12.07.2015 Views

application of frequency-domain system identification techniques in ...

application of frequency-domain system identification techniques in ...

application of frequency-domain system identification techniques in ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

26 Chapter 2. Operational Modal Analysis1995; Maia and Silva, 1997). The multiple coherence mγ 2 o <strong>in</strong> response po<strong>in</strong>t o atcircular <strong>frequency</strong> ω is def<strong>in</strong>ed byN imγo 2 (ω) = ∑ ∑N iH [o,s] (ω) · ŜF F [s,t](ω) · (H[o,t] (ω) ) ∗Ŝ XX[o,o] (ω)s=1 t=1(2.44)This quantity is a measure <strong>of</strong> the correlation between the response signal and all<strong>of</strong> the <strong>in</strong>put forces. A unity value expresses a perfect l<strong>in</strong>ear relationship betweenthe considered signals. A coherence value less than unity, between 0 and 1, can bedue to one or a comb<strong>in</strong>ation <strong>of</strong> the follow<strong>in</strong>g causes• presence <strong>of</strong> uncorrelated noise on the response and/or force measurements• presence <strong>of</strong> leakage <strong>in</strong> the measurements• non-l<strong>in</strong>ear behavior <strong>of</strong> the test structureIn practice, averag<strong>in</strong>g <strong>techniques</strong> are used <strong>in</strong> order to reduce the effect <strong>of</strong> measurementnoise. It should be noted that the values <strong>in</strong> the coherence function aredependent on the number <strong>of</strong> averages that were used to obta<strong>in</strong> the data. S<strong>in</strong>cethe coherence function is a measure for the accuracy <strong>of</strong> the estimate, the functioncan be used for the computation <strong>of</strong> the uncerta<strong>in</strong>ty on the FRFs (see e.g.,(Verboven, 2002)).Parameter confidence <strong>in</strong>tervalsGiven the uncerta<strong>in</strong>ty on the FRF measurements, a good approximation <strong>of</strong> thecovariance matrix <strong>of</strong> the ML estimate ˆθ ML is obta<strong>in</strong>ed by <strong>in</strong>vert<strong>in</strong>g the Fisher<strong>in</strong>formation matrix (P<strong>in</strong>telon and Schoukens, 2001)}cov{ˆθML ≃ [ 2Re ( Jp H J −1p)](2.45)with J p the Jacobian matrix evaluated <strong>in</strong> the last iteration step <strong>of</strong> the Gauss-Newton optimization. As one is <strong>of</strong>ten ma<strong>in</strong>ly <strong>in</strong>terested <strong>in</strong> the uncerta<strong>in</strong>ty on themodal frequencies and damp<strong>in</strong>g ratios, only the covariance matrix <strong>of</strong> the denom<strong>in</strong>atorcoefficients is required. Start<strong>in</strong>g from (2.45), it can be shown (Verboven, 2002)that this matrix is given bycov{ˆα ML } ≃[2N∑oN ik=1(Tk − Sk T R −1kS ) ] −1k(2.46)Hence, it is not necessary to <strong>in</strong>vert the full matrix occurr<strong>in</strong>g <strong>in</strong> (2.45). From (2.46)it is possible to compute the uncerta<strong>in</strong>ty on the modal <strong>frequency</strong> and damp<strong>in</strong>g ratios(Guillaume et al., 1989).

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!