Professional Documents
Culture Documents
Transient heat conduction (or diffusion) equations with nonlinear source terms arise as governing equations in
many different areas of mathematical physics, applied science, and engineering. In this paper we propose an
efficient method, the Laplace transform dual reciprocity method (LTDM), for solving problems governed by
this type of equation. Due to the nonlinearity the source terms are linearized first, followed by a Laplace
transform performed to the linearized differential system. The unknowns are then found in the transformed
space by the dual reciprocity method, after which their values in physical space are obtained through a
numerical inversion. Consequently, problems of this type can be solved without step-by-step calculations,
resulting in a highly eficient way of obtaining numerical solutions, especially at large time steps. The
effectiveness of the method stems from the fact that both time-stepping and domain integrals are avoided. As
examples of our modelling exercises, we used the LTDRM to solve microwave heating and spontaneous ignition
problems; the efficiency and accuracy of the method are thus clearly demonstrated.
Keywords: nonlinear problems, transient heat conduction, Laplace transform, dual reciprocity method
law form; it was not clear if their extension can render where ii is the previously iterated solution at time t,, a =
results of the same level of accuracy if the nonlinearity of l/k, with k being the thermal diffusivity (= K/PC) and
the source term is higher than the power-law form. p = y/K. The LTDRM proposed by Zhu et a1.4 can now
Our recent numerical experiments showed that while be readily implemented. Upon performing the Laplace
the simple linearization scheme worked for the source transformation with respect to t, equation (5) becomes
terms of weak nonlinearity to intermediate-strength of
nonlinearity, it fails when dealing with highly nonlinear V2U=apU-au,- %f,(e) - Pf2PP (6)
cases. For source terms being a transcendental function of
the unknown function, varying exponentially with the un-
where p is the Laplace parameter and U is the Laplace
known function as appears in a spontaneous ignition prob-
transform of u. Then, after applying the DRM in the
lem for instance, the iterations constructed following our
Laplace space, the final matrix equation can be derived as4
previous work5 usually give poor results mainly because
the exponential variation has been linearized as a constant HU - GQ = (Ho - G@F-‘.
if the simple linearization scheme in Ref. 5 is followed.
A remedy to this problem is to adopt a more systematic
linearization scheme based on the first-order Taylor’s se-
ries expansion, which has been used, for example, by Chen
and Lin6 in solving one-dimensional transient problems
[
{“P - Pf2(wJ- au0 - i fl( 4
Ii (7)
with nonlinear material properties and by Ramachandran7 where the subscript i denotes nodal values, and all matri-
in solving one-dimensional nonlinear diffusion-reaction ces are of the conventional sense in the DRM.8 By defin-
problems. In Section 2, the LTDRM formulation with this ing
linearization scheme is described. Several highly nonlinear
problems, including the microwave heating of a square S = (Hfi - G@F -’ (8)
slab and the spontaneous ignition of a circular cylinder are
then examined in Section 3 to demonstrate the numerical equation (7) can be written in the form
efficiency and accuracy of the present method. Our prelim- (H-ST)U=GQ-Sd (9)
inary results show that the linearization scheme based on
the Taylor series ex ansion is more general than the one where T is a diagonal matrix whose diagonal entries are
previously proposed P and thus more recommendable for the nodal values of up - pf2(ii) and d is a vector contain-
the LTDRM. ing the nodal values of uuo + ( /?/p>fi(ii).
The iterative process starts with a guessed value of ii.
The matrix equation (9) is then solved according to the
imposed boundary conditions, and solutions in the time
2. The LTDRM formulation domain are obtained via the numerical inversion algorithm
Transient heat conduction equations are generally of the proposed by Stehfest.’ The value of fi is then replaced by
form that of u, and the iterative process continues until a
convergence is achieved. For all the examples showed in
this paper, a convergence is declared when
Kv*u=pc~-yf(u) (I)
I(%-i-%)/(%-I +%)I <E (10)
in which y is a constant, p, c, K are density, specific heat
and thermal conductivity, respectively, and f represents a is satisfied at every nodal point, where the subscripts
temperature-dependent heat source. Herein p, c, and K denote iteration levels and E is a preset value of tolerance.
