Early sensor fault detection based on PCA and clustering analysis

This paper proposes a novel scoring index for the early sensor fault detection in order to make full use of massive archived spacecraft telemetry data. The early detection of sensor faults is made by using the index constructed by the K-means algorithm and PCA model. The sensor fault detection includes the learning phase and monitoring phase. The amplitude of sensor fault has been always increasing when the performance of sensors deteriorates during a period. The proposed index can detect the smaller sensor faults than the squared prediction error (SPE) index which means it can discover the sensor faults earlier than the later. The simulation results demonstrate the effectiveness and feasibility of the proposed index which can decrease the check-limit as much as 40% than SPE in the same magnitude of bias sensor fault.