Professional Documents
Culture Documents
ABSTRACT: An adaptive finite element method for high-speed flow-structure interaction is pre-
sented. The cell-centered finite element method is combined with an adaptive meshing technique to
solve the Navier-Stokes equations for high-speed compressible flow behavior. The energy equation and
the quasi-static structural equations for aerodynamically heated structures are solved by applying the
Galerkin finite element method. The finite element formulation and computational procedure are de-
scribed. Interactions between the high-speed flow, structural heat transfer, and deformation are studied
by two applications of Mach 10 flow over an inclined plate, and Mach 4 flow in a channel.
of the finite-element flow solution and the aerothermal where the subscript T denotes the structural heat
loads, as well as to reduce the computational time transfer analysis. The vector UT contains the ther-
and memory. The GMerkin finite element method real conservation variable defined by,
is applied to solve the structural energy equation for
temperature distribution and the structural equations UT = pcT (4)
for deformations and stresses. The paper starts by where c is the specific heat of the structure and the
explaining the theoretical formulation of high-speed heat flux components ET and FT are
compressible flow, structural heat transfer, and struc-
tural response. Then the solution procedure for flow- -k ~ and : -k ~
ET = ax Oy (5)
thermal-structural interaction is presented. The basic
idea behind the adaptive meshing technique is then and GT is the heat source.
described. The combined procedure, the cell-centered 2.1.3 Structural Response
finite element method and the adaptive meshing tech- The structural response is governed by the quasi-
nique, and the integrated approach are evaluated by static equilibrium equations given by
analyzing two applications of Mach 10 flow over an
inclined plate and Mach 4 flow in a channel for the OEs+ ~yFs = 0 (6)
fluid-thermal-structural interaction.
where the subscript S denotes the structural analysis.
2 THEORETICAL FORMULATION AND The flux vector components E s and F s are
SOLUTION PROCEDURE
Es = and Fs = (7)
2.1 G o v e r n i n g E q u a t i o n s Txy 0-y
The equations for the high-speed compressible where the stress components ~x, Cry, and Txy are re-
flow, the structural heat transfer, and the structural lated to the strain and the temperature by the gener-
analysis in two dimensions are described below. alized Hook's law [m] .
2.1.1 High-Speed Compressible Flow 2.2 Finite Element Formulation
The equations for high-speed viscous compress- The cell-centered finite element method is ap-
ible flow are represented by the conservation of mass, plied to the Navier-Stokes equations to derive the fi-
momentums, and energy. These equations are written nite element equations. The Galerkin finite element
in the conservation form [1~ as approach is applied to the structural heat transfer
equation and the equilibrium equations to derive the
~-~UF + EF + FF = 0 (1) corresponding finite element equations. The deriva-
tion procedures are briefly described below.
where the subscript F denotes the fluid analysis. The
2.2.1 Finite-Element Flow Equations
vector UF contains the fluid conservation variables The method of weighted residuals [12[ is applied
defined by
to Eq.(1), over the element domain, f2, by using the
unit interpolation function as
pu
UF = pv (2)
/n ff--~UFdX?= -- ~ ~--~EFd ~ -
pc
where p is the fluid density, u and v are the velocity
/ n ~yFFd~? (8)
components in the x and y directions, respectively,
and e is the total energy. The vectors E and F con- The Gauss divergence theorem is then applied to the
sist of the flux components in the x and y directions, flux integral terms of Eq.(8) to yield
respectively[ s] .
2.1.2 Structural Heat Transfer L ~EFd/2 + ~ ~yFF dr2=
The thermal response of the structure is de-
scribed by the energy equation in the conservation
form as
ro(G~ + Gv). r (9)
where the flux vectors GI and G v are the inviscid
~UT + ET + FT = GT (3) and viscous flux vectors of EI + F I and E v + F v ,
Vol.20, No.6 Limtrakarn W & Dechaumphai P: FEM for High-speed Flow-structure Interaction 599
respectively, and fi is the unit vector normal to the where M is the mass matrix, A U ~ +1 = U ~ +1 - U ~
element boundary Fe. Equation (9) is evaluated by at time n + 1. The /i~. 1 and R~. 2 vectors are as-
summing the normal fluxes from all the sides, Fe, of sociated with the thermal fluxes within element and
the element. The fluxes normal to the element sides across element boundary, respectively, and are given
are then approximated by the numerical inviscid and by
viscous fluxes, G1 and O r . By applying an explicit
time marching algorithm[ 1~ Eq.(9) becomes R~,l = fo ~--~zNdOE~ + f~ ~-y Ndf2F~r (15)
A~e(U~+I - u~) = - ~ (~s(GI -~ GV) (10)
At R~r2 = - / N(E~n~ + F~%)d_r (16)
S JG
where vFirrn+l and U F'~are the conservation variables at
2.2.3 Finite-element Structural Equations
time steps n + 1 and n, respectively, Ar is the element
The Galerkin finite element method is applied to
area, 5s is the length of the element side, and At is
Eq.(6) in the same fashion as in the structural heat
the allowable time step following the CFL and viscous
transfer analysis. The finite element equations can
stability requirement [sJ .
