Fault Detection And Diagnosis Of Control Systems For Thermal Power Plant

The control system has become an important part of the large-scale thermal power plants, its reliability and performance play a critical role in reliability and economy of the entire process.Modem control systems are characterized by increasing capacity and complexity with a lager number of control loops and parameters monitored.Meanwhile, the possibility of faults increases.In order to maintain continuous, safe operation and achieve maximum profit, effective measures should be taken to diagnose and recovery faults in the early time, aside from choosing the computer control systems of good quality.However, current maintenance strategies and the chain-protection system cannot meet the higher quirement of on-condition maintenance and the ability to treat emergency.According to the characteristic of thermal processes, theories and approaches of fault detection and diagnosis (FDD) are mainly studied in this paper and applied to real design of FDD design for thermal control systems.Firstly, the reliability of thermal control systems is analyzed, following with the state of art of FDD techniques in thermal power plants.Then the content and tasks of FDD for thermal control systems are defined.For such a complex system as the power plant, there does not exist one method that can suitable for all faults that may suffer.A three-level configuration, which is composed of the component level, the loop level, and the system level, is proposed as the basic framework for FDD system.This configuration gives a general framework of the diagnostic system for large-scale thermal control systems.From the point of component level, a novel method for actuator fault diagnosis using trend analysis of input and output signals related to the component itself is proposed.The simulation model of actuator with spring-diaphgram pneumatic actuator, as the main research object, is discussed.According to the available measurement signals of industrial actuator, the universal signal model for industrial actuators is generalized.Fault detection algorithms for estimating typical deadzone, gain, bias, fix, and stick-slip faults are proposed.Many tests on simulation data and real-world data have been made to demonstrate the effectiveness of the proposed method.Because of its simpleness and independent of process model, this approach is suitable for practical application.In the loop level, model-based FDD methods are very efficient when the process model is available or easy to obtain.A novel scheme for sensor fault diagnosis under whole course condition of control systems is proposed in this paper.The dynamics of the system with nonliearity is described by fuzzy dynamical model that is an aggregation of a set of locallinear models representing the local dynamic behavior of the whole system.Then an observer-based fault diagnosis technique is developed to generate the residuals.A performance index is presented to quantitatively evaluate the robustness of the generated residuals.Genetic algorithm is then employed to optimize the performance index and find theoptimal gain matrices and weighting matrices for the detection and diagnosis observer.Finally, stability of the diagnostic system is discussed.Simulation tests on sensor fault dignosis in the superheated steam control system show its good performance and robustness to model uncertainties and disturbance in different operation points.There exist large numbers of process variables which are highly orrelated with each other.This special correlations provide another redundancy relationships in the system level.A data-driven method based on theory of multivariate statistics for fault diagnosis in thermal control systems is thoroughly studied.Firstly, the difference of operation mode between chemical and thermal plants analyzed.It is showed that conventional multivariate statistical methods appropriate for chemical processes cannot be directly applied to thermal control processes.Taking account the operational characteristics of the boiler process, this paper presents two improved methods to estab...