Escolar Documentos
Profissional Documentos
Cultura Documentos
School of Energy and Power Engineering, Xian Jiaotong University, Xian 710049, China
Department of Mechanical and Aerospace Engineering, Utah State University, Logan, UT 84322, USA
a r t i c l e
i n f o
Article history:
Received 15 February 2015
Received in revised form 29 April 2015
Accepted 29 April 2015
Available online 15 May 2015
Keywords:
Moving particle semi-implicit method
Heat transfer
Heat conduction
Nature convection
Symmetry boundary model
a b s t r a c t
Numerical models of heat conduction and natural convection are developed based on Meshless Moving
particle semi-implicit (MPS) method to study the unsteady complex ow coupling with heat transfer.
Fundamental heat exchange process is systematically investigated in one dimension and extended to
two dimensions without internal heat source. The energy conservation equation is discretized on particles and coupled with ow by momentum change that caused by temperature difference. Both Dirichlet
and Neumann boundary conditions are proposed with mirror image technique and the numerical results
are veried to have high accuracy against benchmark solutions (theoretical or experimental, and or other
numerical solutions) for several representative examples. To improve the calculation efciency, symmetry boundary models are established and applied in transient heat conduction and natural convection
heat transfer. The results using the symmetry boundary are comparable and reliable to the
full-domain results, both of which agree well with the experimental and other numerical solutions.
2015 Elsevier Ltd. All rights reserved.
1. Introduction
Unsteady complex ow including free surface ow, phase
change and multiphase ow is usually accompanied by large deformation of the uid phase (such as crushing and merging).
Simulating these problems is usually a challenge for conventional
mesh-based numerical methods, because the grid reconstruction
can compromise computational efciency and accuracy to a large
extent. In addition, if the mentioned unsteady complex ow is coupled with heat transfer, namely the Unsteady Complex Heat-Fluid
Coupling (UC-HFC) problem, the uid ow and heat transfer would
interact with each other, and it would be rather complicated to
solve them integrally by any existed conventional numerical
methods.
However, Lagrangian particle method, such as the smooth particle hydrodynamics (SPH) method [1] and the Moving particle
semi-implicit (MPS) method [2] have been introduced to address
these computational difculties in ows. Based on the movement
of particles and no mesh-dependent, these methods can be used
for simulating uid ow with arbitrary deformation [3,4].
This advantage has made these methods eligible for analyzing
the UC-HFC problem on the unsteady complex ow part, but
on the other part of heat transfer and their coupling relationship,
Corresponding author.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2015.04.105
0017-9310/ 2015 Elsevier Ltd. All rights reserved.
the numerical models of heat transfer and the high-delity implementation that incorporates multi-physical elds (including the
ow, pressure and temperature elds) should be established based
on SPH and MPS. Such a numerical scheme could be applied to
study the UC-HFC issues, such as lm cooling and evaporation
etc. Whereas, the difculties eventually lie on the discrete numerical model of heat transfer, and the integration of the temperature
eld with the ow and pressure elds within the MPS.
Indeed, heat is exchanged via three primary modes, as the
heat conduction, heat convection and radiative heat transfer. In
this paper, only the heat conduction and heat convection are discussed. In heat conduction, no macroscopic movement of physical particles happens but the motion of phonons and electrons
from the high- to the low-temperature domain. In heat convection, heat is transferred by the movement of uids (when hot uids and cold uids are blended). Convection could be further
divided into two forms, as the natural convection and forced convection. Natural convection happens when the uid is driven to
move by internal temperature or concentration difference, while
forced convection is mainly driven by external forces. As the
basic heat transfer modes, the numerical models of conduction
and natural convection will be studied respectively in the following sections.
