Gearbox Fault Diagnosis Method Based on PSM-BN

The gearbox fault diagnosis method based on deep learning lacks good interpretability, and the complexity of the model leads to prolonged diagnostic times. Based on this, this paper proposes a Compute Unified Device Architecture (CUDA) parallelized Bayesian Network (BN) approach for gearbox fault diagnosis. To address the problem of the Max-Min Hill-Climbing (MMHC) algorithm easily getting trapped in local optima during BN learning, this paper introduces the Snake Optimization (SO) algorithm. This algorithm utilizes multiple individuals to represent potential solutions and updates their positions at each iteration, employing a diversity search strategy to avoid falling into local optima. In response to the high complexity and poor real-time inference of BN fault diagnosis methods, this paper utilizes the CUDA platform as a development framework and employs a CPU+GPU heterogeneous parallel BN to improve the operational efficiency of the model, thereby enhancing the real-time capability of fault diagnosis. Finally, validation is conducted on the gearbox dataset from Southeast University, demonstrating that the proposed method achieves a diagnosis accuracy of 99.7% on 800 fault samples with a training time of 10.4 seconds. Compared to traditional methods, this approach exhibits significant advantages in diagnostic accuracy and training speed, effectively enhancing the accuracy and stability of fault diagnosis.