Fault Diagnosis And Fault-tolerant Control Method Based On Machine Learning For Unmanned Ground Vehicle

In recent years,with the continuous improvement of autonomous driving technology,intelligent vehicle systems have become more and more complex,and the possibility of faults has dramatically increased.To ensure the safe,effective,and reliable operation of intelligent vehicles,efficient fault detection and diagnosis(FDD)and fault-tolerant control(FTC)technologies are becoming more and more critical.On the other hand,with the development of artificial intelligence and data science,machine learning(ML),as an essential branch of artificial intelligence,can optimize its performance through data and experience.Due to the complexities of the intelligent vehicle system,the environment,the control objectives,and the randomness,fuzziness and uncertainty of the faults,the conventional fault diagnosis and fault-tolerant control methods have been difficult to meet the requirements of modern equipment.Meanwhile,the ML methods have been applied in fault diagnosis and fault-tolerant control and achieved some good results,but there are problems of online adaptability and fault-tolerant control performance optimization in learning.In this thesis,adaptive fault diagnosis and optimal fault-tolerant control based on machine learning are studied for intelligent vehicle systems.The main achievements and innovations of this paper are as follows:(1)Aiming at the characteristics of complex correspondence between sensor,actuator fault and symptom data of the intelligent vehicle,and the problem of data-driven fault detection and diagnosis,improve diagnosis accuracy and shorten training time,we designed a fault diagnosis method based on just-in-Time Gaussian process and extreme learning machine(JITGP-ELM).Just-in-Time Gaussian process modelling is used to extract estimated residuals as the input of the extreme learning machine classifier for fault detection and diagnosis.The proposed residual-classifier fault diagnosis scheme provides a new idea for data-driven fault detection and diagnosis without determining the residual threshold while detecting and diagnosing multiple faults through fault classification.Experimental results show that the new method can solve the problem of fault extraction and recognition without the prior model and improve the accuracy of fault diagnosis.(2)Firstly,A Kalman filtering method based on just-in-time learning(JITL-KF)is proposed to estimate the state of unmanned vehicle systems under noise conditions.In the framework of the Kalman filter,the unknown nonlinear system is identified by local modelling using the just-in-time learning method,and the posterior states are estimated by information fusion with sensors.Then,aiming at the problem of model predictive control for unknown systems with sensor faults,we proposed a sensor fault-tolerant control method.Based on the just-in-time Kalman filter,a sensor signal reconstruction mechanism is designed with assuming that the fault information is known.The sensor signal reconstruction can compensate the local faults well with the accurate state estimation of the proposed just-in-time Kalman filter.Finally,the effectiveness of the proposed algorithm is verified by state estimation experiments and fault-tolerant lateral control simulations in the unmanned vehicle system.The proposed algorithm has certain advantages in estimating the unmanned vehicle state under the condition of noise,and the fault sensor signals are reconstructed effectively to achieve fault-tolerant predictive control of the unmanned vehicle sensor.(3)Aiming at the optimal tracking control problem of autonomous vehicle systems with unknown faults and disturbances,to solve the problem that the off-line model may fail due to the change of dynamics of the vehicle when the fault occurs,we proposed a just-in-time learning based dual heuristic programming(JITDHP)algorithm to optimize control performance coping with faults or disturbances.While the fault information and dynamics model are unknown,the changing dynamics of the system is identified online by data-driven method,and the control performance is optimized by dual heuristic programming.The effectiveness of the proposed method is verified by the simulation of fault-tolerant tracking control tasks for the autonomous vehicle system under different reference trajectories and fault modes.The results show that the JITDHP method performs better than the adaptive fault-tolerant control and conventional dual heuristic dynamic programming methods.