Multi-sensor Information Fusion Technology In The Hydraulic System Fault Diagnosis

Hydraulic system is the core part of engineering equipment and its faults may destroy the normal operation of the entire system. However, signal monitoring may be submerged by the noise because of the bad working conditions, which lead to single sensor can not represent the characteristics accurately. Therefore, we should make full use of Multi-Sensor Data Fusion technology, to fix the fault exactly so as to reduce the loss.This paper focus on the fault diagnosis of hydraulic pump, which is the key components of hydraulic power system. Forming effective sensor network and based on the grading idea of improved JDL model. Which make use of pump shell's vibration signals of three directions and temperature signals of the pump's leaked mouth. Then we put forward the information fusion model of hydraulic fault diagnosis with three fusion levels. Data level is mainly responsible for the original data cleaning, calibration and feature extraction. Feature level through the parallel PSO-BP and MPSO-RBF neural network make up vibration subnet and temperature subnet to make local diagnosis. Decision-making level according the modified D-S evidence theory to make the final fusion and decision. In this fusion system, there are data communications and feedback between each sublevel and the final diagnosis. At last, with Visual Studio2005as the platform, calls the matlab COM components, to establish friendly landing interface, realize the C/S structure of fault diagnosis system.In this system, comparing Yager combination rules, D-S combination rules and Sun Quan combination rules, and proposing an improved D-S combination rules aim at the problems that the evidences in D-S theory are difficult to obtain and the classic D-S combination rules are failure when the evidences are conflict. At the same time proposing a decision rule that the joint Pl&Bel combination algorithm of decision rules. Then compared the advantages and disadvantages with kinds of ANNs, putting forward the MPSO algorithm to optimize the RBF neural network in the temperature subnet by using the improved RPCL algorithm. The experimental simulation results prove that the method of this fusion system have strong fault tolerance, and can make full use of sensors' information and fix the position of fault accurately.