Fault Detection For Distributed Networked Control System

Because of the advantages such as low cost, mobility, modularity, resource-sharing and easy maintenance, distributed networked control system (DNCS) becomes one of the hot topics in technical and academic areas. Just the same as the traditional control system does, the DNCS also has various faults, and the features of the DNCS such as time-delay, data drop and the bandwidth constraint aggravate the damage caused by the fault, and even drive the system unstable. Sometimes the fault occurrence frequency is in contradiction with the scale of the DNCS and the network environment. Therefore, it is very important and urgent to raise the safety and reliability of the DNCS. Fault detection of the DNCS is proposed to reduce the influence caused by the time-delay and data loss in this dissertation.This dissertation contains the following main sections:1. For the DNCS based on the packet transmission, the discrete system model is obtained.Based on the analysis about the system characteristic, the packet transmission is introduced to act as the transmission mechanism between the central unit and the subsystem. Due to the different sampling time and different packet size, the whole system is regarded as a multi-rate sampling system and the lifting technique is used to get the discrete system model. The simulation platform is established in Matlab/Simulink environment, and the Truetime tool is applied to realize the packet transmission.2. For the subsystem with time-delay and data drop, the fault detection based on robust filter is proposed.First of all, considering the subsystem with time delay in state, input and output, the effect caused by time delay on the subsystem is analyzed and the system model including time delay is obtained. The fault detection filter design problem is transformed to find the filter parameters such that the extended system is stable and also is satisfied with the robust condition. The residual obtained by the robust filter is used to decide whether a fault happens.Second, considering the subsystem with time delay in state and data drop in input and output, the subsystem is modeled as a markovian jump system, and the similar robust filter is used in the system. Finally, the proposed method is validated by the simulation result. 3. For the central system with time-delay in subsystem, the fault detection based on sliding mode observer is proposed.Firstly, considering the central system with all the states available, the sliding mode observer is designed and the fault detection issue is transformed to find the sliding mode observer parameters to ensure the stability and reachability conditions of the sliding mode movement. The residual is obtained once the observer parameters are available.Secondly, considering the central system with some states unavailable, the transformation matrix is introduced to separate the states available and the ones unavailable, based on which different sliding observers are designed respectively. The stability and reachability conditions are applied to get the filter parameters as well as the fault residual. Finally, the proposed method is validated by the simulation result.4. For the subsystem with periodic communication sequence (PCS), the fault detection based on robust filter is proposed.Considering the principle of PCS, the subsystem with PCS is regarded as a switched system, for which the robust filter is designed. The unknown parameters of the filter are obtained by ensuring the system stable as well as inhibiting the effects of disturbance. The result shows the proposed method still gets a good fault detection performance even with the incomplete information.5. For the central system with deterministic communication logic, the fault detection based on state observer is proposed.First of all, the deterministic communication logic including the identical estimator and the logic judgment unit is added to the central system which becomes an uncertain system correspondingly. The state observer is designed for the uncertain system to obtain the residual used for fault detection.Secondly, consider the case when the data drop happens in the central system with the deterministic communication logic. Whether the information used for fault detection in the central system is updated or not is up to two factors: the logic judgment unit and the data drop. If the information is updated, real information is used, and if not, the output of the identical estimator will be used. Based on the analysis, the central system model is modified and transformed to a markovian jump system, and the state observer is designed for the markovian jump system. Finally, the proposed method is validated by the simulation result.