European Symposium on Computer Aided Process Engineering - 15 79

L. Puigjaner and A. Espufla (Editors)

© 2005 Elsevier B.V. All rights reserved.

The Numerical Solution of Moving Boundary Problems

Using the Moving Finite Element Method
Rui Robalo^, Carlos Sereno, Maria do Carmo Coimbra'^'^* and Alirio Rodrigues^^
^Departamento de Matematica Universidade da Beira Interior
Rua Marques D'Avila e Bolama, 6200 Covilha, Portugal
^Laboratory of Separation and Reaction Engineering
^DEC-Faculdade de Engenharia da Universidade do Porto
^DEQ-Faculdade de Engenharia da Universidade do Porto
Rua Dr. Roberto Frias, s/n, 4200-465Porto, Portugal

Abstract
A moving finite element method (MFEM) is developed for the numerical solutions of
time-dependent partial differential equations involving solid-fluid reactions. Our MFEM
generates an adaptive mesh for each dependent variable so, through spatial domain
decomposition, it can be modified in order to be an efficient solver for a class of
problems showing a moving internal boundary. The algorithm was tested in the
numerical simulation of a non catalytic solid-fluid reaction modelled by the (a)
isothermal, (b) non-isothermal shrinking core model with non-linear kinetics. This work
establishes the effectiveness and applicability of the MFEM in solving moving
boundary problems.
Keywords: Moving Finite Elements; Moving Boundary Problems; Adaptive Grids;
Time-Dependent Partial Differential Equations; Shrinking Core Model

1. Introduction
A large number of problems in science and engineering are formulated in terms of time-
dependent partial differential equations involving a moving interface. The analytic
solutions of mathematical models of dynamic systems are rarely known. This has
motivated many authors and different techniques have been proposed in recent years.
Crank (1987) presents a wide-ranging description on mathematical and numerical
methods for free and moving boundary problems. In this paper we consider the
development of a MFEM to compute numerical solution of one-dimensional problems
involving two phase with a moving interface. The MFEM (Sereno et. al., 1992,
Coimbra et. al., 2001, 2004) generates an adaptive mesh so it can be modified in order
to be an efficient solver for a class of problems showing a moving internal boundary. To
modify the moving finite element equations we perform the decomposition of the
spatial domain by the introduction of a moving node describing the position of the
internal interface. This formulation (Robalo et. al., 2004) is expanded in this work to a

Author to whom correspondence should be addressed: [email protected]

80

larger class of problems by assuming the existence of non linear boundary conditions at
the moving boundary. The paper is organized as follows. In section 2 we present the
problem to be solved. The detailed description of the method is presented in section 3.
Finally, in section 4, we apply the computer code resulting from the numerical
algorithm implementation to the simulation of a non catalytic solid-fluid reaction
(Carabin et.al. 1992) modelled by the (a) isothermal, (b) non-isothermal shrinking core
model with non-linear kinetics.

2. The Problem
Let us define the moving boundary problem that we consider in this work. The general
mathematical model of two-phase system in one-dimensional fixed space domain, with
an internal moving interface, is defined by a system of n parabolic partial differential
equations (PDE) whose m-th equation fill the form

r, Sy W , r, 5y
^.^y-^^ (1)
dx J dx,3 +^™l-'^'y'^
' Jm
dt

on a<x<b, t>0. We consider that the space variable satisfy a<x<X^(^t) if

m<nj and X^(^t)<x<b if n^ <m<n . n^,n^ are the number of dependent variables
in each phase, X^[t^ is the position of the moving boundary at instant t and

y = [ y ' , y ° ] ' = h , . . . , > ' „ ] ' , where r=[yl,...y:'\' and y - = [ 3 ; , ^ . . . , J . „ ° ] ^ L e t

US assume some initial conditions and that equation (1) is subject to Dirichlet or Robin
boundary conditions at the end points of space domain. At the interface we suppose that
the boundary conditions can be expressed by general implicit conditions.

^1 ^V, df = 0. (2)
dx dx

To complete the model let us assume that the interface movement is governed by

df
dt = w ^..'.4,.fi dx dx
^t
(3)

This class of problems is an extension of those considered in (Sereno et. al, 1992,
Coimbra et. al., 2001) where all differential equations have the same fixed one-
dimensional spatial domain but each dependent variable has its own moving mesh. The
main idea to solve this problem is to combine the domain decomposition and these
different moving finite element meshes.

