You are on page 1of 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/279193632

Thermal Fluid-Structure-Interaction of cooling processes in steel forming

Chapter · January 2009

CITATIONS READS

0 47

4 authors, including:

Stefan Hartmann Andreas Meister


Technische Universität Clausthal Universität Kassel
130 PUBLICATIONS   1,437 CITATIONS    154 PUBLICATIONS   864 CITATIONS   

SEE PROFILE SEE PROFILE

Philipp Birken
Lund University
55 PUBLICATIONS   246 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Modeling of the active response of arteries View project

Influence of the material layer thickness on the mechanical properties and the forming of three-layer sandwich composites View project

All content following this page was uploaded by Stefan Hartmann on 26 June 2015.

The user has requested enhancement of the downloaded file.


Thermal Fluid-Structure-Interaction of cooling
processes in steel forming
Stefan Hartmann, Andreas Meister, Philipp Birken und Karsten J. Quint

Abstract

In this article the cooling process in a partially heated metal forced by a blown gas is modelled by
thermal fluid-structure interaction. The thermal coupling of the compressible Navier-Stokes equa-
tions in the fluid region with the heat equation in the solid region is considered. The semi-discrete
coupled system is solved using stiffly stable singly diagonally implicit Runge-Kutta methods of
higher order, where on each stage a thermal fluid-structure-interaction problem is solved. Here, the
idea is exploited that if both solvers, for the fluid and the solid, are able to compute one Backward-
Euler step, the entire partitioned FSI-approach can be treated using diagonally implicit Runge-
Kutta methods to obtain a higher order integration scheme. The applied second-order method is
shown to reach its expected order, which is shown by order diagrams.

1 Introduction
In metal forming processes with local heat treatment one is interested in crystallographic phase
transition effects to predict the local material properties, such as ductility, hardness, yield strength,
or impact resistance. Since a subsequent cooling step using a cold gas has a strong influence on the
evolution of the temperature field, and, accordingly, to the evolution of the material properties, the
realistic numerical treatment of such processes is of large interest. This is the keypoint of the real-
istic representation of phase transitions phenomena in the steel under consideration. Morever, it is
a substantial ingredient to treat and optimize the new process technology. Recently, air cooling, as
opposed to liquid or contact cooling, has received scientific interest, see [1], and experimental
results on this are described in [2]. A numerical simulation of the process needs to take into ac-
count thermo-mechanical coupling effects in the cooling gas (fluid-mechanical part), the heat
transport within the solid (solid mechanical part) and thermo-mechanical coupling effects in the
solid itself. This is an extremely complicated model and therefore, as a stepping stone for further
work, we neglect all mechanical effects in the solid.
Thus, we consider the fluid-structure-interaction (FSI) of the compressible Navier-Stokes-
equations as a model for air, along a non-moving boundary with the linear heat equation as a first
model for the temperature distribution in steel (for the non-linear case of the heat equation solved
by means of higher-order time integration procedures see [3] and for the numerical treatment of
the compressible Navier-Stokes equations alone, see [4]). While a lot of work has been done on
thermal coupling of incompressible fluids with structures, we look at thermal coupling of a com-
pressible fluid and a solid structure. Research on numerical simulation of this problem was so far
mainly driven by problems where hot gas heats the structure, for example supersonic re-entry of
vehicles from space or cooling of gas-turbine blades [5,6]. The results are mainly qualitative, de-
scribing numerical methods and the comparison of numerical results to experimental data, with the
conclusion that the results are not always in agreement with experiments [6]. Accordingly, the
numerical treatment of the cooling process has to be investigated.