Among the previous studies based on SPH or MPS, Cleary [5]
explored integral approximants and effective thermal conductivity
to the thermal conduction equation which could estimate
434
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Dq
qr u 0
Dt
Du
rp lr2 u qf B
Dt
qcp
DT
kr2 T q
Dt
Here, q is the uid density, u is the velocity vector, p is the pressure, l is the viscosity, f is the external force vector, cp is the specic heat capacity, T is temperature, k is the thermal conductivity,
and q is the internal heat source. Note that, here, the momentum
conservation Eq. (2) is modied with a new simplied term B,
which accounts for the effect of temperature eld on the momentum eld. The form of B depends on the physic problems.
2.2. Particle interaction models
The interaction between particles can be dened by the kernel
function:
8
2
<
1 rre
r < r e
wr
:
0
r P r e
where r jri rj j is the distance between the ith and jth particles.
re is a cut-off distance that denes a circular effective area surrounding the target particle. The target particle is then assumed
to interact only with those particles positioned within the effective
area. Compared with the hyperbolic kernel function introduced by
Koshizuka [2], the kernel function described in Eq. (4) has several
advantages and disadvantages. The value of the hyperbolic kernel
function [2] is very large when r gets close to 0, which effectively
prevents particle clustering. However, the absolute value of the
rst-order derivative of such a function is also large when r is close
to 0, which may result in large uctuation of the kernel function. In
contrast, using Eq. (4) could avoid dramatic change of the kernel
function in the vicinity of the target particle, while the relatively
low value of w = 1.0 at r 0 is not effective in preventing particle
clustering. For better simulation results, kernel function needs to
be selected based on the specic problems.
Particle number density is dened based on the kernel function:
hni i
X
wjrj ri j
ji
hr/ii
d X /j /i
rj ri wjrj ri j
n0 ji jrj ri j2
hr2 /ii
2d X /j /i
wjrj ri j
n0 ji jrj ri j2
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
435
2.3. Algorithm
The MPS method utilizes a semi-implicit time marching
scheme. At each time step, the energy conservation equation is calculated implicitly to obtain the temperature eld, as well as the
source term B which is a force term induced by temperature difference. Then the viscosity term and the source term in the momentum equation are calculated explicitly, leading to the temporal
velocities and positions of the particles. With the updated positions
of particles, a temporal particle number density ni can be calculated. Note that the particle number density should be a constant
for incompressible ow, so ni needs to be modied to n0 by the
mass conservation equation. The pressure Poisson equation is then
obtained which can be solved implicitly. This leads to a solution of
the pressure eld, based on which the velocities and positions of
the particles can be updated. Finally, by tracking the ow characteristics of all particles, information about the ow eld can be
obtained.
3. Numerical models and examples
The accuracy and reliability of the MPS method have already
been veried by solving many cases that involve only uid ow.
However, the new computational algorithm coupling with uid
ow and heat transfer requires combined examples with different
systems. In the following sections, several examples that involve
uid ow, heat conduction, and/or heat convection will be discussed, which includes: (1) one-dimensional steady/transient heat
conduction, (2) two-dimensional steady/unsteady heat conduction, (3) natural convection heat transfer, (4) symmetry boundary
with models above.
3.1. Heat conduction
3.1.1. One-dimensional heat conduction
One-dimensional steady heat conduction, as well as the transient heat conduction is studied as two typical cases in this section.
The status of steady heat conduction is actually the same as the
nal stage of the transient heat conduction. The governing equations for heat conduction can be written in the following form:
DT
ar2 T
Dt
wjrn1
rn1
j
j
i
2
n1
n0
Dt
rn1
j
ji jrj
i
Fig. 1. Models of one-dimensional steady heat conduction (left) and transient heat
conduction (right).
1
and the temperature distribution follows Tx T 2 T
x T 1 where
d
Tx denotes the point-wise temperature within the panel. The
results indicate that smaller Dxi (more particles) can achieve
higher calculation accuracy. The absolute error is smaller in the
center of the panel and higher at the two edges. In addition,
the estimated temperature is less than the analytical solution on
the high temperature side but larger on the low side, and the absolute errors distribution presents the symmetric status about the
midline of the panel. The numerical solution eventually agrees well
with the analytical solution in a linear distribution (temperature
with respect to x), and the relative error is less than 0.5% on most
of the points, which shows good accuracy of the MPS method.
