Fault Estimation And Prediction For Nonlinear Stochastic Systems

With the rapid development of science and technology,modern control system is becoming more and more complex.In order to improve the reliability and security of the system,model-based fault diagnosis technology has made great progress.However,the requirement of the reliability and security for dynamic systems is increasing.When the system is in abnormal,the researchers not only want to detect the fault in time,but also want to get some useful information about the fault,such as fault amplitude and occurrence time and so on,which can provide more support for system management and maintenance.Therefore,fault estimation technology has been received much attention by many scholars in recent years.In addition,another important task of improving system reliability is the estimation and prediction of the incipient fault,whose amplitude is usually small.Because of its small amplitude,incipient fault is difficult to detect in its early stage.However,the slow development of the incipient fault can cause very serious consequences.If we can detect the incipient fault and predict its trend as early as possible,then the manager will have more time to take action to avoid the possible serious damage.Therefore,the study about fault estimation and prediction technologies has an important theoretical significance and application values.In this thesis,we mainly analyze the fault estimation and prediction problem from the following four aspects.First,based on the standard particle filter,an improved algorithm is proposed,where the state and the fault are both estimated.According to the fault estimation,each fault is detected by means of hypothesis testing,and then the estimated value of the fault amplitude is obtained by using the average value method.Meanwhile,the relationship of the sample size,the significance level of two types of error,the amplitude of fault and thevariance of the error of preliminary fault estimation are also given.Second,in order to improve the estimation accuracy of the fault,a reasonable assumption about the fault is introduced.In detail,the change of each fault signal between any two time-steps is always bounded,which can be described by the virtual noise.Then,by using the Kalman filter,we enhance the accuracy of fault estimation.Furthermore,in order to effectively deal with the saltation of the abrupt fault,an adaptive fault estimation algorithm is also developed.Third,we investigate the problem of fault estimation and prediction for a class of nonlinear systems with multiplicative incipient fault,which is represented by a nonlinear evolution function with unknown parameters.In order to give the high accuracy fault estimation,we consider the characteristics of incipient fault and introduce the virtual noise.Then,by using the particle filter and Kalman filter,the state and the fault are estimated simultaneously.When the fault is detected,the parameters of the evolution function are obtained by using the Gauss Newton method.According to parameters of the evolution function,the future fault signal can be predicted.Therefore,we can provide more time and information for system maintenance.Fourth,the problem of fault estimation and prediction for nonlinear stochastic systems with intermittent observations is considered.Each component of the incipient fault is represented by a piecewise linear function with unknown parameters.By using the extended Kalman filter and Kalman filter,the fault and state are both estimated.Then it is ectended to the condition of intermittent observations.Meanwhile,the boundedness of the estimation error is also discussed.In the stage of fault prediction,the parameters of fault evolution function are obtained by linear regression method,and then the future fault amplitude is predicted.At last,the conclusion of the whole paper is given and the future work is presented.