A deep learning-based method for hull stiffened plate crack detection

Abstract

Deep learning has attracted the attention of many researchers for structural health monitoring. However, it is difficult to use most of the deep learning-based techniques to detect damage throughout the life cycle of a large or inaccessible structure, especially a ship. Few studies have focused on hull stiffened plate crack damage detection. We propose such a method based on deep learning using a convolutional neural network (CNN). The model is trained on acceleration data, which are calculated by the Abaqus scripting interface. Five crack locations and four crack lengths are considered, as well as the intact condition. The effects of damping ratio, loading area, and load level on the proposed method are considered. The robustness of the proposed approach to noise and stiffener slenderness ratio are also discussed. The proposed method is compared to the multilayer perceptron method by wavelet packet transformation using the same data, so as to quantify its performance. The results show that the proposed method performs better at single- and double-crack detection, and is less sensitive to noise, damping ratio, loading area, and load level.