Real time prognostic strategies: Application to gas turbines .

摘要

Gas turbines are increasingly deployed throughout the world to provide electrical and mechanical power in consumer and industrial sectors. The efficiency of these complex multi-domain systems is dependant on the turbine's design, established operating envelope, environmental conditions, and maintenance schedule. A real-time health management strategy can enhance overall plant reliability through the continual monitoring of transient and steady-state system operations. The availability of sensory information for control system needs often allow diagnostic/prognostic algorithms to be executed in a parallel fashion which warn of impending system degradations. Specifically, prognostic strategies estimate the future plant behavior which leads to minimized maintenance costs through timely repairs, and hence, improved reliability. A health management system can incorporate prognostic algorithms to effectively interpret and determine the healthy working span of a gas turbine. The research project's objective is to develop real-time monitoring and prediction algorithms for simple cycle natural gas turbines to forecast short and long term system behavior.; Two real-time statistical and wavelet prognostic methods have been investigated to predict system operation. For the statistical approach, a multidimensional empirical description reveals dominant data trends and estimates future behavior. The wavelet approach uses second and fourth-order Daubechies wavelet coefficients to generate signal approximations that forecast future plant operation. To complement the empirical models, a real-time analytical, lumped parameter mathematical model has been developed that describes normal transient and steady-state gas turbine system operation. The model serves as the basis to understand a simple cycle gas turbine's operation, and may be utilized in model-based diagnostic algorithms.; To validate the model and the prognostic strategies, extensive data has been gathered for a 4.5 MW Solar Mercury 50 and a 85 MW General Electric 7EA simple cycle gas turbine. For the dynamic gas turbine model, the comparison between the field data and simulation results for five Mercury 50 gas turbine signals (e.g., shaft speed, power, fuel flow, turbine rotor inlet temperature, and compressor delivery pressure) demonstrate a high degree of correspondence. Although there are some deviations between the analytical and experimental results during the transient phase, the estimated steady state results are within 2.0% of the actual data. The direct comparison of the two forecasting methods revealed that the wavelet method is superior since the forecasting error is 2.4% versus 4.0% for the statistical method on the Mercury 50 simple cycle gas turbine steady-state signals (e.g., compressor delivery pressure and turbine rotor inlet temperature). Similarly, the General Electric 7EA steady-state signal (e.g., turbine inlet temperature) offered a forecasting error of 9.23% for the wavelet and 11.47% for the statistical methods, respectively. The developed approaches successfully estimate and predict the system operation and may be used with a diagnostic algorithm to monitor gas turbine system health.