Figure 1: Picture of experimental setup of the air cooling process

For the fluid-structure interaction, a partitioned approach is considered, see [7], where different
codes are used for each sub-problem and the coupling is done by a master program calling the
interface functions of the other codes. This allows the user to use existing software for each sub-
problem, by contrast to a monolithic approach, where a new code is taylored for the coupled equa-
tions. For these particular equations, we use a finite volume method (FVM) for the fluid and a
finite element method (FEM) for the heat equation. Specifically, for the fluid use is made of the
DLR TAU-Code [8] and for the structure the inhouse FE-program TASA-FEM for high order time
integration [9]. Another distinction is made between loose coupling and strong coupling methods.
In the first approach, only one step of each solver is performed in each time step, while the latter
approach adds a convergence criterion and an inner loop. See the monograph [10] for an overview
of the current state of the art.
For the time-discretization, it is common to apply low order implicit methods for the flow, struc-
ture and coupling solver. However, for each sub-solver it has recently been pointed out that high
order time integration methods can offer gains in efficiency for engineering accuracy, in particular
if implicit Runge-Kutta methods are used [11,12]. For the solution of differential-algebraic equa-
tions (DAEs) it is known that stiffly accurate singly diagonally implicit Runge-Kutta (SDIRK)
methods perform well [13]. One major advantage using SDIRK-methods results from the fact that
each stage computation reads like a Backward-Euler step with different initial value and different
time-step. Accordingly, if both the fluid and solid solver are able to compute one Backward-Euler
step, a superordinate driver routine can control each sub-solver in a partitioned manner to obtain a
higher order scheme. Additionally, we think, that a higher order time integration approach is worth
looking at also for fluid-structure interaction problems. Moreover, it offers the possibility to apply
time-adaptive schemes if the higher order is achieved. In a first step, the second-order method of
Ellsiepen [14] is applied and second order is demonstrated through numerical results.
Only limited attention has been paid to the aspect of more sophisticated time integration for FSI
problems so far. In [15], energy conservation for a problem in aeroelasticity is analyzed using the
implicit midpoint rule in a monolithic scheme. Already in [16,17], it is suggested to use an explicit
high-order Runge-Kutta scheme for both sub-problems with data exchange at each stage. Due to
the explicit nature, the resulting scheme has severely limited time steps. The order of coupling
schemes on moving meshes is analyzed in [18], but only convergence of first order is proved for p-
th order schemes. The combination of high order implicit Runge Kutta schemes for problems on
moving meshes is explored in [19] for the one dimensional case and in the subsequent paper [20]
for 3D calculations. There, so-called explicit first stage, singly diagonally implicit Runge-Kutta-
schemes (ESDIRK) are employed and higher order in time is demonstrated by numerical results.
Since the coupling problem can be seen as a DAE-system, the class of stiffly stable singly diago-
nally implicit Runge-Kutta schemes (stiffly stable SDIRK methods) for fluid-structure interaction
is explored, taking the thermal coupling as model problem. To this end, we consider the coupled
semi-discrete equations and apply the implicit method to these. Thus, at each stage a coupled FSI
problem is solved.
Although significant simplifications have been done in the modelling process, the remaining cou-
pled problem is still far beyond current mathematical solution techniques. Therefore, a test case
has been developed, that has a comparably simple solution: a parallel airflow streams along a flat
plate, in which a hot steel block has been inserted. For the validation of the newly developed nu-
merical solver, experiments have been performed in [21].
The outline of the paper is as follows: first, the governing equations are presented and the corre-
sponding spatial and temporal discretizations are summarized. In Section 3 the coupling solver is
described and in Section 4 the test example is explained.

2 Governing equations and discretization


We consider a thermal coupling problem having a fluid domain Ω1 ⊂ R 2 where the compressible
Navier-Stokes equations are valid and a structure domain Ω 2 ⊂ R 2 with the unsteady linear heat
equation. The domains meet at an interface Γ consisting of a curve in R 2 , where we require that
temperature and heat flux are continuous. The remaining interface has no coupling conditions.
To comply with the condition that temperature and heat flux are continuous at the interface Γ , a
so-called Dirichlet-Neumann-coupling is used. Namely, the boundary conditions for the two
solvers are chosen such that we prescribe Neumann data for one solver and Dirichlet data for the
other. Following the stability analysis of Giles [22], temperature is prescribed for the equation with
smaller heat conductivity, namely the fluid and heat flux for the structure.
In the structure, the finite element code TASA-FEM for high-order time-integration, originally
based on stiffly accurate, diagonally implicit Runge-Kutta methods, see [23] is used. Both the
model and the numerical method are described in detail in [3]. On the boundary, we have Neu-
mann conditions. The space discretization leads to a system of ordinary differential equations for
the absolute temperatures Θ :

