Brigham Young University

BYU ScholarsArchive

Theses and Dissertations

2006-03-15

Two Dimensional Friction Stir Welding Model with Experimental

Validation
Charles Blake Owen
Brigham Young University - Provo

Follow this and additional works at: https://scholarsarchive.byu.edu/etd

Part of the Mechanical Engineering Commons

BYU ScholarsArchive Citation

Owen, Charles Blake, "Two Dimensional Friction Stir Welding Model with Experimental Validation" (2006).
Theses and Dissertations. 365.
https://scholarsarchive.byu.edu/etd/365

This Thesis is brought to you for free and open access by BYU ScholarsArchive. It has been accepted for inclusion
in Theses and Dissertations by an authorized administrator of BYU ScholarsArchive. For more information, please
contact [email protected], [email protected].
TWO DIMENSIONAL FRICTION STIR WELDING MODEL

WITH EXPERIMENTAL VALIDATION

by

Charles Blake Owen

A thesis submitted to the faculty of

Brigham Young University

in partial fulfillment of the requirements for the degree of

Master of Science

Department of Mechanical Engineering

Brigham Young University

April 2006
Copyright
c 2006 C. Blake Owen

All Rights Reserved

 BRIGHAM YOUNG UNIVERSITY

GRADUATE COMMITTEE APPROVAL

of a thesis submitted by

Charles Blake Owen

This thesis has been read by each member of the following graduate committee and
by majority vote has been found to be satisfactory.

Date Carl D. Sorensen, Chair

Date Tracy W. Nelson

Date Michael Miles

 BRIGHAM YOUNG UNIVERSITY

As chair of the candidate’s graduate committee, I have read the thesis of Charles
Blake Owen in its final form and have found that (1) its format, citations, and bibli-
ographical style are consistent and acceptable and fulfill university and department
style requirements; (2) its illustrative materials including figures, tables, and charts
are in place; and (3) the final manuscript is satisfactory to the graduate committee
and is ready for submission to the university library.

Date Carl D. Sorensen

Chair, Graduate Committee

Accepted for the Department

Matthew R. Jones
Graduate Coordinator

Accepted for the College

Alan R. Parkinson
Dean, Ira A. Fulton College of Engineering
and Technology
 ABSTRACT

TWO DIMENSIONAL FRICTION STIR WELDING MODEL

WITH EXPERIMENTAL VALIDATION

Charles Blake Owen

Department of Mechanical Engineering

Master of Science

The performance of a coupled viscoplastic model of FSW has been evaluated over

a variety of tool RPMs and feed rates. Initial results suggested that further opti-

mization of the material parameters and an additional ability to model the thermal

recovery of the material would improve the overall performance of the model. There-

fore, an experimental/numeric approach was taken to improve and quantitatively

compare the performance of the model based upon the thermal profile of the work-

piece. First, an experimental method for obtaining real-time temperature measure-

ments during Friction Stir Processing (FSP) of 304L Stainless Steel was developed.

The focus of the method was to ensure that the obtained temperatures were both

accurate and repeatable. The method was then used to obtain thermal cycle data

from nine welds, each at different operating conditions ranging in tool rotational
speed from 300 to 500 RPMs and in feed rate from 0.85 to 2.54 mm/s (2 - 6 in/min).

Then a family of nine numerical models was created, each model corresponding to

one welding condition. The performance due to improved convergence stability and

the added thermal recovery term are also discussed. A gradient following technique

was used to optimization and iteratively adjust nine material parameters to min-

imize the difference between the numerical and experimental temperature for the

whole family of models. The optimization decreased the squared error between the

numerical and measured temperatures by 76%. Recommendations are also made

that may allow the optimization method to return greater dividends.

 ACKNOWLEDGMENTS

I would like to thank all of those who have been influential while writing this

thesis. First, my wife Janica for always believing and encouraging me; second,

Carl Sorensen for countless hours of guidance and direction, as well as allowing me

to learn and gain the confidence that I can accomplish great things; third, Tracy

Nelson for making this work possible.

Financial support for this work was provided by the Office of Naval Research

contract No. N00014-03-1-0792.

Table of Contents
1 Introduction 1
1.1 Friction Stir Welding . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 About this Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Application of a Coupled Viscoplastic Material Model to Friction

Stir Welding with a Variety of Tool RPMs and Feed Rates 5
2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Method for Effective Plate Temperature Measurements during Fric-

tion Stir Processing 25
3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4 Thermocouple Theory . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Thermocouple Time Constants . . . . . . . . . . . . . . . . . . . . 28
3.5.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.6 Measurement of FSW with Multiple Thermocouples . . . . . . . . . 33
3.6.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6.2 Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Friction Stir Welding Model Optimized and

Compared with Experimental Data 51
4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.4 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
viii
 4.4.1 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.2 Numerical Model Improvements . . . . . . . . . . . . . . . . 55
4.4.3 Parametric Optimization . . . . . . . . . . . . . . . . . . . . 60
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.6.1 Temperature Studies . . . . . . . . . . . . . . . . . . . . . . 71
4.6.2 Model Improvement . . . . . . . . . . . . . . . . . . . . . . 72
4.6.3 Parametric Optimization . . . . . . . . . . . . . . . . . . . . 73
4.6.4 Numerical/Experimental Comparisons . . . . . . . . . . . . 74
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

5 Recommendations For Future Work 79

References 81

Appendix A: Mesh Refinement Study 87

Appendix B: State Variable Details 89

Appendix C: FSW Tool Drawing E44016 91

Appendix D: Hickory Code Changes 93

Appendix E: Optimization 99

ix
List of Tables
1 Boundary conditions used in the FSW model. . . . . . . . . . . . . 11
2 Run numbers (italics) listed by parameters used. . . . . . . . . . . . 12
3 The thermocouple hole locations. . . . . . . . . . . . . . . . . . . . 34
4 The maximum measured and estimated temperatures along with the
differences in the repeated measurements. . . . . . . . . . . . . . . . 39
5 Boundary conditions used in the FSW model. . . . . . . . . . . . . 56

x
List of Figures
1.1 Friction Stir Welding schematic. . . . . . . . . . . . . . . . . . . . . 1
2.1 Seidel and Reynolds; streamline plot, along with Egbert’s plan view
that shows the material flow path. . . . . . . . . . . . . . . . . . . . 7
2.2 MegaStir’s parameter study with each respective transverse section.
The mid-plane deformation width of each weld is indicated by a white
line. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 The 2-D Friction Stir Welding mesh and boundaries. . . . . . . . . 10
2.4 The strain rate used to calculate the weld deformation. . . . . . . . 13
2.5 The temperature profile plots from the parametric study. Each plot
is labeled with the run number followed by the maximum and average
temperatures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 The average temperature on a circle 2 mm away from the pin surface
as a function of feed rate and rotational speed. . . . . . . . . . . . . 15
2.7 The temperatures above and below the pin on the advancing (solid)
and retreating side (dotted) of the weld. . . . . . . . . . . . . . . . 15
2.8 Streamline results viewed on a 15 mm (.59in) square. The calculated
boundaries of the deformation zone are shown by the dotted lines
and indicated in millimeters at the top of each plot. . . . . . . . . . 16
2.9 The deformation zone width as a function of feed rate and rotational
speed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.10 The state variable plots from the parametric study. The maximum
and average state variable is indicated above each plot. . . . . . . . 17
2.11 The average state variable value on a circle 2 mm away from the pin
surface is plotted as a function of feed rate and rotational speed. . . 17
2.12 Temperature distribution in front of the pin at 900 RPM. . . . . . . 18
2.13 The calculated (solid lines) and experimental (dotted lines) deforma-
tion zones. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.14 (a) The state variable is plotted along the Outlet boundary, (b) along
with the cross-sectional hardness map of the midplane of two 304L
welds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 The transient non-dimensional temperature result for multiple step
input experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 The measured and estimated instantaneous temperatures for the
0.032 inch grounded thermocouple. . . . . . . . . . . . . . . . . . . 31
3.3 The thermocouple locations represented graphically. . . . . . . . . . 35
3.4 The setup to position thermocouples. . . . . . . . . . . . . . . . . . 35

xi
3.5 The bottom of the backing plate after all the thermocouples have
been placed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.6 Image used to measure weld offset. . . . . . . . . . . . . . . . . . . 37
3.7 The weld’s surface finish. . . . . . . . . . . . . . . . . . . . . . . . . 38
3.8 The real-time temperature data that was collected. . . . . . . . . . 38
3.9 The unadjusted maximum estimated and measured temperatures for
the weld. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.10 The maximum estimated and measured temperatures for the weld. . 40
3.11 The micrograph of the weld cross section. . . . . . . . . . . . . . . . 41
3.12 The measured thermocouple temperature (solid) and the estimated
plate temperature (dotted) with respect to the relative distance from
the pin for the advancing and retreating sides of the weld. . . . . . 42
3.13 The estimated temperature data with respect to the pin’s locations. 42
3.14 The estimated temperature data with bilinear interpolation to give
a full experimental temperature profile. . . . . . . . . . . . . . . . . 43
3.15 The difference in measured temperatures as a function distance from
the weld center. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.1 The feed rate and tool RPM values used in the experiment. The
numbers depict the order in which the experiments were performed. 54
4.2 The 2-D Friction Stir Welding mesh and boundaries. . . . . . . . . 55
4.3 The obtained annealing curves for 1100-H18 sheet. . . . . . . . . . . 58
4.4 The yield strength of aluminum plotted as a function of time (solid)
along with a modeled fit of the data (dotted. . . . . . . . . . . . . . 58
4.5 The yield strength of aluminum plotted as a function of time (solid)
along with a modeled fit of the data (dotted. . . . . . . . . . . . . . 59
4.6 Maximum measured temperatures for the advancing (X, solid) and
retreating side (+, dotted) side of the weld along with the estimated
temperatures (o). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.7 Transient measured (solid) and instantaneous (dotted) temperatures. 63
4.8 Weld micrographs at each operating condition along with a horizontal
line marking the weld deformation at the height of the thermocouples. 64
4.9 The maximum numerical temperatures at realistic RPM values (solid)
and the reduced RPM values (dotted). The advancing (X) and the
retreating (+) temperatures are also indicated. . . . . . . . . . . . . 65
4.10 Stress strain curves for 304L at different temperatures with a strain
rate of 0.01. The numerical equivalent stress strain curves are dotted. 65
4.11 Stress strain curves for 304L at different strain rates with a temper-
ature of 1073 K. The numerical equivalent stress strain curves are
dotted. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.12 Stress strain curves for 304L at different strain rates with a temper-
ature of 1473 K. The numerical equivalent stress strain curves are
dotted. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

xii
4.13 Streamline results viewed on a 15 mm (.59in) square. The calculated
boundaries of the deformation zone are shown by the dotted lines
and indicated in millimeters at the top of each plot. . . . . . . . . . 67
4.14 The calculated (solid lines) and experimental (dotted lines) deforma-
tion zones. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.15 Optimized temperature predictions (solid) with the original numeri-
cal predictions (dotted). . . . . . . . . . . . . . . . . . . . . . . . . 68
4.16 Original measured temperature (solid) with the optimized tempera-
ture predictions (dotted). The percent error between the measured
and calculated maximum temperatures are listed at the top of each
plot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.17 Experimental measured temperatures (Solid) and instantaneous pre-
dicted temperatures (Dotted). The percent error between the mea-
sured and calculated maximum temperatures are listed at the top of
each plot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
A.1 Each mesh used in the refinement study is shown. The number of
elements is indicated below each plot. . . . . . . . . . . . . . . . . . 87
A.2 The maximum obtained strain rate for each grid in the mesh study. 88
A.3 The percent change of the average value 2 mm way from the pin. . 88
C.1 The FSW tool drawing number E44016. . . . . . . . . . . . . . . . 91

xiii
1 Introduction

1.1 Friction Stir Welding

Friction Stir Welding (FSW) is a solid state joining process that can produce supe-

rior material properties to those produced by conventional welding methods. The

advantages of this process include the following: no filler material, no fumes, no

solidification cracks, low porosity, low weld distortion, the ability to weld dissimilar

metals, and greater strength and corrosion properties [1, 2].

Shoulder
Pin

Figure 1.1: Friction Stir Welding schematic [3].

The FSW process consists of a solid rotating tool (pin and shoulder as seen in

Figure 1.1) that is first plunged into the seam of two rigidly mounted plates until

the shoulder contacts the workpiece. When sufficient heat build up is sustained,

the tool traverses along this seam creating a weld. As the tool rotates under severe

loading conditions, frictional and high shear deformation heat the workpiece to

near 80% of the material’s melting temperature [4]. Furthermore, the material is

plastically deformed around the pin at very high strain rates and forged together
1
by the downward force of the shoulder. This environment creates a challenge for

modelers due to the coupled thermo-mechanical nature, the large deformation and

strain rates near the pin, and the strain hardening and thermal softening that takes

place during the weld.

1.2 Objective

The objective of the research is to quantitatively evaluate and improve the perfor-

mance of a 2-D coupled viscoplastic model of FSW. The evaluation is meant to

reveal how the model performs over a variety of tool rotational speed and feed rates

and not at one specific operating condition. This will be done by obtaining thermal

cycle data during Friction Stir Processing (FSP) of 304L Stainless Steel. The tem-

perature data from nine experimental welds and nine numerical models will then

be quantitatively compared to judge the performance of the model.

1.3 About this Thesis

This thesis consists of a collection of three papers that have been submitted for

publication. The first paper is a preliminary evaluation of the 2-D FSW model.

The second paper documents an effective method for obtaining real-time tempera-

ture measurements in Friction Stir Welding. The third paper presents thermal cycle

data in nine different welds and presents a method to quantitatively optimize the

performance of the numerical models. Finally, the thesis concludes by recommend-

ing future work that should be done. The thesis also includes two appendices that

are thought to be beneficial for future work but are not referenced in any of the

papers. The first, Hickory Code Changes in Appendix D, includes a list of most

of the changes that were made to the code in order to improve the understand-

2
ing, reliability and convergence. The second, Optimization Routine Appendix in E,

includes the script necessary to optimize the FSW model results.

3
4
2 Application of a Coupled Viscoplastic Material
Model to Friction Stir Welding with a Variety
of Tool RPMs and Feed Rates

2.1 Abstract

A 2-D Finite Element code named Hickory and a simplified-Hart’s material model

are investigated as a possible means of modeling Friction Stir Welding. To evaluate

the model, a parametric study of FSW in 304L stainless steel is performed by varying

the tool’s feed rate and rotational speed. While the temperature trends agree with

experimental work, the material streamlines predict weld deformation trends that

contradict experimental results in both the tool’s feed rate and RPM. The state

variable trends agreed with hardness maps, however, it is it is recommended that a

thermal recovery term be added to improve the model’s state variable evolution.

2.2 Introduction

Friction Stir Welding (FSW) has been modeled by many since its discovery in 1991.

Modeling FSW is an important tool used to better understand the complex material

interactions that take place in a weld. Current models range from simple analyti-

cal formulas to three-dimensional models that account for intricate tool geometries.

Eventually, it is hoped that these models will help identify optimal operating pa-

rameters and tool geometries. However, many of the published FSW models have

only investigated one parameter set. Thus, the objective of this work is to evaluate

a 2-dimensional FSW model over a wide range of tool feed rates and tool rota-
5
tional speeds. The paper begins by reviewing previous FSW modeling work and

gives some background information about the Hickory code used in this study. The

method for creating and running a family of sixteen numerical models is presented

along with the process for determining the predicted weld deformation width. The

results for all sixteen runs have been plotted together to easily compare the temper-

ature, streamline and state variable data. The numerical trends are discussed and

compared to weld midplane temperatures, weld cross-sections and hardness maps.

Based upon the observations, recommendations are made for future work.

2.3 Previous Work

When modeling FSW the material models used must be able to handle very large

deformations, high strain rates and include coupled relationships to temperature.

Given these challenges, many modelers have started with 2-D models before moving

to 3-D.