are assumed constant, and f is considered to be a nonlin- One of the key issues concerning the success of the
DRM, which is worth mentioning, is the choice of the
ear function of u. Based on the linearization scheme
proposed by Chen and Lin6 and Ramachandran,’ if the basis functions used to interpolate the nonhomogeneous
term of the governing equation. The most widely adopted
solution of the unknown function at a particular time, say
are the functions of the type 1 + a, r + a2 r2 + . . . , where
t,, is to be sought, the nonlinear source term can be
r is the distance between a field and a source point. In
linearized by a first-order Taylor series expansion as
practice, Partridge and Brebbia” have demonstrated that
f(u) =fdfi) +f*W (2) the simplest form of 1 + r is the best choice in terms of
where numerical accuracy and efficiency and a proof of conver-
gence has recently been given by Yamada et al.” There-
fore, this form of the basis functions was used for all the
fdfi> =f(E) -p( g). (3) numerical calculations presented in this paper.
f2@,=(gg. l4 (4)
u
3. Numerical examples and discussions
and thus equation (1) can be written as
To illustrate the LTDRM described in the previous section
and demonstrate its accuracy and efficiency, we shall
VW; -pfi(fi) -Pf2(fi)u
(5) present several numerical examples of heat conduction
Table 1. Temperature at selected points on the slab for the case n= 2: (a)p = 4.7,y= 0,
and t=1.7;(b) p=ll,y=2,and t=1.6;k) p=21,y=4,and t=2.1
X Y DRM LTDRM
a b C a b C
problems including the microwave heating of a square slab from Maxwell’s equations, leading to a simplified model
and the spontaneous ignition of a unit circular cylinder. equation
Two steady-state problems, whose analytical solutions are
available, will also be examined. au
V*U = at - v(u)epY" (13)
3.1 Microwave heating of a square slab Since for many materials used in industry the rate at
which the microwave energy is absorbed (the thermal
As our first example, we adopt a problem studied by Zhu abso tivity) increases with temperature by the power
et al., ’ i.e., the heating of a square slab using microwave law, &! equation (13) becomes
energy. Microwave heating can be modelled by the forced
heat equation V2u+3e-~xu'.
(14)
Table2. Temperature at selected points on the slab forthe case n = 3:(a) p= 2.8,y= 0
and t=l.g;(b) p=6.6,y=2,and t=1.5;(~) /3=12.3.y=4,and t=1.6
X Y DRM LTDRM
a b c a b C
The LTDRM can now be applied to this linearized equa- those obtained by using the Frank-Kamenetskii’s approxi-
tion. The discretization adopted here is similar to that used mation method, especially when n is large. As explained
in Ref. 1; 40 constant elements are placed on the bound- in Ref. 1, the errors involved with using the first eigen-
ary, and 36 internal nodes are uniformly placed inside the function to represent the unknown function are expected to
slab. During our numerical experiments, linear boundary become larger when n is larger.
