Dual Attention-Enhanced Lightweight Shuffle-StarNet Model for Bearing Fault Diagnosis

The key component of the China Spallation Neutron Source-the neutron chopper-relies heavily on the health condition of its core rolling bearings, which directly affects the reliability of the device’s operation and the safety of experiments. Therefore, timely diagnosis of bearing faults is crucial. Traditional fault detection methods perform poorly under complex working conditions and strong noise environments, while classical deep learning models, despite their high accuracy, are characterized by large parameter sizes and high computational complexity, making them unsuitable for real-time requirements. To address this issue, this paper proposes a lightweight Shuffle-StarNet neural network model with a composite attention mechanism, which integrates the advantages of the improved ShuffleNet and StarNet. By combining depthwise separable convolution, star operations, and a composite attention mechanism, the model can accurately capture bearing fault features. Experimental results demonstrate that the model performs excellently in the task of bearing vibration signal fault diagnosis, achieving a test set accuracy of 99.52% with only 140.87K parameters and a computational complexity of 365.34 MFLOPs. The single inference speed is as low as 4.65 ms. Even under a 5 dB noise environment, the accuracy remains at 98.37%, showcasing its exceptional robustness. The model proposed in this study achieves a significant reduction in computational resource requirements while maintaining high precision, thereby offering an efficient and practical technical solution for the health management of equipment such as neutron choppers.