Quantitative damage evaluation of curved plates based on phased array guided wave and deep learning algorithm

Abstract

Recent advances in phased array guided wave (PAGW) have demonstrated the potential of minor damage detection and localization in widely used curved plates, but quantitative damage evaluation remains difficult since effective features that are sensitive to damage size are hard to extract. In this study, a novel integrated framework, GW-SHMnet, is proposed, which leverages the advantages of the PAGW, finite element (FE) modeling, and deep learning algorithm. Firstly, an FE model is constructed to simulate PAGW propagation in curved plates. Secondly, PAGW experiments are performed on a curved aluminum plate to validate the FE model. Thirdly, an FE simulation database considering different sensor locations, testing frequencies, and damage sizes, is constructed and used as the training and testing data. Finally, deep learning is used to automatically extract features to determine damage size. The effectiveness, accuracy, and robustness of GW-SHMnet enable autonomous quantitative evaluation of minor damage in curved plates.