Fault Diagnosis Method for Marine Rotating Machinery Based on Particle Swarm Optimization-Driven Kernelized Cascade Forest

Vollständiger Abstract

Worum geht es in dieser Arbeit?

To address the problem of the fault mode discrimination accuracy of marine rotating machinery under the conditions of noise interference and limited training samples, a fault diagnosis method based on a particle swarm optimization-driven kernelized cascade forest is proposed. This method introduces RBF kernel mapping in the hierarchical structure of the cascade forest and uses PSO to conduct global optimization of the key parameters of the kernel function. Based on the marine fan test platform driven by a three-phase asynchronous motor, the frequency domain features are extracted as the model input, and the predicted probability distribution is utilized during the hierarchical training process. The experimental results show that in a noise-free environment and a noisy environment with Gaussian white noise, the test accuracy of the diagnostic accuracy rate is 98.05% and 97.10%, respectively, with a performance decrease of only 0.95%; it still maintains stable recognition accuracy under the condition of small samples, demonstrating superior small-sample learning ability compared to the benchmark method. Compared with the forest-based baseline method, the proposed method can reduce the performance degradation caused by noise and effectively improve the diagnostic performance of the model under noise and small-sample conditions, and this method may become a potential solution for fault intelligent diagnosis in marine rotating machinery.