Professional Documents
Culture Documents
Method(KMM)
Konark Arora1 and S. M. Deshpande2
1
2
Summary. Meshfree methods are gaining popularity over the conventional CFD
methods for computation of inviscid and viscous compressible flows past complex
configurations. The main reason for the growth of popularity of these methods is
their ability to work on any point distribution. These methods donot require the
grid for flow simulation, which is an essential requirement for all other conventional
CFD methods. However these methods are limited by the requirement of a good
connectivity around a node. Here, a very robust form of the meshfree method called
Weighted Least Squares Kinetic Upwind Method using Eigendirections (WLSKUMED) has been used to avoid the problem of code divergence due to the bad connectivity. In WLSKUM-ED, the weights are calculated to diagonalize the least squares
matrix A (w) such that the x and y directions become the eigen directions along
which the higher dimensional least squares formulae reduce to the corresponding
one dimensional formulae. Here an effort has been made to explain the enhanced
robustness of the WLSKUM-ED meshfree method over the conventional LSKUM
meshfree method. The accuracy of the kinetic meshfree method for the Euler equations has been enhanced by use of entropy variables and inner iterations in the
defect correction method. It is observed that the use of entropy variables and inner
iterations in the defect correction method helps in obtaining the formasl order of
accuracy in case of a non-uniform point distribution.
Key words: LSKUM, WLSKUM-ED, Eigendirections, SVD, LED, Rank deficiency, Entropy variables.
1 Introduction
Meshfree methods are gaining popularity over the conventional CFD methods
for computation of inviscid and viscous compressible flows past complex configurations. The main reason for the growth of popularity of these methods is
their ability to work on any point distribution. These methods do not require
the grid for flow simulation, which is an essential requirement for all other
conventional CFD methods. But they do require point distribution or a cloud
(1)
Define
xi = xi xo ,
Fi = Fi Fo
(2)
p
X
e2i =
p
X
(F i xi Fx o )
(3)
i=1
i=1
Fx (1)
o =
p
P
xi F i
i=1
p
P
(4)
xi
i=1
The least squares matrix A for 2-D and 3-D cases respectively is given by
P4
P3
Po
P1
P2
P5
P
P
2 P
i P xi zi
Pxi y
P xi
A = xi yi
y z
yi 2
P i2 i
P
P
zi a
yi zi
xi zi
P
2 P
x
x
y
i
i
i
P
A= P
yi 2
xi yi
Fx (1)
o =
p
P
wi xi F i
i=1
p
P
(5)
wi xi
i=1
The weights are suitably chosen so that the (x,Y) direction along which the
upwinding is done becomes one of the eigendirections of the LS matrix A
along which the multidimensional LS formulae reduce to the corresponding
one dimensional formulae [2, 3, 4].
P
P
P
wi y i 2 wi xi F i
wi xi F i
=
Fx (1)
=
(6)
P
P
P
o
wi xi 2 wi y i 2
wi xi 2
because by choice
wi xi yi = 0
Further, the weights chosen must be positive so that the Local Extremum
Diminishing (LED) property is satisfied. A very simple and novel way to calculate the positive weights, utilizing the coordinate differentials of the neighbouring nodes in the connectivity in 2-D and 3-D, has been developed for the
purpose [2]. By doing the SVD (Singular Value Decomposition) analysis of
the least squares matrix of a node with good and a bad connectivity[5, 6], it
is observed that the rank deficiency of the LS matrix of the node with bad
connectivity is overcome to a good extent, thus enabling the method to work
effectively on a bad point distribution thereby enhancing the robustness of
the meshfree method.
(8)
(9)
(10)
(11)
Substituting Eq.(11) in Eq.(7), the sum of the squares of the deviation becomes
2
p
X
xi
yi
F i xi Fxo y i Fy o
E=
Fxi
Fy i
(12)
2
2
i=1
Define the modified difference as :
xi
yi
Fx i
Fy i
2
2
In terms of this modified difference, Eq.(12) becomes
Fei = Fi
E=
p
X
i=1
Fe i xi Fx o yi Fy o
2
(13)
Fy o (2) =
Fxo
(2)
P
P
P
y i 2 xi Fe i xi yi y i Fe i
P
P
P
2
xi 2 y i 2 ( xi yi )
P
P
P
xi 2 y i Fe i xi y i xi Fe i
P
P
P
2
xi 2 y i 2 ( xi y i )
(14)
(15)
It is observed that the value of the modified difference Fei in Eqs. 14 and 15
depends upon the value of the second order accurate derivatives at the node
and at its neighbours, thus requiring some inner iterations to get formally
second order accurate derivatives at the node under consideration. However,
this approach is simple as compared to the direct method described above and
the formulae used to get the higher order accurate value of the derivatives has
the same form as first order formula.
