The Early Fault Detection Of Flywheel System Based On Data Driven Method

The data-driven fault detection method is one of important methods for the fault detection of complex spacecrafts,which can achieve the early fault detection of complex spacecrafts.And it can also provide the guarantee for the spacecrafts' safety and reliability in-orbit.In this paper,taking the fault detection of flywheels as the background,we investigate in-depth the early fault detection of flywheels based on data-driven methods including the statistic analysis and pattern recognition technology.As a result,the main study contents and contributions are as follows:In order to implement the early fault detection of flywheels,it is necessary to propose the fault modelling and feature extraction of flywheel's data.Firstly,by analyzing measurement points in the flywheels,it is summarized that the observed variables could be provided by the flywheels and the types of faults required for the early fault detection of the flywheels.Secondly,according to the fault modelling of sensor and system's parameters in the flywheel,the failure data used in the research of ealry fault detection are generated by the simulink model of flywheel.Finally,considering the feature extraction of flywheel's data and disturbing factors of data characteristics,the key issues of ealry fault detection can be dealt with based on the threshold estimation and variation factors of fault detection index.In order to detect the flywheel's faults based on statistic analysis,two methods named as the improved PCA and Gaussian Mixture Model(GMM)are proposed.They can detect the early faults of flywheel with linear correlation and nonlinear correlation respectively.The improved PCA method can divide the values of SPE index in the residual space into several groups or clusters,and decrease the threshold of SPE index by increaseing the number of cluster in residual space.Therefore,the improved PCA method can be used to detect the faults of flywheel as early as possible.However,the GMM method can improve the data fitting precision by increasing the number of Gaussian function,and detect the early faults of flywheel by BIP index.Finally,the effectiveness of the proposed algorithms are verified by the simulation and telemetry data.Based on pattern recognition,the improved IMS(Inductive Monitoring System)algorithm is proposed and the SDF(Symbol Dynamical Filter)method is imployed to detect the early faults of flywheel.In the improved IMS method,the clusters can be used to describe the interactions of physical variables when the amplitude of variables is changed by the fault of flywheel.Compared with the IMS method,the cloud model can be used to offer the confidence of detection results.Thus,the fault detection result of improved IMS can be more reasonable and accurate.While the period of single variable is changed by the fault of flywheel,the SDF method can be used to extract the period characteristics of motor current,and detect the early fault of flywheel by the threshold of probability distance.In this way,the early fault detection of flywheel can be achieved.Finally,the effectiveness of the proposed methods are verified by the simulation data.In order to detect early faults of flywheels with energy storage,a two-step method named as intelligent clustering and PCA method is proposed.This method has a strong ability of self-adaptability without using fault knowledge.The flywheels with energy storage not only can work as an actuator in the attitude control process,but also can work as a battery.The faults of flywheels can be suppressed by the closed-loop control.The faults of flywheels with energy storage are easily mixed by the status switching.Therefore,a method based on the clustering and PCA is proposed to detect the early faults of flywheels with energy storage.Firstly,the reachable-plot in OPTICS can be used to identify the status of the new flywheels with energy storage.Secondly,the K-means algorithm can be used to group the process data according to the status of flywheels with energy storage,Finally,the SPE of PCA can be used to detect the faults in each cluster.The effectiveness of the proposed algorithm is verifed by the simulation data of flywheels with energy storage.