 ) = MΘ
g (t , Θ, Θ  (t ) + K (Θ)Θ(t ) − q(t , u) = 0. (1)

Here, Θ are the unknown nodal temperatures of the solid part. The heat flux q on the coupling
boundary includes a dependence on the fluid data u as explained next. M and K are the heat ca-
pacity and heat conductivity matrices. A Backward-Euler step of Eq. (1) reads

M

+ ∆tn K  Θn+1= M Θn + q(tn+1, u n+1) (2)

implying the solution of a symmetric, sparse linear system of equations to obtain the nodal tem-
peratures at time tn +1 .

Concerning the fluid part, the finite volume scheme used is the DLR-TAU-Code and both the
model and the numerical method are described in detail in [4]. If we write this as an equation for
the complete domain, we obtain

u=
(t ) R(u, Θ). (3)

Here, u is the vector of fluid variables and R represents the space discretization. We have in-
cluded the dependence on the structural temperature on the coupling interface through the vector
of the nodal temperatures Θ in the structure. A Backward-Euler step for equation (3) reads

u n+=
1
(
u n + ∆tn R u n+1, Θn+1 ). (4)

It can easily be seen that each time step within the fluid solver requires the solution of a sparse
non-linear system of equations, which is performed by a dual time-stepping approach, see [24].
Obviously, Eqns.(2) and (4) are coupled.
Time discretization and coupled equations
For the time discretization, we consider SDIRK methods of first and second order, in particular the
implicit or Backward-Euler method of first order and Ellsiepen’s method [14] of second order,
which is an embedded DIRK-method. The latter consists of two subsequent Backward-Euler steps
with specific time steps and starting vectors. The coefficients can be seen in the following
Butcher-tableau where α = 1 − 2 / 2 and the coefficients for the lower order method
5
use αˆ= 2 − 2:
4
α α 0
1 1−α α
1−α α
1 − αˆ α̂

Like all DIRK-methods, Ellsiepen's method is not B-stable and may thus exhibit a loss of order for
particular problems [13]. This was tested numerically for the uncoupled subproblems and found
not to be the case there. The error estimation is done using a classical estimation from [13]. It turns
out that an error estimation based solely on the error in the structure is not able to capture the rele-
vant physics and results in time steps that are too large.
The semi-discrete equations (2) for the domain Ω1 and (4) for the domain Ω 2 represent for a
Backward-Euler step the coupled non-linear system

u n+=
1
(
u n + ∆tn R u n+1, Θn+1 , ) (5)
n +1 n +1
M

+ ∆tn K  Θ = M Θ + q(tn+1, u
n
).

where we assume that both temperature and heat flux are continuous at the coupling interface Γ .
Ellsiepen's method leads to the following algorithm, as described in [25]:
1. Given data u n , Θ n , solve the coupled system (5) using the Backward-Euler method with time
step α∆tn and obtain the stage values u1n , Θ1n .
2. Compute the new starting vector

2  u   u  
n n n
 uS   u 
 =   +   −   
 Θ S   Θ  2α   Θ 1  Θ  
3. Given data uS , Θ S , solve the coupled system (5) using the Backward-Euler method with time
step α∆tn and obtain u n +1 , Θ n +1 .
In steps one and three, a coupled non-linear system has to be solved and in step two, the starting
vector for the second stage is computed.

