Académique Documents
Professionnel Documents
Culture Documents
To cite this article: Xuan-Tan Pham (2013): Two-Dimensional Rosenthal Moving Heat Source Analysis
Using the Meshless Element Free Galerkin Method, Numerical Heat Transfer, Part A: Applications: An
International Journal of Computation and Methodology, 63:11, 807-823
To link to this article: http://dx.doi.org/10.1080/10407782.2013.757089
Xuan-Tan Pham
Department of Mechanical Engineering, Ecole de technologie superieure,
Montreal, Quebec, Canada
The quasi stationary-state solution of the two-dimensional Rosenthal equation for a moving
heat source using the meshless element free Galerkin method is studied in this article.
Node-based moving least square approximants are used to approximate the temperature
eld. Essential boundary conditions are enforced by using Lagrange multipliers. A Gaussian
surface heat source is used for the modeling of the moving heat source. The results obtained
for a two-dimensional model are compared with the results of the nite-element method.
INTRODUCTION
Many manufacturing processes such as fusion welding, heat treatment, or
forming by line heating involve the use of moving heat sources. The studies of
Rosenthal on the heat distribution during arc welding and cutting [1, 2] are considered as the earliest work to solve analytically the problem of moving heat sources.
The works of Rosenthal established the foundation to solve this type of problem in
many research studies in this eld later by both analytical and numerical methods
[39]. The modeling of moving heat sources and deformation of work pieces using
the nite element method (FEM) has been studied by many authors [1012]. To have
a good prediction, this method requires a ne mesh for the heat sources as well as
when the deformation is critical. Particularly when elements are distorted, remeshing
is required to avoid the problem of singularity. Adaptive mesh renement is generally used to solve these problems. However, this method takes a lot of time for
remeshing and projecting results on the new meshes. Meshless methods are new
trends and recently developed to avoid these difculties by using node-based
approximation rather than element-based approximation, such as the smoothed
particle hydrodynamic (SPH) for heat and uid analysis [1315] as well as for solid
mechanics [1618], the meshless local Petrov-Galerkin (MLPG) method [19], the diffuse nite element method [20], the meshless element free Galerkin (MEFG) method
[21], the meshless nite-difference method [22], a space-time meshless method [23],
Received 28 September 2012; accepted 17 November 2012.
cole de
Address correspondence to Xuan-Tan Pham, Department of Mechanical Engineering, E
technologie superieure, 1100, rue Notre-Dame Ouest, Montreal, Quebec H3C1K3, Canada. E-mail:
tan.pham@etsmtl.ca
807
808
X.-T. PHAM
NOMENCLATURE
a(x)
c
dmax
dmI
dI
h
k
n
NI(s)
p(x)
coefcient vector
specic heat, J=kg= C
scaling parameter
size of the support domain of node I, mm
distance between point x and node I, mm
convective heat transfer coefcient,
W=mm2= C
thermal conductivity coefcient tensor,
W=mm= C
surface normal vector
Lagrange interpolant
complete monomial basis
q
Q
QG
t
w
wI(x)
h
h1
_I
h
k
q
UI(x)
and so on. Among these methods, the MEFG method has been successfully used to
solve many different problems, particularly for heat transfer analysis. Singh et al.
[24] studied the heat transfer of two-dimensional ns using the MEFG method.
Three-dimensional heat transfer problems for orthotropic materials were also solved
using the MEFG method by Singh [25]. A numerical solution for temperaturedependent thermal conductivity problems was obtained by Singh et al. [26] when
applying the MEFG method. However, very little work on the modeling of material
processing involving moving heat sources by the MEFG method has been reported.
In this article, the modeling of a two-dimensional Rosenthal moving heat source
using the MEFG method was performed. The moving least square approximant
(MLS) was used in this study. Both regular and irregular meshes were employed
to study the inuence of the support domain on the approximation. Numerical
results of temperature distribution obtained by the MEFG method were compared
with the solutions of the nite element analysis. Very good agreements between these
two methods were found in this study.
