INTEGRATING DEEP LEARNING WITH GUIDED WAVE-BASED SIMULATION TO PREDICT DELAMINATION LOCATION IN COMPOSITE STRUCTURES

Abstract

In the present study, a numerical framework integrating deep learning with guidedwave- based finite element (FE) simulation is developed to accurately predict the delamination location in a composite structure. In the FE model, composite structure is modeled as an 8 layer [02/902]2 laminate. Elliptical delamination with randomly selected orientations and in-plane locations, is placed at the interface between top 0° and 90° layers. The guided wave is excited by a piezoelectric transducer, and the out-of-plane displacement signals are collected at eight prescribed sensor locations. The initial sensor signal is processed using continuous wavelet transform (CWT), and the convolutional neural network (CNN) is utilized to predict the delamination location. With the baseline model trained, the CNN model is extended to predict the location for circular shaped delamination. The use of transfer learning to minimize data needed for CNN model in predicting the location of delamination of a different shape is also investigated. The results show that the CNN-based framework can accurately identify delamination location given the sensor signal. Furthermore, the combination of modeling randomly distributed and oriented elliptical delamination, along with transfer learning, proves to be an effective strategy for reducing the additional data required for a new delamination shape.