Neural-Network-Based Adaptive Fault Tolerant Control For Multi-Variable Discrete-Time Systems

With the advent of industrialization 4.0,the modern industrial automation sys-tems are becoming more and more complex,informationalized as well as intelligent.Meanwhile,modern mechanical manufacturing systems have the characteristics of large scale,high degree of automation and flexibility.Because the structure of the manufacturing systems are more and more complex,and the price is becoming more and more expensive,the downtime caused by various faults is an unbearable burden.In the actual industrial field,due to the large number of the process data,the system variables and the various complex factors of the system itself,the fault diagnosis technique of the existing single-variable system is a challenge problem when they are extencded to the multi-variable systems.The fault,diagnosis system for multi-variable systems can meet the needs in this case,while minimizing the maintenance time of downtime,and responding quickly after the occurrence of fault.Since the modern industrial system has great scale and the communication network is very complexity,the local system component fault will lead to the loss of global control purposes in the actual engineering practice,which will further cause that the loss of econoimic and social is almost impossible to estimate.Although several fault tolerant control methods for discrete-time systems are developed,however,most of the results are concentrated on a single variable system.In addition,a majority of systems should be described in multi-variable forms in the actual industrial process,and the system fault tolerance performance is often reflected in the trade-off be-tween different indicators.Few related report can be found in the existing methods.This encourages the authors to study the fault tolerant control and corresponding optimization problem of the discrete-time multi-variable systems.On the basis of summarizing the previous work,this paper combines the in-telligent approximator with adaptive technology.Based on the advantages of these teclhniques,a neural-based fault estimation method is proposed for the discrete-time multi-variable systems.On the basis of reinforcement,learning algorithm,a new fault tolerant control mechanism is established.The proposed method overcomes the difficulty of the existing fault tolerance technique.The Imain researcch contents in each chapter are outlined as follows:Chapter 1 summarizes the development and main rescearcch methods in fault diagnosis,adaptive control and intelligent approximator.Chapter 2 investigates the adaptive actuator fault tolerant control for a class of uncertain multi-input single-output(MISO)discrete-time systems with triangular forms.The considered actuator faults contain both loss of effectiveness and lock-in-place.By using the radial basis function neural networks to approximate the unknown nonlinear functions,an adaptive fault tolerant control scheme is designed.Compared with the existing methods,by introducing the backstepping technique,one achieves the fault-tolerant control task.It is proved that the proposed control approach can guarantee that all the signals of the closed-loop systems are uniformly ultimately bounded and that the output can track a reference signal in the presence of the actuator faults.Based on Chapter 2,the studied system is extended from MISO systems to multi-input multi-output(MIMO)systems in Chapter 3,while the control method is changed from state feedback manner to output feedback manner.Chapter 3 studies the decentralized neural network(NN)output feedback fault tolerant control problem for a class of MIMO systems with actuator fault.Firstly,in accordance with the diffeomorphism theory,the original system is transformed into an input-output expression,which is suitable for output feedback control and avoids the noncausal problem.Secondly,to obtain a quick response to the fault,the response time will be reduced by adapting the unknown bounds of NN weights.Lastly,stability analysis is provided to guarantee the semi-global uniform ultimate boundedness of all the variables in the resulting closed-loop systems.Chapter 4 considers the optimization algorithm,and the focus is on the incipi-ent fault and abrupt fault rather than the actuator fault.This chapter concentrates on the reinforcement learning based fault tolerant control issue for a class of MIMO nonlinear discrete-time systems.Both incipient faults and abrupt faults are taken into account.Based on the approximate ability of neural networks,a reinforcement learning algorithm is incorporated into the fault tolerant control strategy in which an action network is developed to generate the optimal control signal and a critic network is used to approximate the novel cost function,respectively.Compared with the existing results,a novel fault tolerant controller is achieved based on a reinforcement learning method,which can reduce the total cost or the performance measure after a fault occurs in the system.The meaning of minimizing the cost function after a fault occurs in the MIMO system is that the waste will be decreased and the energy will be saved.Note that the weight values of neural networks are adjusted online instead of being adjusted off-line.Then,it is proven that the adap-tive laws,tracking errors and optimal control signal are uniformly bounded even in the presence of the unknown fault dynamics.Chapter 5 uses the minimum-learning-parameters(MLPs)method,which im-proves the fault tolerant mechanism in the last chapter.Here,it is concerned with a reinforcement learning based adaptive tracking control technique to tolerate faults for a class of unknown MIMO nonlinear discrete-time systems with less learning parameters.Not only abrupt faults are considered,but also incipient faults are taken into account.Based on the approximation ability of neural networks,action network and critic network are proposed to approximate the optimal signal and to generate the novel cost function,respectively.The remarkable feature of the pro-posed method is that it can reduce the cost in the procedure of tolerating fault and can decrease the number of learning parameters and thus reduce the computational burden.Finally,stability analysis is given to ensure the uniform boundedness of adaptive control signals and tracking errors.The former four chapters study the model-based fault tolerant control,but in Chapter 6,the data-based FTC for discrete-time system is given.Chapter 6 presents an echo state networks(ESNs)based adaptive sensor fault detection mechanism and fault tolerant control(FTC)design approach for a class of multi-input single-output(MISO)discrete-time model free systems in the data-based framework.The main feature is the novel realization of fault,detection and estimation(FDE)and FTC for the considered systems.Because of the complexity of fault detection and FTC in practical industrial processes,it will cause more difficulties in the controller design and the stability analysis.Since the ESN has good ability of short-term memory which means that the learning and training of ESN is faster than that of recurrent neural network(RNN),it is utilized to estimate the unknown sensor fault dynamics.After the fault is detected,the FTC strategy is calculated based on the ESN and optimality criterion.Then,it is proved that all the signals including tracking error are bounded.Finally,the results of the dissertation are summarized and further research topics are pointed out.