Deep penetration of renewable energy generation has led to increased periods of operation of industrial gas turbines under part load conditions. The performance fault diagnosis of gas turbine module components such as compressor, turbine, and the combustion chamber is a difficult task for such non-design operating conditions. Hence operational data-based performance health monitoring system is a requirement of gas turbine owners or users. The system must be capable to identify the degradation of gas turbine components, having substantial impact on the performance of the module in any possible operating condition. On time identification of degraded components will reduce the cost of operation, ensure service availability and obtain maximum performance from the turbine. This paper illustrates the application of advanced machine learning techniques to the performance analysis, fault diagnosis and prediction of future performance of gas turbine compressor. Compressor fouling is a primary cause of gas turbine performance deterioration, which accounts for 70% to 85% of the performance loss. The first section of the paper focuses on the residual generation of critical parameters of the compressor. The residual of the critical parameters can be calculated by comparing compressor model output with actual plant parameters. The trained artificial neural network (ANN) classifier uses residual of critical parameters to identify the fast rate of compressor fouling in the early stage. Statistical analysis can estimate the future performance of the compressor from the residual of critical parameters. The predicted values of compressor performance are useful for the planning of offline compressor wash, which in turn improve performance, reliability, and availability of gas turbine module. The residuals can also measure the effectiveness of compressor wash and assess the performance after machine overhauling.

This content is only available via PDF.
You do not currently have access to this content.