Structures
Property 4.2. Closed-Form Controllability and Observability Grammians. In modal coordinates the diagonal entries of the controllability and
4.6 Controllability and Observability of a Second-Order Modal Model a Second-Order Modal Model
4.6.1 Grammians
The grammians and the balanced models are defined exclusively in the state-space representation, and they do not exist in the second-order form. This is a certain disadvantage since the second-order structural equations are popular forms of structural modeling. We will show, however, that for flexible structures one can find a second-order model which is almost balanced, and for which Hankel singular values can be approximately determined without using a state-space representation.
First, we determine the grammians for the second-order modal model.
Property 4.6. Controllability and Observability Grammians of the Second-Order Modal Model. The controllability and observability grammians of the second-order modal model are given as
(wc) (wo)
(4.57)
1 1
1 1
0.25 diag( ),
0.25 diag( ),
c T
o mT
w B
w C
= :
= :
m m m
B C
where diag(B Bm mT) and diag(C CTm m) denote the diagonal part of B B andm mT respectively,
T ,
C Cm m B is given by (2.23),m Cm ª¬Cmq:1 Cmvº¼, and are defined as in (2.24) and(2.25). Therefore, the ith diagonal entries of and are
C ,mq Cmv wc wo
2 2
2 2
4 ,
4 ,
mi ci
i i mi oi
i i
w b
w c
] Z
] Z
(4.58)
and bmi is the ith row of B , and m cmi is the ith column of Cm.
Proof. In order to show this we introduce a state-space representation by defining the following state vector:
m .
m m
x q q ª: º « »
¬ ¼
The following state-space representation is associated with the above vector 0 1
0 , ,
2 m mq mv
A B C C
B
: ª º
ª º
C .
ª º
« »
«: =:» ¬ : ¼
¬ ¼ ¬ ¼
By inspection, for this representation, the grammians are diagonally dominant, in the form
0 0
, ,
0 0
c o
c o
c o
w w
W W
w w
ª º ª
#« » #«
¬ ¼ ¬
º»
¼
where wc and wo are the diagonally dominant matrices, wc#diag(wci) and Introducing the last two equations to the Lyapunov equations (4.5)
we obtain (4.58).
diag( ).
wo # woi
Having the grammians for the second-order models, the Hankel singular values are determined approximately from (4.58) as
2 2, 1,..., . 4
mi mi
i ci oi
i i
b c
w w i n
J # ] Z (4.59)
4.6.2 Approximately Balanced Structure in Modal Coordinates Second-order modal models are not unique, since they are obtained using natural modes that are arbitrarily scaled. Hence we have a freedom to choose the scaling factor. By a proper choice of the scaling factors we introduce a model that is almost balanced, i.e., its controllability and observability grammians are approximately equal and diagonally dominant. The second-order almost-balanced model is obtained by scaling the modal displacement (qm) as follows:
(4.60)
1 ,
ab m
q R q that is,
m a ,
q Rq b (4.61)
and qab is the almost-balanced displacement.
The transformation R is obtained as follows. Denote bmi 2 and cmi 2 as the input and output gains, then
= diag ( ),i 1,..., ,
R r i n (4.62)
and the ith entry ri is defined as a square root of the gain ratio
2 2 mi .
i mi
r b
c (4.63)
Using (2.38) one obtains
2
2 2
2
2 qi ,
mi vi
i
c c
Z
2
c 2 (4.64)
while
2 2
2 T and 2 .
qi qi qi vi vi vi
c c c c c cT (4.65)
In the above equations bmi is theith row of Bm, and are the ith columns of , , and respectively. cqi,c cvi, mi
mq mv
C C Cm,
Introducing (4.60) and (4.62)–(4.65) to the modal equation (2.19) we obtain the almost-balanced second-order modal model
(4.66)
2 2 ,
,
ab ab ab ab
abq ab abv ab
q q q B
y C q C q
=: :
u
where
ab 1 m
B R B (4.67)
and
, R,
abq mq abv mv
C C R C C (4.68)
while the output matrix Cab is defined as
1 .
ab abq abv
C ª¬C : C º¼ (4.69)
This has the following property:
2 1 2
2 2 .