Seidel and Reynolds’ 2-D model [5] displayed the flow field by using material

streamlines. This showed that the weld material is displaced around the pin on the

retreating sided as shown in Figure 2.1(a). However, it was noted that an extremely

high or low tool RPM would cause material to pass the pin on both sides. Xu and

Deng’s model [6] tracked a particle’s initial and final resting positions. Colegrove

and Shercliff [7] investigated how different tool geometries effect the applied loads

and flow paths with both 2 and 3-D models. Ulysse [8] modeled a cylindrical pin

in 3-D and used the flow field predictions to track a given particle’s location and

temperature history. Flow fields have been used by many to track the material’s

history, however, none have used this as a means of predicting the weld deformation

size. Bendzsak et al. [9], and Colegrove [10] have used 3-D models to investigate

6
the effects of threaded Friction Stir Welding pins. While 3-D models are very

informative, they can be very time intensive to run. The long run times may

contribute to the fact that most of the published models have not investigated how

the model performs over of a wide variety of input parameters.

Although much is learned through numerical models, experimental work is also

essential. Seidel and Reynolds investigated the out-of-plane motion by flow visual-

ization experiments using inserted markers with threaded FSW pins in aluminum

[11]. These experiments found that welds with a high advance per revolution (weld

pitch) had little out-of-plane flow. It is thought that the threads on the pin are the

main cause of the out-of-plane material flow. Egbert [12] investigated the material

flow patterns by welding on a confined stack of thin metal strips. Once the specimen

was polished and etched, the residual material interfaces showed the basic flow path

of the material as seen in Figure 2.1(b).

(a) (b)

Figure 2.1: Seidel and Reynolds; streamline plot [5], along with Egbert’s plan view
[12] that shows the material flow path.

MegaStir [13] performed a parameter investigation in A36 mild steel over a wide

range of feed rate and tool RPM. The cross section of each weld along with the

7
respective parameters can be seen in Figure 2.2. It was found that the width of the

deformation area at the material midplane (shown by the white line) decreases as

the travel speed increases. Also, the deformation increases as the rotational speed

increases.

Figure 2.2: MegaStir’s parameter study [13] with each respective transverse section.
The mid-plane deformation width of each weld is indicated by a white line.

Using thermocouple measurements embedded in 304L Stainless Steel, Owen’s

unpublished work [14] found the advancing side of the pin to be 30-80 ◦ C hotter

than the retreating side. Also, Record and Covington’s statistical studies [15] have

shown that the tool’s temperature increased when the RPM increased and when

the feed rate decreased.

Hickory, the 2-D finite element code used in this study was developed by Paul

Dawson at Cornell University. The code development began in the late 1970’s and

has continued with new applications and additional material models since then. It

has been used in both Eulerian and Lagrangian domains to model large deforma-
8
tion processes such as metal rolling [16] and upset welding [17, 18]. In a recent

publication preliminary results of FSW were displayed [19]. This FSW model had

a pin of 0.2 m diameter rotating at 1 RPM. Much of the focus was to incorporate

anisotropic texture evolution into the solution. The Hickory code includes a quasi-

rigid nonlinear viscoplastic material model known as the simplified-Hart’s model,

which is related to Hart’s original work [20]. It includes material evolution equations

with coupled relationships between temperature and strain rate. An internal state

variable (which represents the material’s dislocation density) is integrated along the

material streamlines to account for the work hardening and dynamic recovery that

take place.

2.4 Method

One 2-D model of FSW is developed and applied to a family of sixteen models

with each at a different tool rotational speed and feed rate. The model generates

temperature, streamline and state variable predictions. These numerical predictions

of all sixteen runs are used to investigate trends that are a result of the changing

operating conditions. Furthermore, the numerical trends are then compared with

available experimental trends.

The basic FSW model created with the Hickory finite element program is a 2-D

cross-section in-plane with the weld material. The pin can be viewed as a stationary

rotating cylinder with the workpiece material flowing past it. All material flow is

restricted to in-plane motion. The area modeled is a 25 mm (2 in) square with a pin

diameter of 8.45 mm (0.333 in). This 2-D domain is modeled using an Eulerian mesh

with 6-node triangular elements that are refined near the pin in order to capture

9
the strong gradients that are generated there. The material traverses through the

mesh as shown by the arrows in Figure 4.2.

Material
Travel
Retreating Boundary Direction

Pin
Rotation
Direction

Inlet Outlet
Boundary Boundary

x Advancing Boundary

Figure 2.3: The 2-D Friction Stir Welding mesh and boundaries.

The boundary conditions applied to the domain are outlined in Table 5, where

the velocity, temperature and diameter are represented by u, θ and D respectively.

The material entering the inlet has been modeled at room temperature and given

an initial state variable value that represents the average dislocation density or

hardness of the non-deformed material. The state variable at all other locations

evolves according to its respective flow history. Notice that the same conditions

were applied to the outlet, advancing and retreating boundaries. The pin’s fric-

tional condition applies a tangent traction Ttangent proportional to the difference

between the tool’s prescribed tangential surface velocity vtool and the numerically

calculated tangential material velocity utangent . Notice that this equation in table

5 divides vtool by k. The value of k was experimentally chosen to be 5 in order to

obtain reasonable temperature values throughout the set of runs. The value used

for β is 1.5 × 1011 kg/m2 ·s which was recommended by Cornell University. The

pin’s thermal convection coefficient h is approximated as 1000 W/m2 ·K and θtool

10
 Figure 4 : The 2 -D Frictio n Stir W elding mesh a nd bo unda ries.

Table
Ta ble 1:
1: BBoundary
o unda ry coconditions
nditio ns usedused in FSW
in the the FSW
mo del.model.
Boundary Me chanical T he rm al State V ariable

Inlet ux = F eed R ate m / s θ = 2 5◦ C 1.5 × 10 8

uy = 0 m / s

O utlet ux = F eed R ate m / s

Adv a ncing uy = 0 m / s Adia ba tic Ev o lv es

R etrea ting

Frictio n Co nditio n

Tta n g en t = β(vto o l / k − uta n g en t ) Co nv ectio n

Pin vto o l = R P M · π · D m / s q = h(θma ter ia l − θto o l ) Ev o lv es

D = 8.5 m m (.3 3 3 in)

ur a d ia l = 0 m / s

as 1000 ◦ C. It is important to note that the heat input from the tool’s shoulder
8
has not been accounted for in this model. Other temperature independent parame-

ters used to model 304L stainless steel are density, thermal conductivity, and heat

capacity with respective values of 7900 kg/m3 , 22.6 W/m·K, and 557 J/kg·K. The

specific parameters used by the simplified-Hart’s material model and the method

for determining these values can found in references [17, 19].

A mesh refinement study was performed in order to ensure the mesh used was

refined enough to obtain mesh independent solutions (See Appendix A for details).

Based upon this study the gird chosen had 1238 6-node triangle elements.

The parameter window investigated by this family of models was chosen to be

wider than the actual values that produce defect-free welds. This was done to test

the numerical code and leave the possibility of being able to identify good operating

parameters by the resulting trends observed from the model’s outputs. The chosen

feed rate ranges from 0.42-3.81 mm/s (1-9 in/min) and the RPM ranges from 100-

11
900. Sixteen different cases were run with parameter settings as shown in Table 2.

Table 2: Run numbers (italics) listed by parameters used.

Feed Rate (mm/s)
0.42 1.27 2.54 3.81
900 41 42 43 44
Rotational 600 31 32 33 34
Speed (RPM) 300 21 22 23 34
100 11 12 13 14

Matlab was used to postprocess the data for all sixteen runs. In addition to

creating plots of streamlines, state variables, and temperatures, Matlab calculated

the deformed zone of each run. This was done by inspecting the strain rate along a

line perpendicular to the advancing boundary through the center of the pin. Near

the pin, the strain rate increases as the material deforms around the pin (See Figure

2.4). The location where the strain rate reaches a critical value (0.5 s−1 ) is marked

as the edge of the deformation zone. This distance from the pin center to the edge of

the deformed zone is displayed along with the streamline results for easy comparison

of the deformation width.

The model’s outputs were compared to several different experimental results.

The temperature magnitudes were compared to Owen’s thermocouple work [14].

The temperature trends were compared to Record and Covington’s statistical stud-

ies [15]. The deformation results were compared to the mid plane deformation width

seen in MegaStir’s parameter study [13]. Finally, the state variable values along the

outlet boundary were compared to the cross-sectional hardness of the midplane of

two 304L friction stir welds.

12
 10

9 Strain Rate
Critical Value
8

7
Deformation

Strain Rate
6
Zone
5

0
−25 −20 −15 −10 −5 0 5 10 15 20 25
Y Position mm

Figure 2.4: The strain rate used to calculate the weld deformation.

2.5 Results

The FSW model predicted the material’s temperature, velocity, and state variable

at each of the sixteen operating conditions. Figure 2.5 shows the temperature

profiles for all the runs. Note that the rows represent runs performed at a con-

stant tool rotational speed, and the columns represent runs at a constant feed rate.

Furthermore, the run number, maximum temperature for each run, and average

temperature on a circle 2 mm away from the pin is listed at the top of each plot.

The average value 2 mm away from the pin was found valuable for comparing the

solutions in the mesh study of Appendix A. Therefore, this same method has been

used to quantitatively compare the temperature results in Figure 2.6. Furthermore,

the temperature along a line perpendicular to the advancing boundary through the

center of the pin is shown in Figure 2.7. Note that the temperatures on the ad-

vancing side of the weld are plotted with a solid line while the retreating side are

plotted with a dotted line.

The resulting velocities have been displayed with streamline plots that predict

the path that a particle would travel during a simulated weld. The calculated
13
Figure 2.5: The temperature profile plots from the parametric study. Each plot is
labeled with the run number followed by the maximum and average temperatures.

deformation zone width (in mm) for the advancing and retreating side is displayed

at the top of each graph and visually represented by dotted lines in Figure 2.8. The

total deformation of both sides of the pin is shown in Figure 2.9.

Figure 2.10 shows each state variable profile and Figure 2.11 shows how the

average state variable values 2 mm away from the pin change.

14
 1200
900 RPM
600 RPM
1000 300 RPM
100 RPM
Temperature °C 800

600

400

200

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Feed Rate mm/s

Figure 2.6: The average temperature on a circle 2 mm away from the pin surface
as a function of feed rate and rotational speed.

Run 41 Run 44
1200 1200

1000 1000
Temperature °C

800 800
900

600 600
Rotational Speed rev/min

400 400

200 200

5 10 15 20 25 5 10 15 20 25

Run 11 Run 14
1200 1200

1000 1000
Temperature °C

800 800
300

600 600

400 400

200 200

5 10 15 20 25 5 10 15 20 25
Distance from Weld Center mm Distance from Weld Center mm
0.42 (1) 3.81 (9)

Feed Rate mm/s (in/min)

Figure 2.7: The temperatures above and below the pin on the advancing (solid) and
retreating side (dotted) of the weld.

15
 Run 41, 5.2, −4.9 Run 42, 4.9, −5.3 Run 43, 5.8, −5.3 Run 44, 6.4, −5.3

900

Run 31, 5, −5.2 Run 32, 5.6, −5.3 Run 33, 6.3, −5.3 Run 34, 7.1, −5.3
600
Rotational Speed rev/min

Run 21, 5.3, −5.8 Run 22, 6.2, −5.8 Run 23, 7.5, −5.5 Run 24, 8.3, −5.3
300

Run 11, 6.4, −6.6 Run 12, 7.8, −5.8 Run 13, 9.5, −5.5 Run 14, 10.6, −5.3
100

0.42 (1) 1.27 (3) 2.54 (6) 3.81 (9)

Feed Rate mm/s (in/min)

Figure 2.8: Streamline results viewed on a 15 mm (.59in) square. The calculated

boundaries of the deformation zone are shown by the dotted lines and indicated in
millimeters at the top of each plot.

16
Deformation Zone Width mm

900 RPM
15 600 RPM
300 RPM
14 100 RPM

13

12

11

10
0 0.5 1 1.5 2 2.5 3 3.5 4
Feed Rate mm/s

Figure 2.9: The deformation zone width as a function of feed rate and rotational
speed.

16
Figure 2.10: The state variable plots from the parametric study. The maximum
and average state variable is indicated above each plot.

8
x 10
2.4

2.2
State Varialbe

1.8

1.6 900 RPM

600 RPM
1.4 300 RPM
100 RPM
1.2
0 0.5 1 1.5 2 2.5 3 3.5 4
Feed Rate mm/s

Figure 2.11: The average state variable value on a circle 2 mm away from the pin
surface is plotted as a function of feed rate and rotational speed.

17
2.6 Discussion

The parametric study revealed many expected and unexpected trends. The model

confirmed the expected temperature trends as seen in Figure 2.5.

As the feed rate increases the following is observed:

• The width of the “wake” or “comet tail” of high temperature material follow-

ing the tool decreases.

• The size of the thermally softened region in front of the pin decreases. Figure

2.12 shows the thermal profile in front of the pin at different feed rates.

• The maximum temperature drops. This is because the heat is being conducted

away into cooler material that has not been preheated like it would have been

if the material were traveling at a slower velocity.

0.42 mm/s
1100
1.27 mm/s
1000 2.54 mm/s
3.81 mm/s
900
Temperature °C

800

700

600

500

400

300

200

100

2 4 6 8 10 12 14 16 18 20
Distance in front of Pin mm

Figure 2.12: Temperature distribution in front of the pin at 900 RPM.

Also, increasing the tool’s rotational speed corresponds to higher temperatures

due to the increased amount of frictional and viscous heating. Thus, it was expected

that the maximum temperature would be found in run (14) at the slowest feed rate

18
and the fastest rotational speed, which agrees with Record and Covington’s [15]

work.

It was also observed that the advancing side of the weld had higher peak temper-

atures than the retreating side, which agrees with the experimental work performed

by Owen [14]. However, Figure 2.7 also shows that at high feed rates the retreating

temperatures further away from the weld exceeded the temperatures on the ad-

vancing side of the weld. This result was very unexpected and would require more

experimental temperature data to confirm.

The basic streamlines created from the materials velocity predictions agree with

Seidel and Reynolds’ 2-D results [5]. Essentially the streamlines are straight ex-

cept in the vicinity of the pin itself. Once near the pin, the material is wrapped

around the retreating side of the pin and detaches at a point that is collinear to the

point of attachment. Furthermore, Figure 2.8 shows that as the rotational speed is

decreased, material begins to pass the pin on the advancing side of the pin. The

extreme case would be with no pin rotation, which resulted in equal amounts of ma-

terial flowing past each side of the pin. However, contrary to Seidel and Reynolds’

results [5], the same trend was not seen with high tool rotations.

The model’s deformation trends were compared to MegaStir’s parameter study

in A36 mild steel. It is recognized that comparisons to 304L Stainless Steel would be

the ideal comparison for this paper. Thus, 304L experiments are planned for future

work. However, in the meantime, it is thought that both steels will behave similarly

as the tool geometries are the same. The calculated deformation of each run in

Figure 2.8 was compared to the deformation at the midplane of each cross-section

of MegaStir’s parameter study (Figure 2.2). Figure 2.13 shows the deformation

width versus the feed rate at different RPM values. The differences in the slopes

19
 16

15

14
Deformation (mm)

100 RPM
300 RPM
13 600 RPM
900 RPM
400 RPM
500 RPM
12 600 RPM
700 RPM
800 RPM
11

10

9
0.3 0.8 1.3 1.8 2.3 2.8 3.3 3.8
Feed Rate (mm/s)

Figure 2.13: The calculated (solid lines) and experimental (dotted lines) deforma-
tion zones.

of the solid (numerical) and dashed (experimental) lines indicate that the model’s

results conflict with experimental observations in the following ways:

• Increasing the feed rate increased the deformation width in the model but

increased the width in the experiment.

• Increasing in the rotational speed decreased the deformation width but in-

creased the width in the experiment.

Thus, with these discrepancies, it is apparent that this model cannot currently be

used to predict the weld deformation area. Therefore, the streamlines appear to

provide little help identifying good operating parameters for FSW.

