Deep Learning-Based Functional Assessment Of Piezoelectric-Based Smart Interface Under Various Degradations

The piezoelectric-based smart interface technique has shown promising prospects for electro-mechanical impedance (EMI)-based damage detection with various successful applications. During the process of EMI monitoring and damage identification, the operational functionality of the smart interface device is a major concern. In this study, common functional degradations that occurred in the smart interface are diagnosed using a deep learning-based method. Firstly, the effect of functional degradations on the EMI responses is analytically discussed. Secondly, a critical structural joint is selected as the test structure from which EM measurement using the smart interface is conducted. Thirdly, a numerical model corresponding to the experimental model is established and updated to reproduce the measured EMI responses. By using the updated numerical model, the EMI responses of the smart interface under the common functional degradations, such as the shear lag effect, the adhesive debonding, the sensor breakage, and the interface detaching, are simulated; then, the functional degradation-induced EMI changes are characterized. Finally, a convolutional neural network (CNN)-based functional assessment method is newly proposed for the smart interface. The CNN can automatically extract and directly learn optimal features from the raw EMI signals without preprocessing. The CNN is trained and tested using the datasets obtained from the updated numerical model. The obtained results show that the proposed method was successful to classify four types of common defects in the smart interface, even under the effect of noises.