Data-driven fault detection design for satellite's attitude control system

The object of this paper is to address data-driven fault detection design for satellite's attitude control system with different data characteristics, which includes the establishment and the evaluation of the detection statistics. Firstly, a simple T² detection statistic is constructed for stale data. Then, a mid-value smoothing method based on data-driven is proposed to cope with the unstable data. For the system without baseline data, a predicting and modeling method is applied to generate the residual statistic. Finally, an input-output system identification method is studied to search a model structure for the control system including both tahe control input and the measured output. The proposed approaches are driven by pure data and are evaluated by using an example of a satellite's attitude control system, and excellent results are obtained.