ENCIT-2018-0667

COMPUTATIONAL SYSTEM FOR PHYSICALLY NON-LINEAR

THERMAL ANALYSIS
Guilherme Garcia Botelho
Rodrigo Guerra Peixoto
Roque Luiz da Silva Pitangueira
Programa de Pós-Graduação em Engenharia de Estruturas - PROPEEs, Universidade Federal de Minas Gerais - UFMG, Av. Pres.
Antônio Carlos, 6627, Escola de Engenharia, Bloco 1, Pampulha, Belo Horizonte.
[email protected] ; [email protected] ; [email protected]

Abstract. The present work concerns the inclusion of functionalities into the free software INSANE - INteractive Struc-
tural ANalysis Environment for the physically non-linear thermal analysis in solid bodies by the Finite Element Method
when the physical properties of the material are temperature dependent. INSANE is a multi-platform object-oriented
computational system being developed at the Federal University of Minas Gerais - UFMG. Once the physical properties
are temperature dependent, Newton’s iterative method is used to calculate the desired solution. The implementation was
tested comparing the numerical results to analytical solutions available in the literature. The segmentation and general-
ization of the INSANE’s numerical core allowed the reuse of the existing classes to support the software expansion.

Keywords: Non-linear Thermal Analysis, Finite Element Method, INSANE System

1. INTRODUCTION

In many practical applications industrial machinery is subjected to temperature variations during its operational cycle.
In general, the temperature variation is inherent to the process purposes and has strong influence on the equipment’s
structure behavior, as in the case of molten metal handling equipment into the steel industry, for example.
For common engineering materials, their physical properties express significant changes on wide temperature ranges.
The material’s thermal properties dependence upon the temperature introduces a physical non-linearity into the thermal
analysis (Reddy, 2004). To obtain the temperature field in a situation where the material physical properties depend upon
the temperature itself, a numerical iterative method is usually applied, and Newton’s iterative method is one possible
solution algorithm.
According to Bergman et al. (2011), the knowledge of the temperature field and of the thermomechanical response of
a solid body is of great importance for safe and efficient machinery design, since it allows, for instance, the optimization
of refractory linings and the prediction of the equipment’s behavior under conditions closer to reality.
However, the temperature field calculation is not a trivial task to accomplish, especially when the body geometry is
complex and the medium presents a non-linear behavior. In some situations, this calculation may be even impossible
using analytical methods, being necessary to resort numerical methods (Bergman et al., 2011).
As better described by Zienkiewicz and Taylor (2000), the Finite Element Method (FEM) is a numerical technique
based on the subdivision of the domain into individual components, or ’elements’, whose behavior is readily understood.
These components are then assembled in a proper way to rebuild the original system in order to determine the approximate
behavior of the body.
This subdivision of the domain, called discretization, simplifies the solution by transforming the continuous problem,
governed by partial differential equations, into a discrete problem, governed by a set of algebraic equations. Since the
quality of this approximation improves as the number of individual components increase, it is necessary to solve large
systems of equations in situations of practical application, justifying the computational implementation of the method.
The present work concerns the implementation of functionalities into INSANE’s numerical core to solve non-linear
heat transfer problems. The computational system INSANE (INteractive Structural ANalysis Environment) is a multi-
platform free software, written in Java following the Object-Oriented Programming (OOP) paradigm, intended to be used
as a didactic and researching tool, among other purposes. This software is being developed by the Structural Engineering
Department of the Federal University of Minas Gerais (UFMG) and is available at http://www.insane.dees.ufmg.br. It has
a segmented and generic numerical core independent from the graphical interfaces for pre and post-processing allowing
reuse of the existing source code to support the software expansion.
Problems of different fields of engineering can be solved by INSANE since the software conception is abstract. The
particularities related to each physics can be gradually included into the numerical core by the introduction of new entities
G. Botelho, R. Peixoto and R. Pitangueira
Computational System for Physically Non-linear Thermal Analysis

following an OOP inheritance mechanism.

INSANE’s development initially focused on solid mechanics. The first resources for solving heat transfer problems
by the Finite Element Method were included into the software by Botelho et al. (2015) and the analysis was limited to the
linear case. The work described in this paper is an extension of the referred work by allowing the analysis of physically
non-linear heat transfer problems.

