Damage Identification of Chemical Milling Stiffened Panel Based on Lamb Wave and Inception-Convolutional Neural Network

Mode conversion and wave scattering will occur when ultrasonic guided wave (GW) propagates to the stiffener which makes the received signals complex and poorly interpretable, thus limiting the application of GW in damage detection of chemical milling stiffened panel (CMSP). This article proposes a deep learning (DL) model inception-convolutional neural network (Inception-CNN) to realize damage localization of CMSP. First, a model analysis was conducted to get the resonant frequency of the piezoelectric wafer, and the mode conversion at the stiffener was explored through numerical analysis. Then, the identification effect based on the conventional damage imaging method was discussed. Finally, for training the proposed DL model, the residual signals were collected as a dataset after setting damages in different zones of CMSP. The model was then trained and tested, and its performance was analyzed and demonstrated. The results indicate that S0 mode has a greater conversion degree than A0 mode at the stiffener; GWs do not propagate in the form of a uniform wavefront on CMSP and conventional damage imaging methods based on the wave propagation paths are not applicable to CMSP. The model proposed can automatically extract the signal spatial features, accurately identify the corresponding damage zone, and its accuracy reaches 94% in the testing set. As the input features increase, the classification ability of the model will be further improved. The noise interference experiment shows that the model has good noise resistance performance at noise levels below 15%, indicating the feasibility of the model for practical applications.