elements were also employed. However, no noticeable To visualize the dependence of the temperature distribu-
difference in the results was found and therefore we shall tion on the parameters p and y, the steady-state tempera-
only present the results obtained with constant elements ture profiles for several different values of p and y are
here. The tolerance for stopping the iterations, E, was graphically presented. With the assumption that an expo-
chosen to be 1.0 X 10m3, and the initial condition was nential decay of the electric field is in the x direction, the
always used as the first iteration. Numerical results of the problem is symmetric about the cross-section y = 0.5
temperature, using the LTDRM, at some selected points on where the highest temperature also occurs. Therefore, we
the slab, for the cases when n = 2 and n = 3, are shown in choose to present the steady-state temperature profiles
Tables I and 2, respectively. The corresponding values along this cross-section in Figures l(a-c) for the case
obtained by the method coupling the DRM with a finite n = 2. From these figures, one can see a general pattern of
difference scheme in the time domain,’ for three different the temperature distribution. With a fixed y, the steady-
values of /3, y and t, are also tabulated in Tables 1 and 2 state temperature at every spatial point increases monotoni-
for the purpose of comparison. It is evident that the results cally with /3; the highest steady-state temperature is
from both methods are in good agreement. Also, the reached when /3 = p, (p, = 4.78, 11.27, and 21.09 in
results obtained here by the LTDRM almost match those Figures l(a-d, respectively). Once p > fl,, hot-spots oc-
obtained by the LTDRM with a simple linearization scheme cur eventually. One can also notice that, as y is increased,
shown in Ref. 5. However, while the number of iterations the point of the maximum temperature moves from the
needed to find converged results5 is an average of 16 steps center of the slab toward the left boundary where the heat
for the cases shown in Tables 1 and 2, convergent solu- absorption is the strongest, and thus the gradient of the
tions for the LTDRM with the new linearization scheme temperature field becomes larger and larger in the neigh-
were attained within an average of 5 iterations, thus show- borhood of this edge. Consequently, the material will be
ing the higher efficiency of the new scheme over the damaged first at this edge as one would naturally expect.
previous one. For the case n = 3, a similar pattern was observed as well.
A key achievement in Zhu et al.’ was to use their As far as the numerical efficiency is concerned. Zhu et
DRBEM model to predict the critical /3 value at which the a1.4 demonstrated the excellent efficiency of the LTDRM,
so-called “hot-spots’ ’ occur. Hot-spots are the localized especially when the solution at a large observation time is
areas of high temperature which develop as the material is calculated. When nonlinear iterations are involved, the
being irradiated’; a correct prediction of the hot-spot oc- numerical efficiency of the LTDRM is even further en-
currence is important in any industrial process in which a hanced since iterations are only performed for a single
microwave heating process is involved. It is therefore quite time step in the LTDRM. In contrast, the total number of
interesting to examine whether or not the LTDRM can be iterations in a time-domain method is equal to a double
used to predict correctly critical /3 values as well. To summation over the number of iterations at each time step
calculate the critical /3 values, a simple loop was added to and the number of time steps. Moreover, when only the
our program used to calculate the solution at any observa- solution at a specific time step is needed, all the intermedi-
tion time. Numerically, the steady state of equation (17) is ate solutions and the associated iterations as well as the
deemed to be reached if the difference of the temperature computer storage are eventually discarded and thus wasted;
at two different times is less than a small number, say this worsens if the desired solution is at a large time. This
1.0 X 10-4, at every collocation point and the critical is better illustrated by an example. For the case n = 2 and
value p, is defined as the largest /I such that the steady /3 = 21, Zhu et al.’ had to go through 73 iterations alto-
state of equation (17) can be attained. Our results using the gether before a convergent solution at t = 2.1 was ob-
LTDRM for the critical p values, p,, are tabulated in tained. With a same level of accuracy, we obtained our
Table 3 for the cases n = 2, 3. The p, values obtained by results with only 6 iterations which of course should be
Zhu et al.’ and by using the Frank-Kamenetskii’s approxi- multiplied by the number of solutions needed in the Laplace
mation method’,‘5 are also listed in Table 3. It can be seen space (6 for all examples presented in this paper), resulting
that our results are very close to those obtained by Zhu et in solving equation (9) a total number of 36 times. In
al.’ using their numerical model but slightly larger than addition, as the observation time was increased, the num-
ber of iterations associated with a time-domain method u(r) = u(0) + ru’(0) + gu”(O) + ... (19)
increased at least linearly, whereas the number of iterations
required by the LTDRM virtually remained the same. If we now rewrite equation (18) as
Furthermore, when the critical value p, needs to be
ru,, + u, + pi-u2 = 0 (20)
determined, a higher efficiency of the LTDRM than time-