level. The meshfree upwind scheme is first developed for the Boltzmann equation in the Eulerian limit at the microscopic level. Applying the moment
method strategy to the discretized Boltzmann equation, we get the corresponding meshfree scheme for the Euler equations [11] at the macroscopic
level. The robustness of the meshfree method for the Euler equations can
be achieved obtained as described above by using the weighted least squares
method. The positive weights in the weighted least squares method are suitably chosen that the x-y direction in which upwinding is done becomes one of
the eigendirections of the LS matrix A along which the multidimensional LS
formulae reduce to the corresponding one dimensional formulae. The defect
correction method too can be easily applied to obtain higher order accurate
meshfree method for the Euler equations. However, it is to be noted that by
using the defect correction method (even with inner iterations), we will not be
able to get uniform higher order accuracy in the entire domain. The reason for
this is the fact that the defect correction method requires the computation of
the values of the corresponding derivatives at the neighbouring nodes in the
domain. While this can be easily done for the interior nodes in the domain,
it is not possible for the boundary nodes as the connectivity stencil for calculation of the corresponding derivatives may be outside the computational
domain and hence will be totally empty. This problem in case of the Euler
equations has been solved by making use of the q-variables (also called the
Entropy variables). [9, 7, 10]
(2)
The second order accurate derivatives qxi and qyi in the Eq.16 are obtained by using the full stencil in the least squares formulae. The perturbed
q-variables (ie. qei ) are then used to construct the modified Maxwellians which
are then used to obtain formally second order accurate derivatives in the split
equations. The q-LSKUM thus inherits all the good properties of the first
order LSKUM such as robusness, positivity and convergence characteristics.
and keeping the error in the fluxes bounded, thus enhancing its robustness by
preventing divergence.
Point Distribution and Residue Drop comparison for good and bad
connectivity using conventional and new meshfree method with eigendirections
BAD CONNECTIVITY
0.11
1
0.105
0.5
0.095
0.11
0.105
0.1
0.1
0.09
0.095
0.085
0.09
0.08
0.085
0.075
-0.5
0.07
0.08
0.065
0.075
-1
0.06
-0.5
0.5
1.5
0.055
0.295 0.3 0.305 0.31 0.315 0.32 0.325 0.33
0.07
0.3190.3195 0.32 0.32050.3210.32150.3220.3225
0.1
RESIDUE
RESIDUE
0.01
0.1
0.001
1e-04
1e-05
1e-06
0.01
1e-07
0
10
20
30
40
50
60
70
80
2000
ITERATIONS
4000
6000
8000
10000
12000
14000
ITERATIONS
Figure 4. Residue drop for Conventional Figure 5. Residue drop for Meshfree
Meshfree method
method with eigendirections
The impact of the use of inner iterations coupled with defect correction
method on the accuracy of the results of the meshfree method has been demonstarted by the test case of Linear convective equation (LCE) [17]. The enhancement of the accuracy of the solution of the Eulers equations of gas dynamics
by using entropy variables in place of the primitive variables too has been
demonstrated by the standard test cases of subsonic, transonic and supersonic flow past NACA 0012 aerofoil using an unstructured point distribution.
It is observed that the use of defect correction method coupled with inner
iterations and entropy variables results in an accurate prediction of drag coefficient (close to the theoretical value) and capture of crisper discontinuities.
The first test case consists of a rectangular domain [1, 1]X[0, 1] on which
we consider the linear convective equation :
u
u
u
+y
x
=0
t
x
y
x < 0.65
u(x, 0) = 1,
0.65 x 0.35
u(x, 0) = 0,
u(1, y) = 0,
u(x, 1) = 1,
0<x<1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
-1
-0.5
0.5
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-1
-0.5
0.5
Figure 6. Solution Contours for the LCE Figure 7. Solution Contours for the LCE
with defect correction method without in- with defect correction method with 2 inner iterations
ner iterations
The second test case is of subsonic flow past NACA 0012 aerofoil. The
flow conditions for this test case are a subsonic Mach number of 0.63 at
10
-Cp
0.5
-0.5
-1
-1.5
0
0.2
0.4
0.6
0.8
X/C
pressure 1.35023
1.29949
1.31812
1.26908
1.286
1.23866
1.25389
1.20824
1.22177
1.17782
1.18966
1.1474
1.15755
1.11698
1.12543
1.08657
1.09332
1.05615
1.0612
1.02573
1.02909
0.995311
0.996973
0.964892
0.964858
0.934474
0.932744
0.904056
0.900629
0.873638
0.868515
0.84322
0.8364
0.812801
0.804286
0.782383
0.772171
0.751965
0.740057
11
The third test case is of weak transonic flow past NACA 0012 aerofoil.