2. THEORETICAL BACKGROUND

2.1 FEM Formulation of the Heat Transfer Problem

As better described by Logan (2007), the Finite Element Method approach for heat transfer problems starts by ap-
proximating the temperature field within a finite element by the interpolation of the temperature values at nodes of this
element. This approximation can be expressed mathematically as shown in Eq. 1.
u = Nû , (1)
where u is the temperature field within the finite element domain, û is a vector containing the temperature values at the
finite element’s nodes and N is a matrix containing pre-determined interpolating functions. The matrix N has different
forms depending upon the number of nodes, the spatial domain of the element and the interpolation scheme chosen. The
appropriate matrix N for each type of element can be easily found in the literature (e.g. Zienkiewicz and Taylor, 2000;
Reddy, 2004; Logan, 2007; Weaver and Johnston, 1984).
In the case of isotropic materials, the heat flux per unit area in each direction of the global coordinate system is
determined by the Fourier law and can be expressed by Eq. 2.
q00 = −Dg , (2)
where q00 is the heat flux per unit area in each direction of the global coordinate system, D is the constitutive matrix,
shown illustratively in Eq. 3 for an isotropic material, and g is a vector containing the temperature gradient in each
direction of the global coordinate system.
 
k 0 0
D= 0 k 0  , (3)
0 0 k
where k is the thermal conductivity. It should be pointed out that the conductivity varies with the temperature in non-
linear heat transfer problems. In the present work this relationship between the thermal conductivity and the temperature
is expressed by a polynomial as:
k(x, y, z) = a0 + a1 u(x, y, z) + a2 u2 (x, y, z) + . . . + an un (x, y, z) , (4)
being a0 , a1 , ... , an polynomial coefficients.
The temperature gradient vector mentioned in Eq. 2 can be expressed, for a cartesian system, by the matrix equation:
 T
∂u ∂u ∂u
g= , , . (5)
∂x ∂y ∂z
Writing the derivative operator, L, in matrix form,
 T
∂ ∂ ∂
L= , , (6)
∂x ∂y ∂z
and substituting Eq. 1 into Eq. 5 we get:
g = LNû = Bû , (7)
where B is the internal variables operator calculated as the matrix product between L and N.
The relationship between the actions on a finite element and the nodal temperatures is given by Eq. 8:
cû = f , (8)
where c is the element’s conductivity matrix and f is a vector arising from the potentials for heat transfer acting in the
finite element volume and on its surfaces. The vector f is called an equivalent nodal flux vector and has the three following
components:
f = fQ + fq + fh (9)
 17th Brazilian Congress of Thermal Sciences and Engineering (ENCIT 2018)
November 25th-28th, 2018, Águas de Lindóia, SP, Brazil

where fQ is the potential for heat transfer due to heat sources or heat sinks in the finite element’s volume, fq is the potential
for heat transfer due to flux per unit area prescribed on any surface of the finite element, and fh is the potential due to the
heat transfer by convection prescribed on any surface of the finite element.
The element’s conductivity matrix can be calculated as described in Eq. 10.
Z Z
T
c= B DBdV + hNT NdS (10)
V Sh

where h is the convective heat transfer coefficient.

The equivalent nodal flux vectors are calculated by Eq. 11, Eq. 12 and Eq. 13 :
Z
fQ = NT QdV (11)
V

where Q is the heat sink or heat source per unit volume;

Z
fq = NT q ∗ dS (12)
S

where q ∗ is the heat flux per unit area prescribed on a surface of the element;
Z
fh = NT hUinf dS (13)
Sh

where Uinf is the fluid temperature.

The contributions of each finite element can be properly assembled to obtain an algebraic system of equations repre-
senting the global behavior of the body. The assembling process is based on the principles of equilibrium and continuity
and is described in details by Logan (2007). The resulting set of equations can be written as:

CU = F (14)

where C is the global conductivity matrix of the body, U is a vector containing the temperatures at each node of the finite
element mesh and F is the vector containing the flux prescribed on the nodes of the finite element mesh. This later vector
(F) contains heat flux from the element’s equivalent nodal flux vectors and from concentrated heat flux prescribed directly
onto the nodes of the finite element mesh.