ROSENTHAL MOVING HEAT SOURCE FORMULATION
The heat transfer equation [27] is dened by
qc
qh
Q r:q
qt
where h is the temperature, t is the time, q is the density, c is the specic heat, Q is the
volume heat rate, and q is the heat ux vector of heat conduction in solids [28] determined by the Fourier law as
~h
q k:r
809
qc
qh
qh
vqc kr2 h Q 0
qt
qx
qh
kr2 h Q 0
qx
qh
kr2 h QgdV 0
qx
When choosing the test function w 0 on the essential boundary Ch, Eq. (5)
reduces to
Z
V
wkr2 hdV
wqn dC
Cqn
w;i qi dV
V
where Cqn is the natural boundary for heat ux and=or convection. By replacing
Eq. (6) into Eq. (5), we get the weak form of the Rosenthal equation, as follows:
Z
Z
Z
Z
qh
w;i qi dV
wvqc dV
wqn dC
wQdV
7
qx
V
V
Cqn
V
MEFG METHOD
Moving Least Square Approximation
_1 _2
_N
810
X.-T. PHAM
hh x; xI pT xI ax;
I 1; 2; . . . ; N
where pT(x) [p1(x), p2(x), . . ., pm(x)] is a complete monomial basis and m is the
_I
number of terms in the basis, a(x) is the coefcient vector and h hxI is the
nodal parameter of the eld variable at node I. The coefcient vector a(x) is determined by minimizing a weight discrete L2 norm, which is dened as the following
equation:
Jax
N
X
_I
wI xpT xI ax h 2
I1
where wI(x) w(x xI) is the positive weight function at node I, which quickly
decreases as the distance jjx xIjj increases to preserve the local character of
the approximation. The minimum of Eq. (9) leads to the following set of linear
equations:
_
ax A1 x Bx h
10
where
Ax
N
X
wI xpxI pT xI
11
_1 _2
_N
h h h . . . h
12
13
By substituting Eq. (10) into Eq.(8), the MLS approximant has the usual
following form:
_h
h x pT x A1 x Bx h Ux h
14
15
m X
m
X
j1 k1
pj xA1
jk xBkI x
16
811
Weight Function
Weight functions play an important role in the MEFG method. They should be
constructed according to some conditions as mentioned in references [21, 30]. The
weight function wI(x) should be nonzero over a small neighbourhood of xI, called
the domain of inuence of node I or also the compact support, in order to generate
a set of sparse discrete equations. In this article, the cubic spline weight function [21,
30] was used. This function is dened as follows:
8
< 23 4r2 4r3
wI x wx xI 4 4r 4r2 4r3
3
:3
0
ifr 0:5
if 0:5 < r 1
if r > 1
17
where r dI=dmI is the distance ratio, dI jjx xIjj is the distance between the point
x and node I, and dmI is the size of the support domain at node I. The form of the
cubic spline function and its rst derivative are presented in Figure 1.
kT ChdX
Z
X
dkT ChdX
kT dChdX
18
812
X.-T. PHAM
wT;i qi dV
w qn dC
Cqn
dkT ChdX
wT QdV
kT dChdX 0
19
20
dCh dh
h dh !
! kT dCh kT dh dhT k
21
By using Eqs. (20) and (21), the constrained weak form has the following form:
Z
V
wT;i qi dV
w qn dC
wT QdV
V
dkT h
hdC
Ch
dhT kdC 0
22
Ch
Then, by taking into account the heat ux and convection boundary conditions, the
weak form nally can be obtained as
Z
V
wT;i qi dV
wT q0n dC
Z
Ch
Z
Ch
Cq
wT qhh dC
dk h hdC
T
dhT kdC
Ch
Ch
wT q0h dC
wT QdV
23
where the boundary is composed by qn q0n qhh q0h in which q0n is the constant
heat ux on the boundary Cq; qhh hh and q0h hh1 are two components of the
convection boundary condition on Ch.
