With the continuous development of modern industry, numerous sensors have been ubiquitously deployed in industrial systems. Effectively integrating multi-sensor data is of great significance for improving the accuracy and reliability of fault diagnosis. However, most existing methods are limited to a single-level structure and fail to fully exploit the deep-level features within the data, exhibiting notable limitations. To address this issue, this study proposes a fault diagnosis method that integrates time-domain and frequency-domain graph information to more comprehensively extract and utilize fault features from sensor data. Specifically, graph structures in the time and frequency domains are first constructed by analyzing the correlations among different sensors. Then, graph convolutional neural networks are employed to extract features from both types of graphs, while the mutual information between them is minimized to enhance feature independence. Finally, an attention-based fusion strategy is introduced, and a classifier is constructed to improve the reliability of fault diagnosis. Experimental results on the nickel flash smelting furnace system demonstrate that the proposed method achieves a fault diagnosis accuracy exceeding 95.43%, significantly outperforming methods relying solely on time-domain data.
Jianhua Lin (2025) studied this question.