Research On Fault-tolerant Control And Life Prediction Methods For Uncertain Systems

For safety-critical systems,faults will result in irreparable losses if effective measures are not or cannot be taken.Fault-tolerant control can compensate for the impact of a fault on the system and ensure that the system has a certain level of performance.For the system that effective measures cannot be taken after the fault,a forecast should be given before the fault occurs.An effective way of doing this is life prediction.Faults,as well as the effects of environmental and external forces,will lead to the uncertainty of the system.Therefore,the research on fault-tolerant control and life prediction for uncertain systems has important theoretical significance and application prospects.However,the existing methods of adaptive fault-tolerant control are usually based on certainty equivalence principle.It is shown that the control based on the certainty equivalence principle is far from optimal.For this reason,Feldbaum proposed dual control.This paper introduces the ideal of dual control to fault-tolerant control.And,some problems in modeling and life prediction of lithium-ion batteries are further investigated.The main work is as follows.1.For the variable faults of the system under LQG framework,a multi-model reliable control(MMRC)algorithm is proposed.First,a set of models is used to cover the dynamic behavior of a system with different faults.LQG control is implemented to each model in the model set,and the a-posteriori probabilities of the model innovations are used as the weights information.Second,it is proved that MMRC is able to learn the real model of the system.When the controller is in a deadlock state,a deadlock avoidance strategy is given and the convergence of the corresponding a-posteriori probabilities is proved.Finally,the validity of MMRC is verified by an example simulation of an aircraft lateral-directional control system.The simulation results show that MMRC guarantees an acceptable performance of the closed-loop system.In addition,MMRC does not need to detect the fault time and diagnose the fault model.Since the controller fuses the control law of each model based on the weight information,the controller implements a soft switching when the system model is switched,which avoids the jitter caused by frequent hard switching to the system.2.For abrupt faults of the auto-regressive moving average with exogenous(ARMAX)system,a multi-model fault-tolerant control(MM-FTC)algorithm is proposed.The difference between MM-FTC and MMRC is that each model in the model set in MM-FTC uses minimum variance control(MVC).MVC enables the predicted output of the model in the model set to quickly track the desired output of the system because there is no constraint on the control signal.This feature ensures that MM-FTC can rapidly degenerate into the MVC corresponding to the true model of the system in case of abrupt system faults.Finally,the effectiveness of MM-FTC is verified by a numerical simulation.3.For a class of fault systems with probabilistic output constraints,deterministic control and control incremental constraints,a soft switching model predictive fault-tolerant control(SSMP-FTC)algorithm is proposed.First,the statistical performance index of the models in the model set are transformed into deterministic performance index by expanding the measurable information set.Second,the probabilistic output constraints are transformed into deterministic predictive output constraints according to the chance constraints,and the model predictive control(MPC)for each model in the model set is obtained.Similar to the soft switching strategy in MMRC,SSMP-FTC fuses the MPC of each model based on the model a-posteriori probabilities.Finally,the effectiveness of the SSMP-FTC algorithm is verified by a simulation.4.For stochastic systems with partial loss-of-control effectiveness,a dual fault-tolerant control(DFTC)algorithm is proposed.The algorithm introduces a fault-tolerant objective on the basis of the control objective.By establishing a new objective function,the controller is forced to have the characteristics of active learning to improve the transient performance of the fault system.An example simulation verifies the validity of DFTC.The simulation results show that DFTC can drive the system to the desired output,especially when partial loss-of-control efficiency occurs in the system,it can learn the unknown parameters actively,quickly,and accurately,which is significantly better than the conventional control based on the certainty equivalence principle and improves the reliability of the system.5.For the life prediction problem of Lithium-ion,firstly,the battery capacity degradation model obtained by NASA Ames Research Center in the ground laboratory is modified,and the unknown parameter estimation problem in the modified model is transformed into a large-scale data-based optimization problem.Second,an improved fireworks algorithm(IFWA)is proposed based on the accuracy ε and confidence level β of the optimization index.IFWA is able to determine the minimum number of sparks,and find the optimal estimates of the unknown parameters of the degradation model of battery capacity.Finally,the remaining useful life of Lithium-ion battery is predicted based on the resulting degradation model of the battery capacity.The effectiveness of the proposed method is verified by a simulation.