2.2 FEM Formulation of the Non-linear Heat Transfer Problem

In a physically non-linear analysis the global conductivity matrix C given by Eq. 14 depends upon the temperature,
thus:

[C(U)]U = F (15)

Since the temperatures vector U is not readily known, the matrix C can only be an approximation based on a guessed
temperature distribution. This approximation causes the left side of Eq. 15 to differ from the right side. This inequality is
called residue, R, and can be calculated as the difference between the left and right sides (Reddy, 2004):

R = CU − F (16)

As described in detail by Reddy (2004), Newton’s iteration procedure consists of a technique for minimizing the
residue along the iterations. Since the residue vector is a function of the temperature vector, it can be expanded in a Taylor
series close to the solution Us as:
 
∂R
R(Us ) = R(U(p−1) ) + · δU + . . . (17)
∂U (p−1)

The subscript (p − 1) refers to the previous iteration. In the imminence of the solution the residue is null, thus:

R(Us ) ≡ [C(Us )]Us − F = 0 . (18)

Substituting Eq. 18 into Eq. 17 and omitting the terms of order 2 and higher we obtain:
 
∂R
· δU = −{R(U(p−1) )} (19)
∂U (p−1)
G. Botelho, R. Peixoto and R. Pitangueira
Computational System for Physically Non-linear Thermal Analysis

The first term in the Eq. 19 is called tangent matrix, named Λ:

 
∂R
Λ(U(p−1) ) ≡ (20)
∂U (p−1)

The terms of the tangent matrix of a finite element can be calculated by Eq. 21, where lowercase symbols û and c
accompanied by indices were employed to indicate terms of the corresponding matrices and vectors.
n
X ∂cim
Λij = ûm + cij (21)
m=1
∂ûj

The negative of the residue after the (p − 1)st iteration, present in Eq. 19, can be calculated as:

−{R(U(p−1) )} ≡ F − [C(U(p−1) )]U(p−1) (22)

The approximation for the temperature field at the pth iteration is then given by:

Up = U(p−1) + δU (23)

This iterative process continues until a convergence criterion is reached.

3. IMPLEMENTATION REMARKS

3.1 Analysis of INSANE’s Object-Oriented Design

The analysis of INSANE’s object-oriented design focused on identifying:

• The necessary modifications into the classes for data input and output;

• The execution flow and the entities involved during the solution of physically non-linear problems by the software;

• Available classes for the management of the solution of physically non-linear problems by iterative methods;

• The classes to be extended and the methods to be overloaded to represent the particularities associated to the
physically non-linear problem.

3.2 Expansion of the computational system

To allow the analysis of non-linear heat transfer problems by INSANE it was necessary to implement three new classes
into the numerical core as briefly described below. Besides, modifications were made in the methods responsible for data
input and output from the class PersistenceAsXml.

3.2.1 Expansion in the Problem Driver package

The Problem Driver package groups the classes responsible for calculating, at element level, the matrices and vectors
related to the mathematical problem being solved (Fonseca, 2008). In the field of heat transfer, these matrices are the
element’s conductivity and capacitive matrices and the vectors are the ones arising from the loadings applied into the
element’s volume and over its surfaces (Botelho et al., 2015).
Once in non-linear problems it is necessary to mount a tangent matrix for the iterative solution process, as shown in Eq.
20, a new class descending from HeatTransferPd, called NonLinearHeatTransfer, which can be seen in Fig. 1, overloaded
two methods: getIncrementalC() and getC(). Both classes descend from the Parametric class because the present work
focuses on the parametric formulation of the Finite Element Method. A complete description of the parametric formulation
can be found in Weaver and Johnston (1984) and in Navarra (1995).
The method getC() mounts the element’s conductivity matrix, matrix c in Eq. 8. This method had to be overloaded
because in non-linear problems the thermal conductivity is temperature dependent, and in the general case the temperature
field along the element’s geometrical space is not constant. This implies a different constitutive matrix at each element’s
integration point, what requires the evaluation of the temperature at the integration point before requesting the Constitutive
Model to mount the constitutive matrix. This later process of mounting the constitutive matrix by the Constitutive Model
will be further explained in this paper.
The method getIncrementalC() mounts the tangent matrix as described in Eq. 21.
 17th Brazilian Congress of Thermal Sciences and Engineering (ENCIT 2018)
November 25th-28th, 2018, Águas de Lindóia, SP, Brazil

Figure 1. Problem Driver package expansion.

