LSTM-BASED ESTIMATION OF ATTITUDE SENSOR RUL

Abstract. This paper represents a framework for estimating the remaining useful life of satellite attitude sensor namely the angular rate gyro. First, data collection relies on the output of Kalman filter which estimates both attitude and gyro errors. Next, the LSTM model is fed with these data after processing for a more convenient learning. After training, the model is tested with new data aiming to explore its performance. The RMSE metric is used for assessing model performance where it showed good results with overall RMSE not exceeding 7%.