You are on page 1of 31

Recent advances of Operational Modal

Analysis and applications in Structural


Health Monitoring

Part one (OMA)


Rune Brincker
Motivations
The mechanical engineer The civil engineer

OMA OMA
ODS
ODS
MIMO
SISO MIMO

SISO

2
Classical Modal Analysis
FFT Small structures
Output Signal
[m/s] Time(Response) - Input
- easily tested
Working : Input : Input : FFT Analyzer

80

40 Artificial input provided


Output 0
no problem
-40

-80

0 40m 80m 120m 160m 200m 240m


[s]

[N] Time(Excitation) - Input

200
Output Spectrum
Working : Input : Input : FFT Analyzer

Input 100

0 [(m/s)/N]Frequency Response H1(Response,Excitation) - Input (Magnitude)


[(m/s)/N]Frequency
Working : Input : Input : FFT Analyzer
-100

100
-200
0 40m 80m 120m 160m 200m 240m
[s]

1
Input Signal

FFT 10m

0 400 800 1,2k 1,6k 2k 2,4k 2,8k 3,2k


[Hz]

3
Thinking of Civil Engineers
Larger structures More slender structures Bigger loads

longer service New materials

Codes are getting bigger and bigger

Buildings are falling down

Even civil enginners NEEDS to know reallity

Cheaper and more accurate equipment

December 15, 1967.


September 11, 1916.
Silver Bridge (USA)
Quebec Bridge (Canada)
March 17, 1945.
November 7, 2005.
Ludendorff Bridge (Remagen, Germany)
(Almunecar, Spain)
*Images from http://www.engineeringcivil.com/theory/civil-engineering-disasters

4
Idea of operatinal modal analysis

Combined System

Stationary
Zero Mean Measured
Excitation Filter Structural System
Gaussian (linear, (linear, time-
Responses
White Noise time-variant) invariant)

Unknown excitation
forces
Loading modes Structural modes
Time-variant Time-invariant
Broad banded Narrow banded

5
History on OMA
Bendat & many others: Basic Frequency Domain
G11 G12 K G1N
G G22
G= 21
M O
G GNN
N1 L

Andreas Felber PhD thesis

6
Time domain techniqus
Henry Cole was looking for...
A simple and direct method for translating the time histori into a form
meaningful to the observer (1971)

Sam Ibrahim and the Time Domain


Random Decrement Technique for Identification of Structures, J.
Spacecraft and Rockets, Vol. 14, No. 11, 1977

Vandiver, Brincker and Assmussen and the Random


Decrement (1982-1990)

7
Identification
MIMO

PRCE ERA
H. Vold et al et al,1982 Juang and Pappa, 1985

Singular values of the system Hankel Matrix


3.5

30
3

25 Developed on the basis of traditional 2.5

20
modal testing 2
System Order

15
1.5

Could be used for OMA


10

1
5

0
Use RDD or correlation functions 0.5

0
-5 20 40 60 80 100 120 140 160 180 200
0 2 4 6 8 10 12 index number of the singular values = 2*Model Order
Frequency(Hz)

But it did not (really) happen

Reasons: mode selection problem, no software...


Civil engineers were still sleeping...
8
Begining of aplications

First real test reported in


1993

Felber uses
the frequency domain
successfully

9
Applications
A.J.Felber & R.Cantieni
Introduction of a new Ambient Vibration Testing System, EMPA,
1993-1996.

C.E.Ventura, A,J, Felber, S.F.Stiemer


Determination of the dynamic characteristics of the colquiz Bridge
be full-scale testing, 1992.

Tom Carne 1986


Called it NExT= Correlation
functions+ Polyrefernce and
ERA

10
Applications
M.Hoeklie, O.E. Hansteen J.C.Asmussen, R.Brincker, A.Rytter
Measured and Predicted Dynamic Behavior Ambient modal testing of the vestvej bridge using random
of the Gulfaks A Platform, 1988. decrement, 1998.

Basic freq domain


SSI/ERA? Basic frequency domain
Updating Random decrement + Polyreference

11
Stochastic Subspace Identification: revival of
the time domain
Bart de Moor and Overshees book in 1996 Data driven SSI

Very much like Random Decrement and time domain solved by SVD???
BUT: It sounds like black magic; Kalman gain matrix??? Projection of the Hankel matrix???
If you just need modal parameters forget about the Kalman gain matrix !