ab abq abv
C C : C 22 (4.70)
A flexible structure in modal coordinates is described by its natural modes, Ii, Similarly the almost-balanced modal representation is a modal representation with a unique scaling, and is described by the almost-balanced modes,
1, , . i !n
Iabi, i 1, , .! n The latter ones we obtain by rescaling the natural modes n
, 1, , ,
abi ri i i
I I ! (4.71)
with the scaling factor given by (4.63), andri
ab R,
) ) (4.72)
where )ab [Iab1Iab2!Iabn], and ) is a modal matrix, as in (2.12). In order to show this, note that from (2.18) one obtains q )qm or, equivalently
(4.73)
1
.
n i mi i
q
¦
IqBut, from (4.61) it follows that qmi r qi abi; thus, (4.73) is now
(4.74)
1 1
,
n n
i i abi abi abi
i i
q
¦
r qI¦
I q whereIabi is a balanced mode as in (4.71).Property 4.7. Grammians of the Almost-Balanced Model. In the almost- balanced model the controllability and observability grammians are approximately equal,
2 2
4 .
mi mi
cabi oabi i
i i
b c
w w J
# # ] Z # (4.75)
Proof. From (4.58) we have
2,
4 4
abi abi
cabi oabi
i i i i
b c
w w 2.
] Z ] Z (4.76)
However, from (4.49) and (4.52) it follows that
2
2 2
, 2
mi mi
abi mi abi mi
mi mi
c b
b b c c
b c . (4.77)
Introducing the above equation to (4.76) we obtain approximately equal grammians
as in (4.75).
Define babi 2 and cabi 2 as the input and output gains of the ith almost- balanced mode, respectively, and we find that these gains are equal.
Property 4.8. Gains of the Almost-Balanced Model. In the second-order almost-balanced model, the input and output gains are equal,
2 .
abi abi
b c 2 (4.78)
Proof. The transformationR as in (4.62) is introduced to (4.67) and (4.68) obtaining
2
2 2 2 2
2
1 mi
abi mi mi mi mi
i mi
b b c b b c
r b 2
and
2
2 2 2 2 2
2 mi
abi mi i mi mi mi abi
mi
c c r c b b c b
c 2
º
»»
»¼
.
Example 4.7. Determine the almost-balanced model of a simple structure from Example 2.2.
We obtain the transformation matrixRfrom (4.62) and (4.63) as R=diag(0.6836, 0.7671, 0.8989). Next, we find the almost-balanced input and output matrices from (4.67), (4.68), and (4.69), knowing from Example 2.2 that
hence,
: (3.1210, 2.1598, 0.7708);
diag
1 2 3
1
0.4798
0.7705 ,
0.8198
0.1294 0.2617 0.3825 0 0 0
0 0 0 , 0.4040 0.5653 0.2948 ,
0 0 0 0.2242 0.4534 0.6625
ab
ab ab
ab
abq abv
b
B b
b
C C
ª º
ª º
« »
« »
« » « »
« »
« »
¬ ¼ ¬ ¼
ª º ª
« » «
: « » «
« » «
¬ ¼ ¬
The output matrix Cab is obtained by putting together Cabq:1 and , such that the first column of is followed by the first column of , followed by the second column of
Cabv abq 1
C : Cabv
abq 1
C : , followed by the second column of Cabv, etc., i.e.,
>
1 2 3@
0.1294 0 0.2617 0 0.3825 0
0 0.4040 0 0.5653 0 0.2948 .
0 0.2242 0 0.4534 0 0.6625
ab ab ab ab
C c
ª º
« »
« »
« »
¬ ¼
c c
The almost-balanced mode matrix is obtained from (4.72), 0.4040 0.5653 0.2948 0.5038 0.2516 0.5313 . 0.2242 0.4534 0.6625
ab
ª º
« »
) « »
« »
¬ ¼
Finally, it is easy to check that the input and output gains are equal,
12 1 2
2 2 2 2
3 2 3 2
0.4798, 0.7705, 0.8198.
ab ab
ab ab
ab ab
b c
b c
b c
Also, from (4.76) we obtain wc1 wo1 1.1821, wc2 wo2 6.3628, and which shows that the model is almost balanced, since the exact Hankel singular values for this system are
3 3 55.8920,
c o
w w
1 1.1794,
J J2 6.3736, and
3 56.4212.
J