The simulation results also illustrate a linear relationship
between the computational error and the initial particle separation
Dxi . Fig. 2 plots the computational absolute error at x = 0.06 m and
x = 0.08 m with respect to Dxi (ve cases are selected), and the relationship of which can be well tted by linear equations as
f e 3:05 105 2:92Dxi and f e 9:32 105 8:75Dxi , respectively. The agreement implying that the MPS method has
rst-order accuracy on one-dimensional steady heat conduction.
Similar to the steady case, Fig. 3 shows the point-wise temperature solutions of transient heat conduction case which are evaluated in time advance from simulation and theory, respectively. The
results show that the high temperature region spreads out along
the x coordinate axis as the simulation proceeds. The temperature
distributions agree well with the theoretical solution in the form of :
Tx; t T w
x
erf p
T0 Tw
2 at
10
436
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Table 1
Comparison between the calculation results (Tcal,C) and the analytical results (Tanal,C) and the relative error (e, %) in different initial particle spacing (Dxi , m).
Tanal|x = 0.02 = 16.00
Dxi
Tcal
Tcal
Tcal
Tcal
0.005
0.002
0.001
15.956
15.982
15.991
0.275
0.113
0.056
11.985
11.993
11.997
0.125
0.058
0.025
8.015
8.007
8.003
0.188
0.088
0.038
4.044
4.018
4.009
1.100
0.450
0.225
Fig. 4. Computational error at x = 0.01 m and t = 2.0 s, 3.0 s, 4.0 s, 5.0 s versus the
initial particle spacing (Dxi ). Solid symbols: computational errors. Lines: tted
linear equations.
Fig. 3. Temperature versus x coordinates. Four time instants are considered, i.e.
0.1 s, 0.5 s, 1.0 s, and 5.0 s. Solid symbols: the MPS solutions. Lines: the analytical
solutions.
added to form a symmetry pair about the upper edge of the panel
(Neumann boundary). The paired two particles are assumed to
0
have the same physical quantities, i.e. /i /i , where i and i0
denote the interior particle and the corresponding mirror particle,
respectively. All the particles including the mirror particles will
join the calculation. This technique will guarantee zero gradients
of physical quantities across the Neumann boundary in the MPS
calculation.
The contour map of the temperature distribution and the temperature variations for typical positions as y = 0.5 m for cases 1 and
y = 1.0 m for case 2 are evaluated in time series in Fig. 7 and Fig. 8,
respectively. Fig. 7 shows the temperature rising process with very
smooth isotherms. In Fig. 8, the solid circles represent MPS results
and the lines represent the previously SPH results [22], apparently
these two methods generate very consistent results. When the system is still in unsteady state (when t is small), parabolic temperature distribution is found in both cases. However, when the system
reaches steady state (t = 0.5 s), temperature is nally uniformly
distributed within the panel to match the prescribed boundary
condition.
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Dh @ 2 h @ 2 h
Ds @X 2 @Y 2
437
14
Fig. 5. Settings for two-dimensional heat conduction with different heat boundary
conditions.
stant of 0.71. The Rayleigh number (Ra gbDTD31 =am) is a dimensionless parameter associated with buoyancy lift, the value of
which donates the level of a heat transfer between conduction
and convection. Larger Rayleigh number means more heat convection. Here, g, b, DT, and D1 are the gravity acceleration, volume
expansion coefcient, temperature difference, and the diameter of
the inner cylinder, respectively. Eq. (11) involvesn, the particle
number density, which is proportional to uid density [2] in the
MPS method.
