Fault Diagnosis of Combined Cycle Gas Turbine Components Using Feed Forward Neural Networks

Abstract

A diagnostic tool based on Feed Forward Neural Networks (FFNN) is proposed to detect the origin of performance degradation in a Combined Cycle Gas Turbine (CCGT) power plant. In such a plant, due the connection of the steam cycle to the gas turbine, any deterioration of gas turbine components affects not only the gas turbine itself but also the steam cycle. At the same time, fouling of the heat recovery boiler may cause the increase of the turbine back-pressure, reducing the gas turbine performance. Therefore, measurements taken from the steam cycle can be included in the fault variable set, used for detecting faults in the gas turbine. The interconnection of the two parts of the CCGT power plant is shown through the fingerprints of selected component fault models for a power plant composed of a heavy-duty gas turbine and a steam plant with a single pressure recovery boiler. The diagnostic tool is composed of two FFNN stages: the first network stage is addressed to pre-process fault data in order to evaluate the influence of the single fault variable on the single fault condition. The second FFNN stage detects the fault conditions. Tests with simulated data show that the the diagnostic tool is able to recognize single faults of both the gas turbine and the steam plant, with a high rate of success, in case of full fault intensity, even in presence of uncertainties in measurements. In case of partial fault intensity, faults concerning gas turbine components and the superheater, are well recognized, while false alarms occur for the other steam plant component faults, in presence of uncertainties in data. Finally, some combinations of faults, belonging either to the gas turbine or the steam plant, have been examined for testing the diagnostic tool on double fault detection. In this case, the network is applied twice. In the first step the amount of the fault parameters that originate the primary fault are estimated. In the second step, the diagnostic tool curtails the contribution of the main fault to the fault parameters, and the diagnostic process is reiterated. In the examined fault combinations, the diagnostic tool was able to detect at least one of the two faults in about 60% of the cases, even in presence of uncertainty in measurements and partial fault intensity.