Fault diagnosis of mechanical seals using graph neural networks with multi-sensor data fusion

Abstract

Mechanical seals are critical components in the mechanical industry, and their operational status directly impacts the performance of pumps, compressors, and other machinery. Therefore, conducting research on the fault diagnosis of mechanical seals is essential. To enhance the accuracy of the assessment model, we propose an integrated approach that leverages the fusion of multiple graph neural networks (GNNs). Firstly, recognizing the diversity among different sensors, we utilize multi-channel data to comprehensively represent the operational state of the mechanical seal. These channels include various types of sensors such as acoustic emission and force sensors. Secondly, we employ multiple methods to transform the original multi-channel data into graph data, thereby continuously increasing the diversity of the datasets used for training. Finally, after training GNNs, we output the data of these networks through data fusion to obtain evaluation results. The effectiveness of our assessment approach is demonstrated using mechanical seal test data.