3. Method Development
The MFEM is a discretization process in two stages: first spatial discretization using
finite elements and allowing the movement of the space nodes, in which we focus our
attention, and secondly the time integration of the resulting ordinary differential
 81

equations (ODE) system. To solve this implicit time-dependent ODE system we use the
package LSODI, (Hindmarsh, 1980).
3.1 Spatial Discretization
The grid connected to the m-th dependent variable, y^ , is obtained by partitioning the
spatial domain in N^ finite elements by N^-l interior separation nodes,

p.- ^™,l(0<^-»,2(0<-<^»..iV„(0<^».^„..(0' '^0 (4)

One of the nodes X^, or X ^^ ^, is fixed and the other is defined by the interface
position along time. In each of the finite element of grid P^ we define the approximate
solution by a polynomial obtained by Lagrange interpolation. The relative positions of
the interpolation points are optimized as in the orthogonal collocation method (Sereno,
1991). The degree of the polynomial approximation may be chosen by the user. This
local approximation guides to a global approximation 7^ to y^ in the spatial domain.
It is a continuous piecewise polynomial function that depends, at each instant, on the
value of the the m-th dependent on the i-th collocation node of the y-th finite element,
7^ ^., and on the nodal position, X^.^^ (Sereno et. al., 1992, Coimbra et. al., 2001).

3.2 Residual Minimization, Penalty functions and Integrals Calculation

Following the ideas and implementation of the MFEM expressed in (Sereno et. al.,
1992, Coimbra et. al., 2001, 2004), to estabhsh the semi-discrete variables 7^^ and
WJ+l
we must integrate, in time, the system of ODE's generated by minimization of
the following objective function
r
^=Z dt
Jm
' dx I dx^
-gn.
OX J
+p
enalty
(5)

with respect to time derivatives of nodal amplitudes and nodes positions, where 7^„^^^ is
a penalty term depending on the intemodal viscosity and spring penalty functions of the
7-th finite element of the w-th grid, s^. and S^., respectively (Sereno, 1991). Here we
introduce Miller's penalty functions (Miller, 1981) into the minimization process to
avoid the singularities associated with the method, due to parallelism and element
folding. The penalty functions depend on small positive constants supplied by the user
for each element of each grid. To get the explicit form of general equations of method is
necessary, first, define the partial derivatives of 7^ in order to time variable and also in
order to 7^^. and X^.^^ and, secondly, calculate the integrals that results from
minimization of F . To define the approximations of spatial derivatives of 7^ we use a
smoothing process based on cubic Hermite polynomials in a neighbourhood of the
separation nodes. By this process (Sereno et. al, 1991, 1992, Coimbra et. al., 2001)
show that all the integrals are well defined as a limit when the amplitude of the
neighbourhood tends to zero.
82

4* Numerical Examples
We are interested in solving numerically a fluid-solid non-catalytic heterogeneous
reactions problem described by the Shrinking Core Model (Carabin et. al.) to illustrate
the working and performance of our MFEM. All the numerical results presented are
obtained on a Pentium 4 computer at 1.6 GHz. Some parameters of the method are kept
constant. It is the case of the penalty constants and the ODE solver tolerances. We use
Lobatto's quadrature with 11 interior quadrature points to compute the integrals
appearing in each one of the equations of the ODE systems.
4.1 The Shrinking Core Model, Isothermal Case
Here we take up the simplest model for the reaction of fluid species A with spherical
particles of unchanging size of solid S. Considering the stoichiometric coefficients, 3.,
the reaction of the shrinking coke takes the form:
•9.4/)+^A.)-^'9,5,^) + '9^^.) (6)
The mathematical problem is to determine the values of the dimensionless concentration
of reactant A, C(x,/), X^{t^<x<\, / > 0 , normalized by the bulk molar
concentration of A, C^^.X^(0 is the position of the interface of the soUd unreacted
core. The equation for diffiision of reactant A is, in dimensionless form:

dt 8x' xdx '^^ ^^

subject to the boundary conditions
dC
=DaC", x = X(t\ t>0, (8)
dx
-^=Bi^{\-C\ x = l, t>0, (9)
OX

^ = -r DaC\ x = X^{t), t>0 (10)

and initial conditions C(x,0) = 1, Z^ (O) < x < 1, X^ (O) = 1. Here n is the reaction
order, Bi^is the Biot number for mass transfer and / = DsMgC^j^/OsPg is a
dimensionless number, M^ is the molecular weight of S, p^ is the density of S.
Da = Ae'^R^C^l^ /D is the Damkohler number. Here A is the pre-exponential factor,
j3 = E/(RTi,)is the Arrhenius number at bulk temperature, J^, D is the effective
diffiisivity of A and R^ the radius in sphere. Figure 1(a) shows the conversion of the
soHd reactant, l-[X^) as a function of dimensionless time t/t^ (t^ is the finite time
for complete solid conversion) for different Da numbers and fixed /. We consider
Bi =10^ and first order reaction, n = l. For these runs CPU time take values from
0.42 to 0.78 seconds, using cubic approximations and 4 finite elements. Figure 1(b)
 83

shows the time evolution of the moving boundary position, for higher values of Da and
y using approximations of degree 6. The CPU is 3.3 seconds.

