A new damage feature based on wavelet packet transform for damage detection under ambient vibration

Wavelet packet energies (WPEs) as the input to artificial neural networks for damage detection have been widely studied. However, these methods require that the loads applied on structures be known in that the WPEs extracted from structural responses depend on external loads. Therefore, these methods cannot be used to detect damage occurring in civil engineering structures when ambient vibration is used as the excitation of dynamic tests, since it is difficult to measure the ambient vibration. In this study, the WPEs are extracted from the correlation functions between responses under ambient vibration instead of from responses. The proposed WPEs herein can characterize the natural properties of structures and are independent of external loads. In addition, they are sensitive to structural damage but insensitive to measurement noises. Then, a probabilistic neural network (PNN) with the proposed WPEs as the feature vector is developed for damage detection. Finally, the proposed approach is demonstrated by the steel-frame benchmark structure presented by the IASC-ASCE Structural Health Monitoring Task Group