Research on Intelligent Diagnosis of Wear Faults of Centrifugal Pumps Based on Stacked Autoencoder

Abstract

Mechanical fault diagnosis is very important in industry because early detection can avoid some dangerous situations, and not much research has been done on the diagnosis of wear faults in centrifugal pumps. With the rapid development of data analysis techniques, data-driven diagnosis methods are becoming increasingly popular. In this study, stacked autoencoder based method is proposed to solve the centrifugal pump seal wear fault diagnosis problem. The method extracts power spectral density features directly from the vibration signal and chunks the features, greatly reducing the training difficulty and improving the accuracy of the model. The effectiveness of the method is verified using a centrifugal pump dataset, and the results show that the method can diagnose not only the fault site but also the degree of wear.