Fault diagnosis and control of a thermal power plant.

摘要

The pressing need to improve efficiency, reliability and safety of power plants has resulted in constant effort to upgrade plant systems. The operation of a thermal power plant involves a large throughput of fuel and energy, thus even a small increase in efficiency can result in significant savings. The high pressures and temperatures involved in thermal power plant operation can result in significant damage to property and loss of life in the event of a system failure. The analog control systems which control a majority of the thermal power plants are constantly being replaced by more sophisticated digital distributed control systems. The introduction of modern digital distributed control systems in thermal power plants has facilitated the implementation of complex control and online fault diagnosis algorithms. The design of an online fault diagnosis system and a multivariable sliding mode control system for a thermal power plant is presented in this dissertation. The performance of the fault diagnosis system and the control system was tested by simulating faults in a 21{dollar}sp{lcub}st{rcub}{dollar} order physically based, lumped parameter model of a 235 MW gas fired thermal power plant. The model represents the steady state and transient characteristics of Clifford B. Jones Unit #2 power plant operated by Southwestern Public Services Company in Lubbock, Texas.; The fault diagnosis system is based on a neural network augmented observer and provides robust fault diagnosis even in presence of modeling errors and unmodeled dynamics. A novel technique using a sliding mode observer to characterize the modeling errors and facilitate the training of the neural network is presented. The accurate continuous time model of the system dynamics resulting from this approach is used for model based fault diagnosis. This model can also be used for system simulation and control. The online fault diagnosis system focuses on diagnosing process faults in the water/steam-side of the thermal power plant. The faults simulated in the power plant model were water/steam-side fouling, fire-side fouling, tube leaks, boiler feedpump failure, and a change in the calorific value of the fuel. The neural network augmented observer based fault diagnosis scheme was successful in diagnosing faults in the water/steam-side of the thermal power plant.; The existing multiloop PID control system for the water/steam-side of the power plant is difficult to tune and provides marginal robustness to system faults. The PID control system for the water-steam side of the thermal power plant consists of six loops, controlling the boiler drum level, main steam pressure, superheat temperature, reheat temperature (using spray and burner tilt), and megawatt output. The six PID loops were replaced by a multivariable sliding mode control system. The sliding mode control system was easier to tune and reduced the deviations of the process operating parameters from the setpoints. The inherent robustness of the sliding mode controller also provided improved stability margins and facilitated stable operation in the presence of system faults.