domain methods is even more evident. Generally speaking, letting r + 0, we have
when the parameter to be determined is near its critical
value, the steady-state is approached very slowly. This u’(0) = u,(O) = 0 (21)
creates considerable difficulty in terms of the computa- Differentiating equation (20) n times and letting r + 0,
tional time involved when one tries to find, using any one can show that, for n being even,
time-stepping method, the critical value with a reasonable
accuracy. Take the current microwave heating problem as S(O) =#)(O) = . . . =u(,+ly()) = . . . =()
an example. When p values are in the vicinity of &, the (22)
efficiency of the time-domain method used in Ref. 1
worsens dramatically as the steady-state of equation (14) is and, for n being odd,
reached very slowly. It is therefore very costly to calculate (n + l)u(~+r)(o)
the critical value & since quite a large number of solu-
tions at intermediate time steps needs to be calculated. = -fIn[u(O) z&“-‘)(O) + (n - ~)u’(O)U(“-~)(O)
However, no such problems exist for the LTDRM. As
mentioned before, the LTDRM is very effective especially +(n-l)(n-2)u”(0)u(“-3)(0) + ...I (23)
Table 4. Temperature on the circular disk for the case with the critical value of 1.39 found analytically from
6= 1.2 equation (26). The numerical results at a large time using
r Analytical LTDRM Error (%) the LTDRM representing the steady-state solution were
0.0 1.655 1.620 2.2
obtained within only four iterations and tabulated in Table
0.1 1.647 1.612 2.2 4 for the case p = 1.2. One can clearly see a good
0.2 1.622 1.589 2.2 agreement between the two solutions; the maximum per-
0.3 1.583 1.551 2.1 centage error is less than 2.3%. The results from the
0.4 1.528 1.499 2.0 LTDRM seem to be almost exactly the same as those
0.5 1.462 1.436 1.9 shown in Ref. 5.
0.6 1.383 1.362 1.7
0.7 1.296 1.279 1.4
0.8 1.202 1.189 1.2 3.3 Spontaneous ignition of a unit circular cylinder
0.9 1.102 1.094 0.8 Having now demonstrated, through the previous two ex-
amples, that the new extension of the LTDRM is more
Finally, we have efficient than the LTDRM used in Ref. 5 while rendering
results of the same level of accuracy, we also need to point
out that there is a fatal problem associated with the LT-
u(r) = U(0) + &O) + &4)(O) + ... (24) DRM based on the linearization scheme proposed.5 Our
recent numerical experiments showed that the iteration
or scheme constructed previously5 usually gave poor results
-1.5?“..1...‘1”..!.“‘1’.“‘.“”
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
and employing the Frank-Kamenetskii approximation’8
Figure 2. Boundary and internal nodes used for the circular
e-E/RT = e-E/RT,eu
disk. (29)
v2u+ yeu=a; cess. The value of At was altered such that at each time
step the average temperature change at all interior nodes is
maintained between 5°K and 20”K.2,‘8
where a = l/k, with k being the thermal diffusivity (= To implement the LTDRM, the boundary of the cylin-
K/PC) and der was discretized with 16 linear elements, and 19 inter-
PQEZ nal nodes were placed at an interval of 0.1 m along a
-e-E/RTa
y= KRTZ
(31) diagonal (the y-axis). This discretization is the same as
that adopted in Ref. 2. The LTDRM was then applied to
It is common to define the dimensionless Frank- the linearized form of equation (30). Once the unknown u
Kamenetskii parameter as 6 = 12y, where 1 is a character- has been obtained, the temperature T can be calculated
istic length of the problem being considered. from equation (28). Since the value of SC has already been
For a given geometrical shape of the solid, it has been given2 the main interest is to find the ignition time for the
shownt5 that whether the temperature within the solid cases where 6 > 8,. It would be ideal if the ignition time
eventually reaches the steady state or the value T, depends could be cast into an unknown to be found as part of the
solely on the parameter 6. The critical value SC which solution. However, after a few attempts, it seemed to be
separates these two completely different final states can be impossible to reformulate the problem in a way to achieve
evaluated from equation (30) with the temporal derivative this goal. Therefore, in order to determine the ignition time
term being set to zero. ‘*18 For the cases S < SC, a thermal using the current formulation of the problem, the multistep
equilibrium can be attained with the steady-state tempera- and bisection techniques were necessary to be employed in
ture in the solid being larger than T, everywhere but less conjunction with the LTDRM. In multistep algorithm, the
than T,. However, if 6 is greater than SC, a steady state result from the previous step is used as an initial condition
cannot be reached; the temperature in the solid eventually for the next step which is similar to the time-stepping
reaches T,, at a certain time interval, called the induction technique. However, the time-step size in multistep LT-
time,16 and consequently a spontaneous ignition initiates. DRM can be much larger than that of a time-marching
Therefore, the value of 6, is important in modelling the scheme. When the iteration fails or at least one point inside
spontaneous ignition process and can be determined from the cylinder has a temperature over the ignition tempera-
equation (30) by the LTDRM using a similar technique to ture, the bisection is then used to half the time step before
that described in the microwave heating problem in finding the calculation at that step is repeated. The starting time
p,. Several values of SC for common two-dimensional step was set to be 100 set in all the cases presented below.
