Ms Mechanical

Engineering
2015-2017

[ASSIGHNMENT ON
MODEL REDUCTION
METHOD]
[Type the abstract of the document here. The abstract is typically a short
summary of the contents of the document. Type the abstract of the document
here. The abstract is typically a short summary of the contents of the document.]

8.8 Model Reduction Method

A difficulty with many design and analysis methods is that they work best for
systems with a small number of degrees of freedom. Unfortunately, many
interesting problems have a large number of degrees of freedom. In fact, to
obtain accurate results with finite element models, the number of elements
and hence the order of the vibration Is Increased. Thus finite element models
of practical structures and machines are often very large. One approach to
this dilemma is to reduce the size of the original model by essentially
removing those parts of the model that affect its dynamic response the least.
This process is called model reduction or reduced-order modeling. It is an
attempt to reduce the size of an FEM but still retain the dynamic character of
the system.
Quite often the mass matrix of a system may be singular or nearly singular,
due to some elements being much smaller than others. In fact, in the case of
finite element modeling the mass matrix may contain zeros along a portion
of the diagonal. Coordinates associated with zero or relatively small mass ore
likely candidates for being removed from the model. Another set of
coordinates that are likely choices for removal from the model are those that
do not respond when the structure is excited. Stated another way, some
coordinates may have more significant responses than others. The
distinction between significant and insignificant coordinates leads to a
convenient formulation of the model reduction problem due to Guyan
(1965). Consider the un-damped, forced vibration, finite element model and
mass and stiffness matrices according to significant displacements denoted
by n1 and insignificant displacements n2. This yield

][ ] [ ]

M 11 M 12 } ## {u} rsub {2} rsup { K 11 K 12 u 1

f
u1
+
= 1
[
]
M 21 M 22
K 21 K 22 u 2
f2

Eq. 8.97

Note here that the coordinates u(t) have been rearranged so that those
having the least significant displacement associated with them appear last in
the displacement vector,
u = [u1

u2].

Next consider the potential energy of the system defined by the scalar
V = uT Ku
or, in partitioned form,

[ ][

1 u1
V=
2 u2

][ ]

K 11 K 12 u1
K 21 K 22 u2

(8.98)

Similarly, the kinetic energy of the system can be written as the scalar
T = uT Mu, which becomes
T

[ ][

'
1 u1
T=
2 u'2

][ ]

'
M 11 M 12 u1
M 21 M 22 u'2

(8.99)

in partitioned form. Since each coordinate ui is acted on by a force fi , the

condition that there is no force in the direction of the insignificant
coordinates u2 requires that f2 = 0 and that
V/u2 = 0
This yields

T
T
T
T
(u1 K 11 u1 +u1 K 12 u2+ u2 K 21 u1 +u2 K 22 u2)=0
u 2

(8.100)

Hence the constraint relation between u1and u2 must be (since. K12 = K21T)
1

u2=K 22 K 21 u1

(8.101)

Expression (8.101) suggests a coordinate transformation (which is not a

similarity transformation) from the full Coordinate system to the reduced
coordinate system u1. If the transformation matrix Q is defined by
Q=

1
K

1
22

K 21

(8.102)

Then if u = Qu1is substituted into equation (8.97) and this expression is premultiplied by QT, a new reduced-order system of the form
} + {Q} ^ {T} KQ {u} rsub {1} = {Q} ^ {T} f
Q T MQu 1

(8.103)

results. The vector QT f now has the dimension of u1. Equation (8.103)
represents the reduced-order form of equation (8.97), where

1
T
1
1
QT MQ=M 11 K T21 K 1
22 M 21 M 12 K 22 K 21 + K 21 K 22 M 22 K 22 K 21

(8.104)

and
QT KQ=K 11 K 12 K1
22 K 21

(8.105)

These last expressions are frequently used to reduce the order of finite
element vibration models in a systematic and consistent manner. Such
model reduction schemes are used when a finite element model has
coordinates (represented by u2) that do not contribute substantially to the
response of the system. Model reduction can greatly simplify design and
analysis problem under certain circumstances. If some of the masses in the
system are negligible or zero, the preceding formulas can be used to reduce
the order of the vibration problem simply by setting M22 = 0 in equation
(8.104). This is essentially the model reduction technique referred to as Mass
Contribution.
Example 8.6.1
Consider a four-degree-of-freedom system finite element model with mass
matrix

312 54
0 13
1 54 156 13 22
M=
420 0
13
8 3
13 22 3 4

and stiffness matrix

24 6 0
6
6 12 6 6
K=
0 6 10 4
6 6 4
4

Note that this system is both dynamically and statically coupled. To remove
the effect of the last two coordinates, the sub-matrices of equation (8.97) are
easily identified as
M 11=

1 312 54
420 54 156

M 12=

1 0 13
= M 21
420 13 22

M 22=

1 8 3
420 3 4

K 11= 24 6
6 12

K 22=

10 4
4 4

6 =K
K 12= 0
21
6 6
Using equations (8.101) and (8.105) yields
T

Q MQ=

1.012 0.198
0.198 0.236

[ ]

QT KQ= 9 3
3 3

These matrices form the resulting reduced-order model of the structure.

:- Bookmarks
Model Reduction Method

][ ] [ ]

M 11 M 12 } ## {u} rsub {2} rsup { K 11 K 12 u 1

f
u1
+
= 1
[
]
M 21 M 22
K 21 K 22 u 2
f2

Eq. 8.97