Detection Feature Influence on Optimized Excitations for Structural Health Monitoring

We have recently demonstrated that evolutionary algorithms can be used to modify the excitation applied to a structure such that damage detection resolution on the output will be improved. Recent results indicate that optimized inputs are highly sensitive to the damage detection feature employed in the optimization, implying that the excitation and feature type should be closely matched if improved detection sensitivity is desired. This study compares two variations of the prediction error feature used to discern damage-induced differences in structural response. In particular, we show that the inclusion of time-dependent information in the prediction error feature forces the evolutionary optimization procedure to settle on chaotic excitations for improved detection sensitivity. This contrasts with the multi-tone input preferred for a geometrically-based version of prediction error.