Discretization
The shape function for the temperature can be approximated by the following:
hx; t
N
X
K1
24
813
So,
N
qh X
qUK _
hK
qxi K1 qxi
_
< UK;i >1xN h K
h;i
25
with i 1; 2; 3
Nx1
N
X
_
_
qh
qi k
kh;i k
UK;i h K k < UK;i > h K
qxi
K1
26
with i 1; 2; 3
On the other hand, the test function is a virtual variation of the shape function, as
follows:
N
X
w dh
27
K1
N
X
28
dh K
Nx1
K1
kx
Nk
X
29
I1
where Nk is the number of nodes used for this interpolation, s is the arc-length along
the essential boundary, kI is the Lagrange multiplier at node I on the essential
boundary, and NI (s) can be a Lagrange interpolant used in the conventional nite
element method.
Finally, the discretized form of Eq. (23) has the following form:
_T
30
Kv Kk Kh h Gk qn qh Qr
_
GT h qh
31
814
X.-T. PHAM
< px >< 1
32
y>
33
8
9
< 1 =
wx xN xN
:
;
yN
34
where x (x, y), xI (xI, yI) are, respectively, the coordinates of the point x and of
the node I and the weight function is the cubic spline function dened by Eq. (17).
The weight function at any given point can be determined by a tensor product
weight dened as
wx xI wrx :wry
35
where rx (jjx xIjj)=dmxI, ry (jjy yIjj)=dmyI and dmxI, dmyI are the size of the
support domain at node I, respectively, in x and y directions. The values of dmxI
and dmyI are chosen so that the matrix is nonsingular everywhere. Here, they are
dened by dmxI dmax cxI and dmyI dmax cyI, where dmax is a scaling parameter
and cxI and cyI are the distances, respectively, in x and y directions from node I to the
nearest neighbors.
Gaussian Moving Heat Source
The distribution of the heat rate of surface sources QG (W=mm2) is approximately described by Gausss normal distribution law as
QG r Q0 expr2 =r20
36
where Q0 (W=mm2) is the maximum heat ux in the center of the heat spot, r (mm) is
the spot radius, and r0 (mm) is the radius of the effective circular area through which
the total power of a Gaussian source would be transferred if it were uniformly distributed with a ux equal to the real maximum ux Q0.
NUMERICAL RESULTS AND DISCUSSION
In-house research codes were developed for both the FEM and MEFG methods in which the Rosenthal formulation was implemented. In the FEM code, the
815
Figure 2. Two-dimensional model for Rosenthal moving heat analysis (color gure available online).
Lagrange interpolant was used for interpolation; meanwhile, the MLS approximant
was employed for the MEFG program. Other parameters such as the boundary
conditions, the moving heat source, and its velocity were kept the same for both
FEM and MEFG simulations. The Rosenthal moving heat analyses have been carried out using a two-dimensional model, as shown in Figure 2, where the heat source
moves from the right side to the left side with the velocity v. The essential boundary
condition (T T0) is imposed on the left side of the plate. A constant heat ux
(q q0) is imposed on the three other sides of the plate.
On the one hand, as the essential boundary conditions are imposed by using
the Lagrange multiplier together with the MLS approximant, a series of numerical
tests with different values of dmax were performed to study the inuence of this parameter on the accuracy of the essential boundary condition. The type and values of
the boundary conditions are described in Table 1. There is no moving heat source
in these tests. Both regular and irregular meshes were used. The regular triangular
mesh has 7,381 nodes and 14,400 elements and the irregular triangular mesh has
1,508 nodes and 2,932 elements. Numerical results showed that for both the regular
Table 1. Boundary conditions
Left side
T ( C)
20
Right side
Upper side
q0 (W=mm )
0.001
q0 (W=mm )
0.001
Lower side
q0 (W=mm2)
0.001
Figure 3. Temperature distribution (MEFG, dmax 1.5). a) regular mesh and b) irregular mesh (color
gure available online).
