Parameter Estimation And Fault Tolerant Control For Consensus Of Cooperative Multi-agent Systems

Consensus problems form the foundation of the field of distributed multi-agent systems, a consistent state is reached by cooperative law in the network of agents. Agent’s actuator fault leads to system instability, which often causes quantities of economic losses and casualties. However, control input in traditional fault diagnosis and tolerance does not contain cooperative part, which makes it is not fit for multi-agent consensus. Therefore, a new framework for multi-agent system is proposed for handling fault, in this framework, we study the unknown input based fault detection, extremum seeking based fault parameter estimation and Lyapunov function based cooperative tolerance control.First, this dissertation shows how a multi-agent system can be modeled as first-order system and second-order system, both systems’ control input contains cooperative part, and the relationship between the Lyapunov matrix and system stability is analyzed. On this basis, this dissertation analysis the effects of system fault, sensor fault and actuator fault on system’s stability, in addition, the actuator can be further classification. By reconfiguring the decoupling matrix, a set of unknown input observers for each agent is constructed for fast fault detection. Two simulation scenarios for first-order and second-order system are designed to illustrate the feasibility and effectiveness of the proposed method.Furthermore, by taking advantage of model free of extremum seeking, a novel adaptive fault parameter estimation method is proposed, in which transform the approximation problem into the optimization problem. For the single faulty agent case, the corresponding cost function is developed, and a gradient-based extremum seeking scheme is developed, whose global convergence ability is proved. For the multy faulty agent case, gradient-based and Newton-based extremum seeking scheme are designed respectively. The cost function is created by utilizing the error between the real state and estimate state, and discusses the performance of different excitation function for estimating the parameters. In addition, in comparison with the standard gradient-based scheme, the Newton-based extremum seeking makes a faster convergence rate.Finally, by adjusting some interconnection weights based on Lyapunov potential energy function, this dissertation shows that the consensus state is still reached. For the single faulty agent case, the Lyapunov based potential function is used to derive the stability conditions, on the basis of these conditions the interconnection weights can be set to stabilize the system. A horizontal position cooperative movement example is given to illustrate the feasibility and effectiveness of the proposed method. For the multy faulty agent case, the faulty agents are isolated, a reconfigurable controller is designed to compensate for the impact of faulty agents.