Professional Documents
Culture Documents
with MR = TT MT, CR = TT CT, KR = TT KT being the reduced system matrices and bR = TT B the reduced load vector. The reduction effectiveness and reliability depends on the size of . Based on the choice of
T various techniques have been developed through the last
decades.
II. Reduction Techniques
A. Guyan Reduction
The oldest reduction method, which was introduced by
Guyan [4], is based on the notion of master/external and
slave/internal DOFs [13]. Suppose we have the following
undamped system:
M
x(t) + Kx(t) = f
(3)
I. Introduction
Model reduction is a key issue for the analysis of mechanical systems. For the constant demand of working with
increasingly large models while aiming to control and possibly reduce storage and simulation time needs, the application of the right technique constitutes an important decision.
A common spatial discretization method used for mechanical MBS is the Finite Element Method (FEM). The
Partial Differential Equation (PDE), which describes the
behavior of the elastic body is transformed into a second
order Ordinary Differential Equation (ODE) of the form
M
x(t) + Cx(t)
+ Kx(t) = Bu(t)
(1)
where M, C, K Rnn are the system matrices (mass, damping- and stiffness matrix respectively), Bu(t)
Rn1 the load vector and x Rn1 the unknown vector with n Degrees of Freedom (DOF). In many cases
n (104 , 6 105 ), which leads to large dimension system
matrices and thus to vast storage and simulation time needs.
The general concept of model reduction is to find a lowdimensional subspace T Rnn in order to approximate
the state vector x = TxR + . By projecting (1) on this
subspace a lower dimension 2nd-order ODE is obtained
R (t) + CR x R (t) + KR xR (t) = bR
MR x
E-mail:
E-mail:
panagiotis.koutsovasilis@tu-dresden.de
michael.beitelschmidt@tu-dresden.de
(2)
(4)
By partitioning the system matrices of (3) into block matrices that depend explicitly on the m-set {mm} or s-set
{ss} or a combination of them {ms, sm} the following reordered system is obtained
m
x
xm
fm
+K
=
(5)
M
s
x
xs
fs
Mmm Mms
Kmm Kms
=
=
M
,K
Msm Mss
Ksm Kss
We solve the second equation of (5) for xs and assume
there is no force applied on the external DOFs, i.e fs =
0. By omitting the equivalent inertia terms (static), the
transformation matrix for the static reduction is obtained
according to (2), where the low-dimension subspace is in
this case represented by Tstatic :
xm
I
(6)
=
xm = Tstatic xm
xs
K1
ss Ksm
Generally, Guyan reduction is a good approximation for the
lower eigenfrequencies respectively eigenvectors. For high
frequency motion the effect of inertia terms is significant,
thus the method becomes inaccurate.
B. Dynamic Reduction
This technique [3] is an expansion of the static reduction. Applying a Laplace transformation on (3) we get the
equivalent system
(M 2 + K)X() = B()X() = F()
(7)
which is then re-ordered into the block partitioned master/slave DOFs as defined previously, giving in that way the
transformation matrix for the dynamic reduction:
I
(8)
Tdynamic =
B()1
ss B()sm
(11)
k yk = xm + y
(12)
k=1
Mss K1
ss Ksm
xm
xs
S = Msm
0
= TIRS xm , P =
0
0
K1
ss
(9)
(10)
l nm
Thus, the transformation matrix for the CMS method is obtained:
I 0
xm
xm
x=
= Tcms
(13)
y
y
CMS as IRS delivers good approximation results of the reduced structure with a drawback of having to define what
kind of lower eigenvectors [5] (rigid or non-rigid body) and
how many are to be introduced for the Craig-Bampton set.
E. System Equivalent Expansion Reduction Process (SEREP)
In SEREP [3] the eigenmodes and eigenfrequencies of
the original full model are calculated. Thus x = q, where
is the modal matrix and q the modal coordinate. By
partitioning the displacement x and the modal matrix into
the active (master) and omissive (slave) part we have
m
xm
(14)
=
+
m xm = TSEREP xm
xs
s
where +
:= (Tm m )1 Tm , the m pseudom
inversion.
SEREP approximates high-frequency motion (eigenfrequencies and eigenvectors) perfectly, up to the predefined
limit.
F. Krylov Subspace Method
Rnn , a start vector
Suppose the constant matrix A
n1
+
R
and q Z . The Krylov subspace is defined as
b
follows:
b)
:= span b,
A
b,
...,A
q1 b
,
Kq (A,
A
...,
b,
a subspace spanned by the q column-vectors b,
q1
A b. In the case of undamped systems of the form (3),
K1 M and b
K1 f , i.e
it is proved that A
M
x(t) + Kx(t) = f Tkrylov Kq (K1 M, K1 f )
Tkrylov =
(15)
span K1 f , (K1 M)K1 f , . . . , (K1 M)q1 K1 f .
(16)
Ax = b, A := Kss , x := T, b := Ksm
where the last linearized equation system can be iteratively
solved for as many steps as defined by m, (10, 5 102 )
dim(m) << dim(n). Then T is known, i.e Tstatic is
known. This procedure is also applied for the other reduction methods.
By choosing a right preconditioner depending on the
structure of the matrix (incomplete LU-, Cholesky factorization, block Jacobi, incomplete band-diagonal in case of
band-diagonal matrices) faster convergence is achieved.
There is always the case though, in which the matrix A
is ill-conditioned (large condition number) and the selection
of a preconditioner does not accelerate the convergence resulting in a large number of iteration steps, i.e increase of
the simulation time as shown below
C(A) = max /min
(17)
It is proven that the first q moments for the full and reduced
{(3), (16)} system agree. For the first moment mR0 (reduced model) and m0 (full model) the proof is given according to [11]:
T
T
1 T
Tkrylov f
mR0 = CTR K1
R fR = CR (Tkrylov KTkrylov )
= CT Tkrylov r0 = CT K1 f = m0
(A + Ad )x = b
:= 10(n+k) ,
xm
xs
=
I
T
xm = Tstatic xm
T = K1
ss Ksm
Kss T = Ksm
(18)
(19)
x 10
2
1
Comparison
Static Red.
