 1 and Fig. 2a).
type was carried out as a square edge butt joint and before
welding the plates have been tacked.

Type II: was installed in front of the torch providing a shielding gas
A groove is milled into the outer part of the plasma nozzle. stream into which the laser could enter.
The powder nozzle encloses the plasma nozzle very tight and
thus, a passage for the powder is formed by assembling the Geometric Parameters
plasma into the powder nozzle (Fig. 2b). As other laser arc hybrid welding processes, LPPAW provides
numerous geometric parameters to arrange the plasma powder
Type III: torch to the laser (Fig. 3). The most relevant to this study are:
Plasma and powder nozzle are made of one piece. A powder ∆z , the focal point position of the laser; ∆xB and ∆zB, the
passage is drilled into the massive plasma torch adjacent to the electrode position in relation to the origin of the coordinates
hole for the plasma arc (Fig. 2c). which is represented by the point of laser beam impingement
on the metal plate in this case.

Figure 3: Geometric parameters of the laser plasma process

For most of the welding experiments with the austenitic

stainless steels the focus was set into the plate in a way, that
∆z amounts to one third of the thickness. However, the
experiments showed that positioning the laser focus even a
little lower in the plate (∆z = -3 mm) has positive effects on
the stability of the process and thus, on the weld quality. The
distance between the electrode tip and the laser beam axis ∆xB
was selected as small as possible, i. e. around 7.5 mm which
Figure 2: Different types of plasma powder torches was also the case for the distance between the electrode tip
and the plate surface surface ∆zB.
The most continuous powder flow was achieved by Similar to previous experiments [2], an angle βB of 34°
application of the focusing gas type, i. e. torch Type I. But, as between the laser beam and the axis of the plasma torch was
a disadvantage of this type, the focusing gas represents an also found to be suitable for a stable hybrid process.
additional process parameter which has to be carefully
adjusted for the welding process. In contrast, a slightly pulsed Welding Experiments
powder stream is observed at the plasma powder torches Type In order to achieve the final vertical-up and -down welding
II and Type III which was caused by friction of the powder in positions PF and PG, respectively, the welds were carried out
the powder passage. by gradually increasing the angle from the flat position PA, 0°,
The highest powder deposition rate was achieved by the to 30°, 60° and finally, to 90° for the vertical positions. As a
massive plasma powder torch, Type III (Fig. 2c). At a slightly first approach, the welding parameters of the experiments in
decreased powder feeding rate of 21 g/min as compared to the horizontal position revealing acceptable weld qualities
previous experiments [1], depositions up to 90% were be could be transferred also to out-of-position welding with no
measured for the AISI 316L powder used to weld the steel changes. Two different welding speeds, i. e. 1 m/min and
1.4828. As shown in Fig. 1, Fig. 2a and Fig. 2b, an additional 2 m/min with adjusted laser and arc power, have been
shielding gas outlet for protection of the process zone was selected. Fig. 4 shows the complete test set up.
integrated into the torches Type I and II. Due to limited Out-of position welding with the Nickel base alloy was
assembly space, such an outlet could not be integrated into predominantly carried out with the torch Type II, but also with
the torch Type III. For this reason an additional gas nozzle

the torch Type III, in particular in the 90° positions PG and
PF. Due to the positive reducing effect and a better burn-in, a
mixture of Argon with 2 vol.-% Hydrogen was used as
shielding as well as plasma gas for this material.
With the austenitic steel AISI 309, welding experiments were
carried out using the Type III plasma torch and only Argon
was used as shielding and plasma gas.
Weld quality was investigated by visual inspection and by
respective weld cross sections. High speed video records were
compared to actual welding results and helped to improve the
plasma torch design.

Figure 4: Experimental setup for welding in PF (vertical-up)

position at an inclination angle of 60°
Figure 5: Cross sections of Ni-base alloy 2.4663, welded in
Results position PA (a, b), downhill welded with different inclination
angles of 30° (c,d), 60° (e,f) and in the position PG, (g,h)
During the first out-of-position welding of the Nickel base using the plasma powder torch Type II (welding parameters
alloy 2.4463 it turned out that powder feeding using torch for a linear heat input of 2.9, and 4.7 kJ/cm, respectively: βB =
Type I did not provide a satisfying deposition rate, even 34°, ∆z = -3 mm, PL = 4.4 / 3.0 kW, I =220 / 200 A, vs = 2 /
though it could be enhanced compared to the former plasma 1 m/min)
powder torch design and thus, more concentrated powder
feeding as provided by the other two torch designs appeared as
more convenient to achieve acceptable weld qualities. Fig. 5
shows the cross sections for vertical-down welding of the
Nickel base alloy using the torch Type II. Acceptable results
were achieved for the lower heat input of 2.9 kJ/cm. At an
increased heat input of 4.7 kJ/cm, an increasing undercut is
exhibited at the weld top side, showing also a very flat weld
bead stretched to both sides. One-sided undercut was exhibited
during vertical up welding of this material, as shown in Fig. 6.
Already at the low heat input a remarkable undercut notch can
be observed near the fusion line at the top side at an
inclination angle of 60°. Welding with high heat input in this Figure 6: Weld cross sections of the Ni-base alloy 2.4663,
position increases this undercut towards non-acceptable uphill welded at an inclination angle of 30° using the plasma
dimensions. Gravitational effects have to be addressed as the powder torch Type II (welding parameters see Figure 5)

