State Estimation Fusion And Fault Diagnosability For Multi-sensor Networked Systems

With the rapid development and rise of various sensor network technologies,information transmission,adopting sensor communication network as a pivot,forms networked multi-sensor system,namely multi-sensor network systems(MNSs).The state estimation fusion of MNSs has become an important research direction in the control field,and it has a broad functioning scope and prospective wide applications.The communication network has some inherent characteristics,such as carrying capacity,limited communication bandwidth,limitation of sensor nodes energy,and diversification of network topology.The network provides convenience for multi-sensor state estimation fusion,at the same time,it will inevitably bring a series of new problems and challenges,such as measurement random delays,measurement packet dropouts,sensor faults and complexity of noises.Multi-sensor distributed state estimation fusion has certain capacity of fault-tolerance,but when the number of faulty sensors is too many or the failure is serious,distributed state estimation fusion will not be able to handle the fault.Therefore,fault diagnosis of MNSs is very necessary as well.However,fault diagnosability analysis is the fundamental basis and prerequisite before fault diagnosis.In this thesis,based on the processing of network delays,processing of data packet dropouts,correlation analyses of noises,decoupling of unknown input disturbances and differential geometry knowledge,combining distributed fusion,centralized fusion and fault diagnosis techniques,several effective state estimation fusion algorithms and fault diagnosis conditions and design scheme are given respectively.The main content of this thesis can be summarized as follows:1.The problem of optimal distributed weighted Kalman filter fusion(ODWKFF)has been investigated for a class of multi-sensor unreliable networked systems(MUNSs)with uncorrelated noises,measurement random delays and packet dropouts.Buffers of finite length are proposed to deal with measurement random delays.The independent Bernoulli variables are used to model the process of measurement packet dropouts,and an equation of the redefined noise statistic characteristics to deal with measurement packet dropouts.A novel optimal local Kalman filter with a buffer of finite length and uncorrelated noises is derived for each subsystem.On this basis,the multi-sensor ODWKFF algorithm with buffers of finite length and uncorrelated noises is obtained.Simulation results illustrate that the proposed algorithm is optimal in comprehensive performance aspects and has stronger fault tolerance.2.A distributed weighted Kalman filter fusion(DWKFF)algorithm has been designed for a class of MUNSs with stochastic uncertainties,autocorrelated and cross-correlated noises.The correlation analyses of noises are sufficient.The stochastic uncertainties caused by correlated multiplicative noises exist in the state and observation equations.The process noise and the observation noises are one-step autocorrelated and two-step cross-correlated,respectively.We consider the observation random delays and packet dropouts due to the unreliability of MNSs.Buffers of finite length are proposed to deal with measurement random delays.The independent Bernoulli variables are used to model the process of measurement packet dropouts,and an equation of the redefined noise statistic characteristics to deal with measurement packet dropouts.On the basis of the optimal local Kalman filter with a buffer of finite length and autocorrelated and cross-correlated noises for each subsystem,A multi-sensor DWKFF algorithm with buffers of finite length and autocorrelated and cross-correlated noises has been obtained.Simulation results show that the performance of the proposed DWKFF algorithm with buffers is better that of DWKFF without buffer,and this effect has no relationship with the length of buffers within a certain length.Simulation results also demonstrate that the optimal length of buffers is obtained,and the proposed algorithm with buffers has better fault-tolerance ability.3.The problem of distributed federated Kalman filter fusion(DFKFF)is proposed for a class of MUNSs with uncorrelated noises,measurement random delays and packet dropouts.First,an optimal DFKFF algorithm of MUNSs without buffer is presented,and rigorously proved to be equivalent to centralized optimal Kalman filter fusion(COKFF)algorithm of MUNSs without buffer.Then,the methods of the above two chapters are used to deal with measurement packet dropouts.A DFKFF algorithm of MUNSs with finite length buffers is derived based on the optimal local Kalman filter with a buffer of finite length for each subsystem.Finally,two simulation examples are given to illustrate the effectiveness and superiority of the proposed algorithm.4.Centralized scaled unscented Kalman filter(SUKF)state estimation fusion algorithms have been designed for a class of nonlinear MNSs with uncorrelated noises.First,three multi-sensor centralized SUKF fusion algorithms are proposed,including augmented measurements,measurements weighted and sequential filtering fusion.Then,these three algorithms and the corresponding centralized extended Kalman filter(EKF)fusion algorithm are analyzed with theoretical filtering accuracy.Finally,Simulation comparison results and comprehensive analysis show that the multi-sensor centralized augmented measurements SUKF fusion algorithm is optimal in comprehensive aspects among six algorithms.5.Determining conditions and design scheme of fault diagnosability are given for a class of nonlinear affine MNSs with unknown input disturbances.Based on the differential geometry theory,two coupling relationships of outputs with uncertainties and faults are presented,and the sufficient and necessary conditions of fault detectability and isolability are obtained.For the problem that uncertainties influence the system outputs and make the faults undetectable,a design scheme of fault diagnosability is proposed by using state expansion method.The unknown input observers are designed to generate the residual signals,which verify that the faults are diagnosable.Simulation results and numerical analysis are proposed to demonstrate the effectiveness and practicality of the analysis and design scheme of fault diagnosability.In this way,a relatively complete theoretical system of the analysis and design of fault diagnosability for nonlinear affine MNSs with unknown input disturbances.6.Simulation results are presented to illustrate the effectiveness and superiority of the proposed state estimation fusion algorithm.Simulation examples and numerical analysis are presented to demonstrate the effectiveness and practicality of the proposed analysis and design scheme of fault diagnosability.In the end,main results in this thesis are summarized,and the prospects for the future research are presented.