Site-Specific Defect Detection in Composite Using Solitary Waves Based on Deep Learning

Abstract

AbstractWe propose a real-time non-destructive evaluation technique for defect detection in composites using highly nonlinear solitary waves (HNSWs) and a deep learning algorithm based on the convolution neural network (CNN). This technique implements deep learning to identify the presence of defects and classify the defect locations in the thickness direction of composites through HNSWs with strong energy intensity and non-distortive nature. To collect HNSW datasets for training and validation of the deep learning algorithm, AS4/PEEK composite specimens with artificial delamination are fabricated and HNSW datasets are generated from the experimental setup of a granular crystal sensor. Testing pretrained CNN based algorithms verifies the performance of detecting and classifying defects by location in composite plates.KeywordsNon-destructive evaluationDelaminationGranular crystalConvolution neural networkMachine learningAS4/PEEK