3.2.2 Expansion in the Material package

The Material package contains classes to represent the physical properties of the material that constitutes the medium.
The new class NonLinearThermalIsotropicMaterial, shown in Fig. 2 extended the class ThermalIsotropicMaterial. Heir
classes of the existing classes weren’t shown in Fig. 2. In the new class the material’s physical properties temperature
dependence is described by polynomials. The new class also contains two new methods for obtaining the secant and the
tangent physical properties at a given temperature.

Figure 2. Material package expansion.

3.2.3 Expansion in the Constitutive Model package

The Constitutive Model package arranges the classes engaged in describing the relationship between the inner vari-
ables. In heat transfer problems this law is the relationship between the heat flux per unit area and the temperature
gradient. Since in non-linear heat transfer the constitutive law has a dependence upon the temperature, a new class named
NonLinearHeatTransferCm was implemented extending the class LinearHeatTransferConstitutiveModel.
G. Botelho, R. Peixoto and R. Pitangueira
Computational System for Physically Non-linear Thermal Analysis

Figure 3. Constitutive Model package expansion.

This new class overloaded the methods responsible in charge of mounting the tangent and the secant constitutive
matrices. The overloaded methods receive the temperature at the integration point and then request the Material entity to
calculate the physical properties at the given temperature. Having the physical properties, the constitutive model requests
an entity called Analysis Model to mount the constitutive matrix based on the problem’s spatial dimension. A description
of the entity Analysis Model for heat transfer problems can be found in Botelho et al. (2015).

4. NUMERICAL EXAMPLES

4.1 Reference problems for validation

To validate the software expansion, three different physically non-linear heat transfer problems whose analytical so-
lutions are readily known were solved by INSANE. These problems, presented in the following subsections, consist on
calculating the temperature field in situations where the temperatures are prescribed at the boundaries and the thermal
conductivity has a linear dependence on the temperature.

4.1.1 Reference problem 1: Heat conduction in a rod

The rod shown in Fig. 4 has length L and cross section area Ac . The temperatures at the ending A, where x = 0
and at the ending B, where x = L, are Ta and Tb . The thermal conductivity is ka and kb for temperatures Ta and Tb ,
respectively. The analytical solution of this problem is explained in detail by Danish et al. (2011) and consists of solving
eq. 24:
 
d dT
Ac k(T ) =0 . (24)
dx dx
In eq. 24, k(T ) is the temperature-dependent thermal conductivity. For a linear interpolation between the values of ka
and kb , k(T ) can be written as:
 
T − Ta
k(T ) = ka 1 + β (25)
Tb − Ta
where β is calculated by:
kb
β= −1 , (26)
ka
To allow the solution, the problem is transformed to the dimensionless variables ξ and θ:
x
ξ= , (27)
L
 17th Brazilian Congress of Thermal Sciences and Engineering (ENCIT 2018)
November 25th-28th, 2018, Águas de Lindóia, SP, Brazil

Figure 4. Rod with temperatures prescribed at the endings.

T − Ta
θ= . (28)
Tb − Ta

The solution obtained by direct integration after the coordinate transformation is given by eq. 29.
p
−1 + 1 + 2βξ + β 2 ξ
θ= . (29)
β

4.1.2 Reference problem 2: Heat conduction in a hollow disk

The hollow disk present in Fig. 5 has thickness td , inner raius Ra and outer radius Rb . The temperature at the inner
radius is Ta , where the thermal conductivity is ka , and Tb at the outer radius, where the conductivity is kb . We can arrive
at the analytical solution to this problem by solving eq. 30, following the technique presented by Danish et al. (2011):

Figure 5. Disk with temperatures prescribed at the inner and outer radii.

