Experimental investigation on electro-hydraulic actuator fault diagnosis with multi-channel residuals

Electro-hydraulic actuator (EHA) is a commonly used critical component in many important occasions. The enhancement of fault diagnosis accuracy of EHA can greatly increase the reliability of whole equipment. However, EHA fault diagnosis is quite difficult due to complicated multi-channel monitoring data. Besides, experimental test data are lacking and numerical simulations are always used for method validation. In order to deal with these challenges, this paper proposes a novel EHA fault diagnosis method through residual generation and deep learning. First, several observers based on back propagation neural network (BPNN) are constructed to generate multi-channel residuals. Subsequently, one dimensional (1-D) convolutional neural network (CNN) is adopted to achieve accurate fault diagnosis taking advantage of multi-channel residuals. The performance of the proposed method is experimentally validated through comparisons with several classical fault diagnosis methods. Results reveal that the proposed method has good robustness and can greatly enhance diagnosis accuracy of EHA.