DSmT-based three-layer method using multi-classifier to detect faults in hydraulic systems

Abstract

Fault identification in hydraulic valves is essential in maintaining the reliability and security of hydraulic systems. Due to the nonlinear characteristics of hydraulic systems under noisy working conditions, it is difficult to extract fault features from vibration signals collected from the surface of the valve body. Therefore, a DSmT-based three-layer method using multi-classifier is proposed to detect multiple faults occurred in hydraulic valves. Firstly, the raw signals are personalized to construct the training samples and the unknown testing samples. Secondly, a three-layer structure of the hybrid model called the layered hybrid model is constructed, which is suitable for hydraulic valves to detect the faults of different fault groups (including coil fatigue in the actuator and the abrasion inside the valve) and improve the diagnosis accuracy obviously. Finally, classification methods are selected to classify fault groups in the first two layers, and then the fault types are identified in the third layer by the fusion results using the Dezert-Smarandache Theory (DSmT). Experimental investigations are performed to validate the performance of the present method using a hydraulic valve (solenoid controlled pilot operated directional valve) controlled the hydraulic test rig. The results show that the average accuracy of detecting twelve types of faults is about 98.1%, which are better than those using other methods. It is expected that the present DSmT-based three-layer method using multi-classifier can be applied to more complex hydraulic systems.