Research on CNN Based Ultrasonic Guided Wave Multi-bolt Connection Looseness Detection

Ultrasonic guided wave is one of the most potential structural health monitoring technologies. In recent years, artificial intelligence technologies such as deep learning have flourished, and it is possible to establish more effective structural damage detection techniques using deep learning combined with guided wave damage detection principles. However, for complex multi bolt connection structures, when bolts become loose, the guided wave signal changes more complex, often requiring multiple sets of sensors to detect simultaneously. The traditional machine learning methods have limited feature extraction capabilities, and the determination of the bolt loose position of the bolt group is limited. Based on this research background, this paper focuses on the construction and verification of guided wave damage detection methods for structures based on deep learning. Taking bolt looseness detection as the research object, a detailed study has been conducted. Two multi-sensor information processing methods based on convolutional neural networks have been proposed: multi-channel input convolution method and high-level feature fusion model method. Taking a 14-bolt-connection component as the research object, using Hankel matrix to convert one-dimensional data to two-dimensional matrix, the effectiveness of the two methods was verified, and their performance was compared.