Professional Documents
Culture Documents
Frequency Domain
System Identification
Gang Jin 6.1 Introduction ....................................................................................... 1069
Ford Motor Company, 6.2 Frequency Domain Curve-Fitting ........................................................... 1069
Dearborn, Michigan, USA 6.2.1 M a t r i x Fraction Parameterization • 6.2.2 Polynomial Matrix Parameterization •
6.2.3 Least Squares O p t i m i z a t i o n A l g o r i t h m s
6.3 State-Space System Realization ............................................................... 1074
6.3.1 M a r k o v Parameters Generation • 6.3.2 The ERA M e t h o d
6.4 Application Studies .............................................................................. 1075
6.4.1 Identification o f a 16-Story Structure • 6.4.2 Identification o f the Seismic-Active Mass
Driver B e n c h m a r k Structure
6.5 Conclusion ......................................................................................... 1078
References .......................................................................................... 1078
/ \
/ \
/ Experimental \ I--
I input-output ~- -- ..~.1 FRF MF/PM
,,measurement/ IL__ Calculation Curve-fitting
\ /
\
\
\
X
Model Markov param-
evaluation eters generation
FIGURE 6.1 GeneralProcedure of Frequency Domain System Identification. Bolded components imply critical steps; dashed components
imply steps may be excluded.
m output measurement channels. U s e {~((Oi)}i_l,...1 to denote Substituting the Q(z -1) and R(z 1) polynomials in equations
the observed FRF data based on which the TFM G(z 1) will be 6.2 and 6.3 and vectorizing the summation, equation 6.4 is
estimated. To evaluate G(z 1) at discrete frequencies coi, use changed into the form:
the map z(coi)= ej%~#w', with ws being the sampling fre-
quency. The curve-fitting error is measured by Frobenius G*(z-1) = arg min
G=Q-IR
llano ÷, IIF, (6.5)
norm ]]-]]F for matrices and by Z-norm ]]. ][2 for vectors.
Use I, to denote an identity matrix of dimensions n × n. where:
-Z I(OJ1)GT(o)I) .. z P(COl)GT((DI) --Ir --z-l(oJ1)Ir ... --Z P(COI)/r]
6.2.1 Matrix Fraction Parameterization
The matrix fraction (MF) parameterization of a TFM takes
~=
:
-Z-I(o~i)~T("~t)
i i i
z-P(toI){~T(toI)--It --Z-l(tOl)Ir ... --Z P((ol)IrJ
i /
the following form:
(6.6)
G(z-1) = Q l(z-1)R(z 1), (6.1) TT = [G(tOl)..-G(o~l)]. (6.7)
Q(z -1) =Im + QIz 1 + Q z z - 2 + . . . + Qqz-q, (6.2) Thus, the MF curve-fitting has been reduced to a standard
LLS problem, which can be solved by various efficient algo-
R ( z 1) = Ro + R1 z-1 + R2z - 2 + - ' ' + Rpz -p. (6.3) rithms (e.g., the QR factorization approach quoted in
algorithm of equations 6.29 through 6.31).
The constant matrices Q1 . . . . . Qq and R0, R1. . . . . Rp are re-
ferred to as observer Markov parameters in ]uang (1994). 6.2.2 Polynomial Matrix Parameterization
From the same reference, the reader may find more detailed
material for the MF curve-fitting discussed here and the ERA The polynomial matrix (PM) parameterization of a TFM has
realization algorithm discussed in a later section. To simplify the form:
the notation, without loss of generality, assume p = q.