 
1 d dT
rk(T ) =0 . (30)
r dx dr

The temperature-dependent thermal conductivity varies linearly in the same way described for the rod in equations 26
and 25.
G. Botelho, R. Peixoto and R. Pitangueira
Computational System for Physically Non-linear Thermal Analysis

In the case of the disk, a different coordinate transformation is employed. The dimensionless coordinate ξd for the
disk can be written as:
r − Ra
ξd = . (31)
Rb − Ra

The dimensionless temperature θ remains as described in eq. 28.

After the coordinate transformation and solution by direct integration the non-linear temperature distribution is:
s
  ln 1+ Rb −Ra ξd

−1 + 1 + 2β 1 + β2 R
Ra


b
ln Ra
θ= . (32)
β

4.1.3 Reference problem 3: Heat conduction in a hollow sphere

The hollow sphere seen in Fig. 6 has inner radius Ra and outer radius Rb . The temperature at the inner surface is Ta
and at the outer surface is Tb . As in the cases of the rod and disk , the thermal conductivity at the temperature Ta is ka and
at the temperature Tb is kb . Similarly to the solution of the problem of heat conduction in a disk, the analytical solution
of this problem is obtained by the solution of the eq. 33 using the methodology described by Danish et al. (2011):

Figure 6. Sphere with temperatures prescribed on the inner and outer surfaces.

 
1 d dT
r2 k(T ) =0 . (33)
r2 dr dr

The same coordinate transformation as the employed for the disk, eq. 31, was performed for the sphere, as described
by equation 34:

r − Ra
ξs = . (34)
Rb − Ra

The dimensionless temperature θ remains as described in eq. 28.

The non-linear temperature distribution obtained by direct integration of eq. 33 after transformation to dimensionless
variables is:
s  
 
−1 + 1 + 2β 1 + β2 R R−R b
1 − R −R1a

a b
b
Ra ξs +1
θ= . (35)
β

4.2 Validation of the implementation

The reference problems described in the subsections 4.1.1 , 4.1.2 and 4.1.3 were modeled according to the Finite
Element Method using INSANE. The new classes were then employed to solve these problems.
 17th Brazilian Congress of Thermal Sciences and Engineering (ENCIT 2018)
November 25th-28th, 2018, Águas de Lindóia, SP, Brazil

4.2.1 Solution using one dimension elements - Heat transfer in a rod

Figure 4 shows a rod modeled using 100 one dimensional straight elements of two nodes as depicted in Fig. 7. The
temperature prescribed at the left ending was 0◦ C and at the right ending, 1000◦ C. The thermal conductivity at 0◦ C was
considered 20 W/(m · K) and at 1000◦ C, 1020 W/(m · K). These values lead to the polynomial k(T ) = 20 + T . The
bar length is 10m.

Figure 7. Finite Element Model for the rod.

The comparison between the analytical and the numerical results is shown in Fig. 8. As it can be seen, a good
agreement between the solutions was obtained.

1,000

800
Temperature (◦ C)

600

400

200
Analytical solution
FEM solution calculated by INSANE
0
0 2 4 6 8 10
Distance from the left ending (m)
Figure 8. Results comparison between analytical and numerical solutions

4.2.2 Solution using two dimension elements - Heat transfer in a hollow disk

Figure 5 depicts a hollow disk modeled using 500 two dimensional quadrilateral elements of four nodes as depicted
in Fig. 9 - (a). The boundary conditions, prescribed at the inner and at the outer radii, were the temperatures 0◦ C and
1000◦ C, respectively. The temperature-dependent thermal conductivity was described by the same polynomial that was
employed for the rod. Since the problem is symmetrical, there is no heat flux in the circumferential direction, allowing
the modelling of only a sector of the disk. It was chosen to model a fraction of 11.25 degrees.
The temperature distribution calculated by INSANE can be seen on Fig. 9 - (b), and the comparison between the
analytical and numerical results is shown in Fig. 9 - (c). The numerical solution calculated by the software converged to
the analytical result.

