Semi-supervised small sample fault diagnosis under a wide range of speed variation conditions based on uncertainty analysis

The wide variety of rotational speeds and the small sample dataset make it challenging to mine the fault-related features in the signal, thereby limiting the generalization capability of the model. To address this problem, a semi-supervised method for fault diagnosis through feature perturbation and decision fusion is proposed in this paper. Firstly, a dual correlation model is constructed, and then the model's structural parameters are adjusted to increase the diversity and uncertainty of the diagnostic results. Finally, the conditional probability and penalty term of the sub-models are analyzed based on high-confidence samples, enabling the achievement of the final fusion diagnosis. To verify the effectiveness of the proposed method, variable speed experiments containing 20 kinds of speeds were conducted based on the Bearing Prognostics Simulator test bench. The validity of the proposed method is demonstrated through the constructed dataset.

Automated summary

This paper proposes a semi-supervised fault diagnosis method through feature perturbation and decision fusion. To improve the generalization capability of the model, a dual correlation model is constructed, and the structural parameters are adjusted. The final fusion diagnosis is achieved by analyzing high-confidence samples.