Changes in the Statistics of Ambient Excitations in the Performance of Two Damage Detection Schemes

Data driven damage detection is often carried out by computing a metric that can be extracted from measurements and whose statistics depend on the structural state. A well known difficulty arises because changes in environmental conditions also lead to changes in the metric and these fluctuations need to be minimized for the methods to have adequate resolution. Two techniques that have been investigated for some time in damage detection are: (1) the Kalman Filter innovations whiteness test and (2) a subspace approach based on the left kernel of the output covariance. Recent work by the authors showed that both techniques lost significant resolution when the statistics of the ambient excitation changed from the time the filters were formulated to the interrogation time. Work to formulate modified versions that can operate more robustly under a changing loading environment was undertaken and this paper reports on the modifications developed as well as on the effectiveness of the new procedures compared to the original strategies.