The flow conditions for this test case are a Mach number of 0.80 at angle
of attack() = 1.25o . The number of nodes in the computational domain
of this test case is 10058. Fig. 11 shows the comparison of the Cp plot
over the NACA 0012 aerofoil for this test case, when the defect correction
method has been applied using q-variables with and without use of inner
iterations. The pressure contours for this test case are shown in Figures 12
and 13. It is observed that a crisp strong shock on the upper surface and a
weak shock on the bottom surface of the aerofoil are captured when defect
correction is applied using q-variables coupled with inner iterations.
The fourth test case is of supersonic flow past NACA 0012 aerofoil. The
flow conditions for this test case are a Mach number of 1.20 at angle of
attack() = 0.00o. The number of nodes in the computational domain
of this test case is 10058. Fig. 14 shows the comparison of the Cp plot
over the NACA 0012 aerofoil for this test case, when the defect correction
method has been applied using q-variables coupled with and without inner iterations. The pressure contours for the corresponding test case are
shown in Figures 15 and 16. Table 1 shows that the lift coefficient value is
accurately predicted when inner iterations coupled with defect correction
method using q-variables are used to obtain higher order accuracy.
CL
CD
WEAK
CL
TRANSONIC CD
SUPERSONIC CL
CD
0.301666
0.003745
0.355055
0.023945
-0.001278
0.095882
0.281740
0.000897
0.371242
0.022119
0.000676
0.094695
0.329-0.336
0.003-0.0007
0.3632
0.0187-0.02698
0.0000
0.0946-0.0960
7 Conclusion
The enhancement of the accuracy of the robust meshfree method using
eigendirections has been demonstrated by coupling the inner iterations with
the defect correction method using q-variables. It is observed that the inner
iterations coupled with defect correction method using entropy variables not
12
-Cp
0.5
-0.5
-1
-1.5
0
0.2
0.4
0.6
0.8
X/C
Figure 11. Comparison of Cp using q-LSKUM with and without inner iterations
Pressure Contours for Weak Transonic flow test case over NACA 0012
Aerofoil using qLSKUM
pressure 1.55831
pressure 1.54502
1.50426
1.49166
1.4502
1.43829
1.39614
1.38493
1.34208
1.33157
1.28802
1.2782
1.23396
1.22484
1.1799
1.17147
1.12584
1.11811
1.07178
1.06475
1.01773
1.01138
0.963666
0.958018
0.909607
0.904654
0.855549
0.85129
0.80149
0.797926
0.747431
0.744562
0.693372
0.691198
0.639313
0.637834
0.585254
0.58447
0.531195
0.531106
only results in accurate prediction of lift and drag coefficients but also enables
the capture of crisper discontinuities.
References
1. K. Anandhanarayanan, Development and Applications of a Gridfree Kinetic Upwind Solver to Multibody Configurations ,PhD. Thesis, Department of
Aerospace Engineering, Indian Institute of Science, Bangalore, India.
13
Cp distribution for Supersonic flow test case over NACA 0012 Aerofoil
1.5
Aerofoil
qLSKUM without inner iterations
qLSKUM with inner iterations
1
-Cp
0.5
-0.5
-1
-1.5
0
0.2
0.4
0.6
0.8
X/C
Figure 14. Comparison of Cp using q-LSKUM with and without inner iterations
Pressure Contours for Supersonic flow test case over NACA 0012
Aerofoil using qLSKUM
pressure 2.42379
pressure 2.41001
2.33341
2.32037
2.24302
2.23074
2.15264
2.1411
2.06226
2.05147
1.97187
1.96183
1.88149
1.87219
1.79111
1.78256
1.70072
1.69292
1.61034
1.60329
1.51995
1.51365
1.42957
1.42402
1.33919
1.33438
1.2488
1.24474
1.15842
1.15511
1.06804
1.06547
0.977652
0.975837
0.887268
0.886201
0.796884
0.796565
0.7065
0.706929
14