816
X.-T. PHAM
Values of parameters
Length (L)
Width (B)
Thickness (H)
Specic heat (c)
Density of the material (q)
Thermal conductivity (k)
Type of the heat source
Maximum heat rate (Q)
Radius of heat source (r0)
Velocity of the heat source (v)
100 mm
50 mm
1 mm
658 J=kg= C
7.6 106 kg=mm3
0.025 W=mm= C
Gaussian
5 W=mm2
2 mm
2 mm=s
and irregular meshes, the values of dmax from 1.5 to 1.7 gave the exact values of the
temperature imposed on the essential boundary as depicted in Figure 3. On the other
hand, the two-dimensional quasi-stationary Rosenthal moving heat analyses were
carried out by both the FEM and MEFG methods with the parameters indicated
in Table 2. The cubic spline weight function and the value dmax 1.5 were used
for the computation with the MEFG method. Figures 4 and 5 present the temperature distributions obtained by the MEFG and FEM methods for the regular and
Figure 4. Temperature distribution. a) MEFG-regular mesh and b) FEM-regular mesh (color gure
available online).
Figure 5. Temperature distribution. a) MEFG-irregular mesh and b) FEM-irregular mesh (color gure
available online).
817
Figure 6. Temperature comparison (FEM-MEFG regular mesh) (color gure available online).
818
X.-T. PHAM
Table 3. Temperature comparison in the case of regular mesh
Regular mesh
MEFG
FEM
Difference
x(mm)
T ( C)
T ( C)
(%)
50.00
40.00
30.00
20.00
10.00
0.00
10.00
20.00
30.00
40.00
50.00
20.01
20.00
19.99
20.04
24.56
483.80
251.29
188.30
159.28
142.21
133.06
20.00
20.00
19.99
20.04
24.49
474.41
251.37
188.34
159.30
142.21
133.05
0.07%
0.00%
0.00%
0.01%
0.31%
1.98%
0.03%
0.02%
0.01%
0.00%
0.01%
FEM
Difference
x(mm)
T ( C)
T ( C)
(%)
50.00
39.93
29.86
19.78
9.71
0.36
11.15
20.50
29.86
39.93
50.00
20.00
20.00
19.99
20.04
24.96
485.93
231.89
180.26
154.59
137.85
128.97
20.00
20.00
19.99
20.05
25.30
488.24
234.77
182.54
156.47
139.56
130.54
0.01%
0.00%
0.00%
0.01%
1.33%
0.47%
1.22%
1.25%
1.20%
1.23%
1.20%
as depicted in Table 3 and Table 4. The differences between two methods are very
small except at the center of the heat source, where there is a discrepancy of about
2% and 1.3% for the regular and irregular meshes, respectively. It was observed
that there is no singularity in the simulation with the MEFG method; the essential
boundary conditions are well imposed, and there is very good similarity between
FEM and MEFG results.
CONCLUSION
The results in this study showed that the two-dimensional Rosenthal moving
heat analysis can be successfully modeled using the MEFG method. Good agreement between the numerical results computed by both FEM and MEFG was
obtained. The following points were also observed:
819
. In the MEFG method, the boundary conditions were well imposed using the
Lagrange multiplier with the scaling parameter dmax having the range from 1.5
to 1.7 and for both regular and irregular triangular meshes.
. The MLS approximant gave good results in comparison to the FEM method
using the Lagrange interpolation for a moving heat source problem.
. The use of the irregular triangular mesh for the MEFG method showed the
robustness of the algorithm used in this article, which automatically denes the
support domain based on the irregular distribution of nodes in the mesh. It is very
promising for further applications.
REFERENCES
1. D. Rosenthal, Mathematical Theory of Heat Distribution in Welding and Cutting,
Welding J., vol. 20, no. 5, pp. 220234, 1941.
