A Hybrid Physics and Machine Learning Based Approach for Guided Wave Based Detection of Delaminations in FRP Composites

This paper proposes a hybrid physics and machine learning based approach for detection of delamination in composite laminates through a simulation-based study. A glass-epoxy cross ply laminated plate has been considered as an example structure. Delaminations of various lengths have been considered as damages. Piezoelectric patches have been used as actuators and sensors. A plain strain finite element model has been developed in Abaqus incorporating the effect of delamination and electro-elastic coupling in the piezoelectric patches. Contact nonlinearity in the surfaces of delamination has been incorporated in the model. The piezoelectric actuators are excited with tone bust signal for generation of guided waves for several damage cases. Symmetric and antisymmetric components of the wave propagation response captured by the sensors have been separated. The separated components have been analyzed through fast Fourier transform, and wavelet transform and damage features have been extracted. Presence of super harmonics in both symmetric and antisymmetric components generated due to contact nonlinearity at the damaged surfaces has been observed through the analysis. Identification of delaminated layers has been solved as a classification problem using probabilistic neural network (PNN) classifier. These trained PNN have been tested for several unknown damage cases and a satisfactory performance has been observed. Localization of the delaminations has been done by analyzing the time of arrival of the symmetric wave component.