3 Fluid-Structure-Coupling
As described in the introduction, we pursue a partitioned approach. The technical difficulty of
different programming languages (FORTRAN for TASA-FEM and C++ for TAU) in the parti-
tioned approach is dealt with using a C++-library called Component Template Library (CTL) [26].
It is assumed that at time tn the fluid data u n , the structure data Θ n and a global step size ∆tn are
known. As described above, the fluid and the structure equations are both treated implicitly with
associated solvers for the time stepping procedure. In the coupling context, it is useful to regard
the two solvers as mappings that, for given fixed data u n respectively Θ n take an approximation
to the boundary data at tn +1 from the other solver and provide a new approximation to their data at
tn +1 . Thus, a new approximation to the boundary data at tn +1 for the other solver is obtained [27].
The fluid solver provides a solution to (4) and can be written as

F ( P(Θ)) =u n+1, (6)

whereas the structure solver provides a solution to (2) and can be written as

S (qΓ (u)) = Θn+1. (7)

P is a projection onto the boundary of Ω 2 and qΓ provides the boundary heat flux in the fluid.
Using this notation, it is possible to define coupling methods. The most simple coupling proce-
dures are loose coupling methods, where no convergence criterion is used in the coupling iteration.
In particular, there is Gauss-Seidel-coupling

u n+=
1
F ( P(Θn )), (8)
Θn+1 = S (qΓ (u n+1)).

As a fixed point equation this is given by

=
P(Θ) S (qΓ ( F ( P(Θ)))), (9)

which can be used as a convergence criterion for the fixed point iteration.
4 Numerical Example
To analyze the properties of the coupling method, we choose a test case that is as simple as possi-
ble. The reason is that this comparably simple coupling problem is already beyond the possibilities
of current mathematical analysis. Therefore, we want to make sure that at least the exact solutions
for the uncoupled problems are known, in order to make sure that no additional side effects are
present, which cannot be controlled.
Accordingly, the cooling of a flat plate resembling a simple work piece is considered. A similar
example has been studied by other authors [28] in conjunction with the cooling of structural parts
in hypersonic vehicles. There, localized heating was of special interest. In our case the work piece
is initially at a much higher temperature than the fluid and then cooled by a constant air stream.
The latter is modeled in a first approximation as a laminar flow along the plate, see figure 1.

Figure 1: Geometry of the test case

At the left is the inlet, where air enters the domain with an initial velocity of Ma ∞ = 0.8 in hori-
zontal direction and a temperature of 273 K. Then, there are two succeeding regularization regions
of 50mm to obtain an unperturbed boundary layer. In the first region, 0 ≤ x ≤ 50, symmetry
boundary conditions are applied. In the second region, 50 ≤ x ≤ 100, a constant wall temperature
of 300 K is specified. Within this region the velocity boundary layer fully develops. The third part
is the solid (work piece) of length 200 mm, which exchanges heat with the fluid, but is assumed
insulated otherwise, i.e. q = 0 . Therefore, Neumann boundary conditions are applied throughout.
Finally, the flow domain is closed by a second regularization region of 100 mm with symmetry
boundary conditions and the outlet.
The grid is chosen cartesian and equidistant in the structural part, whereas in the fluid region the
thinnest cells are on the boundary and then become coarser in y -direction. The points of the pri-
mary fluid grid, where the heat flux is located in the flow solver, and the nodes of the structure
grid match on the interface, which avoids additional difficulties from interpolation.
To specify reasonable initial conditions within the fluid, a steady state solution of the flow with
constant wall temperature Θ =900 K is computed. Due to the flat plate, we are able to validate
the results using the theoretical solution of Blasius for the velocity boundary layer and the experi-
mental results of van Driest for the temperature boundary layer [29]. In the structure, an initial
constant temperature of 900 K is chosen throughout.
In figure 2 one can see the temporal evolution of the temperature at the coupling interface. As
expected, the temperature decreases monotonously with a large gradient in the beginning which
then decreases. At t = 1s, the temperature has dropped to 895 K. Furthermore, at the beginning of
the solid part with the coordinates (x,y) = (100,0) the temperature decrease is much more than in
the middle region, e.g. (x,y) = (200,0), where the gas has already been heated. I.e. from the left to
the right the heat absorption is strongly inhomogeneous.
Since no exact solution is available, we computed reference solutions using a time step of
∆t =0.001 . They were all obtained using the Backward-Euler method using fixed point coupling
and a fixed tolerance for all involved equation solvers.
As a first test case we use a steel block, where the following constant material properties are as-
sumed: mass density ρ = 7836 kg/m3 , specific heat capacity cD = 443 J/(kgK) and thermal con-
ductivity λ = 48,9 W/(mk). The computation is done on a grid with 9660 cells in the fluid region
and nx × n y= 120 × 9 = 1080 elements with 121 × 10 = 1210 nodes in the structure. In the fluid
domain, the thinnest cells have an aspect ratio of 1:200 and then become coarser in y-direction.
In the following, the Backward-Euler method is compared with Ellsiepen’s method. Thus, we have
a method of second order that has double the computational effort per step compared to the first
order method. Figure 2 shows the relative error at the interface, compared to the reference solution
for different time-step sizes. Here, only the interface is considered in view of practical interests.
All tolerances for the termination of the non-linear solver were set to ε = 10 −7 .

