An Online Wear State Identification Method for Axial Piston Pump Key Friction Pair based on FSANN

Abstract

Wear state identification of the axial piston pump is of great importance to secure the modern hydraulic system. Offline intelligent fault diagnosis methods are significant to solve the wear state identification problems. Nevertheless, these methods cannot meet the demand of real-time wear state identification. In this paper, an accurate and online wear state identification method based on edge computing using feature selected artificial neural network (FSANN) is proposed for the axial piston pump key friction pair. To reduce latency, an edge end node including data collection, signal pre-processing, feature extraction, fault classification is established. To cut down the amount of calculation and transmission while retaining accuracy, features sensitive to the fault are selected. The embedding performance of one-against-all support vector machine (OAA-SVM), artificial neural network (ANN), deep belief network (DBN) is compared and ANN is chosen as the embedded diagnostic model. The experimentally verified accuracy of this method is 99.0%. The single diagnosis time(SDT) is about 0.24s. Compared to transmitting raw data to the host computer, this method cut down the amount of data by about 200 times. The proposed method could diagnose the slipper wear state accurately and quickly and provide a potential way for real-time fault diagnosis for axial piston pump.