A Systematic Approach to Sensor Selection for Gas Turbine Exhaust Gas Temperature Prediction

Abstract The performance of a gas turbine deteriorates over time due to various primary flow component degradations such as fouling, erosion, corrosion, and foreign object damage. Exhaust gas temperature, commonly used for engine control, condition monitoring, fault detection, and maintenance decisions, effectively indicates the severity of engine performance degradation. Although EGT can be directly measured by placing a few probes at the exit of the turbine, EGT sensors themselves are subject to frequent failures, providing a fairly inaccurate indication of the gas turbine hot-section status. This paper proposes an active-subspace-based approach to sensor selection for EGT prediction based on active subspace approach. Instead of selecting a subset of parameters, the active subspace approach identifies a few primary directions along with EGT changes fastest. The method’s effectiveness is examined using the Advanced Geared Turbofan (AGTF30) engine, representing a 30,000 lbf thrust-class turbofan engine for civil aviation. All the numerical simulations are performed using the open-source Toolbox for the Modeling and Analysis of Thermodynamic Systems (T-MATS) covering the operating conditions, including takeoff, climb, cruise, descent, and landing. Results show that the proposed method can effectively search for an optimal sensor suite for EGT prediction. The optimal suites yield reasonably good EGT predictions with majority predictions of 95% confidence interval and are less sensitive to input measurement uncertainties. Furthermore, intelligent selection of sensor suites is crucial for EGT prediction, and the inclusion of additional sensor counts can sometimes undermine the precision of the prediction.