Figure 2: Order diagram (left) and temperature evolution at the interface (right)

As one can see, the coupled scheme using the Backward-Euler method is of first order for both
loose and strong coupling of the codes. Ellsiepen’s method with strong coupling reaches second
order, but if only lose coupling is employed, the method reduces to first order.
5 Summary and Conclusions
A coupling of the compressible Navier-Stokes equations and the heat equation was considered,
where the fluid cools the structure. Thereby, we developed a novel stiffly stable SDIRK time inte-
gration approach to obtain a higher order FSI method in time. The new method is suitable for the
solution of DAEs and reaches higher order in time.
In the near future, we will validate the method with the available experimental results [21], per-
form time adaptive calculations and apply the method to a more complex structural model that
includes nonlinearities in the heat equation for the phase change in the steel, as well as the impor-
tant factor of radiation. Further work includes the extension to thermomechanical coupling be-
tween fluid and structure, the implementation of even higher orders, moving grids and improve-
ment of the fixed point solver.

References
[1] Weidig, U., Saba, N., Steinhoff, K.: Massivumformprodukte mit funktional gradierten Eigenschaften durch eine
differenzielle thermo-mechanische Prozessführung, 2007, WT-Online, S. 745-752.
[2] K. Steinhoff, U. Weidig, N. Saba: Investigation of plastic forming under the influence of locally and temporally
variable temperature and stress states, in (Eds.) K. Steinhoff, K.-J. Maier, B. Svendsen: Functionally graded materials
in industrial mass production, Verlag Wissenschaftliche Scripten, Auerbach 2009
[3] S. Hartmann, D. Kuhl, K.-J. Quint: Time-adaptive computation of finite strain thermo-viscoplastic structures, in
(Eds.) K. Steinhoff, K.-J. Maier, B. Svendsen: Functionally graded materials in industrial mass production, Verlag
Wissenschaftliche Scripten, Auerbach 2009
[4] A. Meister, P. Birken: Asymptotics based simulation of thermo-mechanical cooling processes, in (Eds.) K. Steinhoff,
K.-J. Maier, B. Svendsen: Functionally graded materials in industrial mass production, Verlag Wissenschaftliche
Scripten, Auerbach 2009
[5] M. Hinderks, R. Radespiel: Investigation of hypersonic gap flow of a reentry nosecap with consideration of fluid
structure interaction, AIAA Paper 06-1111.
[6] R. C. Mehta: Numerical computation of heat transfer on reentry capsules at mach 5, AIAA-Paper 2005-178.
[7] C. Farhat: CFD-based nonlinear computational aeroelasticity, in: E. Stein, R. de Borst, T. J. Hughes (eds.), Encyclo-
pedia of Computational Mechanics, Volume 3: Fluids, John Wiley and Sons, 2004, pp. 459-480.
[8] T. Gerhold, O. Friedrich, J. Evans, M. Galle: Calculation of complex three-dimensional configurations employing
the DLR-TAU-code, AIAA Paper 97-0167.
[9] S. Hartmann, TASA-FEM: Ein Finite-Elemente-Programm für Raum-Zeitadaptive gekoppelte Strukturberechnungen,
Mitteilungen des Instituts für Mechanik 1.
[10] S. Hartmann, A. Meister, M. Schäfer, S. Turek (Eds.): Fluid-Structure-Interaction: Theory, Numerics and Applica-
tions, Kassel University Press, Kassel, 2009
[11] H. Bijl, M. H. Carpenter, V. N. Vatsa, K. C. A.: Implicit time integration schemes for the unsteady compressible
