An approach to fault diagnosability quantitative evaluation for a class of nonlinear systems

To improve the fault diagnosis ability of satellite control systems in the design phase, a novel approach to quantitatively evaluate fault diagnosability of nonlinear models is proposed. This approach is the extension of the previous researches on qualitative fault diagnosability analysis. Considering the faults caused by nonlinear factors are ignored because of linearization, satellite control systems are described a class of affine nonlinear models. The definition of fault diagnosability based on the minimal invariant dual-distribution is presented through the differential-geometric approach. According to the discriminant function of subspace similarity, the relationships of the angle included between the fault directional vector and the minimal invariant dual-distribution and the angle included between directional vectors of different faults are used to design quantitative criteria of detectability and isolability respectively. The criteria can actually exploited for answering an question like How difficult is it to detect/isolate a fault? and for guiding the design of fault diagnosis algorithms and the optimal placement of sensors. And then a specific evaluation process is provided. Finally, the momentum wheel is taken as an illustrative example applied to verify the correctness of the proposed approach. The simulation results show that this approach can exactly quantify the difficult levels to detect and isolate a fault before designing a fault diagnosis algorithm. ©, 2015, China Spaceflight Society. All right reserved.