Health Assessment of Gas Turbine Compressor Using Process History Based Modelling Approach

Gas turbine based combined cycle power plants are found to play vital role in electric power generation in recent years. Modern gas turbines are integral elements in these plants operating at very high temperature with high efficiency. Improvements in plant reliability, availability and maintainability (RAM) have been major areas of concern for power producers to ensure competitive positions. In gas turbine based power generation systems, the performance of the turbine drops due to several reasons like compressor fouling and inlet filter clogging. For improving plant RAM, advanced methods of health monitoring are vital for gas turbine plant components such as inlet air filter, compressor, combustion chamber and turbine. This paper focuses on health monitoring of gas turbine compressor considering major fault condition of compressor fouling. The health monitoring is achieved by developing a compressor model, to predict the performance of the compressor at design and off-design operational conditions. A thermodynamic model of the gas turbine system has limited applicability for health monitoring applications. The modelling framework has to incorporate the complex assembly of various components that make up the overall system and the real time off-design operations of the system. With the recent developments in computational methods and availability of vast computing power, process history data models are found to be convenient options for the system modelling. Among the process history data based methods, Artificial Neural Networks (ANN) have proved to be effective for modelling non-linear and complex processes. Hence, ANN is used as the modelling platform for this study and the model is developed from process data of a GE frame 9E machine. Residuals generated from the model are used for analysing the health of the system. The prediction of future events achieved through the model is found to provide vital information for the decision making and planning of maintenance actions. Principal Component Analysis (PCA) is a suitable method that is efficient in accounting the variability of the data. It derives the loading vectors and is suitable for improving the effectiveness of the ANN model. Different fault conditions relevant to the gas turbine compressor are demonstrated with actual plant data using the ANN based health monitoring system. Effect of compressor fouling and recouping of fouling effect with off line compressor water wash are also analysed in this paper.