Research On Fault Detection And Diagnosis Method For Satellite Momentum Wheel

The momentum wheel is one of the most important subsystems of the satellite,and its reliability is the basic guarantee for the normal operation of the satellite. The momentum wheel is a high speed rotating component, its life and reliability is limited, so it is the most prone to failure. Under this background, the momentum wheel is taken as the research object. So this paper mainly studies the fault detection and diagnosis of momentum wheel using telemetry data.Firstly, the structure and working principle of the momentum wheel are analyzed in detail. The relationship between data is analyzed by the working principle, then according to the telemetry data, the least square method is used to estimate the parameters of the measurement parameters. Depending on parameter correlation to establish system of equations,, and estimate the another data using particle filter, and compared with the actual measurement data.Verify the accuracy of the relationship.Residual error of estimated value and real measured data are used as the measure of momentum wheel running state. We will use Shewhart control chart andCUSUM control chart to set the threshold. Because the residual value is not subject to the requirement of the detection method, so the residual error need to be processed. Finally, the corresponding control charts of different residual errors are given.Because of the failure data of the momentum wheel,In order to get other fault data, the simulation model of momentum wheel is established by considering the relationship values. And the accuracy of the simulation model is verified by the telemetry data.The common failure modes of momentum wheel are analyzed, and the typical faults are simulated.According to the fault simulation data,and the characteristic parameters of different faults are estimated by kalman filter and kalman tracking filter, at the same time the corresponding residuals of differentfaults are given by the method of particle filter. To establish the mapping relation between characteristic parameters, the residuals and the corresponding fault. Finally,the BP neural network is used to train to give the fault diagnosis model. At the end,the accuracy of the diagnostic method is verified.