Research On Data-Driven Fault-Tolerant Control For Unknown Linear Dynamic Systems

Due to the increasing complexity and scale of control systems,it is getting harder to guarantee their long-term reliable operation.An extremely small malfunction that is not tackled appropriately may cause catastrophic consequences,particularly in the domains closely related to personal and property safety,such as aviation,aerospace,and nuclear power.To ensure that the performance of the whole system is still acceptable when faults occur,fault-tolerant control(FTC)techniques have attracted more attention in the above domains.During the past three decades,analytical redundancy-based FTC methods have been studied extensively.Instead of employing the hardware redundancy,such methods make use of the mathematical models of the controlled plants to achieve the functional redundancy of the components in control systems and further develop controllers for the purpose of FTC.However,since it is very difficult to build the accurate models of certain complex industrial plants(e.g.,large-scale chemical processes)in terms of the physical and mathematical knowledge,the model-based FTC approaches cannot handle properly the faults arising in these plants.On the other hand,with the aid of the rapid development of information technology,a great number of input-output data of the aforementioned plants can be derived in the industrial field.Because the intrinsic characteristics of the plants are contained in the obtained data,how to exploit effectively these data to fulfill the FTC(namely data-driven FTC)has become a research hotspot in both the academic and industrial communities.Therefore,the study of data-driven FTC has theoretical and practical significance.Unfortunately,the relevant research is still in its early stage and many open problems remain.To this end,by virtue of approximate dynamic programming(ADP)and subspace identification techniques,the dissertation is concerned with the problem of data-driven output-feedback FTC for unknown linear dynamic systems.As a result,Chapters 3-5 present fault-tolerant optimal control methods for the controlled plants in the design phase,i.e.,the open-loop systems;Chapters 6 and 7 describe fault-tolerant compensa-tion schemes for the PID systems that have been put into operation,i.e.,the closed-loop control systems;in the absence of the stochastic measurement and process noise,that is,under the condition that the aforesaid noise is very small and therefore can be neglected,Chapters 3 and 4,respectively,devote themselves to the solutions to the output-feedback fault-tolerant linear quadratic regulator(LQR)and L2 control problems considering both the actuator and plant failures;by means of a proposed ADP methodology,Chapter 5 fur-ther gives an approach to output-feedback fault-tolerant nearly optimal control in the pres-ence of random noise;Chapter 6 discusses how to compensate for the actuator and plant faults of digital PID systems with unknown dynamics;Chapter 7 presents a compensation strategy for multiple simultaneous sensor drift faults in such PID systems.Furthermore,all the theoretical results have been tested by experiment or simulation.The main contents of each chapter in the dissertation are summarized below.Chapter 1 covers the background and significance of this study,and introduces cur-rent developments of data-driven FTC technology.Chapter 2 includes a nomenclature part and describes the software and hardware environment of two experimental platforms,i.e.,a dual-chamber electric heating furnace and a DC servo system.In Chapter 3,the data-driven design of a residual generator(via a full-rank transfor-mation matrix)is given and a fault detection mechanism is constructed.With the help of the above transformation matrix,the relation between system state vectors and input-output data is built.As a result,an output-feedback ADP algorithm is devised so as to accomplish the LQR optimal control of unknown linear time-invariant(LTI)discrete-time systems with several outputs.Besides,the chapter presents a time-varying value function approximation structure that is based on both the input-output information and the fault detection mechanism.On this basis,an output-feedback fault-tolerant LQR optimal con-trol scheme without employing any model parameters is proposed.Finally,two numerical examples and a simulation example of a DC motor control system are used to demonstrate the effectiveness and advantages of the suggested methods.In Chapter 4,an output-feedback L2 controller for unknown LTI plants is developed in terms of the data-driven form of residual generators;for the implementation of the tracking control of the closed-loop system constructed with the L2 controller,a pre-filter is realized through resorting to the parameterized matrices of the residual generators;by dint of the above-obtained results and by modifying the value function approximation structure in Chapter 3,a data-driven output-feedback fault-tolerant L2 control algorithm is given,and subsequently,its effectiveness and merits are validated by two simulation examples.In Chapter 5,taking into account the unknown linear dynamic systems with stochas-tic measurement and process noise,a dithered Bellman equation with the innovation co-variance matrix is constructed via the expectation operator given in the form of a finite summation.On this basis,a novel output-feedback-based ADP methodology is devised.Compared with the existing ADP approaches,the methodology can ensure that the value function converges to the optimal one in the case of stationary white noise with zero mean.Besides,a fault detection mechanism,together with a tracking control policy,is presented.A resultant data-driven fault-tolerant nearly optimal control strategy consider-ing the aforementioned noise is proposed and then applied to the DC servo system for the verification of the strategy by experiment.In Chapter 6,an algorithm with less memory and computational resources is sug-gested for the online recursive identification of the residual generator,state observer,and observability canonical form of the controlled plants;a data-driven fault-tolerant compen-sation controller is further designed;a resulting fault-tolerant compensation plan depend-ing on fault detection is put forward in order to handle the actuator and plant malfunctions of closed-loop PID systems.Different from the FTC approaches developed for the open-loop systems,the plan is devised for the closed-loop systems,so that the traditional PID systems with slight modifications can possess the fault-tolerant capability,and the original PID controllers,which have been tested in long-term practice,are retained.In addition,the FTC experiments on the DC servo system are carried out with the aforesaid plan.In Chapter 7,a novel residual generator is constructed with the state vector and coefficients of the PID controller;the data-driven realization of the residual generator is accomplished;a data-driven iteration algorithm is given,by means of which the full decoupling identification of the multiple simultaneous sensor drift failures can be fulfilled such that the failures are estimated continuously and without bias(whereas the existing data-driven methods can only deliver the biased estimates);by correcting the tracking error signal with the estimated faults,the effect of such failures on the tracking properties is eliminated;a resultant data-driven sensor-fault-tolerant control scheme is provided for variable-gain PID systems with extremely large time constants and long dead time;finally,the simulation verification of the proposed approach is implemented via the continuous stirred tank heater benchmark process,and furthermore,the above scheme is applied to the dual-chamber electric heating furnace,so that its practicability is proven by experiment.Both the main contributions of the dissertation and the further research topics are pointed out in Chapter 8.