In recent years, with the increasing complexity and operational intensity of industrial equipment, the importance of mechanical equipment fault diagnosis has become increasingly important. Especially in the fields of aerospace, energy and manufacturing, how to quickly and accurately diagnose failures of key components with limited data has become a central issue in ensuring equipment reliability and safety.

Recently, the Intelligent Detection and Equipment Research Team at the Shenyang Institute of Automation (SIA), Chinese Academy of Sciences (CAS), proposed a lightweight domain fusion network for enhancing fault diagnosis under few-shot conditions.

Schematic diagram of the network structure (Image by the research group)

The network architecture handles the time domain, frequency domain and statistical features of vibration signals separately, and enables multi-sensor data fusion through an improved channel attention mechanism. In addition, the research team proposed two model-independent enhancement strategies for few-shot learning. One is, optimizing the sampling interval to reduce sample variance; the other is, maximizing inter-class separation by regularizing the loss function. These methods can effectively enhance the diagnostic accuracy of the model under few-shot conditions.

Optimization of feature distribution with fewer samples (Image by the research group)

Experimental results show that the new method outperforms existing techniques on two publicly available datasets. In particular, its diagnostic accuracy is significantly higher than that of other methods when only one to five samples are provided for each fault, while the computation speed is faster. The results not only provide an efficient solution for few-shot fault diagnosis in real industrial scenarios. It also lays the foundation for the deployment of real-time diagnosis systems in resource-constrained environments.

This research result was published in the international journal IEEE Transactions on Reliability as A Lightweight Triple-Stream Network With Multisensor Fusion for Enhanced Few-Shot Learning Fault Diagnosis. Hao-Tian Peng, a PhD student of Shenyang Institute of Automation, is the first author of the paper, and researchers Wei Wang and Jinsong Du are the corresponding authors. This research was supported by the National Natural Science Foundation of China, Liaoning Provincial Applied Basic Research Programme and Liaoning Provincial Natural Science Foundation.

Link to paper: https://doi.org/10.1109/TR.2025.3540500