also be derived in the form
The basic concept behind the cell-centered finite
element method used in this paper is to determine K U s = Rs + RT (17)
the flux across element interfaces by Roe's averaging
procedure [9]. The average inviscid flux GI is given by where K is the stiffness matrix, Us is the nodal dis-
placement vector, Rs is the external load vector, and
Oi= ~[G~ § (11) RT is the thermal load vector. These matrices are
defined by
where the superscripts L and R denote the left and
right elements, respectively. The last term in Eq.(11) K = ~ BTCBdY2 (18)
may be viewed as the artificial diffusion needed for
solution stability.
By substituting E q . ( l l ) into Eq.(10), the incre- Rs = fv~ N T F s d F (19)
ments of the conservation variables, AUF = U~ +1 -
U~, can be computed explicitly from RT = f o BTCc~(T - T0)dO (20)
(V~ § - U~:)de 1
at = - 2 z as +aiR+ where B is the strain-interpolation matrix, C is the
s elastic modulus matrix, F s is the surface traction ma-
trix, c~ is the thermal expansion coefficient vector, and
IA*I(U)- sea)] - ~ asOv (12) To is the reference temperature for zero stress state.
S
2.3 S o l u t i o n S e q u e n c e
2.2.2 Finite-Element Structural Heat Transfer
For high-speed compressible flow, the flow be-
Equations
havior normally approaches the steady state in a much
The method of weighted residuals is applied to
shorter time than those for the thermal and structural
Eq.(3), over the element domain, f2, by assuming a
response of the structure. Typically, the heating rate
linear distribution of the conservation variable UT,
approaches the steady state in about a few millisec-
and the flux components ET and FT in the form
onds. At this period, the structural configuration re-
UT(x,y,Q = N(x,y)UT(t) (13a) mains nearly undeformed at temperature only slightly
higher than the initial temperature. After a few sec-
ET(x,y,Q = N(x,y)ET(t) (13b) onds, the structural temperature begins to rise appre-
ciably and significant deformation may occur. At this
FT(x,y,t) = N(x,y)FT(t) (13c)
time, the thermal and deformation coupling effect can
where N(x, y) is the linear interpolation function ma- alter the flow field. The coupling effect continues to
trix. The finite element equations can then be derived alter the flow and structure behavior until the struc-
in the form ture reaches the state of the thermal equilibrium.
Based on the fact that the high-speed flow be-
M A U ~ +1 = RT~~+ RT~ (14) havior normally reaches the steady-state condition in
600 ACTA MECHANICA SINICA 2004
a much shorter time t h a n the structural response, the ter a short interval of time at t = 11, the predicted
analysis procedure of the flow-structure interaction aerodynamic heating rate is applied to the structural
presented in this p a p e r consists of the solution se- configuration and structural heat transfer analysis as
quence as described in Fig.1. This solution sequence denoted by TA (Thermal Analysis) is used to solve
can reduce the total computational time by avoiding for the structural temperature. Both the structural
detailed transient flow analysis which requires signifi- t e m p e r a t u r e and the fluid pressure are then used to
cant computational effort. At the initial time, t = to, predict the structural response for deformation and
the adaptive cell-centered finite element method is stresses as denoted by SA (Structural Analysis). The
first used to predict the high-speed flow behavior as same sequence is repeated to predict the new flow field
denoted by FA (Flow Analysis). The flow analysis behavior, the aerothermal loads, the structural tem-
generates aerothermal loads t h a t include the heating perature, as well as the new structural deformation
rate and pressure along the structural surface. Af- and stresses.
update / - - - computational
domain update
domain / fluid domain
/
3 ADAPTIVE MESHING TECHNIQUE the number of elements and their connectivities the
same but relocates the nodes [15].