Other insights were gained from the state variable. The hardness represented

by the state variable along the outlet boundary (Figure 2.14(a)) correlates well

20
with experimental hardness maps of the midplane of the material (Figure 2.14(b)).

Notice that two parameter settings are compared. The model predicted the weld

area to increase in strength at 300 RPM and decrease in strength at 900 RPM.

Similar behavior is found in the experimental hardness maps indicating that the

model’s trends were correct. Comparing modeled temperatures with state variable

8
x 10
1.9
900 RPM, 1 in/min 225
1.8 300 RPM, 1 in/min

215
1.7
205
1.6
State Variable

195

Hardness (HV)
1.5
185
1.4

175
1.3

165
1.2
900 RPM, 1 in/min
155 300 RPM, 1 in/min
1.1 Pin

145
1
−25 −20 −15 −10 −5 0 5 10 15 20 25 -10 -8 -6 -4 -2 0 2 4 6 8 10
Distance From Weld Center (mm) Distance From Weld Center (mm)

(a) (b)

Figure 2.14: (a) The state variable is plotted along the Outlet boundary, (b) along
with the cross-sectional hardness map of the midplane of two 304L welds.

values in the stir zone along the outlet boundary reveals that runs with higher

temperatures resulted in lower state variables. This is expected because at higher

temperatures there is less work hardening and more dynamic taking place. Thus,

the state variable or hardness in the stir zone on the outlet boundary changed in

the following manner:

• Increasing the feed rate increased the hardness of the material.

• Increasing in the rotational speed decreased the hardness of the material.

Runs (13), (14) and (21) of Figure 2.10 were the only cases where the state vari-

able went below the initial state variable value. All the softening took place in

the deformation zone. The model allows no static thermal recovery outside the
21
deformation zone as shown in Appendix B. This should be modified to allow static

thermal recovery with a zero strain rate.

While the temperature results were reasonable, the streamline and state variable

results have left a few things to be investigated and explored:

• More experiments in 304L stainless steel will need to be conducted. This

data could then be used in turn to optimize the material parameters for the

Simplified Hart’s model.

• The thermal conductivity and heat capacity should be modified to be tem-

perature dependent.

• The state variable evolution model needs to be modified to allow for static

thermal recovery.

• Finally, different pin boundary assumptions may need to be investigated to

better model the pins frictional and thermal conditions.

2.7 Conclusion

A 2-D Parametric study of Friction Stir Welding has been performed by varying

the tool feed rate and rotational speed being between 0.4-3.8 mm/s (1-9 in/min)

and 100-900 RPM respectively. The model predicts temperature trends observed

by FSW researchers in actual experiments-

• Increasing rotational speed predicts higher temperature

• Increasing feed rate predicts lower temperature

• The advancing side has higher peak temperatures than the retreating side

22
 The basic velocity streamlines show that all the material is displaced around

the retreating side of the pin, which coincides with Egbert’s [12] experiments and

the work performed by Seidel and Reynolds [5]. However, the streamlines show

deformation trends that conflict with experimental results.

• Increasing rotational speed predicts less deformation

• Increasing feed rate predicts more deformation

Thus, this model can not currently be used to predict the weld deformation area or

to determine good operating parameters for FSW.

It was found that the state variable hardness decreased with increasing temper-

ature. Also, the state variable values along the Outlet boundary correlated well

with experimental hardness maps. However, it was also discovered that the model

does not currently allow static thermal recovery to take place when the strain-rate

is zero.

With the inconsistencies in the deformation trends, it is recommend that the

following be done:

• Complete more FSW experiments performed in 304L stainless steel

• Add temperature dependent properties

• Optimize the Hart’s material parameters

• Allow for thermal recovery even with a zero-strain rate

• Investigate different frictional and thermal boundary conditions for the tool

23
24
3 Method for Effective Plate Temperature Mea-
surements during Friction Stir Processing

3.1 Abstract

Accurate temperature measurements of the FSP process require accurate placement

of thermocouples, compensation for machine deflection and adjustment for thermo-

couple time constants. The time constant of a 0.032 inch grounded thermocouple

embedded in a plate was found to be 0.385 seconds. This type of thermocou-

ple was embedded at eight different distances away from the weld line to monitor

temperatures during FSP. The thermocouple time constant was used to estimate

the material’s transient temperature. During the process, FSW machine deflection

changes the location of the weld centerline relative to the thermocouples. Digital

images of the weld surface were used to measure this shift. Overall, the accuracy

and repeatability of the obtained temperature measurements are found to be ac-

ceptable for FSW. The results indicate that near the weld centerline the advancing

side of the weld has peak temperatures that are higher but farther from the weld,

the temperatures on the retreating sides exceed those on the advancing side.

3.2 Introduction

The thermal profile and heat transfer involved in Friction Stir Welding (FSW) are

important factors in understanding this welding process. Current methods used to

evaluate thermal predictions include thermocouple measurements at various loca-

tions and infrared surface images. While this data is informative, there is gener-
25
ally little information presented on how the temperatures were obtained and how

accurate the measurements really are. Thus, this work seeks to evaluate thermo-

couple performance and create a robust method for obtaining accurate and precise

workpiece temperature measurements in FSW. The paper begins by reviewing pre-

vious experimental work and some basic thermocouple theory. Two experiments

are described in this paper, a thermocouple time-constant experiment, and FSW

temperature measurements with multiple thermocouples. The time-constant sec-

tion presents the method used, along with a discussion of how the time constant

can be used to estimate the instantaneous temperature of the material during tem-

perature transients. The FSW temperature section covers the experimental setup

and necessary post-weld steps to ensure accurate interpretation of the data. The

maximum observed temperatures are shown with the instantaneous temperature

estimates. Also, the temperature results are used to create an experimental steady

state temperature profile of FSW.

3.3 Previous Work

While many have modeled the heat transfer in FSW, different experimental ap-

proaches have been used to evaluate the model’s performance. Lambrakos et al.

[21] used an inverse modeling approach with thermocouple measurements on the

back of a workpiece to create a thermal boundary condition for the model. Song

and Kovacevic [22] used two thermocouples on the top and bottom surfaces as a

means to show that the model’s predictions were in agreement with experimental

results. Similarly, McClure et al. [23] and Askari et al. [24] used temperature read-

ings from multiple thermocouples that were embedded in the weld at various depths

and distances from the weld to validate the models. Dickerson [25] focused his work

26
on the tool and used thermocouples on the tool’s shank to validate the heat flux

into the tool. Covington [26] used a combination of thermocouples and an infrared

camera of the tool’s shank as a means to optimize the tool’s heat flux and boundary

conditions. While published temperature data is valuable, generally there is little

presented about the method, accuracy or repeatability of the obtained tempera-

tures. For example, the effects of the thermocouple’s time constant have not been

investigated and the accuracy of the weld placement has largely been overlooked.

Furthermore, almost all of the published experimental temperature data has

been generated in aluminum. Favorable material weldability is thought to be the

biggest factor contributing to this. However, the ability to weld materials with

high melting temperatures like Stainless Steel has recently grown with the use of

Polycrystalline Cubic Boron Nitride (PCBN) tools [27, 28].

This publication addresses details that have been generally ignored in previous

reports of experimental work. While others have used thermocouples, the focus here

is to analyze the effect of the thermocouple time constant on the measured tem-

perature data. Also, in preliminary experimental work [14], machine deflection was

significant enough to shift the planned weld location by about 0.1 inches. Therefore,

a method to accurately account for the machine deflections is essential in obtaining

FSP temperatures. Finally, the experiment was performed in 304L Stainless Steel

which currently has very little published temperature data.

3.4 Thermocouple Theory

In order to understand how thermocouples perform in transient experiments, one

must first understand the basic characteristics and responses exhibited by ther-

mocouples themselves. Generally, thermocouples behave with a first order response

27
[29] as shown by Equation 3.1 , where θtc is the current thermocouple measurement,

θpl is the true plate temperature and τ is the time constant of the thermocouple.
dθtc 1
= (θpl − θtc ) (3.1)
dt τ
When a step input in temperature is applied to the system, the first order equa-

tion can be solved with the following initial conditions t = 0, θpl = θ∞ , θtc = θ0 .

This essentially means that the moment the step input is applied, θpl is instan-

taneously at the final resting temperature θ∞ while θtc is still at the initial room

temperature θ0 . Solving for the step input with these conditions yields Equation

3.2.
θtc(t) − θ0
θnd = = 1 − exp(−t/τ ) (3.2)
θ∞ − θ0
For sake of simplicity the left hand side of Equation 3.2 will be referred to as

the non-dimensional measured temperature θnd . Therefore, when the t = τ the

non-dimensional temperature will be equal to 0.632. Knowing this allows for an

easy method to determine the time constant of thermocouples. With any given

non-dimensional temperature curve, the time constant is easily found by noting the

time that it took for the temperature to reach a value of 0.632.

Once the time constant is known, it can be used to predict the true instantaneous

temperature of the material. Solving Equation 3.1 for θpl , and using a central-
dθtc
difference numerical approach to estimate dt
allows an estimate θ̂pl for the true

plate temperature θpl given by Equation 3.3.

θtc(i+1) − θtc(i−1)
 
θ̂pl(i) = τ + θtc(t) (3.3)
t(i+1) − t(i−1)

3.5 Thermocouple Time Constants

The thermocouple time constant can vary according to the heat transfer that takes

place at the thermocouple bead or tip. This means the conductivity of the measured
28
substance and a number of contact resistances can change the thermocouple perfor-

mance. Therefore, it is important to evaluate the time constant with a method that

is similar to the actual use of the thermocouple. As a result, this section evaluates

the performance of 0.032 inch grounded type K thermocouples when placed into a

hole in 304L Stainless Steel. The time constant of the thermocouple is then used

as a means of estimating the material’s instantaneous temperature.

3.5.1 Method

The time-constant experiment applies a step input in temperature to the ther-

mocouple. This is done by quickly inserting thermocouples into a hot plate. By

analyzing the response, the time constant describing the thermocouple performance

can be obtained. This time constant is then used with the measured temperature

to calculate a predicted temperature during the transient experiment.

Time constants were measured as follows:

• Number 65 (0.889 mm) holes were drilled 3.4 mm into a 1/4 inch 304L Stain-

less Steel plate.

• The plate was heated to 1000 ◦ C in a furnace.

• The plate was removed from the furnace.

• A thermocouple was inserted into one hole on the plate to monitor the tem-

perature of the plate.

• While recording the temperature with a computer data acquisition system, a

thermocouple was inserted into a second hole in the plate.

• The temperature was recorded until the thermocouple reached a steady state

value.
29
 θ(t) −θ0
• The non-dimensionalized temperature θ∞ −θ0
was plotted versus to time.

• The time constant was found by noting the time where the non-dimensional

temperature reached 0.632.

As a means to check the accuracy of this first order equation, the temperature

results and time constant from the 0.032 inch grounded thermocouple were used to

estimate the true instantaneous temperature using Equation 3.3. A 10 point moving

average was applied to the estimated data to reduce the noise of the solution.

3.5.2 Results

The transient non-dimensional temperature results from several repeated experi-

ments are shown in Figure 3.1. The black horizontal line indicates a non-dimensional

temperature of 0.632. Therefore, the time constant is found by noting the time that

each curve intersects with the black horizontal line and is indicated in seconds on

the right side of the legend.

1
Non−Dimensional Temperature

0.8

0.6

0.4

Test 1, 0.348 sec

0.2 Test 2, 0.385 sec
Test 3, 0.398 sec
Test 4, 0.417 sec
Test 5, 0.458 sec
0
0 0.2 0.4 0.6 0.8 1 1.2
Time, sec

Figure 3.1: The transient non-dimensional temperature result for multiple step
input experiments.

30
 Figure 3.2 shows the estimated and filtered instantaneous temperature using the

measured thermocouple data and time constant.

Non−Dimensional Temperature 0.8

0.6

0.4
Measured Temperature
First Order Approximation
0.2 Estimated Temperature
Filtered Estimated

0
0 0.2 0.4 0.6 0.8 1 1.2
Time, sec

Figure 3.2: The measured and estimated instantaneous temperatures for the 0.032
inch grounded thermocouple.

31
3.5.3 Discussion

While the object is to obtain accurate temperature measurements, the validity of

the time constant and the method for predicting the instantaneous temperatures

was evaluated.

For this experiment to be a true step input in temperature, it is important to

consider how the temperature of the 304L Stainless Steel plate varied after it was

taken out of the furnace. Due to the thermal mass of the plate, the temperature

only decreased roughly 5 ◦ C during the time of each thermocouple test. Considering

that the thermocouples were started from room temperature and approached 1000
◦
C the 5 ◦ C change in the plate can be considered insignificant.

The exponential response of the thermocouple can be seen in the non-dimensional

temperature curves of Figure 3.1. While all the tests generally created good expo-

nential curves, many of the tests deviated slightly from a true exponential curve.

This is thought to be due to the thermocouples’ sensitivity to very small shifts in

location. Changes in position can potentially change the heat transfer rate which

affects the resulting temperature-time gradient. As a result, the time constants

obtained at a non-dimensional temperature of 0.632 ranged from 0.348 to 0.458

seconds. However, Test 2 from Figure 3.1 was found to make the most consis-

tent and smooth exponential curve. As a result, Test 2 is thought to describe the

thermocouple performance the best and the time constant of 0.385 will be used.

After evaluating the time constants, one question remains: Does Equation 3.3

accurately predict the material’s true temperature at any given point in time? This

was answered by predicting the temperature of a simple step input where the solu-

tion is known. Thus, the temperature data from Test 2 along with the time constant

of 0.385 was used to predict the instantaneous plate temperature in Figure 3.2. No-

32
tice that the filtered solution is essentially at the known solution of one for the

whole length of time. Therefore, the first order equation accurately models the

thermocouple performance and Equation 3.3 accurately predicts the instantaneous

plate temperature within 5% error of the true solution. This knowledge allows the

same process to be applied to other transient experiments where the true material

temperature is not known.

3.6 Measurement of FSW with Multiple Thermocouples

The object of this experiment was to evaluate both the accuracy and repeatability of

temperature measurements near the mid-plane of the workpiece. This was accom-

plished by embedding multiple thermocouples outside the stir zone throughout the

length of the weld. Therefore, items that can potentially cause misinterpretation

of the data were addressed, including the consistency of the weld itself, deflection

of the FSW machine during the weld, and the effects of the thermocouple time

constant.

3.6.1 Setup

The thermocouple setup consisted of drilling 0.89 mm diameter holes 3.4 mm deep in

a 1/4 inch thick 304L Stainless Steel plate at each of the locations shown in Table

3. A diagram of the hole locations is shown in Figure 3.3. Note that the order

of the hole locations has been randomized and two different thermocouples were

placed at each distance away from the weld. A 1/2 inch thick mild steel backing

plate was machined with 1.78 mm diameter through-holes at each of the specified

thermocouple locations. A shallow channel was then machined to each of these

holes that allowed for the thermocouples to be placed through the backing-plate

33
and up into the workpiece as shown in Figure 3.4. The shallow channel coupled

with the bend radius of the thermocouple wire produced a spring load to force

the thermocouple firmly into the plate when the whole assembly is clamped down.

Figure 3.3 depicts alignment holes at the beginning and the end of the weld. A

dowel pin was pressed into both of these holes to ensure that the workpiece and

backing plate were aligned properly. After each plate was instrumented with 0.032

inch grounded type K thermocouples, aluminum foil tape was placed as shown in

Figure 3.5 to prevent the thermocouple from moving during the rest of the setup

procedure.

Table 3: The thermocouple hole locations.

Hole number X-Position, mm Y-Position, mm Depth, mm
1 171.45 3.99 3.4
2 171.45 -3.99 3.4
3 215.9 12.7 3.4
4 215.9 -12.7 3.4
5 260.35 5.72 3.4
6 260.35 -5.72 3.4
7 304.8 8.64 3.4
8 304.8 -8.64 3.4
9 349.25 12.7 3.4
10 349.25 -12.7 3.4
11 393.7 5.72 3.4
12 393.7 -5.72 3.4
13 438.15 3.99 3.4
14 438.15 -3.99 3.4
15 482.6 8.636 3.4
16 482.6 -8.636 3.4

34
 Hole Locations Weld Line Alignment Holes
0.6

0.4

Y Position
0.2
Workpiece Thermocouple Depth
0

-0.2
Backing Plate
Weld Line
-0.4

-0.6
0 5 10 15 20

XThermocouple
Position Depth
Figure 3.3: The thermocouple locations represented graphically.

