Research On Bayesian Filtering Diagnosis Methods Based On Gaussian Process Regression Model

The complicated system can not be monitored and managed by the traditional fault diagnosis method since there is no accurate system equation.However,lots of data which reflects the operating mechanism and state of the system are produced all the time.How to use these data to improve the security and reliability of the system is a hot topic in fault diagnosis.In this paper,the Bayesian filtering methods based on Gaussian process regression model are used into fault diagnosis.The methods overcome the disadvantage in traditional Bayesian filtering methods of high dependence on system equation by data-driven method,and improve efficiency of fault diagnosis by recursive method.The new methods have great significance to system security and reliability.The thesis emphasized on several aspects as follows:(1)Fault diagnosis methods are summarized and supplemented.Based on traditional classification,the data-driven fault diagnosis method is added.The Neural Network(NN)method and the Fuzzy Logic Controller(FLC)method are classified into the data-driven method,which are knowledge-based method in original calssification.Further more,the Gaussian Process Regression(GPR)method is added into the data-driven method.(2)To overcome the disadvantage of heavy dependence on system equation in traditional Bayesian filtering method,GP-UKF and GP-CKF are used into fault detection,and the accuracy and efficiency of the two methods are compared.The simulation results show that both GP-UKF and GP-CKF methods can detect faults successfully without system equation since the two methods based on only off-line data.In addition,GP-CKF method performs better in efficiency and accuracy.(3)To improve the efficiency of fault detection in non-linear system,the two new filtering algorithms:ISRGP-UKF and ISRGP-CKF algorithms are propoesd.The new algorithms use recursive method to improve efficiency,and introduce Importance Sampling(IS)to select the subset of training inputs,or basis vectors,which is used to ensure accuracy.The new algorithms are used in aircraft tracking system and real-time air temperature sensor fault detection.The results show that the new algorithms are computationally more efficient,and guarantee accuracy in fault detection.