Figure 1. Conversion (a) and evolution of the interface (b) as functions of scaled time

4.2 The Shrinking Core Model, Non-Isothermal Case

In the non-isothermal case the dependent variables are the dimensionless concentration
of reactant A, C and the dimensionless temperature, v = {T-T^)/Tj^. For simphcity we
denote by v^ the temperature for Q<x<X^{t) and by ^2^^^ temperature for
X^ (/) < X < 1. The model is now defined by the mass balance equations (7)~(10), taking
into account that the Damkohler number is now a function of temperature.
Da = Ae'^'^^^^^^^R^C'^ ID, combined with the dimensionless energy balance equations:

dv. d^v. 2 dv.

-^=—^^-—L^ i = l2, 0<x<l, />0 (11)
dt dx X dx

with initial conditions, v^=V2=0 and boundary conditions

^^-0, ^^^ (12)
dx dx -Bi,C, x=l

(13)
dx dx
Bif^ is the Biot number for heat transfer, a is the dimensionless heat of reaction and
A: = 1.4 . Figure 2 shows the temperature (a) and concentration (b) profiles for n = 2.

0.012

0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Figure 2. Temperature (a) and concentration (h) profiles at different times

84

This run test was performed considering y^ = 20.9, y = 4AxlO~^, Bi^ =^h =1^%
a = -7.5 X10^ and Da(t = 0) = 0.31. We use 3 finite elements in the grid associated to
Vi, and 4 in the two other grids associated to V2 and C. Using cubic approximations the
CPU time for this run is 2.6 seconds. Figure 3(a) and 3(b) present the trajectories of
nodes separation associated to temperature and concentration respectively. Initially
nodes are concentrated near x=l.

0.012 1.00
^1.3
(a)
0.01 h2/ ^'^ 0.99
\ \x^
0.008 0.98

E>-t].0O6 0 0.97
v"'
/x^
0.004
v' 0.96

0.002 0.95

0 0.94
0. DO 0.25 0.50 0.75 1.00 G.DO 0.25 0.50 0.75 1.00
X. S..

Figure 3. Temperature (a) and concentration (b) at nodes position and interface

S. Conclusions
In this paper we have presented a MFEM for moving boundary problems. The main
advantages of the method are its simplicity and performance, which account for its
widespread use. The complexity of the numerical examples considered requires a robust
and efficient algorithm. In fact, the numerical results obtained by this formulation of
MFEM exhibit very good agreement with those obtained by previous methods so, this
work establishes the effectiveness and applicability of the MFEM in solving moving
boundary problems.

References
Carabin, P. and D. Berk, 1992, Analysis and modelling of the isothermal and non-isothermal
shrinking core model with non-linear kinetics, Chem. Engng. Sci., 47,2499-2504.
Coimbra, M.C., C. A. Sereno and A. E. Rodrigues, 2001, Applications of a moving finite element
method. Chemical Engineering Journal. 84, 23-29.
Coimbra, M.C., C. A. Sereno and A. E. Rodrigues, 2004, A. E., Moving finite element method:
application to science and engineering problems, Comp. Chem. Engng., Vol.28, pp.597-603.
Crank J., 1987, Free and Moving Boundary Problems, Clarendon Press, Oxford.
Hindmarsh, A. C , 1980, LSODE and LSODI, Two New Initial Value Ordinary Differential
Equation Solvers, ACM-SIGNUM Newsletter, Vol. 15, No. 4, pp. 10-11.
Miller, K. and R.N. Miller, 1981, Moving Finite Elements. I, SIAM J. Numer. Anal, Vol. 18, No.
6, pp. 1019-1032.
Robalo, R., M.C. Coimbra., Sereno, C. A. and Rodrigues A. E., 2004, Aplica9ao do MEFM a
problemas com fronteiras moveis, Congresso de Metodos Computacionais em Engenharia,
edited by C. Mota Scares, A. Batista, G. Bugeda, M. Castaleiro, J. Goicolea, J. Martins, C. Pina
and H. Rodrigues, Editions LNEC, ISBN 972-49-2008-9, pp.406.
Sereno, C. A., A.E. Rodrigues, and J. Villadsen, 1991, The Moving Finite Element Method with
Polynomial Approximation of Any Degree, Comp. Chem. Engng., Vol. 15, No. 1, pg. 25-33.
Sereno, C. A., A.E. Rodrigues, and J. Villadsen, 1992, Solution of Partial Differential Equations
Systems by MFEM, Comp. Chem. Engng. Vol. 16, 583-592.