Health assessment for hydraulic system based on GRNN and metric learning

Performance degradation of the hydraulic system may result in failure or even catastrophic accidents. Therefore the health assessment for the hydraulic system is of great importance. However, the aim of most existing methods is fault detection or diagnosis, and few studies focus on the health assessment method. A scheme of health assessment based on the general regression neural network (GRNN) and metric learning is proposed in this study. First, GRNN is employed to generate the estimated output of the hydraulic system. And the residual error is obtained by calculating the difference between the actual and estimated output. Then, degradation features are extracted from residual error using a sliding window. After that, Mahalanobis metric learning (MML) is applied to adaptively find a suitable metric between testing samples and normal samples. The derived distance is normalized into confidence value (CV) as the constructed health indicator, which provides available information for Condition-based Maintenance (CBM). A simulation model of the hydraulic system is established in HyPneu and Matlab/Simulink. And two gradual faults including contamination and leakage are injected to validate the proposed method. The results of the simulation analysis demonstrate the effectiveness and adaptability of the health assessment method.