2. D. Rosenthal, The Theory of Moving Source of Heat And its Applications to Metal
Treatments, Trans. of the American Society of Mechanical Engineers, vol. 68, pp.
849865, 1946.
3. N. Christensen, V. de L. Davies, and K. Gjermundsen, Distribution of Temperatures in
Arc Welding, British Welding J., vol. 12, no. 54, pp. 5475, 1965.
4. R. Weichert and K. Schonert, Temperature Distribution Produced by a Moving Heat
Source, Mech. Appl. Math., vol. 31, pt. 3, pp. 363379, 1978.
5. T. W. Eager and N. S. Tsai, Temperature Fields Produced by Traveling Distributed Heat
Sources, Welding J., vol. 62, no. 12, pp. 346355, 1983.
6. J. Goldak, A. Chakravarti, and M. Bibby, A New Finite Element Model for Welding Heat
Sources, Metallurgical Trans. B, vol. 15B, pp. 299305, 1994.
7. N. T. Nguyen, A. Ohta, K. Matsuoka, N. Suzuki, and Y. Maeda, Analytical Solutions for
Transient Temperature of Semi-Innite Body Subjected to 3-D Moving Heat Sources,
Supplement to the Welding J., pp. 265274, 1999.
8. R. Komaduri, and Z. B. Hou, Thermal Analysis of the Arc Welding Process: Part I. General Solutions, Metallurgical and Materials Trans. B, vol. 31, no. 6, pp. 13531370, 2000.
9. R. Komaduri, and Z. B. Hou, Thermal Analysis of the Arc Welding Process: Part II.
Effect of Variation of Thermophysical Properties with Temperature, Metallurgical and
Materials Trans. B, vol. 32, no. 3, pp. 483499, 2001.
10. J. Ascough, A Single Step Finite Element Analysis of the Temperature Distribution
around a Moving Laser Heat Source, Optics and Laser in Engi., vol. 6, pp. 137143, 1985.
11. C.-Y. Yang, The Determination of Two Moving Heat Sources in Two-Dimensional
Inverse Heat Problem, Applied Mathematical Modelling, vol. 30, pp. 278292, 2006.
12. A. Nisar, M. J. J. Schmidt, M. A. Sheikh, and L. Li, Three-Dimensional Transient Finite
Element Analysis of the Laser Enamelling Process and Moving Heat Source and Phase
Change Considerations, Proc. of the Institution of Mechanical Engineers (Proc. Instn.
Mech. Engrs.), Part B: J. of Engi. Manufacture, vol. 217, pp. 753764, 2003.
13. J. J. Monaghan, An Introduction to SPH, Comput. Phys. Commn., vol. 48, pp. 8996,
1988.
14. J. J. Monaghan, Simulating Free Surface Flows with SPH, J. Comput. Phys., vol. 110,
pp. 399406, 1994.
15. P. W. Cleary and J. J. Monaghan, Conduction Modelling using Smoothed Particle Hydrodynamics, J. of Computational Physics, vol. 148, pp. 227264, 1999.
16. L. D. Libersky, A. G. Petschek, T. C. Carney, J. R. Hipp, and F. A. Allahadi, High Strain
Lagrangian Hydrodynamics, J. Comput. Phys., vol. 109, pp. 6775, 1993.
820
X.-T. PHAM
17. W. Benz and E. Asphaug, Simulation of Brittle Solid using Smoothed Particle Hydrodynamics, Comput. Phys. Comm., pp. 25365, 1995.
18. J. P. Gray, J. J. Monaghan, and R. P. Swift, SPH Elastic Dynamics, Comput. Methods
Appl. Mech. Eng., vol. 190, pp. 66416662, 2001.
19. M. Shibahara and S. N. Atluri, The Meshless Local Petrov-Galerkin Method for the
Analysis of Heat Conduction due to a Moving Heat Source in Welding, Int. J. of Therm.
Sci., vol. 50, pp. 984992, 2011.
