A Deep Learning Based Fault Detection Method for Rocket Launcher Electrical System

This paper proposes a fault detection method for a rocket launcher electrical system by using 1D convolutional neural network. Compared with the method based on analysis of mechanism model and the method based on knowledge, this end-to-end data-driven fault detection method, which only relies on the rich data generated during the running of the system, has the ability of automatic extraction of hierarchical features. The experimental results show that the 1D convolutional neural network designed in this paper achieves the accuracy of 98.66% in the practical fault detection for a rocket electrical system, which is improved by 29% and 13% higher than the traditional fully connected shallow neural network and support vector machine, respectively, which further verifies the feasibility and effectiveness of data-driven deep learning method in fault detection applications.