navier-stokes equations: Laminar flow, J. Comp. Phys. 179 (2002) 313-329.
[12] S. Hartmann: A remark on the application of the Newton-Raphson method in non-linear finite element analysis,
Computational Mechanics 36 (2005)100-116.
[13] E. Hairer, G. Wanner: Solving Ordinary Differential Equations II, Series in Computational Mathematics 14, 3. edi-
tion, Springer, Berlin, 2004.
[14] P. Ellsiepen: Zeit- und ortsadaptive Verfahren angewandt auf Mehrphasenprobleme poröser Medien, Dissertation,
University of Stuttgart, Institute of Mechanics II (1999).
[15] R. Massjung: Discrete conservation and coupling strategies in nonlinear aeroelasticity, Comp. Meth. Appl. Mech.
Engrg. 196 (1-3) (2006) 91-102.
[16] O. Bendiksen: A new approach to computational aeroelasticity, AIAA Paper AIAA-91-0939-CP (1991) 17120--
1727.
[17] G. A. Davis, O. O. Bendiksen: Transonic panel flutter, AIAA paper AIAA 93-1476.
[18] H. Guillard, C. Farhat: On the significance of the geometric conservation law for flow computations on moving
meshes, Comp. Meth. Appl. Mech. Engrg. 190 (2000) 1467-1482.
[19] A. H. van Zuijlen, H. Bijl: Implicit and explicit higher order time integration schemes for structural dynamics and
fluid-structure interaction computations, Computers & Structures 83 (2005) 93-105.
[20] A. H. van Zuijlen, A. de Boer, H. Bijl: Higher-order time integration through smooth mesh deformation for 3d fluid-
structure interaction simulations, J. Comp. Phys. 224 (2007) 414-430.
[21] Stempkovskyy, S.: Konzeption, Konstruktion und Aufbau eines Versuchsstandes zur Messung von konvektiven
Abkühlprozessen, Diploma Thesis, Institute of Mechanics, University of Kassel, 2009
[22] M. Giles: Stability analysis of numerical interface conditions in fluid-structure thermal analysis, Int. J. Num. Meth. in
Fluids 25 (1997) 421-436.
[23] P. Ellsiepen, S. Hartmann: Remarks on the interpretation of current non-linear finite-element-analyses as differential-
algebraic equations, Int. J. Num. Meth. Eng. 51 (2001) 679-707.
[24] A. Jameson: Aerodynamics, in: E. Stein, R. de Borst, T. J. Hughes (eds.), Encyclopedia of Computational Mechanics,
Volume 3: Fluids, John Wiley and Sons, 2004, pp. 325-406.
[25] P. Birken, S. Hartmann, A. Meister, K. Quint: On Higher Order Time Integration for Thermal Coupling, in T. Simos,
G. Psihoyios, Ch. Tsitouras (eds.), Numerical Analysis and Applied Mathematics, International Conference on Nu-
merical Analysis and Applied Mathematics 2009, Volume 2, AIP Conference Proceedings, 2009, pp. 1184-1187.
[26] H. G. Matthies, R. Niekamp, J. Steindorf: Algorithms for strong coupling procedures, Comput. Methods Appl. Mech.
Engrg. 195 (2006) 2028-2049.
[27] P. Birken, K. J. Quint, S. Hartmann, A. Meister: On Coupling Schemes for Heat Transfer, in FSI applications, Pro-
ceedings of the International Workshop on Fluid-Structure Interaction, Herrsching, 2008, S. 21-30.
[28] P. W. Yarrington, E. A. Thornton: Finite element analysis of low-speed compressible flows within convectively
cooled structures, J. of Thermophysics and Heat Transfer, S. 8-4.
[29] E. Van Driest, National Advisory Commitee for Aeronautics (NACA) - Investigation of laminar boundary layer in
compressible fluids using the crocco method, NACA.

View publication stats

You might also like