20. B. Nayroles, et al. Generalizing the Finite Element Method: Diffuse Approximation and
Diffuse Element, Comput. Mech., vol. 10, pp. 307318, 1992.
21. T. Belytschko, et al. Element-Free Galerkin Methods, Int. J. Numer. Meth. Eng., vol. 37,
pp. 229256, 1994.
22. C. Varanasi, J. Y. Murthy, and S. Mathur, Numerical Schemes for the ConvectionDiffusion Equation using a Meshless Finite-Difference Method, Numer. Heat Transfer
B, vol. 62, pp. 127, 2012.
23. T. Sophy, A. Da Silva, and A. Kribe`che, A Space-Time Meshless Method that Removes
Numerical Oscillations when Solving PDEs with High Discontinuities, Numer. Heat
Transfer B, vol. 62, pp. 5070, 2012.
24. I. V. Singh, K. Sandeep, and R. Prakash, Heat Transfer Analysis of Two-Dimensional
Fins using Meshless Element Free Galerkin Method, Numer. Heat Transfer A, vol. 44,
pp. 7384, 2003.
25. I. V. Sigh, Meshless EFG method in Three-Dimensional Heat Transfer Problems: A
Numerical Comparison, Cost and Error Analysis, Numer. Heat Transfer A, vol. 46,
pp. 199220, 2004.
26. A. Singh, I. V. Singh, and R. Prakash, Numerical Solution of Temperature-Dependent
Thermal Conductivity Problems using a Meshless Method, Numer. Heat Transfer A,
vol. 50, pp. 125145, 2006.
27. R. W. Lewis, et al. The Finite Element Method in Heat Transfer Analysis, pp. 113, John
Wiley & Sons, 1996.
28. H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids, pp. 68, Oxford, 1959.
29. P. Lancaster and K. Salkauskas, Surface Generated by Moving Least Squares Methods,
Math. Comput., vol. 37, no. 155, pp. 141158, 1981.
30. G. R. Liu, Mesh Free Methods: Moving Beyond The Finite Element Method chapt. 5,
pp. 83161, CRC Press, Boca Raton, 2003.
37
where a k=qc is the thermal diffusivity of the material (mm2=s). Now, for a point
source Q (W=mm2) that moves along the axis x with the speed v (mm=s), the heat
equation can be described in the coordinates which attach to this point source by
making coordinate changes as follows:
X x vt; Y y; Z z and S t
38
821
By referring to [1], from Eq. (38) the following relations can be obtained as
qh
qh qX
qh
qx qX qx qX
39
qh qh qY
qh
qy qY qy qY
40
qh qh qZ qh
qz qZ qz qZ
41
qh qh qX qh qS
qh qh
v
qt qX qt qS qt
qX qS
42
2 2
2
qX
qy
qz
a qX a qt k
43
By noting that the distinction between qh=qXand qh=qx can be ignored, this
equation can be reformulated as
qh
qh
vqc kr2 h Q 0
qt
qx
qc
44
APPENDIX B: DISCRETIZATION
Velocity Term
wT vqc
_
qh
dV < dh I >
qx
R
where (Kv)IJ UIvqcUJ, xdV.
V
Diffusion Term
V
_
< dh I > 4
fUI;i gqi dV
R
where (Kk)IJ UI, mkmnUJ,
V
dV.
822
X.-T. PHAM
Ch
Ch
_
Ch
< d h I > Kh f h J g
R
where Kh IJ
hUI UJ dC:
Ch
Ch
where qh I
_T
Ch
R
Ch
UI UJ q0hJ dC:
Cq
where qn I
_T
Cq
R
Cq
UI UJ q0nJ dC:
Ch
where GIJ
_T
Ch
R
Ch
UI NJ dC:
dkT h
hdC < dkI >
Ch
Ch
T
Ch
dk G h qh
T
where GIJ
R
Ch
NI UJ dC and qh I
R
Ch
NI hdC:
fNI ghdC
_T
823