Wise-local response convolutional neural network based on Naïve Bayes theorem for rotating machinery fault classification

摘要

Fault identification is a vital task to ensure the integrity and reliability of rotating machinery. The vibration signals produced by the defective system components typically bear a significant amount of noise, including non-linear and non-stationary characteristics induced by the intricate operational environment. Advanced signal processing technologies still have difficulty in detecting faults in mechanical systems. This paper presents a wise local response convolution neural network-based Naïve Bayes algorithm (WCNN-NB) to identify multiple faults in rotating machines. In WCNN-NB, the WCNN structure is first used to characterize the features of the gray-scale images transformed from the original vibration signals. The nonlinearity parameter value of WCNN is explored in order to enhance learning efficiency. The NB algorithm is then used as a robust steady-state method to identify the learned features. Experimental data regarding helical gears and bearing test rigs are used to validate the feasibility of the WCNN-NB model. The superiority of the classification results is verified by comparing the current model with the WCNN-support vector machine, WCNN-random forest, standard CNN, the support vector machine (SVM) and the neural backpropagation network (BP) models. The results demonstrate that the classification accuracy is 99.68%, 92.5% and 97.5% for three data sets with tolerable misclassification rates under all operational conditions.