Spacecraft hitch detection and health evaluation based on Multivariable Time Series

Abstract

Abstract In this paper, unsupervised method is studied to solve the problem that it is difficult for spacecraft to obtain tagged hitch data. Firstly, the incremental cross-correlation filtered attribute selection algorithm (ICF) is used to complete the selection of feature subsets of spacecraft multivariate time series (MTS);Then the unsupervised learning model of LSTM-SAE is trained on a large number of normal data; Finally, a complete hitch monitoring and health evaluation system is established and verified on the data of three fault degrees. The feature space of normal data is constructed, and the health state of spacecraft is measured by the reconstruction error of faulted data and the distance of feature space. It solves the problem of predicting and avoiding catastrophic failures of spacecraft under the conditions of lack of prior knowledge, unbalanced data distribution and incomplete failure modes.