Fault Diagnosis of Electro-mechanical Actuator Based on Deep Learning Network

An efficient and accurate fault diagnosis method based on deep learning network is proposed to solve the problems in the traditional fault diagnosis methods of electro-mechanical actuator (EMA). In this method, several denoising autoencoders are stacked to generate a neural network with multiple hidden layers, in which fault features are automatically extracted from original signals, getting rid of the dependence on signal processing technologies and diagnostic experiences. The greedy algorithm is adopted to carry out the pre-training of network to avoid problems of local extremum and gradient diffusion. Then, the back propagation (BP) algorithm is used to fine tune the whole network, in which the weights and biases of each layer are corrected to minimize the classification errors. The experiment results show that the fault diagnosis accuracy of the method in this paper can reach 100% with appropriate parameters, which can realize the accurate fault diagnosis of EMA.