On-line Intelligent Fault Diagnosis Of Hydraulic AGC Servo Valve

This paper mainly studies the fault diagnosis of electro-hydraulic servo valve in hydraulic AGC (Automatic Gauge Control) system of 4200mm heavy plate mill. Electro-hydraulic servo valve is one of the most easy-to- failure components and it's performance and reliability will directly affect AGC system's performance and reliability. Therefore, the research in fault diagnosis of electro-hydraulic servo valve has great significance in assurance of product quality, reducing maintenance costs, improving equipment maintenance style and promoting the modernization of fault diagnosis.The failure modes and their mechanism of servo-valve were studied and it's internal flow field was analyzed by simulation. The parametric model of fluid within servo-valve which was used to calculate and simulate the distributions of flow fields through FLUENT was created in Pro / E software .The simulation validated the severe non-line relationship between pressure and flow due to wear of spool working edge. Meanwhile, the simulation result also lay foundation for one of the servo valve's feature extraction methods that based on it's dynamic characteristics.A new feature extraction method based on the valve's dynamic characteristics was proposed. One class support vector machine was adapted to identify the working state of the valve, the parameters of which were optimized by cross-validity estimation learning means. The results indicate that the proposed feature extraction method can effectively extract the fault features of the valve and it's abnormal state can be more effectively identified by taking advantage of one-class support vector machine's generalization ability than by the traditional BP neural network, which lays the foundation for on-line faults diagnosis of electro-hydraulic servo valve.In order to overcome the shortcomings of traditional BP neural network's poor generalization ability, a bayes regularization method was used to improve it. Diagnostic results of the improved BP neural network combined with different feature extraction methods were compared. The compared result indicated that feature extraction method that based on wavelet packet decomposition was more effective in diagnostic accuracy than that based on time-domain signal digital characteristics under the improved BP neural network method.A continuous description method that based on distributions of the servo valve's state characteristic parameters was proposed to give quantitative description of the servo valve's performance degradation .This can lay a solid foundation for prediction of the servo valve's performance degradation .