Dyn.Reduction
CMS
IRS
SEREP
Krylov
0.6
0.4
0
1
0
dif
0.8
Static Red.
Dyn.Reduction
CMS
IRS
SEREP
Krylov
0.2
10
12
14
0
0
0.2
5
IRS
SEREP
Krylov
0.15
10
12
14
x 10
SEREP
Krylov
4
3
0.1
2
0.05
0
0
1
2
10
12
14
0
0
10
12
14
modM ACk,l =
selected according to standard criteria [13] made for this
purpose. The number, though, of master nodes is restricted
(m = 10); this is done in order to prove that a possible inappropriate master node selection vastly affects the results
of certain reduction methods.
The re-ordering into block matrices according to (5) produces a different sparsity pattern (Fig.2) for the system
matrices, because of the wavefront solver implemented in
ANSYS for the solution of the linearized system.
The sparsity pattern of the stiffness matrix K is the
same, but with notedly more non-zero elements.
(Tk Ml )2
T
(k Mk )(Tl Ml )
1
0.8
0.6
0.4
Static Red.
Dyn.Reduction
CMS
IRS
SEREP
Krylov
0.2
0
0
10
12
14
10
12
14
1
0.8
0.6
IRS
SEREP
Krylov
0.4
0.2
0
0
A. Eigenfrequency Comparison
Theory implies the eigenfrequencies of the reduced
model to be higher as the eigenfrequencies of the
original (full) model. Thus, in Fig.3 the difference
eigdif := eigsub eigf ull and normalized relative difference reigdif := |eigsub eigf ull |2 / |eigsub |2 for the first
14 of the non-rigid body eigenfrequencies are presented.
B. Modified Modal Assurance Criterion (modMAC)
modMAC gives information concerning the eigenvectors angle; by this criterion (in comparison to MAC) the
ther the reduced are expanded to the dimension of the original via the transformation matrix or the opposite. A value
modM AC = 1 means absolute correlation; the less this
value becomes, the worst the eigenvector correlation is, as
shown in Fig. 4. The correlation of the first six eigenvectors is unimportant, since it concerns the rigid body motion
eigenvectors, which are generally of no interest. Thus the
x-axes of Fig. 4,Fig. 5, Fig. 6 depict the 14 eigenvectors
starting from the 7th up to the 20th.
C. Mass Normalized Vector Difference (MNVD)
This criterion gives the relative vector difference of massnormalized eigenvectors.
f ull
Msub M
2
M N V Dk,l =
Msub
2
makes it an important tool for modal correlation. The calculation is based on Modal Scale Factor (MSF), which is a
scale factor according to the principle of least-square error.
14
x 10
Static Red.
Dyn.Reduction
CMS
IRS
SEREP
Krylov
12
10
8
6
M SFi,j =
Ti j
Ti i
4
2
0
|k (r) M SF k (r)|2
k (r)
M DMk,r =
10
12
14
which gives minor deviations due to numeric implementation, in comparison to the other.
D. Stiffness Normalized Vector Difference (SNVD)
Analogously to the MNVD criterion, SNVD gives information about the relative vector difference of stiffnessnormalized eigenvectors.
f ull
Ksub K
2
SN V Dk,l =
Ksub
2
1.5
Eigenvector 7 Static
x 10
15
10
1.5
1
0.5
0.5
0.5
1
0
5
0
6
x 10
Eigenvector 7 SEREP
Eigenvector 7 Krylov
200
100
1.5
1
8
4
x 10
Eigenvector 7 IRS
x 10
1
0
8
4
x 10
0.5
In this case results are sensitive, concerning small deviations of the modal stiffness for both the original and reduced
model as shown in Fig. 6.
Eigenvector 7 CMS
x 10
2.5
0.5
2.5
Eigenvector 7 Dynamic
x 10
100
0
2
0.5
1
0
200
4
0
300
0
x 10
8
4
x 10
x 10
1
0.8
0.6
0.4
Static Red.
Dyn.Reduction
CMS
IRS
SEREP
Krylov
Eigenvector 8 Dynamic
x 10
15
0.5
10
2
3
10
12
14
0.5
4
5
0
0.35
1
0
0.3
0.25
60
0.15
50
50
40
40
30
30
20
20
10
10
12
14
1
0
8
4
x 10
10
0
10
0
2
10
0
x 10
Eigenvector 8 Krylov
70
60
0.05
70
0.1
x 10
x 10
Eigenvector 8 SEREP
Eigenvector 8 IRS
x 10
5
0
x 10
IRS
SEREP
Krylov
0.2
0
0
Eigenvector 8 CMS
x 10
0.2
0
0
Eigenvector 8 Static
x 10
8
4
x 10
x 10
Eigenvector 18 Static
0.5
0
4
x 10 Eigenvector 18 Dynamic
x 10
as well as other combined methods is a matter of actual research, since the obtained results constitute these
techniques competitive to the already standardized CMS
method.
Eigenvector 18 CMS
0.5
1
1.5
2
0
10
15
20
0
x 10
4
0
500
500
400
400
300
300
200
200
100
100
2
0
100
0
x 10
100
0
References
[1]
[2]
0
2
8
4
x 10
600
Eigenvector 18 Krylov
Eigenvector 18 SEREP
x 10
Eigenvector 18 IRS
600
x 10
5
[3]
2
x 10
8
4
x 10
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]