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

74 Chapter 3. Identification from multi-patch non-stationary operational dataTable 3.3: Overview of the stairway estimation results obtained for the complete dataset.ω (Hz) ξ (%) MAC7.0469 0.52 0.98759.0788 0.48 0.996813.4248 0.56 0.999017.6422 0.26 0.998218.6027 0.28 0.965519.3346 0.71 0.9864Table 3.4: Overview of the stairway estimation results obtained from the derived multipatchdata set.PoGER approachω (Hz) ξ (%) MAC— — —9.0750 0.52 0.966613.4282 0.54 0.999017.6466 0.27 0.996618.6104 0.24 0.996219.3349 0.46 0.5666PreGER approachω (Hz) ξ (%) MAC— — —9.0794 0.54 0.989113.4299 0.58 0.999017.6490 0.27 0.998018.6107 0.28 0.998919.3478 0.75 0.9914the mode shape estimates of modes 5 and 6 when no such re-scaling is taken intoaccount. A comparison with the corresponding high quality mode shapes identifiedin Chapter 2 clearly shows the presence of a distortion on the mode shape resultsfor the multi-patch data. In order to deal with this problem, both the PoGERand PreGER approach were tested.An overview of the ML estimation results, on the multi-patch data set, is shownin Table 3.4. For both tested methods, all modes were identified apart from thefirst one. This fact can be explained by the smaller amount of data present in themulti-patch data set and the related SNR of the spectral estimates (5 averagesused opposed to 65 with the original data). A comparison of the natural frequencyestimates of the 5 remaining modes shows a good agreement between boththe PoGER and PreGER method and the estimates from the original single patchdata set. Fig. 3.16–3.19 show the results of the 5 th and 6 th re-scaled mode shapeestimates for both approaches. A comparison of the results with those representedin Fig. 3.14–3.15 clearly indicates the presence of non-stationarities in the data set.