To fit the TFM G(z 1) as in equation 6.1 into the observed C(z_l) _ B(z-')
~x(z-') ' (6.9)
FRF data {G(coi)}i=1,..., l, one may solve the following parameter
optimization problem: where:
1 B(Z - 1 ) = Bo + B1 z-1 ÷ BzZ -2 ÷ "'" -}- BpZ -p. (6.10)
G*(z 1) = arg min Z IlQ(z-l(~°'))G(~°i) - R(z l(co,))ll~. (6.4)
G=Q- R i=1 ot(z -1) = 1 + alz -1 ÷ a2 z - 2 + . . . + aqz -q. (6.11)
6 Frequency Domain System Identification 1071
These equations are the numerator polynomial matrix and the Finally, the right-hand side of equation 6.18 is vectorized for
minimal polynomial of the TFM, respectively. Assume that standard optimizations:
the orders of the numerator and denominator are equal
(i.e., p = q). The goal of parameter optimization is to find F(0) = IIW(a)(y - Ho)ll~, (6.19)
G*(z -~) with a prespecified order p such that the estimation
error is minimized: where 0 = [b~ , b~, .... bmr,
-W aT] T, y is a vector containing
the measured FRF data, and W(a) is a weighting function
l
G*(z 1) = argmin ~'w2(o3i)llG(toi) _ G(z 1( o ))1l 2 (6.12) with variate a.
G z.....a
i=1
Readers should have no difficulties to derive the detailed
expressions of W(a) and H. For the special case when there is
Here, an additional term w( • ) is included to allow desirable no fixed zeros (i.e., t~Yb = 1 in equation 6.14), the results are
frequency weighting on the estimation error. In the following, given in Bayard (1992). It is important to point out that W(a)
G(z -1) in equation 6.12 will be parametized in a way that and H have the following structure:
allows the inclusion of the fixed zeros. This is done first for
the single input/single output (SISO) case; then, it is general- - W(a) 0 • •" 0
ized to the MIMO case.
Let the numerator polynomial of the SISO system be the 0 W(a)
following: W(a) = (6.20)
1 n
F(O)-- ~ Z ) ~ ( 0 ) 2 = llf(O)ll , (6.24)
i=1
g(O)=l(O)Tf(o) (6.25)
G(O)=J(O)rJ(O)+Q(O), (6.26)
(l(Ok))Tj(Ok)+Q(Ok))Pk=--J(Ok)Tf(Ok). (6.27)
A = QR, (6.29)
FIGURE 6.2 Picture of the 16-Story Structure
6 Frequency Domain System Identification 1073
16-Story MF 16 16 32 32 10
Benchmark PM 8 8 12 30 8 In these equations, ]~2 = (I - Q1QT)A2.
In practice, ](Ok) is divided into ](Ok) = [h(0k), h(0k)],
with ll(Ok) corresponding to the block-diagonal terms and
out that the ]acobian matrix I(0 k) of f has an identical structure 12(0k) the last block matrix column in ](Ok). Solving the LLS
as H in equation 6.21. Thus, instead of solving equation 6.28 problems with 12(0 k) and 11(0 k) thus corresponds to updating
directly, the following decomposition of LLS problems may be the denominator a and the numerators bjk estimations. More-
applied. over, due to its block-diagonal structure, the LLS problem with
11(Ok) should be further decomposed (i.e., the LLS solutions of
Algorithm 2 (Decomposition of LLS Problem) bjk are independently solved for each j and k). Thus, the
Let A E R m×n have full column rank. Decompose A into computational cost of the optimization algorithm is signifi-
A = [At, A2], with Al C R mxp and A2 C R mxq. Let the unique cantly reduced.
QR factorization of A1 be A1 = Q1R~. Then the unique solu- To complete the discussion of the algorithm, note that the
tion x* of the LLS problem of equation 6.30 has the form initial values for the Gauss-Newton iteration may be generated
x* = [x~T, x~ T] r, with xq and x~ being the unique solutions by the classical Sanathanan-Koerner (SK) iteration composed
of the following LLS problems: of a sequence of reweighted LLS problems:
f16 f12 f8 f4
-40 :. ! • ..... 40[~ -- 1 -40 • : : I[ z • I
8o
-60 • : i -60
-60 : • :. . . . : -60
- 8 o ~ -80
-100 -100
-,OOl y: V f ,1 1 -,oo _120i : , \f j-120
-120 L . . . . . . " .... " ....... ~ -120
20 40 60 20 40 60 20 40 60 20 40 60
-100
-80
-100 ook,
-80
-12orry ....
.... : :
Y
-80 i : ....
