Fault Detection Of Networked Markov Jump Systems With Communication Constraints

Driven by the dual-scale characteristics of time and events,networked Markov jump systems（MJSs）can well model randomly occurring autonomous abrupt changes or switch-ing in system parameters or structures,and their system components are usually connected by a shared communication network,which frees them from the constraints of geographi-cal and spatial distribution and brings many advantages such as low cost,easy deployment,etc.As a result,they have increasingly become the preferred choice for studying large-scale complex industrial systems.Although networked MJSs have many advantages,the limited mode access and the problems of bandwidth limitation,communication security,and mode non-synchronization caused by the intervention of communication networks also bring a series of difficulties to the analysis and design of the systems.In addition,due to the long-term high-load operation of the actual industrial system,the internal com-ponents or subsystems will inevitably encounter failures,such as actuator failures,sensor failures,etc.The occurrence of these failures seriously affects the control system perfor-mance and may even cause system instability,resulting in serious production accidents and huge economic losses.With the steadily increasing requirements of modern industry for operational reliability and safety,fault diagnosis technology,as an effective means to cope with such unexpected failures,is essential to ensure the safe and smooth opera-tion of control systems.Therefore,this dissertation mainly focuses on the fault detection problem of MJSs with limited information access in the limited network communication environment.The main research contents of this dissertation are summarized as follows:The asynchronous fault detection problem for nonlinear MJSs with uncertain prob-abilities is investigated under limited network bandwidth and mode mismatch.A hidden Markov model with partially unknown probabilities is established to describe the mode mismatch between the filter and the original system,and the hidden mode information of the system is estimated for constructing the filter.In order to improve the utilization of network resources,a random logarithm quantizer is designed based on the hidden Markov model to encode the system output signal.Considering the random measurement missing phenomenon,the dissipativity analysis criterion of the Markov jump system with limited mode information access and quantized input is established,and a consistent asynchronous fault detection filter design method is proposed,which can be used for a variety of designs such as passive,_∞performance.This approach gives a unified framework for solving the filter synthesis problem based on the hidden Markov model with partially unknown probabilities,i.e.,the method is available regardless of whether the partially unknown probabilities exist in the transition probability matrix of the Markov chain,in the mode detection probability matrix of the observed signal,or in both of them.For continuous-time MJSs,a hidden Markov model with partially unknown proba-bilities is introduced to describe the mode mismatch phenomenon between the original system and the filter.By formulating boundary conditions for the frequency and duration of aperiodic Denial-of-Service（Do S）attacks,the exact mathematical model in stochastic form is adopted to characterize the network attack phenomenon in actual communication links.A new resilient dynamic event-triggered scheme is designed to efficiently save com-munication resources while resisting the impact of aperiodic Do S attacks on network trans-mission.Considering the effects of event-triggered,asynchronization and Do S attacks,a stochastic stability criterion based on dissipativity is established for MJSs according to switched system theory,and a co-design criterion of asynchronous dissipative resilient filter and the dynamic event-triggered scheme is given.Traditional residual evaluation mechanisms are mostly based on the residual root-mean-square design and usually rely on human experience to pre-select a fixed evaluation function threshold.However,it is often difficult to choose an appropriate prior threshold in practice,which directly affects the fault detection performance.Aiming at this problem,a zonotope-based dynamic event-triggered fault detection method is investigated with the help of set-membership estimation theory.Different from the previous fault detection filter design using the single performance index,considering the unknown but amplitude-bounded noise and disturbance characteristics,a joint design method of hybridl₁/H_∞asyn-chronous fault detection filter and dynamic event-triggered mechanism under deception attacks is proposed based on hidden Markov model with partially unknown probabilities.Among them,optimizing thel₁performance can improve the joint attenuation capability of the filter for noise and disturbances,while optimizing theH_∞performance can improve the sensitivity of the filter to system faults.Then,the residual interval estimation for the fault-free case is given via the zonotope-based method,and the corresponding dynamic threshold residual evaluation algorithm is proposed.Most of the previous fault detection schemes were designed with full-order filters,but for high-order systems,the computational complexity will increase dramatically with the growth of the system order,which is difficult to solve.Aiming at this problem,consider-ing the conic-type nonlinear Markov jump system with general probability,the zonotope-based asynchronous fault detection problem is studied under an adaptive event-triggered mechanism.For this system,the probability information may be known,unknown,or un-certain.In order to describe the limited mode information access and general probability information,a comprehensive hidden Markov model is established,in which the transi-tion probabilities and the mode detection probabilities contain both uncertainty and un-known peculiarity,so it is more general.By introducing the double variables-based decou-pling principle and variable substitution principle,a co-design method for asynchronous reduced-order fault detection filter and adaptive event-triggered schemes is derived under the proposed mixedl₁/H_∞performance framework in the previous chapter,and a dynamic threshold residual evaluation mechanism is established based on the zonotope technique.Considering that the full-frequency domain filter design method is relatively con-servative,the finite-frequency fault detection problem for fuzzy MJSs with generally hy-brid probabilities is studied,and a bandwidth-aware adaptive event-triggered scheme is proposed to schedule data transmission in the communication network.The automatic adjustment of the dynamic threshold coefficient in the event-triggered scheme is accom-plished through a bidirectional adjustment mechanism,while the bandwidth status ac-knowledgement parameter balances the bandwidth occupancy and system performance according to the traffic load in the network.Based on the comprehensive hidden Markov model,the general case of non-symmetric boundary for probability estimation error is considered,and a double-boundary method based on lossless transformation technique is proposed to deal with the estimation error with non-symmetric boundaries,so as to reduce the conservatism of the filtering solution.Considering the constraints caused by mismatched premise variables in fuzzy rules,an event-triggered finite-frequency L_∞/H_∞ asynchronous reduced-order fault detection filter design criterion is deduced with the aid of a new cross-decoupling method.On this basis,a dynamic threshold residual evaluation scheme with asynchronous premise variables is designed by using zonotope technique.