Structural damage localization based on wavelet packet analysis under varying environment effects

A novel method for structural damage localization under varying temperature conditions is proposed, utilizing wavelet packet analysis combined with an autoencoder neural network. Dynamic response data collected from sensors are decomposed through wavelet packet transform to extract damage-related features, specifically the wavelet packet energy ratio. To eliminate the influence of environmental factors such as temperature, an autoencoder is employed to filter temperature effects from the extracted features. Structural damage localization is achieved through the Mahalanobis distance, which identifies anomalies by comparing the residuals of the predicted autoencoder output with those under baseline conditions. The proposed method is validated through experimental testing on a steel beam and numerical simulations on a spatial latticed shell structure. Results demonstrate robustness in accurately detecting and localizing structural damage under varying temperature conditions, even with noise interference.