4.2.3 Solution using three dimension elements - Heat transfer in a hollow sphere

In Fig. 6 we can see a hollow sphere that was modeled using 306 three dimensional hexahedral elements of eight
nodes as depicted in Fig. 10. The temperature prescribed on the inner surface was 0 ◦ C and on the outer surface 1000◦ C.
Since the problem is symmetrical, as was also the case of the disk, there is no heat flux neither in the circumferential nor
in the azimuthal directions. We chose to model only a fraction of 5.625 degrees of the sphere. The thermal conductivity
is k(T ) = 20 + T , the same used for the rod and for the disk.
Figures 11 - (a) and 11 - (b) exhibit, respectively, the temperature distribution in the sphere calculated by INSANE
and the comparison between the analytical and the numerical analyses. The results obtained validated the solution for the
three dimensional case, since they presented good agreement.
G. Botelho, R. Peixoto and R. Pitangueira
Computational System for Physically Non-linear Thermal Analysis

Figure 9. Analysis of the disk. FEM Model (a) - Temperature Distribution (b) - Results comparison (c)

Figure 10. Finite Element Model of the sphere.

5. CONCLUSIONS

Classes for managing and solving physically non-linear heat transfer problems were implemented into INSANE’s
numerical core. These classes were supported by the existing source code and they were introduced in the software with
minimum impact on the generality and modularity. Once the inheritance mechanism was used, only the particularities
associated to the non-linear heat transfer problem had to be described.
 17th Brazilian Congress of Thermal Sciences and Engineering (ENCIT 2018)
November 25th-28th, 2018, Águas de Lindóia, SP, Brazil

Figure 11. Temperature distribution on the sphere and comparison.

Since the solution algorithm focused on the parametric formulation of the Finite Element Method, the classes written
were able to analyze models containing parametric elements of any type: line, plane and solid elements.
Reference problems whose analytical solutions are readily known were modeled and solved by INSANE and the
results validated the implementation presented in this paper.

6. ACKNOWLEDGEMENTS

The authors gratefully acknowledge the important financial support of the Brazilian research agencies CAPES (in
Portuguese "Coordenação de Aperfeiçoamento de Pessoal de Nível Superior"), CNPq (in Portuguese "Conselho Nacional
de Desenvolvimento Científico e Tecnológico") and FAPEMIG (in Portuguese "Fundação de Amparo à Pesquisa do Estado
de Minas Gerais").

7. REFERENCES

Bergman, T.L., Lavine, A.S., Incropera, F.P. and Dewitt, D.P., 2011. Fundamentals of Heat and Mass Transfer. John
Wiley & Sons, New Jersey, 7th edition.
Botelho, G.G., da Fonseca, F.T. and Pitangueira, R.L.S., 2015. “Sistema computacional orientado a objetos para análises
acopladas termo-estruturais pelo método dos elementos finitos”. In Proceedings of the XXXVI Ibero-Latin American
Congress on Computational Methods in Engineering - CILAMCE 2015. Rio de Janeiro, Brazil.
Danish, M., Kumar, S. and Kumar, S., 2011. “Exact analytical solutions of three nonlinear heat transfer models”. In
Proceedings of the World Congress on Engineering. London, UK.
Fonseca, F.T., 2008. Sistema Computacional para Análise Dinâmica Geometricamente Não Linear Através do Método
dos Elementos Finitos. Master’s thesis, UFMG - Universidade Federal de Minas Gerais, Belo Horizonte, MG, Brasil.
Logan, D.L., 2007. A First Course in the Finite Element Method. Thomson, Ontario, Canada, 4th edition.
Navarra, E.O.I., 1995. Calculo de Estructuras por el Metodo de Elementos Finitos. Centro Internacional de Métodos
Numéricos en Ingenieria, Barcelona, Spain, 2nd edition.
Reddy, J.N., 2004. An Introduction to Nonlinear Finite Element Analysis. Oxford University Press, New York, 1st edition.
Weaver, W. and Johnston, P.R., 1984. Finite Elements for Structural Analysis. Prentice-Hall Inc., 1st edition.
Zienkiewicz, O.C. and Taylor, R.L., 2000. The Finite Element Method - Volume 1: The Basis. Butterworth Heinemann,
Massachusetts, 5th edition.

8. RESPONSIBILITY NOTICE

The authors are the only responsible for the printed material included in this paper.