reason for such undercut. During vertical-up welding, liquid material can be welded with such conditions in the vertical of
weld metal piled up in the rear weld pool and led to a lack of vS = 1.5 m/min resulting in a total heat input of 3.9 kJ/cm. As
material in front of the process zone. Thus, minor shown by the respective cross sections in Fig. 8, this down as
misalignments between the laser and the plasma process as well as in the vertical up position, PG and PF, respectively.
well as irregularities of powder feeding caused one-sided For these experiments, also the plasma powder torch Type III
undercut which became worse with increasing inclination was used avoiding undercut as compared to the torch Type II,
angle. By rearrangement of the powder feeding through the but exhibiting a similar reinforcement on the top side of the
hole underneath the plasma nozzle, as designed for the torch welds during uphill welding as already observed with the
Type III (Fig. 2c), powder feeding was acting against the Nickel base alloy.
gravity caused melt flow out of the pool. Thus, the total hybrid
welding process was more stabilized. As shown by the cross
sections in Fig. 7, even during uphill welding at the high heat
input in the vertical position, no undercut was observed. In
contrast, nearly all welds exhibited a reinforcement at the top
side which can be controlled more easier than weld metal

Figure 8: Weld cross sections of the stainless steel type AISI

309 welded in PG- (a) and PF-Position (b) using the plasma
powder torch Type III (βB = 34°, ∆z = -3 mm, PL = 4.4 kW, I
=220 A, vs = 1.5 m/min)

These results generally show that LPPAW uphill welding with

a trailing plasma powder torch appears as more complicated
than downhill welding. By respective high speed video
records, the gravitationally piled-up liquid in the rear of the
weld pool was identified as the predominant reason for the
poor weld shape. During downhill welding, the liquid metal
does not pile up due to the proceeding laser keyhole. More
plane welds on the top side can certainly be achieved by
stabilization of the liquid pool by a more precise adjustment of
the plasma gas stream and the powder feeding rate acting
against gravity on the liquid weld pool. Such parameter
variations for further weld pool stabilization are currently
continued and, at the present state, the torch Type III turned
out as the most convenient design to achieve this.

As with other Nickel base alloys [5], liquation cracking has

been detected in nearly all welds of the Ni base alloy 2.4463
Figure 7: Weld cross sections of the Ni-base alloy 2.4663, which was found to be independent of the welding position.
uphill welded with different slope angles using the plasma As shown by Fig. 9, the number of micro cracks at the fusion
powder torch Type III (welding parameters for a linear heat line in the weld cross sections is reduced with increasing heat
input. Such liquation cracking might be prevented by
input of 2.9, and 4.7 kJ/cm, respectively: βB = 34°, ∆z = -
adjustment of the filler material and/or shielding gas
3 mm, PL = 4.4 / 3.0 kW, I =220 / 200 A, vs = 2 / 1 m/min)
composition. In this context, it is an important requirement for
laser arc hybrid welding that the filler material can be
The steel 1.4828 (AISI 309) was welded with similar
transported deeply towards the weld root not only for hot
parameters as the Nickel base alloy at the same welding speed
cracking avoidance, but also to improve corrosion resistance.

A comparison of the cross section of the sole laser weld in the
position PG in Fig. 10 to that of the hybrid weld shown in Fig.
6g shows that a much wider root is established during LPPAW
and thus, the hybrid welding process actually affects the weld
root formation. As a control that the alloying elements of the
filler material are sufficiently transported into the weld depth
and thus, could be used for hot cracking avoidance, in Fig. 11
the EDX analyzed Molybdenum content in a LPPAW weld of
the steel 1.4828 is plotted versus the distance from the weld
top side. By the average line it can be seen that the
Molybdenum enriched powder of the 316 L type (Table 4) is
transported at least to a depth of half the thickness of the lower
Molybdenum alloyed base material (Table 3).


From the present state of out-of-position laser arc hybrid

Figure 9: Number of microcracks at the fusion line of a weld welding the following conclusions can be drawn:
made of 2.4663 1. LPPAW of various CRAs is possible with a trailing plasma
torch in the positions PF and PG at a total heat input of 2.9 and
4.7 kJ/cm, respectively.
2. Since both positions could be welded with the same
parameter set as for the horizontal position, the process
generally appears as quite tolerant to out-of-position welding.
3. The plasma torch should be designed with a powder outlet
that stabilizes the liquid weld pool against gravitational effects
in the position PF.
4. Liquation cracking in the Nickel base alloy welds is
independent of the welding position, but dependent on heat
input and might be avoided by respective adjustment of the
filler powder composition.
Figure. 10: Cross section of a laser weld with 2.4663 (βL = 0°, 5. The current results represent a significant step towards an
PL = 3.8 kW, vS = 1.5 m/min, ∆z = -3 mm) enormous widening of the industrial application limits for high
energy laser hybrid welding processes with the perspective to