Research On Hydraulic Pump Fault Diagnosis Based On Gray Neural Network

With the development of hydraulic technology, the Hydraulic systems was used in various fields, its function increased continually and the structure more and more complex, thus the possibility of failure in the hydraulic system increased.Pump work continuous, and its structure is complicated, so it is prone to failure. According to the study in recent years, more attention has been paid to fault diagnosis of hydraulic pumps. As new technologies emerge, the hydraulic pump fault diagnosis technology develops toward automation, intelligent direction. Application of such methods to diagnose fault effectively is current research focus. This paper presents a method of combining neural network theory and gray theory to diagnose fault in hydraulic pump.Neural network and Gray theory are two kinds of mature theoretical methods in the field of fault diagnosis. The two theories combined in many ways, and more and more people study such field. In this thesis, the neural network theory and the grey relational grade theory combined. A customized network model was designed with the network toolbox of Matlab software. The transfer function and other attributes of network were set. This network model could diagnosis typical fault of the axial piston pump and calculate the grey relational grade between the unknown failure mode and known failure mode. Based on the calculated result, the network can judge the fault type of the unknown failure mode and output the result.This article is based the hydraulic servo system of Materials Testing Machine. Take the axial piston pump as research object. Monitoring the pump's status and collecting vibration signals by sensor which was fixed to the lid of pump. Wavelet theory and Hilbert transform are applied for deal with signals. Then feature vectors can be extracted from amplitude domain and time-frequency domain. These feature vectors will input the gray network, then simulate the model in order to diagnose fault. The simulated result shows effectiveness of the new method to diagnose fault.