The parameters are all non-dimensional-normalized, as the
radiuses of the cylinders are R1 = 0.5 and R2 = 1.3, besides,
L=D1 = 0.8, T 1 = 1.0, T 2 = 0.0, and Pr m=a = 0.71. Six simulations
have been performed in different RaD1 with respect to
RaL 1 102 , 1 103 , 1 104 , 2 104 , 3 104 and 5 104 , i.e.
Dn
nr U 0
Ds
11
DU
@P
@2U @2U
Pr
Ds
@X
@X 2 @Y 2
DV
@P
@2V @2V
Pr
Ds
@Y
@X 2 @Y 2
!
12
!
Ra Pr h
t=0.01 s
t=0.01 s
13
t=0.05 s
(a) Case 1
t=0.10 s
t=0.05 s
(b) Case 2
t=0.10s
438
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Fig. 10. Local equivalent conductivities for both inner and outer cylinders under
different particle size.
nite difference method (FDM) [23]. The local equivalent conductivity is dened as a ratio of the actual Nusselt (Nu) number which
includes both convection and conduction over the Nu number with
conduction only. In this example, keq is evaluated for both cylinders at different angles. The MPS results show good consistency
to the FDM results and as well as the experimental results. We note
that for the inner cylinder, the MPS and FDM results diverge from
each other when h > 150 . The FDM curve rises and the MPS curve
drops, while the experimental data show slight declination. The
Fig. 8. Temperature distribution variation in time series along given lines (a) case1, y = 0.5 m. (b) case2, y = 1.0 m. Solid circles: MPS results. Lines: FO-SSPH solutions [22].
(a)
(b)
(c)
Fig. 9. Model and setting of the convective heat transfer problem (a) basic parameters (b) initial particle discrete (c) initial temperature eld.
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
439
Fig. 11. Isotherms for different RaD1 for nal steady status.
440
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Fig. 12. Radial variation of dimensionless temperature at six different angles (Inset denes the angle, h, and the radial position, R, for any point of interest).
Fig. 13. Local equivalent conductivities for both inner and outer cylinders.
Fig. 14. Local equivalent conductivities for the (a) inner and (b) outer cylinders.
Case1 through Case5 correspond to RaD1 1:95 102 , 1:95 103 , 1:95 104 ,
3:91 104 and 5:86 104 , respectively.
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
441
Table 2
Comparison of the MPS Simulation with the Benchmark Solution.[24].
3
Ra 10
Umax
YUmax
Vmax
XVmax
Nuave
Nu1/2
Nu0
Numax
YNu-
Ra 10
Benchmark
[24]
MPS
Error(%)
Benchmark
[24]
MPS
Error
(%)
3.649
0.813
3.697
0.178
1.118
1.118
1.117
1.505
0.092
3.718
0.827
3.753
0.178
1.124
1.127
1.120
1.556
0.100
1.89
1.72
1.52
0.00
0.54
0.81
0.27
3.39
8.70
16.178
0.823
19.617
0.119
2.243
2.243
2.238
3.528
0.143
15.869
0.830
19.431
0.115
2.298
2.302
2.280
3.656
0.150
1.91
0.85
0.95
3.36
2.45
2.63
1.88
3.63
4.90
0.692
1.000
0.725
1.000
4.77
0.00
0.586
1.000
0.623
1.000
6.31
0.00
max
Numin
YNumin
MPS method, assuming the uniform particles distribution throughout the calculation process, as long as the conditions of U i U i0 ,
V i V i0 , P i P i0 , hi hi0 are met, the expressions of U 0,
@V=@X 0, @P=@X 0, @h=@X 0 on the symmetry line can be
guaranteed.
Fig. 22 illustrates the isotherms at RaD1 9:76 104 using the
full- and half-eld calculations. The two isotherms agree well with
each other except little differences on the upper part which is
marked by an ellipse. The error may come from particle clustering
on the upper domain, and the post-processing of particle method
could be another potential source to generate the deviation.
