Hybrid deep convolutional networks for the autonomous damage diagnosis of laminated composite structures

Abstract

This article presents a robust autonomous damage diagnosis method using hybrid deep convolutional networks for the damage diagnosis of laminated composite structures. Inspired by the potential of deep learning models to autonomously extract deep discriminative features and machine learning models that provide better diagnosis on limited data, the current research integrates deep convolutional networks, namely convolutional neural networks (CNN) and convolutional autoencoder (CAE), with support vector machines (SVM) to build hybrid damage detection models. The proposed hybrid models incorporate the advantages of both convolutional operations to extract deep features, and SVM to diagnose using limited feature data. The proposed hybrid models are validated using random vibrational signals for one healthy and two delamination states of laminated composites. The results showed improved damage detection performance compared to the conventional methods, with lower computational costs. Additionally, the hybrid methods autonomously extracted deep discriminative features, eliminating the need for manual damage-sensitive feature extraction.