Quantitative Monitoring of Bolt Looseness Using Multichannel Piezoelectric Active Sensing and CBAM-Based Convolutional Neural Network

The bolted connection is widely utilized in engineering to practically and rigidly couple structural components. The integrity of the connection is paramount to the safety of the structure and has prompted the development of many monitoring methods, including the piezoelectric-based active sensing method. However, the active sensing method cannot quantify bolt looseness due to the unclear relationship between bolt looseness and the single monitoring index typically used in the active sensing method. Thus, the authors propose the unique combination of a one dimensional convolutional neural network (1DCNN) and multi-channel active sensing for quantitative monitoring of bolted connections. In an experiment, piezoelectric ceramic transducer (PZT) patches are bonded on steel plates connected by a bolt. Each patch is wired to a multi-channel active sensing monitoring system. After obtaining multi-channel stress wave signals at different looseness levels, a looseness vector is calculated to generate training and validation datasets. A baseline 1DCNN model and a novel model improved with Convolutional Block Attention Module (CBAM) are used to monitor the bolt looseness. Finally, the authors verify that the multi-channel active sensing method combined with 1DCNN model can accurately perform quantitative monitoring of bolt looseness and the monitoring accuracy of the baseline 1DCNN model is above 91.07% in three different specimens. Compared with the baseline 1DCNN model, the monitoring accuracy of the CBAMCNN model approximately improved by 5%. Overall, the method proposed in this paper offers a new and highly accurate approach for quantitative monitoring of bolted connections.