-120
- 1 4 0 F ; F ....... -120I " ¢" "', --" i l~ 1
-120
20 40 60 20 40 60 20 40 60 20 40 60
,oo
• . i . . . . . ....
-80 i . i. . -80 i . i. . . .
x -100 -100
FIGURE 6.3 Comparison of Experimental and Model FRF for the 16-Story Structure: Magnitude Plot. ~ denotes the input force on the jth
floor; xj denotes the output displacement of the jth floor; dotted lines are for measurement data; solid lines are for model output.
1074 Gang fin
0k+a = arg rnoin][W(ak)(y-- H0)[[~, (6.34) tem realization algorithm (ERA), is selected to construct a
model in the state space form. First presented are the formulas
to generate the Markov parameters from the TFM, which are
with initial condition 00 = 0.
the starting point for the ERA method.
System realization is a technique to determine an internal To calculate the Markov parameters Y0, Y1, Y2. . . . from the
system TFM, first note that:
state-space description for a system given with an external
description, typically its TFM or impulse response. The name
reflects the fact that if a state-space description is available, an G(Z-1) = ~ yi Z i. (6.35)
electronic circuit can be built in a straightforward manner to i=o
realize the system response. There is a great amount of litera-
ture on this subject both from a system theoretical point of For the case when G(z-1) is parameterized in the MF form
view (Antsaklis and Michel, 1997) and from a practical system (i.e., G(z 1) = Q - l ( z 1)R(z-1)) ' the system Markov param-
identification point of view (]uang, 1994). In the following, a eters can be determined from:
well-developed method in the second category, the eigensys-
f16 f12 f8 f4
0 0
20 40 60 20 40 60 20 40 60 20 40 60
n
0
lOO~ -.! ........ i ........ i ....
-ioo -50 ...... o~ i- i.... i. - -200
-1 O0 ........ ......
-2°°vqF 73 iN -400
-+OlLjL
. . . .
oI +: : -200 i.
0i ... : ~........~ i ....... 0
-50 . . . . . . . i ...i..... -200
~
20 40 60 20 40 60 20 40 60 20 40 60
-200
O'
-2001~,
0 ~ ...... ; ..... ~ ...... ~ .....
-1O0
-400 i : : :, -1 O0
=g -400 -200
-600'
-600 -300 -200
-800
20 40 60 20 40 60 20 40 60 20 40 60
Frequency(Hz) Frequency (Hz) Frequency (Hz) Frequency (Hz)
FIGURE 6.4 Comparison of Experimental and Model FRF for the 16-Story Structure: Phase Plot. Same notations are used as in Figure 6.3.
6 Frequency Domain System Identification 1075
Yk Yk+ l •. Yk+f3 1 ]
!
Yk+l Yk+2 • • Yk+f3 ] .
H(k- 1) = (6.39)
Um XgddOt
40
.~ 20 20
-o 0
"(3 0
-20
-20
-40,
5 10 15 20 25 30 35 5 10 15 20 25 30 35
40 .... ' " i ' i .... i .... '. ' 40
20
0
-20
-40 iiiiiiiiiiiii IIYIIIIIIIIIIIIIIIIII -40
5 10 15 20 25 30 35 5 10 15 20 25 30 35
20
"5
"0 0
"(3 0
x~ - 2 0 -20
-40
-40
5 10 15 20 25 30 35 5 10 15 20 25 30 35
20
~5
"0
0
"(3 -20
E
X
-40
5 10 15 20 25 30 35 5 10 15 20 25 30 35
10
E -50
X
0
-100
5 10 15 20 25 30 35 5 10 15 20 25 30 35
F r e q u e n c y (Hz) F r e q u e n c y (Hz)
FIGURE 6.6 Comparison of Experimental and Model FRF for the Seismic-AMD Benchmark Structure: Magnitude Plot. Um denotes the input
command to the AMD; xg ddot denotes the input ground acceleration to the structure; xjddot denotes the output acceleration of the jth floor;
xmddot denotes the output acceleration of the AMD; Xm denotes the output displacement of the AMD; dotted lines are for measurement data;
solid lines are for model output.
