Research On Analytical Model-based Fault Detection For Networked Control Systems

NCS is a growing research topic in control field with the development of automatic control and computer networks technologies. Due to the increasing complexity and demands for safety and reliability, research on fault detection has become more important and challenging in the area of automatic control. Because of the existence of network-induced time delay, packet dropouts, packets out of sequence and multirate sampling, traditional fault detection approaches to LTI systems usually loss efficiencies for NCSs and more attention has been received in recent years. However, the fault detection problem for NCS has not been solved effectively in the presence of uncertain network-induced time delay and multirate data sampling.This dissertation focuses on the model-based fault detection for NCSs with network-induced time delay and multirate data sampling. The main research results consist of the following parts.(1) For NCS with unknown network-induced time delay less than one sampling period, a discrete-time system model is firstly established and the influence of the unknown network-induced time delay is handled as an unknown input. Then the problem of observer-based fault detection for NCS can be solved by applying an H_∞, optimization approach. Both the observer gain matrix and the post-filter have been obtained by solving a matrix Riccati equation.(2) Based on a discrete-time system model of NCS with network-induced time delay less than one sampling period, the problem of parity space-based robust fault detection is investigated for NCS. The main purpose is to minimize the error between residual and fault (or weighted fault) and simultaneously guarantee the robustness to network-induced time delay and unknown input. Applying a new proposed performance index, the design of parity space-based fault detection is formulated as a multi-objective optimization problem and a solution to parity matrix can be obtained by solving a set of LMIs.(3) A new discrete-time uncertain system model is first introduced for the NCSs under consideration of network-induced time delay. Then a parity relation is established to generate residual and the design of parity vector is formulated as a minimization problem and a feasible solution can be obtained by solving a generalized eigenvalue problem. Especially in the case of a matrix rank condition being satisfied, an analytic solution can be further derived by applying a singular value decomposition method. (4) By applying the lifting technique, a lifted LTI system is firstly established to model the multirate sampled-data system and, based on this, the design of the residual generator is formulated as an optimization problem. By utilizing the non-uniqueness of its optimal solution, the causality of the residual generator can be guaranteed. Then the fast rate residual updated at the same rate with the measurement output can be implemented via the inverse lifting technique.(5) Under the assumptions of network-induced time delay being unknown but bounded, packet dropouts and packets out of sequence being unavoidable, a state feedback controller, a system model for NCSs is firstly introduced in the continuous-time domain. Then an observer-based H_∞fault detection is formulated and, by applying the Lyapunov-Krasovskii functional approach, a delay-dependent sufficient condition on the existence of an H_∞fault detection filter is derived in terms of matrix inequality. Furthermore, an algorithm is proposed to get a feasible solution to the H_∞, fault detection filter gain matrices in terms of LMIs using a cone complementary technology.(6) Using a tree-tank system as a plant, we apply the proposed fault detection approaches to a NCS with the aid of MATLAB Toolbox. Both the computer simulations and experiment results demonstrate the effectiveness of the proposed methods.