geometrical shapes can be found, for example, in Refs. 2 Results show in Figure 3(a) are temperature profiles for
and 18. It is worth noting that although the two problems, the case 6 = 1. The temperatures at time t = 130 and 670
i.e., microwave heating and spontaneous ignition prob- set were calculated separately from the multistep-bisection
lems, are mathematically the same in essence, the source algorithm in order to be able to compare the results with
term in the latter is of a much higher nonlinearity than that those shown in Ref. 2. The multistep-bisection algorithm
in the former. It is therefore desirable to see if the LTDRM was used and the procedure converged to a steady-state
with the new linearization scheme can be applied to such a solution T, with T, < T, < T, at all internal points at time
highly nonlinear problem. t = 2,400 set as a result of S being less than SC in this
To demonstrate the application of the LTDRM to a case. The results obtained from the LTDRM and DRM
diffusion equation with a highly nonlinear source term, the agree well with each other. One should however notice
spontaneous ignition of a long cylinder of unit radius taken that the elapsed time for the temperature to reach the
from Ref. 2 is considered. Initially, the cylinder is of thermal equilibrium is different in the two methods. This
uniform temperature To = 298°K and is abruptly sub- can be explained as follows. The incremental time used in
merged at time t = 0 in a bath at temperature T, = 400°K. the DRM is doubled if the average temperature change at
The cylinder is of a uniform isotropic reactive material, the all interior nodes is less than 5°K before the procedure
ignition temperature of which is T, = 425°K. The problem proceeds to the next time step. Since the change in temper-
is subject to an essential boundary condition where T = T, ature becomes very small when the temperature is above
is imposed at all boundary nodes. The other numerical the ambient value or the steady-state solution is near, the
values of the physical parameters are a = L285.71 incremental time is always doubled thus resulting in a
set/cm*, E = 47,500 kcal/M, and R = 1.987 cal/(MK). large time increment and also a large observation time. In
The critical value, SC, was numerically estimated, by the contrast, there is no requirement for the minimum tempera-
LTDRM, to be 2.08 which agrees very well with the ture change when using the LTDRM; therefore, the
analytical value of 2.00 given in Ref. 2. steady-state temperature is attained at a smaller observa-
Time-domain methods in conjunction with the FEM and tion time. It should also be noted here that such a large
DRM have been adopted to solve the spontaneous ignition difference in time, at which the steady-state solution is
problems by Anderson and Zienkiewicz” and Partridge declared to be reached, is expected for different numerical
and Wrobel,* respectively. They all pointed out that at the models or even the same model with different runs when
earlier time steps, when the heat generated by the reaction 6 < 8, and the approach to the steady-state is very slow.
is obvious, a large At may be used but it must be reduced However, it is absolutely necessary for any numerical
quickly as t gets closer to the ignition time. Therefore, a model to be able to predict accurately the ignition time
variable time step was employed due to some varying when 6 > SC, as will be shown below.
A2
(32)
v2u = se-U
is an important equation appearing as the governing equa-
tion in many different areas of mathematical physics,
applied science, and engineering.“p2’ Since it has a well-
known special analytical solution,21 i.e.,