Adaptive mesh generation techniques may be The remeshing technique, the second adaptive
classified into two major categories: (1) refinement/ mesh-generation category, generates an entirely new
derefinement, and (2) remeshing. The first category, mesh based on the solution obtained from an earlier
the adaptive refinement/derefinement technique, can mesh [16'17]. The technique is combined with the cell-
be further classified into three subcategories: (a) the centered finite element method in this paper to solve
h method, (b) the p method, and (c) the 1" method. high-speed compressible flow problems. The idea is
In the h method, the elements in the initial mesh are to construct a new mesh that consists of small el-
refined into smaller elements or derefined into larger ements in the regions with large change in solution
elements[13]. The p method maintains the geometry of gradients and large elements in the other regions with
the elements of the initial mesh but increases (or de- small changes in solution gradients. As an exam-
creases) the order of the polynomials used for the ele- ple, small elements are needed in the regions of shock
ment interpolation functions [141. The r method keeps waves to capture shock resolution, whereas hrger el-
Vol.20, No.6 kimtrakarn W & Dechaumphai P: FEM for High-speed Flow-structure Interaction 601
ements can be used in the free-stream region because 4.1 Mach 10 F l o w o v e r a n I n c l i n e d Plate
the flow behavior is uniform. To determine proper The problem statement of a Much 10 flow over
element sizes at different locations in the flow field, an inclined plate is illustrated in Fig.2(a). The flow
the solid-mechanics concept of determining the prin- enters through the left boundary of the computationM
cipal stresses from a given state of stresses at a point fluid domain and creates an oblique shock wave as
is employed. Since the fluid density changes abruptly highlighted in the figure. The figure shows that a
across the shock waves, the density distribution can panel is supported by the panel holder in an initial
be used as an indicator in the determination of proper flat orientation and is being heated at time t = 0 by
element sizes. high-speed flow. At the left and right ends of the
Because small elements must be placed in the re- panel, conduction heat transfer is permitted to tile
gion of shock waves with large changes in the density panel holder. The b o t t o m surface of the panel is as-
gradients, the second derivatives of the density at a sumed perfectly insulated and at the top surface an
point with respect to global coordinates x a n d / j are aerodynamic heating rate is applied as illustrated in
to be computed Fig.2(b). Temperature distribution on the panel is
2p 02p ] then computed by the structural heat transfer analy-
4 APPLICATIONS msulated
40
/r~ -A undeformed
30
c~
20
10
0 I I J
4.2 M a c h 4 F l o w in a C h a n n e l
/ / To further evaluate the analysis procedure for
the high-speed flow-structure interaction, the problem
of a Mach 4 flow in a channel is studied. The prob-
lem statement and the sketch of the flow behavior are
shown in Fig.7. The flow behavior is more complex
than that shown in the preceding example. Due to
symmetry of the problem, only the upper half of the
entire domain is used for the analysis. The flow en-
ters from the left boundary and creates oblique shock
from the wedge that impinges at the middle of the
upper wall resulting in a reflecting shock wave. The
(b) Density distribution reflecting shock wave also intersects with the Mach
waves generated from the expansion ramp. It should
Fig.4 Adaptive mesh and corresponding
be noted that the flow behavior of the intersected zone
density contours (kg/m a) for Mach
is not known a priori. The combined procedure of the
10 flow over an inclined plate with
convex deformation cell-centered finite element method and the adaptive
meshing technique is used to predict the flow beha-
6 ~ -0.3m~O.25m~i~[ 0.15m ~ -
- - undeformed ~ I i ~ 0.002m
4 ~ f o ~ m e _ d deformed ( t > 0 ) ~ 4
mdeformed (t ~0)~'f_panel / j / 0.2m
M=4 /~,, panelholder
~2
0
0.0
_1
0.1
x/m
I
0.2
J
0.3
0,m 0.09m
vior in the same manner as described in the preceding static structural analysis is then performed to pre-
example. Small elements are generated automatically dict the structural deformation shape. The computa-
in the regions of high change of the solution gradient tional fluid domain and its boundary conditions are
to provide the solution accuracy as shown in Fig.8(a). then updated. The entire process is repeated every
The total 20 272 triangular elements are generated in 20s. Figure 9(a) shows the adaptive mesh that con-
the inviscid region and 8 770 quadrilateral elements sists of 13 634 triangular elements and 7 050 quadrilat-
in the boundary layer along the upper wall. Again, eral elements for the flow field after 60 s. The corre-
ten graded layers of quadrilateral elements are used sponding predicted density distribution is shown in
in the boundary layer to capture steep t e m p e r a t u r e Fig.9(b). The predicted heating rate and pressure
gradients for accurate aerodynamic heating rate pre- distributions on the deformed structure at this time
diction. The corresponding predicted density distri- are compared with those for the initial undeformed
bution represented by the contour fines is presented shape as shown in Figs.10 and 11, respectively. The
in Fig.8(b). For structural heat transfer and struc- m a x i m u m aerodynamic heating rate and pressure oc-
tural response analysis, the boundary conditions of curring at the impinged shock location of the upper
the panel are used in the same manner as presented wall are plotted. These figures highlight the effect of
in the preceding application. With the predicted aero- the deformed structure that can significantly alter the
dynamic heating rate from the flow analysis, the tem- flow field. T h e y also highlight the need of the flow-
perature distribution of the panel is computed by us- thermal-structural analysis for predicting high-speed
ing the structural heat transfer analysis. The quasi- flow-structure interaction phenomena.
