Similarity Based Modeling for Turbine Exit Temperature Spread Monitoring on Gas Turbines

Abstract

Failures in the gas path of a Gas Turbine will cause a deviation in the measured performance parameters. One of the most important parameters is the Turbine Exit Temperature (TET) and refers to the hot gas temperature at the exhaust of a Gas Turbine (GT). However, TET is not uniform at the turbine outlet and the temperature is therefore sometimes measured at several axial and radial positions. The TET has what can be considered a natural variation, an effect of operation in different ambient and operational conditions which influences the internal flow field. It can be informative on the health status of the GT by monitoring the TET variation during operation, as a number of failures or abnormal operation conditions will affect the TET distribution. A regular way of monitoring the TET is to use the average value from different sensor readings, or compare the highest deviating sensor to the average value of all sensors. However in order to detect anomalies as early as possible deviations from the healthy profile should be detected more finely across the section. In this paper, a data-driven similarity based algorithm called Auto Associative Kernel Regression is applied to the issue of monitoring the TET spread variation on an industrial gas turbine. A case study is supplied to show the practical usefulness of the algorithm to a field failure.