Earthquake Engineering Laboratory (SDC/EEL) at University fication is to capture accurately the first five pairs of the
of Notre Dame. This section only presents the results perti- complex poles of the structure. For this purpose, a DSPT
nent to identification studies discussed so far in this chapter. Siglab spectrum analyzer is used to measure the FRF data.
For detailed information about these experiments, includ- The sampling rate is set at 256 Hz, and the frequency reso-
ing experimental setups and/or control developments, the lution is set at 0.125 Hz. The experimental FRF is precondi-
reader may refer to Jin et al. (2000), Jin (2002), and Dyke tioned to eliminate the second order direct current (dc) zeros
et aL (1994), respectively. from acceleration measurement. The MF parameterization
is chosen for the curve-fitting, which is complemented by
the ERA method for state-space realization. The key identifi-
6.4.1 Identification of a 16-Story Structure
cation parameters are given in Table 6.1. The final discrete-
The first identification target is a 16-story steel structure model time state-space realization has 10 states. The magnitude
shown in Figure 6.2. The system is excited by impulse and phase plots of its transfer functions are compared to
force produced by a PCB hammer and applied individually the experimental FRF data in Figures 6.3 and 6.4. Excellent
at the 16th, 12th, 8th, and 4th floors. The accelerations of agreements are found in all but the very high frequency
these floors are selected as the system measurement outputs range. The mismatch there is primarily due to the unmodeled
and are sensed by PCB accelerometers. The goal of the identi- high-frequency dynamics.
6 Frequency D o m a i n System Identification 1077
Um xgddot
0
-1 O0 -200
"0o
% -200
x -400 ! i i .... i ! l
-300
5 10 15 20 25 30 35 5 10 15 20 25 30 35
t
-100
-200
-300
5 10 15 20 25 30 35 5 10 15 20 25 30 35
0
-200
-400
-100
-600
5 10 15 20 25 30 35 5 10 15 20 25 30 35
× -lOO
_4oot ..... ...... t
-200
5 10 15 20 25 30 35 5 10 15 20 25 30 35
FIGURE 6.7 Comparison of Experimental and Model FRF for the Seismic-AMD Benchmark Structure: Phase Plot. Same notations are used
as in Figure 6.6.
6.4.2 Identification of the Seismic-Active mental data accurately in all but the low-frequency range of
Mass Driver Benchmark Structure channels corresponding to the AMD command input and the
acceleration outputs. A detailed analytical modeling of the
The second identification target of this discussion is a three-
system reveals that there are four (respectively two) fixed dc
story steel structure model, with an active mass driver (AMD)
zeros from AMD command input to the structure (respectively
installed on the third floor to reduce the vibration of the
AMD) acceleration outputs:
structure due to simulated earthquakes. A picture of the
system is given in Figure 6.5. This system has been used
recently as the ASCE first-generation seismic-AMD bench- lira Gx~um(S) -- ki, i = 1, 2, 3. (6.42)
s-e0 S4
mark study. The benchmark structure has two input excita-
tions: the voltage command sent to the AMD by the control
lim G~um ( S) - kin. (6.43)
computer and the ground acceleration generated by a seismic s-*O S2
shaker. The system responses are measured by four accelerom-
eters for the three floors and the AMD and one linear variable The G~,~ and G~,,u,~ are the transfer functions from
differential transformer (LVDT) for the displacement of AMD command input u m to structure and AMD acceleration
the AMD. The sampling rate is 256 Hz, and the frequency outputs, respectively. The ki and km are the static gains of
resolution is 0.0625Hz. Due to noise and nonlinearity, these transfer functions with the fixed dc zeroes
only the frequency range of 3 to 35Hz of the FRF data is removed. These fixed zeros dictate the use of the PM curve-
considered to be accurate and, thus, this range is used for the fitting technique to explicitly include such a priori informa-
identification. tion.
A preliminary curve-fitting is carried out using the MF Again, key identification parameters are presented in Table
parameterization. The identified model matches the experi- 6.1. The outputs of the parameter optimization iterations are
1078 Gang Jin