2D-CNN-Based Fault Diagnosis of Internal Leakage in Electro-Hydrostatic Actuators

Abstract

Electro-Hydrostatic Actuators (EHA) are used extensively to produce displacements and high forces in various industrial applications, such as aircraft and ships. The internal leakage of EHA can lead to economic loss and personal injury. Convolutional neural network (CNN) is a basic method of deep learning, which has strong autonomous learning capability. In this paper, a two-dimensional convolutional neural network (2D-CNN) based fault diagnosis method for EHA internal leakage is proposed. Firstly, the one-dimensional pressure signals collected by sensors are converted into two-dimensional signals, and then these two-dimensional signals are directly fed into a 2D-CNN model, features are extracted through convolution and pooling operations, and the model is optimized using the reset learning rate to improve the fault diagnosis accuracy of the model, and then the diagnostic results are output using a classifier. The results of the study show that the accuracy of this method in diagnosing the internal leakage of EHA reaches 95.75% Compared with the traditional 1D-CNN, the accuracy of this method in fault diagnosis has been improved to a large extent.