Workpiece Weld Line

Backing Plate

Figure 3.4: The setup to position thermocouples.

Figure 3.5: The bottom of the backing plate after all the thermocouples have been
placed.

35
3.6.2 Operation

The FS weld was performed on a modified Kearney & Trecker knee mill [30]. First,

the alignment holes on the plate were used to align the plate with the tool on the

FSW machine. The tool was placed 1.2 mm (0.050 inch) to the retreating side of the

center to accommodate for machine deflection and bring the true weld location very

close to the planned center location. The weld was then run at 300 rpm and 150

mm/min (6 in/min) with Z-load control of 33.3 kN (7500 lbf). The data acquisition

system had eight thermocouple channels sampling at 18 Hz each. As the weld

progressed, the channels were switched to capture the transient temperature data

as the FSW tool passed each of the 16 thermocouples.

Once the weld was finished, the true thermocouple position relative to the tool

position during the weld was determined as follows.

• A line was scribed down the weld based upon the center of the two alignment

holes at each end of the plate.

• Close-up pictures were taken of the weld with a machinist ruler in each picture.

• A line perpendicular to the scribed mark was added using Adobe Photoshop.

• The picture was inspected further to reveal where the curved surface marks

on the weld were tangent to the perpendicular line (See Figure 3.6). This

location is the true weld center.

• The distance from the true weld center to the scrolled mark was recorded as

the weld offset value.

As a means to check the accuracy of the offset value, a weld cross section was

obtained in the pane of two thermocouple holes. The planned weld center was then

found and the offset value was used to overlay the tool geometry onto the weld.
36
 Figure 3.6: Image used to measure weld offset.

Matlab was then used in the following manner to post-process all the tempera-

ture data.

• The known thermocouple locations were shifted to account for the weld offset.

• An estimate of the instantaneous temperature was generated using the method

presented in Section 3.5.1.

• The maximum estimated and measured temperatures were found and plotted.

• The estimated and measured transient temperatures were plotted with respect

to the relative X-location of the pin.

• The estimated temperatures were also plotted with respect to their relative X

and Y-locations.

• A steady state thermal profile was created using the previous plot by inter-

polating the temperatures throughout the whole domain.

37
3.6.3 Results

The weld surface finish created by the tool is shown in Figure 3.7.

Figure 3.7: The weld’s surface finish.

The unprocessed real-time temperature data collected by the data acquisition

system can be seen in Figure 3.8. Note that the shoulder temperature of the PCBN

tool is indicated in this plot as well.

1000
Plate Temperatures Tool Temperature
900

800

700
Temperature, °C

600

500

400

300

200

100

0
100 150 200 250 300 350 400 450 500

X-Position, mm

Figure 3.8: The real-time temperature data that was collected.

38
 The maximum thermocouple temperature and the maximum estimated plate

temperature for each thermocouple position are listed in Table 4. The differences in

repeated measurements are listed along with the difference in temperature between

the estimated and measured values.

The maximum estimated and maximum measured temperatures at each distance

from the weld center are shown in Figures 3.9 and 3.10. Figure 3.9 ignores the effect

of the weld offset while Figure 3.10 compensates for the offset. The advancing,

retreating and estimated temperatures are indicated with a X, +, and O respectively.

A cubic polynomial fit is also applied to the advancing and retreating temperature

data. In Figure 3.10, this same polynomial fit has also been shifted horizontally

to represent upper and lower bounds of the worst case error in the weld offset

measurement.

Table 4: The maximum measured and estimated temperatures along with the dif-
ferences in the repeated measurements.
Distance
From the Measured Estimated Measured Estimated Estimated -
Weld Temperatures, °C Temperatures, °C Difference Difference Measured
Center
mm First Second First Second First Second
13.5 343.2 364.6 343.5 365.2 -21.4 -21.7 0.2 0.5
Retreating

9.4 572.3 497.6 575.1 499.2 74.7 75.9 2.9 1.6

Side

6.5 774.7 763.7 777.5 767.5 10.9 10.1 2.9 3.7

4.8 842.9 832.7 846.2 836.8 10.2 9.4 3.3 4.0
Weld Center
3.2 874.8 902.2 876.0 903.7 -27.4 -27.7 1.3 1.6
Advancing

4.9 826.2 829.6 828.6 832.1 -3.4 -3.6 2.4 2.5

Side

7.8 598.7 602.7 601.6 605.7 -4.0 -4.1 2.9 3.0

11.9 356.8 392.9 357.3 393.7 -36.1 -36.4 0.5 0.9
Absolute
Average 23.5 23.6 2.0 2.2

The geometry of the FSW pin has been overlaid and shifted according to the

weld offset on the weld micrograph in Figure 3.11.

39
 Advancing
900
Retreating
Estimated Adv.
Maximum Temperature, °C Estimated Ret.
800 Polynomial Fit Adv.
Polynomial Fit Ret.

700

600

500

400

300
0 2 4 6 8 10 12 14 15
Distance from Weld Center, mm

Figure 3.9: The unadjusted maximum estimated and measured temperatures for
the weld.

Advancing
900
Retreating
Estimated Adv.
Maximum Temperature, °C

Estimated Ret.
800 Polynomial Fit Adv.
Polynomial Fit Ret.
Adv. Location Error
700 Ret. Location Error

600

500

400

300
0 2 4 6 8 10 12 14 15
Distance from Weld Center, mm

Figure 3.10: The maximum estimated and measured temperatures for the weld.

40
 Figure 3.11: The micrograph of the weld cross section.

The estimated and measured transient temperature for the advancing and re-

treating side of the weld are shown in Figure 3.12. Note that the longitudinal-

position of the data is shifted and plotted relative to the location of the pin.

The estimated temperature data used in the previous figure is shaded according

to magnitude and plotted according to its respective spacial location relative to

the pin in Figure 3.13. Finally, the same data is taken and used with a bilinear

interpolation function to generate an estimation of the temperature everywhere

within the whole domain in Figure 3.14.

41
 Advancing Retreating
1100 1100

1000 1000

900 900

800 800
Temperature, °C

700 700

600 600

500 500

400 400

300 300

200 200

100 100

0 0
−20 −10 0 10 20 −20 −10 0 10 20
Distance Relative to Pin, mm

Figure 3.12: The measured thermocouple temperature (solid) and the estimated
plate temperature (dotted) with respect to the relative distance from the pin for
the advancing and retreating sides of the weld.

Figure 3.13: The estimated temperature data with respect to the pin’s locations.

42
Figure 3.14: The estimated temperature data with bilinear interpolation to give a
full experimental temperature profile.

43
3.6.4 Discussion

To ensure repeatability of temperature measurements, transient effects should be

avoided. Figure 3.7 shows that the weld surface was uniform the whole length

of the weld. Also, the tool temperature in Figure 3.8 varied only 10 ◦ C during

the length of the weld where temperature measurements were being taken. In the

absence of transients, the repeatability can be inspected by analyzing the difference

in the two separate temperature measurements that were taken at each distance

away from the weld. Table 4 indicates that the maximum measured temperature

at each respective location differed from 3.4 to 74.6 ◦ C. Figure 3.15 shows that

the magnitude of the differences is uncorrelated with the relative distance to the

weld center. Therefore, it is reasonable to average the observed differences into one

value representing the average repeatability error. This value is found to be 23.5 ◦ C

(See Table 4). This amount of resolution in repeatable temperature data is deemed

acceptable for drawing reliable FSW conclusions.

80

70
Retrating Side
Advancing Side
60
Temperature Difference

50

40

30

20

10

0
0 2 4 6 8 10 12 14

Distance from Weld Center, mm

Figure 3.15: The difference in measured temperatures as a function distance from

the weld center.

44
 The accuracy of the thermocouple cannot be estimated because the true solu-

tion is not known. However, the estimated plate temperature can be compared to

the measured temperature at each location to ensure that the thermocouples are

capturing data that is believed to be the true temperature. Table 4 shows that the

estimated temperatures were 0.2 to 4.0 ◦ C higher than the measured temperatures.

The table also reveals that the difference between the measured and estimated tem-

peratures increases with proximity to the weld center. However, this difference is

so small that it would be safe to assume that the maximum temperatures are accu-

rately measured by the thermocouples. Nevertheless, it is also important to see the

comparison between the transient measured and estimated temperatures. The tran-

sient temperature data for both the advancing and retreating sides of the weld can

be seen in Figure 3.12. This figure shows how the estimated (dotted) temperatures

reach a maximum value about 1 mm before the measured values do. Given that

the weld is traveling at a constant velocity of 150 mm/s (6 in/min), this distance

correlates to 0.4 seconds. This lead time is consistent with the thermocouple time

constant of 0.385 seconds.

The importance of accounting for the true weld offset is illustrated in Figures 3.9

which ignores the weld offset, and Figure 3.10 which accounts for the weld offset.

In Figure 3.9, it appears that the temperature profiles on both the advancing and

retreating sides are similar in shape, with the advancing side 50 ◦ C hotter than

the retreating side. However, with the weld offset adjustment in Figure 3.10, the

temperature profiles on the retreating and advancing sides appear to have different

shapes. Near the pin, the peak temperature on the retreating side is probably

lower than the advancing side, but the elevated temperature zone is wider on the

retreating side.

45
 The worst case location shift is shown in Figure 3.10 by shifting the third order

polynomial curve for both the advancing and retreating sides. While this error does

not affect the accuracy of the measured temperatures, it does effect the accuracy of

the way that the results are interpreted. This shift on one extreme would allow the

advancing temperature far from the pin to be very similar to or slightly above the

retreating temperatures. However, the other extreme would result in the retreating

temperatures far from the pin exceeding the advancing temperatures by nearly 100
◦
C.

While current results may be counterintuitive, Figure 3.11 suggests that the

value obtained for the offset is reasonable. The planned weld center is defined as

the center point between the thermocouple holes. The measured weld offset shifted

the weld center to the right. When the tool geometry (black line) is centered on

the offset weld line the advancing side of the tool lines up very closely with the

edge of the deformation zone. This agrees with experimental work; the deformation

zone on the advancing side of the pin is very small while it is much larger on the

retreating side of the pin. This also illustrates that the weld centerline is not at the

geometrical center of the stir zone.

Another item of interest shown in Figure 3.12 is the cooling rate behind the pin.

This cooling rate is higher than that obtained for the 2-D models in Chapter 2.

Thus, the heat loss to the backing plate and free surface is significant.

The same estimated temperature information can also be seen in a more quali-

tative fashion in Figure 3.13. This figure shows the temperature by color variations

as if one were looking down on the weld from above. This can be used as qualita-

tive tool to compare numerical and experimental results. However, this data can be

taken one step further by interpolating the temperature throughout the domain as

46
in Figure 3.14. Here the experimental isothermal lines around the pin can be seen

based upon the color variation.

Overall, the method used to obtain the workpiece temperature profile has many

strengths.

• By using a machined backing plate, subsequent tests only required drilling

thermocouple holes in the plate to be welded.

• The thermocouples are spring fit into the backing plate and do not require

glue for securing them into place.

• The initial 170 mm of plate travel was just enough time to allow the FSW

tool to reach steady state before recording workpiece temperatures.

• With small spaced-out thermocouple holes, the natural thermal profile of the

workpiece is thought to be preserved.

• Two temperature readings at each distance from the weld provide a means of

inspecting the repeatability of the data.

• The maximum temperatures were accurately measured at feed rate of 150

mm/s.

3.7 Conclusion

Two experiments were covered in this paper. The first focused on obtaining the

time-constants for a 0.32 inch grounded thermocouples when making embedded

temperature measurements. The second sought to obtain FSW temperature mea-

surements in 304L Stainless Steel. The main emphasis here was to analysis the

repeatability and accuracy of the data.

47
 The following conclusions can be drawn based upon the results from the time-

constant experiment.

• The 0.32 inch grounded thermocouple had a time constant of 0.385 second.

• The time constant and measured temperature data can be used to approxi-

mate the temperature within 5% of the true instantaneous material tempera-

ture.

The following can be concluded based upon the FSW temperature measurement

experiment:

• The repeatability of the temperature data was measured by comparing two

temperature readings at the same distance from the weld. The average re-

peatability error is 23.5 ◦ C.

• The time constant and measured temperature were used as a means to esti-

mate the true instantaneous temperature and judge the performance of the

thermocouples. The difference between the measured and estimated tempera-

tures increases with proximity to the weld centerline. However, the difference

was less than 5 ◦ C and is believed to be insignificant.

• The true weld location or machine deflection is found to be critical for accurate

interpretation of the data and a method for accounting for the deflection is

presented.

• The weld centerline is not the geometrical center of the stir zone.

• The advancing side of the weld is believed to have higher peak temperature

but the retreating side is nearly 50 ◦ C greater further away from the pin.

48
 Based upon these results, it is thought that this method accurately measures

the temperature during FSP. It is recommended that this same process be used

to obtain workpiece temperatures over a variety of tool rotational speeds and feed

rates.

49
50
4 Friction Stir Welding Model Optimized and
Compared with Experimental Data

4.1 Abstract

Friction Stir Welding temperature data has been obtained in 304L Stainless Steel

at nine different operating conditions. The tool rotational speed was varied from

300 to 500 RPM and the feed rate ranged from 0.85 to 2.54 mm/s (2-6 in/min).

The experimental temperature data and weld cross-sections obtained at each oper-

ating condition were used to evaluate the performance of a two-dimensional finite

element model of FSW. This model was enhanced to enable modeling of realistic

tool rotational speeds. The ability to model thermal recovery was investigated but

is believed to be unnecessary for FSW of Stainless Steel. The predicted weld defor-

mation is also compared with the experimental weld deformation. An optimization

routine was used to iteratively adjust nine material input parameters in order to

minimize the difference between the numerical and experimental temperatures. The

optimization decreased the squared error between the numerical and measured tem-

peratures by 76%. However, the current friction and material models are insufficient

for accurate 2-D modeling of FSW over a large range of operating conditions.

4.2 Introduction

One of the challenges in evaluating the performance of Friction Stir Welding (FSW)

models is the difficulty of obtaining experimental data to compare with model pre-

dictions. Some of the experimental methods currently used to validate FSW mod-
51
els include embedded thermocouple measurements, infrared temperature measure-

ments, force data, microstructure data, and particle displacement data. While these

are all valid methods, much of the published FSW modeling work has only eval-

uated a single operating condition. The objective of this study is to obtain 304L

Stainless Steel temperature measurements over a variety of operating conditions,

and use this data to evaluate and improve the performance of a numerical model

over a range of operating conditions. The paper begins with a summary of prior

modeling and experimental work that has led up to the current publication. A brief

outline is given of the experimental procedure to obtain workpiece temperatures,

followed by an overview that shows how the numerical stability of the code was

increased to handle actual spindle speeds rather than scaled spindle speeds. Also,

an explanation is given of why thermal recovery may not be needed to accurately

model FSW. The predicted weld deformation size is also compared to the experi-

mental weld deformation. Next, the process used to optimize the model over the

whole range of operating parameters is presented. Both the experimental data and

numerical temperature predictions are displayed and discussed. The un-optimized

and optimized numerical predictions are compared. The paper concludes by re-

viewing what was learned from this work along with some recommendations that

may allow the optimization method to be used to fine tune a frictional and thermal

boundary condition.

4.3 Previous Work

While many have modeled the heat transfer in FSW, rarely are the models vali-

dated and optimized over a wide range of operating conditions. In fact, the only

known publication to report on this type of optimization was done in aluminum by

52
Covington [26]. In his work, the heat flux and boundary conditions on the tool were

changed to minimize the error between the modeled and experimental temperatures

in the tool over a wide range of operating conditions.