Again, for the quantitatively comparison, Figs. 23 and 24 show
comparisons on the non-dimensional temperature and local equivalent conductivities within the full-, half-eld and experimental
results [23]. The maximum relative error of the dimensionless temperature between the full-eld and half-eld calculation appears at
the curve of 120. For local equivalent conductivities, the difference
increase to the maximum 4.51% at the location of 30 in inner
cylinder, and it is 3.87% in outer cylinder at the position of 90.
The qualitative and quantitative comparisons in Figs. 2224
demonstrate the effectiveness of symmetry boundary model in this
ow-heat coupled issue. Indeed, the application of symmetric
boundary model can greatly reduce the computational time with
reliable results, which is of great signicance in dealing with
Large-scale problems.
442
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Fig. 18. Comparison of isotherms of full- and half- eld simulation. (t = 0.05 s) (a) case1 (b) case2.
Fig. 19. Comparison of the temperature distribution for full- and half- eld simulation. (a) case1, y = 0.5 m. (b) case2, y = 1.0 m.
Fig. 20. Half-domain model and symmetrical boundary setting for natural convection heat transfer in concentric cylinders.
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
443
Fig. 24. Comparison of local equivalent conductivities for both inner and outer
cylinders.
4. Conclusions
Fig. 22. Comparison of the isotherms between full-(left solid line) and half-eld
(right dash line) calculations.
Fig. 23. Comparison of the radial variation of dimensionless temperature at ve different angles.
444
Y. Liang et al. / International Journal of Heat and Mass Transfer 88 (2015) 433444
Conict of interest
None declared.
Acknowledgment
Thanks to Xiao Chen and Wenjin Cao for providing helpful discussions. This work is supported in part by the Natural Science
Foundation of China (NSFC) Project (No. 51106125, 51236006)
and in part by the Fundamental Research Funds for the Central
Universities.
References
[1] R.A. Gingold, J.J. Monaghan, Smoothed particle hydrodynamics: theory and
application to non-spherical stars, Mon. Not. R. Astron. Soc. 181 (1977) 375
389.
[2] S. Koshizuka, Y. Oka, Moving particle semi-implicit method for fragmentation
of incompressible uid, Nucl. Sci. Eng. 123 (3) (1996) 421434.
[3] Z.G. Sun, Y.Y. Liang, G. Xi, Numerical study of the owing sequence of a pouring
liquid, Sci. Chin. Phys. Mech. Astron. 54 (8) (2011) 15141519.
[4] M. Kondo, S. Koshizuka, Improvement of stability in moving particle semiimplicit method, Int. J. Numer. Methods Fluids 65 (6) (2011) 638654.
[5] P.W. Cleary, J.J. Monaghan, Conduction modelling using smoothed particle
hydrodynamics, J. Comput. Phys. 148 (1) (1999) 227264.
[6] A.K. Chaniotis, D. Poulikakos, P. Koumoutsakos, Remeshed smoothed particle
hydrodynamics for the simulation of viscous and heat conducting ows, J.
Comput. Phys. 182 (1) (2002) 6790.
[7] V. Vishwakarma, A.K. Das, P.K. Das, Steady state conduction through 2D
irregular bodies by smoothed particle hydrodynamics, Int. J. Heat Mass
Transfer 54 (13) (2011) 314325.
[8] K. Szewc, J. Pozorski, A. Tanire, Modeling of natural convection with
Smoothed Particle Hydrodynamics: non-Boussinesq formulation, Int. J. Heat
Mass Transfer 54 (2324) (2011) 48074816.
[9] A. Leroy, D. Violeau, M. Ferrand, A. Joly, Buoyancy modelling with
incompressible SPH for laminar and turbulent ows, Int. J. Numer. Methods
Fluids 2015 (online).
[10] M.A. Mansour, M.A.Y. Bakier, Free convection heat transfer in complex-wavywall enclosed cavity lled with nanouid, Int. Commun. Heat Mass Transfer 44
(2013) 108115.