3~ 2
Fig.8 Adaptive mesh and corresponding density contours (kg/m a) for a Mach
4 flow in a channel at initial time
Vol.20, No.6 Limtrakarn W & Dechaumphai P: FEM for High-speed Flow-structure Interaction 605
3~
50 1.0
- - undeformed
deformed 0.8
40 -- undelbrmed
0.6 deformed
3O
0.4
20 0.2
LI
lution gradients to provide the high accuracy and 7 Dechaumphai P, Limtrakarn W. Adaptive cell-
large elements are generated in the other regions to centered finite element technique for compressible
minimize the computational time and memory. The flows. Journal of Energy, Heat and Mass Transfer,
Galerkin finite element method was used to predict 1999, 21:57~65
the structural heat transfer and structural response 8 Limtrakarn W, Dechaumphai P. Computations of
high-speed compressible flows with adaptive cell-
behaviors. The finite element formulation, the com-
centered finite element method. J Chin Inst Eng,
putational procedure and the basic idea behind the
2003, 26:553~563
adaptive meshing technique were described. Two ap-
9 Gnoffo PA. Application of program LAURA to three-
plications of high-speed flow-structure interaction, the dimensional AOTV flowfields. AIAA Paper 86-0565
Mach 10 flow over an inclined plate and the Mach 4 10 Hirsch C. Numerical Computation of Internal and Ex-
flow in a channel, were studied to assess the capability ternal Flows. Vol. 1, New York: Wiley, 1988
and efficiency of the proposed procedure. These ap- 11 Beer FP, Johnston ER, DeWolf JT. Mechanics of Ma-
plications demonstrate that the proposed procedure terials. Third edition, New York: McGraw-Hill, 2002
can provide analysis solution accuracy and computa- 12 Zienkiewicz OC, Taylor RL. The Finite Element
tional efficiency for predicting complex flow-thermal- Method. Fifth ed. Woburn: Butterworth-Heinemann,
structural behavior of the fluid-structure interaction 2000
problems. 13 Ramakrishnan R, Bey KS, Thornton EA. Adaptive
quadrilateral and triangular finite-element scheme for
REFERENCES compressible flows. AIAA Journal, 1990, 28(1):
51~59
1 Glass DE, Merski NR, Glass CE. Airframe research 14 Dechaumphai P. Improved finite element methodology
and technology for hypersonic airbreathing vehicles. for integrated thermal-structure analysis. NASA CR
AIAA Paper 2002-5137 3635, 1982
2 Anderson JD Jr. Modern Compressible Flow with His- 15 Lohner R, Morgan K, Zienkiewicz OC. Adaptive grid
torical Prospective. New York: McGraw-Hill, 1982 refinement for compressible Euler and Navier-Stokes
3 Anderson JD Jr. Fundamentals of Aerodynamics. equations. The International Conference on Accu-
Second ed. New York: McGraw-Hill, 1991 racy Estimates and Adaptive Refinements in Finite
4 Baum JD. Development of a Coupled CFD/CSD Element Computations, New York: Wiley, 1984, 2:
Methodology Using an Embedded CSD Approach. 189~202,
Conference on Computational Physics 2002 16 Dechaumphai P. Adaptive finite element technique for
5 Aftosmis MJ, Berger M J, Adomavicius G. A paral- heat transfer problems. Journal of Energy, Heat &
lel multilevel method for adaptively refined cartesian Mass Transfer, 1995, 17(2): 87~94
grids with embedded boundaries. AIAA Paper 2000- 17 Peraire J, Vahjdati M, Morgan K, et al. Adaptive
0808 remeshing for compressible flow computation. J Corn-
6 Lohner R, Baum JD, Mestreau E, et al. Adaptive put Phys, 1987, 72:449~466
embedded unstructured grid methods. AIAA Paper 18 Oden JT, Carey GF. Finite Elements: Mathematical
03-1116 Aspects. New Jersey: Prentice-Hall, 1981