Owen and Sorensen [31] addressed the performance of a whole family of FSW

models each with different tool rotational speeds and feed rates. The results were

compared with the limited amount of available experimental data. The modeled

strain rate was also used to define a predicted weld deformation width that could

be compared to experimental cross-sections. The work also revealed that the model

did not allow the state variable (a scaler value representing the hardness of the

material) to thermally recover. Therefore, if the material was not deforming, the

model did not allow the material to soften when it was held at high temperatures for

a given amount of time. In addition, for the previous work, the surface velocity of

the tool-workpiece interface was scaled by 0.2 to maintain reasonable temperatures

and to avoid limit cycling of the numerical solution that took place at high surface

velocities. Owen [32] used similar methods as described in this paper to achieve true

tool surface velocity at 900 RPMs and found the maximum temperature predictions

exceeded the melting temperature of the material by more than 600 ◦ C.

This paper uses data from a parametric study of in-process welding temperatures

using the technique described by Owen and Sorensen [33] to experimentally calibrate

and improve a 2-D FSW model.

4.4 Method

4.4.1 Experimental

Following the techniques from reference [33], sixteen thermocouples were embedded

throughout the workpiece at eight different distances from the weld center all at the

53
same elevation. After the weld was performed the true weld center was measured

and the thermocouple locations were analytically shifted to represent the actual

distance from the measured weld center. The temperature data was offset relative

to the instantaneous pin location in order to create an experimental temperature

profile near the FSW tool. Furthermore, an instantaneous material temperature was

analytically predicted by using the measured time constant of the thermocouple.

In this work these steps are applied to nine separate welds each run at different

operating conditions. The feed rate and tool rotational speeds used for each weld

are shown in Figure 4.1. The order in which the welds were run is indicated by the

number in each circle.

500 5 8 3
Tool Rotational Speed, RPM

400 9 2 6

300 1 7 4

0.846 1.69 2.54

Feed rate, mm/s

Figure 4.1: The feed rate and tool RPM values used in the experiment. The numbers
depict the order in which the experiments were performed.

To achieve consistency from weld to weld, the tool depth was adjusted to give a

consistent weld surface width. This works because the FSW tool used has a convex

scrolled shoulder (The actual tool drawing can be seen in Appendix C). With this

tool design, very small differences in tool height make large differences in the surface

weld width.
54
 Micrographs of each weld cross section were also examined and the width of the

deformation zone at the height of the thermocouples was recorded.

4.4.2 Numerical Model Improvements

The basic FSW model created with the Hickory finite element program is a 2-D

cross-section in-plane with the weld material. The pin can be viewed as a stationary

rotating cylinder with the workpiece material flowing past it. All material flow is

restricted to in-plane motion. The area modeled is a 25 mm (2 in) square with a pin

diameter of 4.25 mm (0.171 in). This 2-D domain is modeled using an Eulerian mesh

with 6-node triangular elements that are refined near the pin in order to capture

the strong gradients that are generated there. The material traverses through the

mesh as shown by the arrows in Figure 4.2.

Material
Travel
Retreating Boundary Direction

Pin
Rotation
Direction

Inlet Outlet
Boundary Boundary

x Advancing Boundary

Figure 4.2: The 2-D Friction Stir Welding mesh and boundaries.

The boundary conditions applied to the domain are outlined in Table 5, where

the velocity, temperature and diameter are represented by u, θ and D respectively.

The material passing through the inlet-plane has been modeled at room tempera-

ture and given an initial state variable value that represents the average dislocation

55
 Table
Table 5:
5: Boundary conditions
B o undary co usedininthe
nditio ns used theFSW
FSWmomodel.
del.
Boundary Me chanical T he rm al State V ariable
Inlet ux = F eed R ate m / s θ = 25◦ C 1.5 × 10 8
uy = 0 m / s
O utlet ux = F eed R ate m / s
Adv ancing uy = 0 m / s Adiabatic Ev o lv es
R etreating
Frictio n Co nditio n
Tta n g en t = β(vef f − uta n g en t ) Co nv ectio n
Pin vto o l = R P M · π · D m / s q = h(θma ter ia l − θto o l ) Ev o lv es
D = 8 .5 m m (.3 3 3 in)
ur a d ia l = 0 m / s

Theormaterial
density hardnesspassing thro
of the ugh the inlet-plane
non-deformed has been
material. Themo deled
state at ro o mattemper-
variable all other
ature andevolves
locations giv en an initial state
according v ariable
to its v alue flow
respective that represents the av that
history. Notice eragethe
dislosame
catio n
con-
density o r hardness o f the no n-defo rmed material. The state v ariable at all o ther
ditions were applied to the outlet, advancing and retreating boundaries. The pin’s
lo catio ns ev o lv es acco rding to its respectiv e flo w histo ry . No tice that the same co n-
frictional condition applies a tangent traction Ttangent proportional to the difference
ditio ns were applied to the o utlet, adv ancing and retreating bo undaries. The pin’s
between the tool’s effective prescribed tangential surface velocity vef f and the nu-
frictio nal co nditio n applies a tangent tractio n Tta n g en t pro po rtio nal to the difference
merically calculated tangential material velocity utangent . In previous work [31] vef f
between the to o l’s effectiv e prescribed tangential surface v elo city vef f and the nu-
was limited to 0.2vtool due to convergence problems with the code. In this work the
merically calculated tangential material v elo city uta n g en t . No te that vef f = vto o l in
convergence was improved so that vef f could be equal to vtool . The value used for β
this paper but in prev io us publicatio ns vef f = 0 .2 ∗ vto o l . This was req uired prev i-
is 1.5 × 1011 kg/m2 ·s which was recommended by Dawson [34]. The pin’s thermal
o usly fo r numerical co nv enience in o rder to allo w co nv ergence, ho wev er, this will be
◦
convection coefficient h is assumed to be 1000 W/m2 ·K and θtool as 1000
11
C.2 The
addressed and reso lv ed in this paper. The v alue used fo r β is 1.5 × 10 k g/ m ·s
heat input from the tool’s shoulder has not been accounted for in this model. The
which was reco mmended by Co rnell Univ ersity . The pin’s thermal co nv ectio n co -
specific
efficientparameters
h is appro xused in the
imated simplified-Hart’s
as 10 0 0 W/ m2 ·K andmaterial
θ as 10model
0 0 ◦ C.and
It the method
is impo rtant for
to o l

determining
to no te thatthese values
the heat canfrofound
input m theinto references [17,
o l’s sho ulder 19].
has no t been acco unted fo r in
The
this moabove model
del. The was used
specific to analyze
parameters used abyfamily of nine differentmaterial
the simplified-Hart’s cases where
mo delthe

feed
andrate
the and tool
metho d forotational speed
r determining coincide
these v alues with
can fothose shown
und in in Figure
references [17 , 194.1.
].

Two modifications were made to the code to improve the performance of the

model.
56

56
 • Under-relaxation was applied to the state variable and temperature solutions.

• A thermal recovery term was investigated.

Under-relaxation: The equation used to under-relax the state variable and tem-

perature solutions are shown in Equations 4.1 and 4.2, where κ, θ, and α represent

the state variable, temperature and relaxation constant respectively. The subscript

i represents the solution of the previous iteration, therefore, with a value of α be-

tween zero and 1, convergence of the new solution is slowed allowing the numerical

solution to slowly approach the true solution in a stable fashion.

κ = (1 − α) ∗ κi+1 + α ∗ κi (4.1)

θ = (1 − α) ∗ θi+1 + α ∗ θi (4.2)

The advantages of using under-relaxation is that models with higher RPM values

are stable enough to converge to a solution. Therefore, the true RPM values can

be modeled which means that vef f = vtool in the friction model.

Thermal Recovery: While sufficient thermal recovery data was not found for

304L Stainless Steel, the needed information was found for aluminum and can be

seen in the isochronal plot shown in Figure 4.3. Therefore, the approach taken to

model the recovery of Stainless Steel was to first develop a simple model to charac-

terize the observed behavior in aluminum. Then use the same form of equations but

with different constants for Stainless Steel. The values for these constants would

then be obtained by comparing the modeled stress strain curves to experimental

stress strain curves.

The aluminum recovery behavior is better understood when it is plotted against

time as shown in Figure 4.4. These thermal recovery curves were then fit using

57
 Figure 4.3: The obtained annealing curves for 1100-H18 sheet [35].

1.80E+08

1.60E+08

1.40E+08
Yield Strength (Pa)

1.20E+08 422 K
1.00E+08 505 K
533 K
8.00E+07 562 K
6.00E+07 588 K

4.00E+07

2.00E+07

0.00E+00
0 5000 10000 15000 20000 25000
Time (sec)

Figure 4.4: The yield strength of aluminum plotted as a function of time (solid)
along with a modeled fit of the data (dotted.

58
Equation 4.3 and 4.4, where τ is the time constant obtained from the hottest ma-

terial curve at 588 K, and κf inal represents the final softening value of the material

if the material were left at the given temperature for a long period of time. The

value of κf inal is determined by Equation 4.4 which is a linear fit dependent upon

the temperature of the low and high κ values at 3000 seconds in Figure 4.4. The

modeled aluminum recovery performance can be seen in Figure 4.5 overlaid onto the

original recovery data. It is apparent that the model does not match the data ex-

actly. Especially at 533 ◦ C the experimental data deviates from the proposed form

of the solution. However, it does give a simple equation with only three parameters

to model the recovery of the material.

D −1
(κ) = (κ − κf inal ) (4.3)
Dt τ

κf inal = m ∗ θ + b (4.4)

1.80E+08
1.60E+08
422 K
1.40E+08 505 K
Yield Strength (Pa)

1.20E+08 533 K
562 K
1.00E+08 588 K
422 K
8.00E+07
505 K
6.00E+07 533 K
562 K
4.00E+07
588 K
2.00E+07
0.00E+00
0 5000 10000 15000 20000 25000
Time (sec)

Figure 4.5: The yield strength of aluminum plotted as a function of time (solid)
along with a modeled fit of the data (dotted.

59
 The thermal recovery term (Equation 4.3) was then added to the current evo-

lution equation. The new evolution equation is shown in Equation 4.5 and 4.6.

D  κ no −1
(κ) = ho 1 − sat D+ (κ − κf inal ) (4.5)
Dt κ τ

κf inal = m ∗ θ + b (4.6)

The parameters for the recovery term were evaluated with 304L stress strain curves

that were obtained by London [36] and Venugopal et al. [37]. Similar compression

tests were modeled to simulate the experimental curves. The parameter values were

then changed to minimize the differences between the numerical and experimental

stress strain curves.

The model was also used to predicted the weld deformation width as described

by Owen and Sorenson [31].

4.4.3 Parametric Optimization

The numerical model was then optimized over a range of operating parameters us-

ing the measured temperature data. Nine parameters in the model were adjusted to

improve the family of numerical predictions. (Eight parameters from the material

model and one from the frictional model). Furthermore, a Matlab function was cre-

ated to quantitatively compare the error between the numerical and experimental

results. While many different error functions were tried, the one used in this paper

squares the difference in the maximum numerical temperature θnum and experimen-

tal temperature θexp at the closest location on the advancing and retreating sides of

the weld. This error value was calculated for five numerical runs that correspond

with the operating conditions represented by the four corners and center of Figure

4.1. The error of each run was summed into one value that describes the total error

60
for the whole family of numerical predictions as shown in Equation 4.7.
5models
X 2X dist. 
θexpj − θnumj 2

T otal Error = (4.7)
i=1 j=1
scalar

The numerical optimization was conducted in the following manner:

• A script inputs the initial parameters and updates all the parameters in five

of the numerical models and runs the FEA calculations.

• Upon successful completion of all the numerical runs, a Matlab function is

called to calculate the error between the numerical and experimental results

as described above. If by chance the parameters do not allow for a converged

solution, a high error is returned and the process continues.

• The script receives this error and uses a gradient following technique to change

the parameters in a way that minimizes the error. This whole process was

repeated until the error could not be minimized further.

Matlab functions were also created to post-process, analyze and display all the

results. These functions were also used to compare the experimental and numerical

results in the same figures for easy comparison.

4.5 Results

The results have been organized to first show the experimental temperature and

deformation results, followed by the numerical temperature differences as a result

of under-relaxation. Experimental and numerical stress-strain curves are compared

and the deformation predictions are shown. Thereafter, the optimized numerical

solution is compared to the experimental results.

The maximum measured temperature at each distance from the weld center is

plotted in Figure 4.6. Note that the thermocouple locations are shifted in each plot
61
to represent the true distance to the weld center. The advancing, retreating and es-

timated plate temperatures (predicted temperatures determined from the measured

temperatures by accounting for the thermocouple time constant) are indicated with

a X, +, and O respectively. Also, a third degree polynomial fit is applied to the

advancing and retreating temperatures. Note that rows represent runs performed

at a constant tool rotational speed, and columns represent runs at a constant feed

rate.
Advancing
Retreating
Maximum Temperature, °C

Estimated Adv.
1000 1000 1000 Estimated Ret.
Polynomial Fit Adv.
800 800 800 Polynomial Fit Ret.
500

600 600 600

400 400 400

0 5 10 15 0 5 10 15 0 5 10 15
Rotational Speed rev/min

Maximum Temperature, °C

1000 1000 1000

800 800 800

400

600 600 600

400 400 400

0 5 10 15 0 5 10 15 0 5 10 15
Maximum Temperature, °C

1000 1000 1000

800 800 800

300

600 600 600

400 400 400

0 5 10 15 0 5 10 15 0 5 10 15
Distance from Weld Center, mm Distance from Weld Center, mm Distance from Weld Center, mm
0.85 (2) 1.69 (4) 2.54 (6)
Feed Rate mm/s (in/min)

Figure 4.6: Maximum measured temperatures for the advancing (X, solid) and
retreating side (+, dotted) side of the weld along with the estimated temperatures
(o).

The transient measured (solid) and instantaneous (dotted) temperatures at each

distance from the weld are plotted in Figure 4.7. Note that the X-position of the

temperature is shifted and plotted relative to the pin’s current location.

62
 1000 1000 1000
Temperature, °C

800 800 800

500

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20
Rotational Speed rev/min

1000 1000 1000

Temperature, °C

800 800 800

400

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20

1000 1000 1000

Temperature, °C

800 800 800

300

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20
Distance Relative to Pin, mm Distance Relative to Pin, mm Distance Relative to Pin, mm
0.85 (2) 1.69 (4) 2.54 (6)
Feed Rate mm/s (in/min)

Figure 4.7: Transient measured (solid) and instantaneous (dotted) temperatures.

63
 The weld cross-sections of all nine welds have been shown in Figure 4.8. The

width of the weld deformation at the hight of the thermocouples has been indicated

by the white horizontal line in each cross-section.

Figure 4.8: Weld micrographs at each operating condition along with a horizontal
line marking the weld deformation at the height of the thermocouples.

In Figure 4.9, the numerical maximum temperatures are shown in a similar

fashion to Figure 4.6, but the realistic RPM values (where vef f = vtool ) are shown

by solid lines and the reduced RPM values (where vef f = 0.2 ∗ vtool ) for numerical

convenience) are shown by the dotted lines.

