Intelligent Fault Diagnosis Method of Inertial Sensors for Space Gravitational Wave Detection

Space inertial sensor is one of the key loads in space gravitational wave detection mission. Once it fails, the entire mission is likely to be affected or even fail. The existing data-driven intelligent fault diagnosis methods can effectively diagnose some sensor faults, but it is still difficult to solve the problem that measurement data of space inertial sensor is strong coupling and includes much noise. To solve this issue, this paper proposes a convolutional recurrent variational encoder (CRVAE) for fault diagnosis of space inertial sensors. Specifically, a multilevel feature matrix that represents different time scales is firstly constructed based upon sensor raw data. Subsequently, CRVAE trained by health sensor data encodes the feature matrix, then reconstructs the matrix by decoding. Decoded matrix should restore the original feature matrix as much as possible. Intuitively, the decoded matrix of fault data will hardly restore the original state. By analyzing the residual feature matrix generated by CRVAE, the fault diagnosis of space inertial sensors can be realized. In addition, a fault evaluation function is given in order to estimate the fault severity. The result shows the method of this paper can detect fault timely and accurately, and the proposed fault evaluation function can achieve precisely quantitative analysis of fault severity.