Health Monitoring of Bolted Structures Based on Multivariate Information Fusion and Attention-Convolutional Neural Network

The structural integrity of bolt connections significantly influences the safety and reliability of aircrafts. Monitoring the health of these connections is critical, yet single-sensor data often lacks sufficient information, resulting in reduced classification accuracy. This paper presents an innovative method for monitoring the health of bolt structures using multivariate information fusion and attention-convolutional neural networks. Initially, multiple sensors acquire vibration signals under diverse operational conditions of the structure. These one-dimensional signals are transformed into two-dimensional time-frequency maps via the Fourier synchronous compression transform method. Furthermore, the time-frequency maps from multiple sensors are integrated into the neural network. In addition, the attention mechanism is employed to highlight salient features, thereby facilitating effective health monitoring of bolt structures. Experimental results demonstrate that the proposed method excels in feature extraction, achieving high accuracy and stability in state recognition.