The stress strain curves obtained by London [36] and Venugopal et al. [37] are

plotted (solid) along with corresponding numerical stress strain curves (dotted) in

Figure 4.10, 4.11 and 4.12

64
 1200 1200 1200

Maximum Temperature, °C
1000 1000 1000

800 800 800

500
600 600 600

400 400 400

200 200 200

0 0 0
0 5 10 15 0 5 10 15 0 5 10 15
Rotational Speed rev/min

1200 1200 1200

Maximum Temperature, °C

1000 1000 1000

800 800 800

400

600 600 600

400 400 400

200 200 200

0 0 0
0 5 10 15 0 5 10 15 0 5 10 15

1200 1200 1200

Maximum Temperature, °C

1000 1000 1000

800 800 800

300

600 600 600

400 400 400

200 200 200

0 0 0
0 5 10 15 0 5 10 15 0 5 10 15
Distance from Weld Center, mm Distance from Weld Center, mm Distance from Weld Center, mm
0.85 (2) 1.69 (4) 2.54 (6)
Feed Rate mm/s (in/min)

Figure 4.9: The maximum numerical temperatures at realistic RPM values (solid)
and the reduced RPM values (dotted). The advancing (X) and the retreating (+)
temperatures are also indicated.
Strain rate = 0.01, Different Temperatures

1.40E+09

1.20E+09 T=293 K
T=473 K
1.00E+09 T=673 K
T=873 K
T=1073 K
Stress, pa

8.00E+08
T=1273 K
T=293 K
6.00E+08 T=473 K
T=673 K
4.00E+08 T=873 K
T=1073 K
2.00E+08 T=1273 K

0.00E+00
0 0.2 0.4 0.6 0.8 1
Strain

Figure 4.10: Stress strain curves [36] for 304L at different temperatures with a strain
rate of 0.01. The numerical equivalent stress strain curves are dotted.

65
 Experimental, Temperature = 1073 K, different strain rates

5.00E+08

100
4.00E+08 10
1
0.1
3.00E+08 0.01
Stress, pa

0.001
100
2.00E+08 10
1
0.1
1.00E+08 0.01
0.001

0.00E+00
0 0.1 0.2 0.3 0.4 0.5 0.6
Strain

Figure 4.11: Stress strain curves [37] for 304L at different strain rates with a tem-
perature of 1073 K. The numerical equivalent stress strain curves are dotted.

Experimental, Temperature = 1473 K, different strain rates

2.00E+08

100
10
1.50E+08 1
0.1
0.01
Stress, pa

0.001
1.00E+08 100
10
1
0.1
5.00E+07 0.01
0.001

0.00E+00
0 0.1 0.2 0.3 0.4 0.5 0.6
Strain

Figure 4.12: Stress strain curves [37] for 304L at different strain rates with a tem-
perature of 1473 K. The numerical equivalent stress strain curves are dotted.

66
 The resulting velocities have been displayed with streamline plots that predict

the path that a particle would travel during a simulated weld. The calculated

deformation zone width (in mm) for the advancing and retreating side is displayed

at the top of each graph and visually represented by dotted lines in Figure 4.13.

The total deformation width of both the experimental (dotted) and numerical (solid)

have been shown in Figure 4.14.

Run 31, 3.3, −3 Run 32, 3.4, −3.1 Run 33, 3.9, −3.1
500

Run 21, 3.3, −3 Run 22, 3.5, −3.2 Run 23, 4, −3.2
Rotational Speed rev/min
400

Run 11, 3.3, −3.1 Run 12, 3.8, −3.2 Run 13, 4.1, −3.3
300

0.85 (2) 1.69 (4) 2.54 (6)

Feed Rate mm/s (in/min)

Figure 4.13: Streamline results viewed on a 15 mm (.59in) square. The calculated

boundaries of the deformation zone are shown by the dotted lines and indicated in
millimeters at the top of each plot.

The non-optimized and optimized numerical solutions can be seen overlaid on

one another in Figure 4.15.

In Figure 4.16, the optimized numerical maximum temperatures are plotted

along with the maximum measured temperature.

67
 8

Deformation Width, mm 6
300 RPM
5 400 RPM
500 RPM
4 300 RPM
400 RPM
3 500 RPM

0
0.85 1.35 1.85 2.35
Feed Rate, in/min

Figure 4.14: The calculated (solid lines) and experimental (dotted lines) deforma-
tion zones.

1200 1200 1200

1000 1000 1000

Temperature °C

800 800 800

500

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20
Rotational Speed rev/min

1200 1200 1200

1000 1000 1000

Temperature °C

800 800 800

400

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20

1200 1200 1200

1000 1000 1000

Temperature °C

800 800 800

300

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20
X Position Relative to the Pin X Position Relative to the Pin X Position Relative to the Pin
0.85 (2) 1.69 (4) 2.54 (6)
Feed Rate mm/s (in/min)

Figure 4.15: Optimized temperature predictions (solid) with the original numerical
predictions (dotted).

68
 Adv. Error 29.6 Ret. Error 0.973 Adv. Error 0.445 Ret. Error 1.32 Adv. Error 4.85 Ret. Error 9.4
1200 1200 1200
Maximum Temperature, °C

1000 1000 1000

800 800 800

500

600 600 600

400 400 400

200 200 200

0 0 0
0 5 10 15 0 5 10 15 0 5 10 15

Adv. Error 4.8 Ret. Error 12.5 Adv. Error 6.85 Ret. Error 10.9 Adv. Error 1.29 Ret. Error 11.2
Rotational Speed rev/min

1200 1200 1200

Maximum Temperature, °C

1000 1000 1000

800 800 800

400

600 600 600

400 400 400

200 200 200

0 0 0
0 5 10 15 0 5 10 15 0 5 10 15

Adv. Error 2.56 Ret. Error 6.92 Adv. Error 10.6 Ret. Error 14.5 Adv. Error 15.1 Ret. Error 22.5
1200 1200 1200
Maximum Temperature, °C

1000 1000 1000

800 800 800

300

600 600 600

400 400 400

200 200 200

0 0 0
0 5 10 15 0 5 10 15 0 5 10 15
Distance from Weld Center, mm Distance from Weld Center, mm Distance from Weld Center, mm
0.85 (2) 1.69 (4) 2.54 (6)
Feed Rate mm/s (in/min)

Figure 4.16: Original measured temperature (solid) with the optimized tempera-
ture predictions (dotted). The percent error between the measured and calculated
maximum temperatures are listed at the top of each plot.

69
 The optimized numerical solution overlaid onto the instantaneous measured tem-

peratures can be seen in Figure 4.17.

Adv. Error 29.6 Ret. Error 0.973 Adv. Error 0.445 Ret. Error 1.32 Adv. Error 4.85 Ret. Error 9.4
1200 1200 1200

1000 1000 1000

Temperature, °C

800 800 800

500

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20

Adv. Error 4.8 Ret. Error 12.5 Adv. Error 6.85 Ret. Error 10.9 Adv. Error 1.29 Ret. Error 11.2
Rotational Speed rev/min

1200 1200 1200

1000 1000 1000

Temperature, °C

800 800 800

400

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20

Adv. Error 2.56 Ret. Error 6.92 Adv. Error 10.6 Ret. Error 14.5 Adv. Error 15.1 Ret. Error 22.5
1200 1200 1200

1000 1000 1000

Temperature, °C

800 800 800

300

600 600 600

400 400 400

200 200 200

0 0 0
−20 −10 0 10 20 −20 −10 0 10 20 −20 −10 0 10 20
Distance Relative to Pin, mm Distance Relative to Pin, mm Distance Relative to Pin, mm
0.85 (2) 1.69 (4) 2.54 (6)
Feed Rate mm/s (in/min)

Figure 4.17: Experimental measured temperatures (Solid) and instantaneous pre-

dicted temperatures (Dotted). The percent error between the measured and calcu-
lated maximum temperatures are listed at the top of each plot.

70
4.6 Discussion

4.6.1 Temperature Studies

The maximum temperatures in Figure 4.6 reveal the magnitude of temperatures

experienced by the workpiece. The maximum temperatures reached in all the welds

ranged from 900 - 1125 ◦ C which corresponds to 70 - 83 % of the absolute melting

temperature. Also, two expected trends shown in the figure include the following:

• The workpiece temperature decreases with increasing feed rate.

• The workpiece temperature increases with increasing rotational speed.

However, changes in feed rate affected temperature more than changes in RPM

values. Also, it is interesting that temperature differences were larger farther away

from the weld center as the feed rate increased. This suggests that welds approach a

maximum temperature near the weld center that is not easily exceeded. The larger

difference farther away from the weld center makes sense because the higher feed

rate means that the weld is being heated for a shorter period of time and the heat

is being conducted away into the backing plate before it reaches the thermocouples

further away from the weld. Note that the maximum estimated plate temperatures

are also indicated on the plot by the circles. There is very little difference between

the measured and estimated temperatures. This reveals that the thermocouples

had a short enough time constant to accurately capture the maximum temperatures

throughout the weld.

The advancing and retreating temperatures are also depicted in Figure 4.6. Gen-

erally, speaking the advancing and retreating temperatures are very similar in mag-

nitude (within 15 ◦ C). However, at the highest and lowest weld pitch (500 RPM

0.85 mm/s and 300 RPM 2.54 mm/s), the retreating temperatures exceeded the
71
advancing temperatures by nearly 50 ◦ C farther away from the pin. However, it is

still thought that the advancing side of the pin had a higher peak temperature near

the pin.

The estimated temperatures in Figure 4.7 also reveal how well the thermocouples

performed. At low feed rates the estimated and measured temperatures are very

similar to one another. However, as the feed rate increases the estimated (dotted)

temperature rises slightly before the measured (solid) temperatures. At higher feed

rates this discrepancy may become larger but at the current feed rates the difference

is basically insignificant.

4.6.2 Model Improvement

The under-relaxation that was applied to the state variable and temperature solu-

tions had a large impact on temperature solutions. Essentially, this allows the nu-

merical solutions to remain stable at higher RPM values and allows realistic RPM

values to be modeled. The difference in the maximum temperatures at each distance

away from the weld can be seen in Figure 4.9 for the true RPM (vef f = vtool ) and

reduced RPM (vef f = 0.2 ∗ vtool ). The realistic RPM values brought the maximum

temperatures up substantially but did not change the shape of the temperature

curves at all. This reveals how dependent the numerical temperatures are upon the

modeled RPM value or surface velocity at the tool-workpiece interface. If fact, Owen

[32] found that at 900 RPM the maximum temperatures reached 2048 ◦ C which is

not reasonable considering the fact the melting temperature of 304L Stainless Steel

is 1421 ◦ C. While, this strong interface velocity dependence can produce reasonable

results for specific values of the parameters, it is not adequate for modeling FSW

over a large range of operating parameters. Therefore, other friction models should

be investigated that may model FSW better over a large range of conditions.
72
 The stress-strain curves obtained by London [36] and Venugopal et al. [37] are

shown with the corresponding numerical stress strain curves in Figures 4.10, 4.11

and 4.12. Different optimization methods were used to pick thermal recovery con-

stants that would minimize the squared difference between the numerical and ex-

perimental stress-strain curves. However, almost all of the numerical stress-strain

curves are already below the experimental curves. By adding thermal recovery the

material softens and stresses decrease which in effect increases the error between the

numerical and experimental stress-strain curves. Therefore, this method suggests

that there would be no benefit to include a thermal recovery term.

4.6.3 Parametric Optimization

The difference resulting from the parametric optimization can be seen in Figure

4.15. The optimized solution increased the maximum temperatures 300 to 600 ◦ C

depending upon the operating conditions. The un-optimized and optimized solution

had total squared errors of 751,899 and 178,275 respectively. The total squared error

decreased by 76% from the non-optimized solution.

While the optimization process did reduce the overall temperature error, it is

important to increase the stability of the numerical calculations as much as possi-

ble. As the optimization proceeded by changing variable values many un-converged

solutions resulted. This is detrimental to the optimization because the optimization

process must stay within bounds that always produce converged solutions. There

could be parameter combinations that would cause better temperature predictions,

but those combinations may only provide converged solutions near the given values.

Essentially, the whole parameter domain doesn’t result in converged solutions and,

as the optimization proceeds, many parameters may not be reached because the

73
optimization will shift the variable away from values that resulted in un-converged

solutions.

4.6.4 Numerical/Experimental Comparisons

The numerical experimental deformation width comparison can be seen in Figure

4.14. The calculated deformation widths are 1 to 2 mm larger than the measured

widths. The experimental weld deformation width (dotted) decreases in size at 300

and 400 RPM as the feed rate increases. However, this is contrary to the numerical

deformation trends that show the width increasing as the feed rate increases. While

the model does predict a deformation within 2 mm of the observed, it is believed

that the conflict in deformation trends will prevent the model from adequately

predicating the weld deformation width over a wide range of operating parameters.

The performance of the maximum numerical and experimental temperatures

is compared in Figure 4.16. The optimized numerical temperatures match the

measured temperatures the best closest to the pin due to the error function used

in the optimization process. The largest error in these temperatures occur at the

lop left (500 RPM at 0.85 mm/s) and the bottom right (300 RPM at 2.54 mm/s)

of Figure 4.16. While the model is capable of predicting reasonable temperatures

on a limited range of parameters it is believed that the the current friction and

material model are insufficient for predicting the temperature over a large range of

operating conditions. It is also interesting to note the shape of the numerical and

experimental maximum curves differ. The experimental temperatures near the pin

seem to approach a maximum and the temperature gradient decreases. However, the

numerical temperature gradient increases to a maximum right at the pin itself. This

is believed to happen because the 2-D model has an adiabatic boundary condition

above and below the workpiece. Therefore, the heat can only dissipate out away
74
from the pin where in actuality heat also goes into the backing plate. Consequently,

it is recommended that a thermal boundary condition be added for the backing

plate. However, this may cause the predicted temperature to substantially lower,

which would in turn, require the frictional condition to be modified.

The transient performance of the model is shown in Figure 4.17. At low feed

rates, it is easy to see the difference in the preheating of the material in front of

the pin. The measured temperatures increase at a slower rate and affect a larger

area. However, at higher feed rates this difference is much less prominent. This can

possibly be explained in the following two ways:

• Lack of heat input from the shoulder of the FSW tool in the model

• Conductivity of the material may be too low

The model’s lack of heat input from the shoulder influenced the thermocouple

setup that was used. The height of all the thermocouples was selected to be away

from the top surface and in effect, minimize the effect of the shoulder on the ther-

mocouples and allowing the pin itself to play a more dominate roll on the resulting

temperatures. However, the heat from the shoulder may affect the weld tempera-

ture more dominantly at slow feed rates than it does at higher feed rates. Therefore,

neglecting heat from the shoulder would explain why the model predicts much less

preheating at slow feed rates and reasonable preheating at higher feed rates.

The model currently applies a fixed thermal conductivity value of 22.6 W/m·K

throughout the whole mesh. This value was obtained based upon a material tem-

perature of 800 K. However, calculating the temperature-dependent thermal con-

ductivity throughout the whole domain would not be expensive computationally

and may improve model performance.

75
 Another observation from Figure 4.17 is that the low numerical maximum tem-

peratures take place much farther behind the tool than they do experimentally. In

fact, some of the numerical temperatures continue to rise behind the tool clear up

to the exit boundary of the modeled domain. This is probably due to the adiabatic

boundary condition used in the model. The heat near the pin can only dissipate

throughout the workpiece which continually increases the temperature farther away

from the pin. Therefore, this example also suggest that it would be beneficial to

add a thermal boundary condition to account for the heat to the backing plate.

Ultimately, it required both under-relaxation and optimization were required to

match the experimental temperature data. While the model did accurately predict

the temperature and deformation of a limited parameter range, the model in the

current form is insufficient for modeling FSW over a large range of parameters.

However, the optimization method detailed in this paper may provide a good way

to fine other frictional and thermal boundary conditions.

4.7 Conclusion

FSW temperature results were obtained in nine different welds performed at dif-

ferent conditions. The tool rotational speeds vary from 300 to 500 RPMs while

the feed rates range from 0.85 to 2.54 mm/s (2 - 6 in/min). Nine 2-D numerical

models were run at the corresponding conditions. The numerical performance was

changed by using realistic tool rotational speeds and by optimizing nine adjustable

parameters in the material model.

The following conclusions can be drawn from the experimental temperature data:

• The maximum temperatures for all nine welds range from 900 to 1000 ◦ C which

corresponds to 70 to 83 % of the absolute material melting temperature.

76
 • The temperature of the advancing and retreating side of the weld are very

similar in magnitude (Generally within 25 ◦ C) .

• The temperature gradient is not at a maximum value near the pin of the FSW

tool.

• The workpiece temperature increases with increasing tool rotational speed.

• The workpiece temperature decreases with increasing feed rate.

The following conclusions can be drawn from the numerical temperature predic-

tions:

• The temperature gradient is steepest right next to the pin.

