Fault Detection And Reconfiguration For The Control Of Hybrid System Using MLD

With the emergence of complex process industries and wide use of computers, control systems become hybrid with increasing computational complexity. According this situation, this dissertation studies the control of hybrid systems using Mixed Logical Dynamical (MLD) theory. An improved algorithm for solving Mixed Integer Quadratic Programming (MIQP) problems is developed and then applied in predictive control scheme to control hybrid systems. Based on MLD, modeling of fault and Fault Detection of hybrid systems are studied. Two strategies for controller reconfiguration are presented while hybrid systems have redundant manipulated variables. Simulation results of the algorithms studied in the dissertation show promising performance. Main research works and achievements are summarized as following:Transformation approach from propositional logical to mixed integer inequality is studied, and several strategies for reducing the number of logical variables in MLD model are presented. Considering the exhibition of limit cycle, the method of determining steady states for hybrid systems by solving mixed integer programming problems is researched. Then predictive control for hybrid systems with adjustable performance based on MLD model is analyzed.The fault model of actuator and sensor fault for hybrid system is analyzed and established. The method of fault detection for hybrid systems is formulated as a state estimation problem. A cost function of Moving Horizon Estimation (MHE) is proposed and used to estimate the states of hybrid systems. Two strategies of controller reconfiguration for hybrid systems using MLD framework are studied, then applied in case of failure to maintain stability and acceptable performance of hybrid control system.The resulting optimization problems of predictive control, fault detection and reconfiguration can be formed into MIQP problems. In order to reduce the number of Quadratic Problems to be solved, a new tree exploring strategy for Branch and Bound method is proposed. A new approach of solving MIQP using Discrete Particle Swarm Optimization (DPSO) algorithm is developed, and several improving methods in refreshing the inertia weight and discrete variables are proposed, the simulation result shows the effectiveness of proposed DPSO algorithm.The strategies and algorithms in this dissertation are studied by simulation in a three tank laboratory model. Simulation results show effectiveness and promising performance in predictive control, fault detection and controller reconfiguration for hybrid systems.