Multi-sensor Data Fusion Diagnosis Method Based on Interpretable Spatial-temporal Graph Convolutional Network

Abstract

With the development of big data and artificial intelligence technologies, deep learning and its derived algorithms have achieved fruitful results and have been widely used in the field of fault diagnosis. However, intelligent fault diagnosis methods also face many challenges：1) existing multi-sensor data fusion methods have difficulty fully exploring the spatial-temporal information between multiple sensors to optimize diagnostic performance; 2) the decision-making process of intelligent diagnostic models has weak interpretability and cannot meet the reliability requirements in actual industrial scenarios. Therefore, we propose a multi-sensor data fusion diagnostic method based on an interpretable spatial-temporal graph convolutional network(ISTGCN). Firstly, a gate convolution layer is constructed to learn and enhance time features. Secondly, by combining the spatial layout relationship of sensors and the multi-sensor information fusion ability of graph convolutional networks, the spatial features of multi-sensor data are learned and extracted, and the effectiveness of the model is verified through the planetary gearbox fault diagnosis task. Finally, the improved gradient-based activation mapping algorithm is used to analyze the importance of each sensor data to the model diagnostic decision-making process, thereby improving the interpretability of the model update process. The experimental results show that the proposed method not only has good diagnostic performance but also provides an effective post-explanation method for multi-source sensor fusion diagnosis.