• The advancing side of the weld is always roughly 35 - 50 ◦ C hotter than the

retreating side of the weld.

• Using under-relaxation of the state variable and temperature solutions allowed

for modeling the actual RPM values. This enables the predicted temperatures

to be more realistic.

• The ability to model thermal recovery is believed to be unnecessary for FSW

of Stainless Steel.

• Optimizing the material parameters for the whole family of models reduces

the numerical experimental error by 76%.

Comparison of the numerical and experimental results revealed the following:

• The model can not adequately predicate the weld deformation width over a

wide range of operating parameters.

77
• The model shows less heating taking place in front of the pin. This may be

due to the lack of heat input from the shoulder of the FSW tool.

• To numerically predict correct temperature gradients and maximum tempera-

tures, it is important for the model to account for the heat loss to the backing

plate.

• The model can fit a limited parameter set but is insufficient for modeling a

large range of parameters.

The following is recommended for future work:

• Add boundary condition to account for the heat loss to the backing plate.

• Allow the thermal conductivity to be calculated based upon temperature

throughout the whole domain.

• Investigate new friction models.

• Optimize a new friction model and thermal boundary condition in a similar

method as performed in this paper.

78
5 Recommendations For Future Work

Based upon the results of the previous three chapters, it is recommended that the

following work be performed.

• Improve the convergence of the numerical code. This may be accomplished by

adding a function that automatically adjusts the under-relaxation constants

to optimal values based upon the current convergence data. Not only will this

decrease the execution time of the code but it should allow for more explo-

ration to be done without being limited due to limit cycling of the numerical

solutions.

• Allow the thermal conductivity to be calculated based upon temperature

throughout the whole domain.

• Add a boundary condition to account for the heat loss to the backing plate.

• Investigate and model FSW with new friction models.

• Optimize the parameters contained in a friction model and a non-adiabatic

thermal boundary condition to the backing plate using a process similar to

that performed in this work.

Due to the similarities of the 2-D Hickory and 3-D Isaiah codes, all of the above

mentioned items would be beneficial to the 3-D code as well. However, due to

the increase in needed computational power for 3-D models, it is suggested that

preliminary studies be preformed in the 2-D code and then applied to the 3-D code.
79
80
References
[1] S. M. Packer, T. W. Nelson, and C. D. Sorensen. Tool and equipment re-
quirements for friction stir welding ferrous and other high melting temperature
alloys. In Proceedings of the 3rd International Symposium on Friction Stir
Welding, Kobe, Japan, 27-28 September 2001.

[2] K. Yoshikawa. A joining criterion for lap joining of dissimilar metal materials
of aluminum and stainless steel by friction stir. In Proceedings of the 4th
International Symposium on Friction Stir Welding, Park City. Utah, May 2003.

[3] TWI Website. Friction stir welding, 2003. Technology File

http://www.twi.co.uk/j32k/unprotected/band 1/tffricst.html.

[4] Y. J. Chao and X. Qi. Thermal and thermo-mechanical modeling of friction

stir welding of aluminum alloy 6061-t6. Journal of Materials Processing and
Manufacturing Science, pages 215–233, October 1998. Technomic Publishing
Co., Inc.

[5] T. U. Seidel and A. P. Reynolds. Two-dimensional friction stir welding process

model based on fluid mechanics. Science and Technology of Welding and Join-
ing, 8(3):175–183, 2003.

[6] S. Xu and X. Deng. Two and three-dimensional finite element modes for the
friction stir welding process. In Proceedings of the 4th International Symposium
on Friction Stir Welding, Park City, Utah, May 2003.

[7] P. A. Colegrove and H. R. Shercliff. Modelling the friction stir welding of

aerospace alloys. In 5th International Symposium on Friction Stir Welding,
Metz, France, 14-16 September 2004.

[8] P. Ulysse. Three-dimensional modeling of the friction stir welding process.

International Journal of Machine Tools and Manufacture, 42:1549–1557, 2002.

[9] G. J. Benzsak, T. H. North, and C. B. Smith. An experimentally validated

3d model for friction stir welding. In Proceedings of the 4th International
Symposium on Friction Stir Welding, Park City, Utah, May 2003.

[10] P. Colegrove, M. Painter, D. Graham, and T. Miller. Three dimensional flow

and thermal modeling of the friction stir welding process. In Proceedings of the
2nd International Symposium on Friction Stir Welding, Gothenburg, Sweden,
June 2000.

81
[11] T. U. Seidel and A. P. Reynolds. Visualization of the material flow in aa2195
friction-stir welds using a marker insert technique. Metallurgical and Materials
Transactions A Physical Metallurgy and Materials Science, 32A:2879–2884,
2001.

[12] J. R. Egbert. Material Flow In Friction Stir Welding. Master’s thesis, Brigham
Young University, 2002.

[13] MegaStir. Process window analysis. May 2003.

[14] C. B. Owen. Temperature measurments in 304l stainless steel during friction

stir processing. June 2004.

[15] J. H. Record, J. L. Covington, T. W. Nelson, C. D. Sorensen, and B. W. Webb.

Fundamental characterization of friction stir welding. In Proceedings of the 5th
International Symposium on Friction Stir Welding, France, September 2004.

[16] P. R. Dawson. On modeling of mechanical property changes during flat rolling

of aluminum. International Journal of Solids and Structures, 23(7):947–968,
1987.

[17] G. M. Eggert and P. R. Dawson. On the use of internal variable constitutive

equations in transient forming processes. International Journal of Mechanical
Science, 29(3):95–113, 1987.

[18] G. M. Eggert and P. R. Dawson. Assessment of a thermoviscoplastic model of

upset welding by comparison to experiment. International Journal of Mechan-
ical Science, 28(9):563–589, 1986.

[19] J. Cho, P. R. Dawson, and D. E. Boyce. 2-D modeling of friction stir welding by
eulerian formulation. In American Institute of Physics Conference Proceedings,
volume 712(1), pages 1326–1331, June 10 2004.

[20] E. W. Hart. Constitutive relations for the non-elastic deformation of metals.

Journal of Engineering Materials and Technology, pages 193–202, July 1976.

[21] S. G. Lambrakos, R. W. Fonda, J. O. Milewski, and J. E. Mitchel. Analysis of

friction stir welds using thermocouple measurements. Science and Technology
of Welding and Joining, 8(5):385–390, 2003.

[22] M. Song and R. Kovacevic. A coupled heat-transfer model for workpiece and
tool friction stir welding. In Proceedings of the 4th International Symposium
on Friction Stir Welding, Park City, Utah, May 2003.

[23] J. C. McClure, W. Tang, L. E. Murr, X. Guo, Z. Feng, and J. E. Gould. A

thermal model of friction stir welding. In Proceedings of the 5th International
Conference on Trends in Welding Research, pages 590–595, Pine Mountain
Ga., 1998.
82
[24] Abe Askari, S. Silling, B. London, and M. Mahoney. Modeling and analy-
sis of friction stir welding processes. In Proceedings of the 4th International
Symposium on Friction Stir Welding, Park City, Utah, May 2003.

[25] T. Dickerson, Q. Shi, and H. Shercliff. Heat flow into friction stir welding tools.
In Proceedings of the 4th International Symposium on Friction Stir Welding,
Park City, Utah, May 2003.

[26] J. L. Covington. Experimental and Numerical Investigation of Tool Heating

During Friction Stir Welding. Master’s thesis, Brigham Young University,
August 2005.

[27] C. D. Sorensen, T. W. Nelson, and S. M. Packer. Tool material testing for FSW
or high-temperature alloys. In Proceedings of the 3rd International Symposium
on Friction Stir Welding, Kobe, Japan, 27-28 September 2001.

[28] T. W. Nelson and C. D. Sorensen. Microstructural evaluation and tool develop-

ment for FSW/FSP aluminum 7075 and hsla-65. In Friction Stir Technologies
Principal Investigators’ Meeting, Sedona, Arizona, 28-30 September 2005.

[29] D. D. Pollock. Thermocouples: Theory and Properties. CRC Press, 1991.

[30] Brigham Young University. FSW machine, 2005.

[31] C. B. Owen. Application of a coupled viscoplastic material model to friction

stir welding with a variety of tool rpms and feed rates. Jan 2006.

[32] ONR. Friction Stir Technologies Principal Investigators’ Meeting, Sedona, Ari-
zona, September 2005.

[33] C. B. Owen. Method for effective plate temperature measurements during

friction stir processing. Jan 2006.

[34] P. R. Dawson. Private communication. Cornell University.

[35] J. E. Hatch. Aluminum : properties and physical metallurgy. American Society

for Metals, Metals Park, Ohio, 1984, p. 125.

[36] Blair London. Mechanical testing results from third wave systems. DARPA
Friction Stir Processing Program, December 2002.

[37] S. Venugopal, S. L. Mannan, and P. Rodriguez1. Optimum design of a hot

extrusion process for aisi type 304l stainless steel using a model for the evo-
lution of microstructure. Modelling and Simulation in Materials Science and
Engineering, 10:253–256, 2002.

83
84
APPENDIX

85
86
Appendix A: Mesh Refinement Study
The mesh refinement study compared the solutions of eight meshes each with dif-
ferent element sizes near the pin (See Figure A.1). A high feed rate and high RPM
were used to simulate a case that would generate steep gradients near the pin.

Figure A.1: Each mesh used in the refinement study is shown. The number of
elements is indicated below each plot.

The grid study revealed that the maximum strain rate continued to increase
as the number of elements increased (See Figure A.2). The location of this maxi-
mum strain rate is on the advancing side of the pin and affects an extremely small
region. Thus, the average value for the strain rate, temperature, velocity, stress
and state variable at a distance of 2 mm away from the pin has been calculated to
quantitatively compare the solutions. The percent change in this value as the mesh
was refined is plotted in Figure A.3. Notice that the percent change in the total
deformation is also plotted in this Figure as well. Based upon the these results, the
mesh used in all sixteen runs had 1238 6-node triangle elements.

87
 700

600
Max. Strain Rate
500

400

300

200

100

0
0 500 1000 1500 2000 2500 3000
Number of Elements

Figure A.2: The maximum obtained strain rate for each grid in the mesh study.

60
Strain Rate
50 Temperature
X Velocity
40 Stress
State Variable
30 Total Deformation
Percent Change

20

10

−10

−20

−30

−40
1000 1500 2000 2500 3000
Number of Elements

Figure A.3: The percent change of the average value 2 mm way from the pin.

88
Appendix B: State Variable Details
Equations B.1 through B.3 are used to evolve the state variable along each stream-
line. The constants are defined below.

κsat = State V ariable saturation value

C = Constant
Φ = F isher F actor
mo = Constant
θ = T emperature
Do = Constant
D = StrainRate
ho = Constant
no = Constant
D
(κ) = Change in the State V ariable according
Dt
to time along the current streamline

 mo
sat C
κ = (B.1)
Φ
 
Do
Φ = θ ln (B.2)
D
D  κ no
(κ) = ho 1 − sat D (B.3)
Dt κ

89
90
 REVISIONS
Locking Collar Set Screw ZONE REV. DESCRIPTION DATE APPROVED
PCBN Tip
Tip Detail
F E
E

1.00
1.45
20 TPI RH
2.49 Stepped
DETAIL F Spiral
SCALE 5 : 1
0.93
3.74 2 starts
0.190 CW .017 TPI
2 thread Starts

Insert thermocouple
here

91
0.284

Back View 30
° 3.500 Radius
Convex

E-E (3 : 1)
NAME DATE
DIMENSIONS ARE IN INCHES
TOLERANCES: DRAWN
FRACTIONAL 1/32
CHECKED
ANGULAR: +/- .5 Degrees
PROPRIETARY AND CONFIDENTIAL TWO PLACE DECIMAL .01 ENG APPR.
MegaStir TECHNOLOGIES IS A BUSINESS
SP 6/10/05
THREE PLACE DECIMAL .001
MFG APPR.
ALLIANCE OF ADVANCED METAL
PRODUCTS AND SII MEGADIAMOND. MATERIAL Q.A.
THE INFORMATION CONTAINED IN THIS MS80 COMMENTS: DESCRIPTION

Figure C.1: The FSW tool drawing number E44016.

DRAWING IS THE SOLE PROPERTY OF
Appendix C: FSW Tool Drawing E44016

E44016 Customer Print

ADVANCED METAL PRODUCTS AND FINISH
SII MEGADIAMOND. ANY REPRODUCTION
-- SIZE REV.
A DWG. NO.E44016
IN PART OR AS A WHOLE WITHOUT
WRITTEN PERMISSION IS PROHIBITED. DO NOT SCALE DRAWING SHEET 3 OF 3
SCALE:1:1 WEIGHT:
92
Appendix D: Hickory Code Changes
This document details the modifications made to the Hickory code. The file names
of the modified files have been listed along with a brief summary of the changes and
reasons behind each change.

hickory.F

• Commented the code for better understanding.

• Modified the iteration routine to rotate through the following function: vis-
cflow, galev, and strlyn from the beginning. This improves convergence.

• Added code to cause the post processing to take place even if the maximum
number of iterations is reached. This gives a visual representation of the
current solution even if the convergence criteria is not met.

• All post processing calls were placed in a new function called postprocess.

indat.F

• A subroutine called getinputline was added by C. D. Sorensen. This parses

the input file and allows for comments in the input file.

• An option was added to allow for non-uniform heat generation. This was
done to simulate the heat that is conducted out of the work piece by using
a negative heat generation proportional to the temperature. Currently, it is
thought that convergence issue limit the effectiveness of this function.

file-ops.F

• File creation was corrected to create dynamic and unique file unit number for
several files. example: .supg1 etc.

• Several output files were created for better understanding:

filename.corner - The corner nodes are output to this file.

filename.strlen - The streamline lengths are output to this file and use when
plotting the streamlines in Matlab.
filename.check - This file was used to write many different things to check
what was taking place in the code

93
 filename.converge - Convergence information is written to this file. The file
can be parsed and graphed to display the convergence history throughout
the run.
filename.Corrector - The predictor corrector method used in the code was
checked using this file.
filename.shape - The accuracy of the shape functions were checked using this
file.
filename.shapedata - This file can be used to graph variable changes along
the streamline by using the Matlab function oneline.
• Each of the user prompts for file names were hard coded in as “steel”. Thus
each run should be labeled by the directory that contains the files.
streamline.F
• The code was commented for better understanding.
• The element order is sorted and reordered by the average X coordinate of
all the nodes in the element. Thus, the state variables for elements with the
smallest X-values will be calculated first. In the FSW case the mesh is solved
from left to right. This was done with the hope of allowing fast streamline
calculations. See faster calculations below for more detail.
• Modified bkstp to handle the streamlines when the code is both running and
post processing the streamlines. This eliminates possible coding errors due
to duplicated functions for the same purpose. No more need for the post
processing sltrackpt function, because bkstp now does this work as well. The
function bkstp now needs an input xie and eta for the very first step of the
streamline because post processing streamlines start on the boundary and
fndele fails to find the correct element.
• All streamline loop checks were moved within bkstp. The streamline type is
flagged and returned.
• The loop and spiral checks were changed. It sums the angle made between
each streamline segment and checks that it is less then 2-pie. If greater than
2-pie it is flagged as a loop/spiral.
• A few variable names were changed for better understanding.
• The constant saturation value called by the function compute ss sat was com-
mented out and the saturation values are found using the fisher factor with
the function compute sat sv1. However, experimentation with compute ss sat
may still be helpful.
• When a streamline returns to previously entered element it is no longer flagged
as a loop in the function fndele. All streamline loop checks were moved into
the function bkstp.
94
• Minor fixes were made to enable faster streamline calculations. Rather than
stepping all the way up to the inlet to obtain the initial state variable value,
steps were made until an element was entered that had already calculated the
state variable values. Using the shape functions and the nodal values of that
element, the current streamline starting state variable value was found. This
is enabled by using the variable iopt. However, this method for fast streamline
calculations does not work because the local nodal points are only transferred
to global nodal points at the end of function strlyn and inside of gsmooth.
Thus, for this idea to work, the current element and all of its neighboring
elements need to have calculated state variable values that are transfer to
global nodal point values. These global nodal point values could then be used
to generate the first state variable value along the streamline.

