Fault Detection Based on RBF Neural Network in a Hydraulic Position Servo System

Due to the inherent nonlinearity existed in the hydraulic system, the failure mechanism become complex and the failure characteristics are difficult to extract. Model-based fault diagnosis method depends heavily on the accuracy of mathematical model. An accuracy mathematical model of the process, however, is difficult to avail because of the nonlinearity and ripple coupling in actual hydraulic servo system. Therefore, robustness of fault diagnosis method based on approximate linear model is worse. A failure observer based on RBF neural network is developed to realize failure detection. The Gaussian function is used for hidden node function, whose centers are adjusted by improved K-means clustering algorithm presented. The weights of output layer are obtained based on improved LS (least square) presented. The trained RBF observer, working concurrently with the actual system, accepts the input voltage signal to the servo valve and the measurements of the ram displacements, then rebuilds the system states. The output of the system is accurately estimated. By comparing the estimated output with the actual measurements, residual signal is generated and then analyzed to report the occurrence of faults. The experimental results demonstrate that the failure observer based on RBF neural network is effective in detecting the failure of the hydraulic position servo system