12
Frequency domain decomposition: revival of the
frequency domain
N k 1 2ik
Ri , j (k ) = E {yi , y } N 1 k y (y )
N
Gi , j ( ) = Ri , j (k )e
* i j
N
j +k n n+k
n =0 k =0

G11 G12 L G1n 1 0 L 0


G G22 L G2 n 0 L 0
G = 21 =U 2 U
M M O M M M O

n
Modal coordinate
1 ( ) Gn1 Gn 2 L Gnn 0 auto spectral density

PSD Mag. Example SVD of PSD Matrix Decoupled Modes

13
Frequency domain decomposition: revival of the
frequency domain
Invented in 1998, published at IMAC in 2000
Easy to use and understand
Works well even with closely spaced modes
Enhanced FDD in 2001

14
Overview

What do we have (around year 2000)?

Reliable frequency domain techniques

Reliable time domain techniques

Using several data sets mode shapes

Many applications

Still problems: mode shape scaling... software, harmonics....

15
Scaling mode shapes (Mass change method)
Nomenclature :

Un-scaled mode shapes T = 1



Scaled mode shapes T M = 1 2 = 2T M

Scaling factor =

Method: Parloo et. al

M12 = K 12 22 2 =
2
Classical equation TM
Approximation
Solve for
(M + M) 22 = K 2
1
Make a mass change
Brincker et. al

M(12 22 ) = M 222 12 22
Solve for =
Subtraction 22T M

16
Approximation errors

12 22 2
Simulation study
Mathematical Model

1 0 . 2 1
0 1 1 2 .
M = m K = k
. . . . 1
1 2
1

20 DOF chainlike system

Mass change 0-20 % of DOF mass

17
Approximation errors
Errors should vanish for mass changes in increasing DOFs

2 12 22
Parloo equation = Improved equation =
TM 22T M

18
Error reduction

Error types How to do it!


Random Error on frequency shift Do good ID and make large shift
Linearization error Use improved equation
Random Error on mode shape Do good ID and use many DOFs
Mode shape change error Distribute mass change M = M

How can it be done (in an easy way) ?


Make a mass change and estimate the frequency shift
Easy: Shift masses around while using several data sets

What is the accuracy ?


In a lausy test the typical uncertainty will be
around 10-20 %
In a well prepared test the typical uncertainty will
be around 2-5 %

19
Harmonic removal

Without Harmonic
With Harmonic

Harmonic Removal
Algorithm

20
Harmonic removal
Structural Modes:

( x )2
y = f (x | , ) =
1 2 2
e
2 2
Harmonics:

y = f (x | a ) =
1
{ ( )}
cos sin 1 x a
sin 1 (.) = Arc sin (.)

(x | , ) =
[
E (x )
4
] 3
4
median ( meas

. ) 1.5

21
Harmonic removal
Natural Frequency Damping Ratio

Including Harmonic

Harmonic Removal
Harmonic Removal
Algorithm

Algorithm
Excluding Harmonic
22
Known Harmonic Removal
P.Mohanty & D.J.Riexen (2003-2005)
Modifications of Time domain Algorithms for Removing known Harmonic components

* Notation adopted from P.Mohanty & D.J.Riexen

23
Known Harmonic Removal

24
Automated OMA

Finding modes by FDD

Modal coherence
[si ]

Modal domain

Harmonics Removal 1
d
f
f

f1 f 2

FDD automated f0 f
f

25
Modal Coherence Requirements:
Discriminator function: d1 ( f 0 ) = u1 ( f )T u1 ( f 0 ) . MAC level
Low modal coherence: Noise . Averaging
High modal coherence: Modal dominance
Number
Random vectors: { }
E u1 ( f 0 )T u1 ( f ) = 0 { }
Var u1 ( f 0 )T u1 ( f ) = 1 / N

26
Modal Domain Requirements:
Discriminator function: d 2 ( f ) = u1 ( f )T u1 ( f 0 ) MAC level
The modal domain is found as the Region around the peak where

[si ]

f
f0 f
d

Mode property defined for all modes


d2 2
f1 f 2

f Defines the frequency region dominated by


f0 f
f the mode

27
Example: Plate with Harmonics

28
Example: Heritage Building

29
Example: Z24 Highway Bridge

30
Conclusions

Valuable tools:
Modal Coherence
Modal Domain
Harmonic discrimination
Dynamic headroom etc.

Important to note:
It works..
Easier OMA
Applications in health monitoring

31

You might also like