• The function precor had an “if” statement that would not allow for thermal
softening once the saturation value has been reached. This was changed to
allow for thermal softening

• The function slwrite outputs more streamline variables to the iunitvgrad file
for better inspection and understanding.

• The function slbndry was modified to generate streamlines that are equally
spaced.

• The function sldeformation was written to do a binary search for the stream-
line with a critical change in the Y position along the streamline. This value
was flagged and output to a file as the deformation size of the weld. How-
ever, Matlab is now used to inspect this deformation size using the X-velocity
gradient in the Y-direction. Thus, this function is no longer needed.

• The function saturation was created to calculate the saturation values every-
where. However, other functions already do this and it is currently an un-
needed function.

sortbd.F

• Code was added to output: the boundary numbers, element numbers and
corner node numbers to file filename.corner

*.h

• “Implicit None” was added to all files ending in .h. This was done to unsure
that no errors were taking place by variables that were mistyped and thus
implicitly created. This also required variable definitions to be added before
the common area in each *.h file for it to compile.

viscflo.F

• The code was commented for better understanding.

95
galev.F
• The code was commented for better understanding.
matlev.F
• Code was added to included a non-uniform heat generation option. This was
done in hopes of showing more accurate temperature contours by effectively
accounting for heat loss into the backing plate. A negative heat source is
applied proportionally to the temperature difference. Currently convergence
problems limit the effectiveness of this function.
errnorm.F
• The code was commented for better understanding.
gsmooth.F
• Under-relaxation is applied to the state variable to help prevent limit cycling
and improve convergence.
evvar.F
• Under-relaxation is applied to the state variable to help prevent limit cycling
and improve convergence.
npquant.F
• First the call to npquant was added to several functions to ensure that the
correct local element properties were being used in each of the functions.

• The variable lastcall was added to ensure that the function was not being run
multiple times unnecessarily for the same element.
friction-models.F
• The first relative velocity friction model has divided the the true input velocity
by 5. This was done so all input files would not need to be changed.

• Several experimental friction models were added.

cleanup
• Non-essential write statements were commented out to decrease the output
file sizes.
Architecture Changes Needed to compile
• HP-ux platform needed a -DALPHA LINUX flag added to CFlAGS.

• All manually created file unit numbers were changed to be automatically gen-
erated by the function newunit.
96
• All file unit numbers need to be below 99. The function newunit was modified
to return unit values above 50 and below 99.

• Mac os X platform needed a“-I.” flag in the hsupg.x compile command in order
to handle the included files correctly. This “-I.” flag causes it look in the cur-
rent directory for the needed include file. See the following example. hsupg.x:
$(OBJECTS) $(MAIN.o) $(FC) -o $@ $(FFLAGS) $(MAIN) $(OBJECTS)
-I. -lc

• Also the malloc function is the stdlib.h on a mac. Therefore, “#include

<stdlib.h>” need to be added to nrutil.c and reorder.c. Also “#include
<malloc.h>” and “<alloc.h>” should be commented out or excluded.

Other Convenient Changes

• The -ffixed-line-length-none flag was added to FFLAGS. This removes the 72

character length restriction on each line of code.

97
98
Appendix E: Optimization
#! / u s r / l o c a l / b i n / b a s h
#
# S c r i p t used w i t h OptdesX t o o p t i m i z e t h e Hickory Code
# By : B l a k e Owen

# Summary : Read i n t h e new v a l u e s c r e a t e d by OptdesX and

# read i n t h e v a l u e s l a s t used . Search and r e p l a c e t h e
# needed i n f o r m a t i o n i n t h e Hickory i n p u t f i l e s ( s t e e l . i d ) .
# Batch P r o c e s s a l l t h e needed runs t o maryloux . Every so
# o f t e n c h e c k Hickory ’ s o u t p u t f i l e t o know when a l l t h e
# b a t c h e s have f i n i s h e d . When f i n i s h e d run t h e Matlab f i l e
# t h a t c a l c u l a t e s t h e e r r o r between t h e n u m e r i c a l and t h e
# e x p e r i m e n t a l r e s u l t s . Matlab w r i t e s t h i s e r r o r t o a f i l e
# f o r OptdesX t o read i n . Based upon t h i s v a l u e OptdesX
# w i l l a d j u s t t h e p a r a m e t e r s and t h i s s c r i p t w i l l be c a l l e d
# again .

# Bach j o b s −y e s do not−no
batch=” y e s ”
#b a t c h=”no”

f i n i s h=” y e s ”
#f i n i s h =”no”

# Pause time between c h e c k s , i n s e c o n d s

s l e e p t i m e =600

# Number o f p a r a m e t e r s t h a t w i l l be changed
num parameters=9

# l i n e numbers t o s e a r c h and r e p l a c e d a t a f i r s t i s t h e z e r o

# must match t h e num parameters + t h e z e r o a t t h e b e g i n n i n g

l i n e n u m b e r =( 0 143 143 143 145 145 145 145 146 185 )

a w k p o s i t i o n =( 0 1 2 3 2 3 4 5 1 1 )
99
s t a r t l o c a t i o n=${HOME}/FSW/ study3 / o p t i m i z e 3
r u n l o c a t i o n=${HOME}/FSW/ study3
#l o o p t h r o u g h =” l s −d f f ∗” # Run a l l t h e d i r e c t o r i e s
l o o p t h r o u g h=” l s −d f f 1 1 f f 1 3 f f 2 2 f f 3 1 f f 3 3 ” # Run s p e c i f i e d
#l o o p t h r o u g h =” l s −d f f 1 1 f f 1 3 ” # Run s p e c i f i e d d i r e c t o r i e s

# S t a r t from t h e r i g h t d i r e c t o r y
cd ${ s t a r t l o c a t i o n }

# Open t h e i n p u t n e w f i l e and read i n t h e new v a l u e s

exec 6<&0
exec < input new . t x t
# Read and s t o r e v a l u e s from t h e i n p u t f i l e
f o r ( ( i =1; i<=num parameters ; i +=1));
do
# Read a l l l i n e s o n l y s t o r e e v e r y o t h e r l i n e
read linenumber [ ${ i } ]
read newvalue [ ${ i } ]
echo ”New ”${ i } : ${ newvalue [ ${ i } ] }
done
# Close the input new f i l e
exec 0<&6 6<&−

# Loop t h r o u g h needed f o l d e r , change i n p u t f i l e s , and run

# t h e code
cd ${ r u n l o c a t i o n }
f o r i i n ‘ ${ l o o p t h r o u g h } ‘
do
echo ${ i }
cd ${ r u n l o c a t i o n }/ ${ i }

# I c o u l d no g e t ’ awk ’ and ’ p r i n t ’ t o work i n a f o r loop

# t h u s we g e t t h e f o l l o w i n g .
o l d v a l u e [ 1 ] = $ ( sed −n ${ l i n e n u m b e r [ 1 ] } p s t e e l . i d | \
awk ’ { print $1 } ’ )
o l d v a l u e [ 2 ] = $ ( sed −n ${ l i n e n u m b e r [ 2 ] } p s t e e l . i d | \
awk ’ { print $2 } ’ )
o l d v a l u e [ 3 ] = $ ( sed −n ${ l i n e n u m b e r [ 3 ] } p s t e e l . i d | \
awk ’ { print $3 } ’ )
o l d v a l u e [ 4 ] = $ ( sed −n ${ l i n e n u m b e r [ 4 ] } p s t e e l . i d | \
awk ’ { print $2 } ’ )
o l d v a l u e [ 5 ] = $ ( sed −n ${ l i n e n u m b e r [ 5 ] } p s t e e l . i d | \
awk ’ { print $3 } ’ )

100
 o l d v a l u e [ 6 ] = $ ( sed −n ${ l i n e n u m b e r [ 6 ] } p s t e e l . i d | \
awk ’ { print $4 } ’ )
o l d v a l u e [ 7 ] = $ ( sed −n ${ l i n e n u m b e r [ 7 ] } p s t e e l . i d | \
awk ’ { print $5 } ’ )
o l d v a l u e [ 8 ] = $ ( sed −n ${ l i n e n u m b e r [ 8 ] } p s t e e l . i d | \
awk ’ { print $1 } ’ )
o l d v a l u e [ 9 ] = $ ( sed −n ${ l i n e n u m b e r [ 9 ] } p s t e e l . i d | \
awk ’ { print $1 } ’ )
# Update t h e Hi c kory i n p u t f i l e w i t h t h e new v a l u e s
# P o s i b l e e r r o r ! ! ! ! i f two numbers are t h e same
# This might be done b e t t e r u s i n g P e r l
# Simple f i x w r i t e t h e numbers a d i f f e r e n t l e n g t h
f o r ( ( k=1;k<=num parameters ; k +=1));
do
sed −e ${ l i n e n u m b e r [ ${k } ] } s /${ o l d v a l u e [ ${k } ] } / \
${ newvalue [ ${k } ] } / s t e e l . i d > s t e e l . f i x
echo ${ l i n e n u m b e r [ ${k } ] } s /${ o l d v a l u e [ ${k } ] } / \
${ newvalue [ ${k } ] } /
cp s t e e l . f i x s t e e l . i d
rm s t e e l . f i x
done
# Run t h e H i ck o ry Code
# run t h e code on maryloux b a t c h system
# i f y e s b a t c h t h e run i f not do not
i f [ [ ” $batch ” == y e s ] ]
then
make c l e a n ;
#cp ˜/FSW/ h i c k o r y /h/ hsupg . x . ;
#echo ; echo c o p i e d ; echo ;
sleep 5
s s h maryloux ” hostname ; cd ${ r u n l o c a t i o n }/ ${ i } ; \
pwd ; qsub run . sh ; ”
fi
done

i f [ [ ” $ f i n i s h ” == y e s ] ]
then
# Check t h e s t a t u s o f t h e b a t c h j o b s e v e r y few minutes
# When f i n i s h e d run matlab t o c a l c u l a t e t h e e r r o r
# 1=not done 0=done
r u n s t a t u s =1
l o o p c o u n t=0
s e c =60

101
t o t a l t i m e =0
( ( m i n s l e e p t i m e=s l e e p t i m e / s e c ) )

sleep 5

while ( ( r u n s t a t u s != 0 ) )
do
( ( l o o p c o u n t +=1))
echo
echo
i f (( loop count < 2 ))
then
echo S l e e p f o r 10 s e c o n d s b e f o r e next check
s l e e p 10
else
( ( t o t a l t i m e=t o t a l t i m e+s l e e p t i m e / s e c ) )
echo S l e e p f o r ${ m i n s l e e p t i m e } minutes b e f o r e next \
check
echo Estimated t o t a l run time : ${ t o t a l t i m e } minutes
s l e e p ${ s l e e p t i m e }
fi

count=0
c o n v e r g e d=0
n o c o n v e r g e d=0
n o t f i n i s h e d =0
n o t c r e a t e d =0
f a i l e d =0

# Check which j o b s a re s t i l l running on t h e b a t c h system

running=$ ( s s h maryloux ” q s t a t | grep cbo4 ” | \
awk ’ { print $3 } ’ ) # i f above command d i d not r e t u r n a n y t h i n
none
i f [ −n ” $ru nning ” ]
then
echo run ning c o n t a i n s : ${ running }
else
# I f n u l l s e t v a l u e t o none t o o b v o i d p r o b l e m s
run ning=” none ”
echo run ning c o n t a i n s : ${ running }
fi

cd ${ s t a r t l o c a t i o n }

102
# Write t h e r e s u t s t o a f i l e
exec 7>&1
exec > s t a t u s c u r r e n t . t x t
echo ’ ’ # Add a b l a n k l i n e
cd ${ r u n l o c a t i o n }
# Loop t h r o u g h and c h e c k each j o g s s t a t u s
f o r i i n ‘ ${ l o o p t h r o u g h } ‘
do
cd ${ r u n l o c a t i o n }/ ${ i }

( ( count +=1))
# I f t h e f i l e s t e e l . done i s c r e a t e d
i f [ [ −e s t e e l . done ] ]
then
# Open t h e s t e e l . done f i l e
exec 4<&0
exec < s t e e l . done
# Check i f t h e f i l e c o n v e r g e d
read r e s u l t
read r e s u l t
read r e s u l t
read i t e r a t i o n
read i t e r a t i o n
# Close the input f i l e
exec 0<&4 4<&−

# Go b a c k t o t h e s t a r t d i r e c t o r y
cd ${ s t a r t l o c a t i o n }

i f ( ( r e s u l t == 2 ) )
then
echo ${ i } CONVERGED ${ i t e r a t i o n }
( ( c o n v e r g e d +=1))
e l i f ( ( r e s u l t == 1 ) )
then
echo ${ i } NOT CONVERGED ${ i t e r a t i o n }
( ( n o c o n v e r g e d +=1))
else
# Checks i f t h e run i s s t i l l running
i f [ [ ” $running ” == ∗ $ i ∗ ] ]
then
echo ${ i } not f i n i s h e d
( ( n o t f i n i s h e d +=1))
else

103
 echo ${ i } RUN FAILED
( ( f a i l e d +=1))
fi
fi
else
echo ${ i } f i l e not c r e a t e d y e t
( ( n o t c r e a t e d=n o t c r e a t e d +1))
fi
done

# Print out the s t a t u s of the check

echo ’ Converged : ’ ${ c o n v e r g e d } ’ o f ’ ${ count }
echo ’ Not Converged : ’ ${ n o c o n v e r g e d } ’ o f ’ ${ count }
echo ’ F a i l e d : ’ ${ f a i l e d } ’ o f ’ ${ count }
echo ’ Not F i n i s h e d : ’ ${ n o t f i n i s h e d } ’ o f ’ ${ count }
echo ’ Not Created : ’ ${ n o t c r e a t e d } ’ o f ’ ${ count }
echo Estimated t o t a l run time : ${ t o t a l t i m e } minutes
echo

# Close the output f i l e

exec 1>&7 7>&−

i f ( ( c o n v e r g e d+n o c o n v e r g e d+f a i l e d == count

&& t o t a l t i m e > 3 0 ) )
then
# Set v a r i a b l e to s t e p out of the while loop
r u n s t a t u s =0
# Go b a c k t o t h e s t a r t d i r e c t o r y
cd ${ s t a r t l o c a t i o n }
# Write t h e run run r e s u l t s t o a f i l e
# Amend each f i l e s r e s u l t t o one f i l e
i f [ [ −e s t a t u s a l l . t x t ] ]
then
cat s t a t u s a l l . t x t s t a t u s c u r r e n t . t x t \
> junk1
mv junk1 s t a t u s a l l . t x t
else
cp s t a t u s c u r r e n t . t x t s t a t u s a l l . t x t
fi
fi
done

# Go bac k t o t h e s t a r t d i r e c t o r y

104
cd ${ s t a r t l o c a t i o n }

if (( failed > 0 ))
then
#Run t h e F a i l e d Error Code
#S e t t h e Error t o a l a r g e v a l u e
echo ONE OR MORE RUNS FAILED
echo F a i l u r e S c r i p t e x e c u t e d
# Write a l a r g e e r r o r t o t h e o u t p u t
exec 6>&1
exec > o u t p u t e r r o r . t x t
echo 98888888 988888888
exec 1>&6 6>&−
else
# Run m a t l a b and c a l c u l a t e t h e num . exp . e r r o r
# matlab w r i t e s t h i s error to a f i l e
echo Matlab c a l l e d t o c a l c u l a t e th e e r r o r
matlab −nojvm −r a l l e r r o r 3

# Remove o l d copy and copy t h e new r e s u l t s t o

# s t u d y t e m p f o r temporary v i e w i n g w h i l e t h e code
# c o n t i n u e s t o run
cd ${ r u n l o c a t i o n }
cd . . /
rm −r study temp3
cp −r study3 study temp3
fi

fi

echo
echo The s c r i p t has f i n i s h e d
echo
exit #

105
106