Diagnostics of power setting sensor fault of gas turbine engines using genetic algorithm

Gas path diagnostics is one of the most effective condition monitoring techniques in supporting condition-based maintenance of gas turbines and improving availability and reducing maintenance costs of the engines. The techniques can be applied to the health monitoring of different gas path components and also gas path measurement sensors. One of the most important measurement sensors is that for the engine control, also called the power setting sensor, which is used by the engine control system to control the operation of gas turbine engines. In most of the published research so far, it is rarely mentioned that faults in such sensors have been tackled in either engine control or condition monitoring. The reality is that if such a sensor degrades and has a noticeable bias, it will result in a shift in engine operating condition and misleading diagnostic results. In this paper, the phenomenon of a power-setting sensor fault has been discussed and a gas path diagnostic method based on a Genetic Algorithm (GA) has been proposed for the detection of power-setting sensor fault with and without the existence of engine component degradation and other gas path sensor faults. The developed method has been applied to the diagnostic analysis of a model aero turbofan engine in several case studies. The results show that the GA-based diagnostic method is able to detect and quantify the power-setting sensor fault effectively with the existence of single engine component degradation and single gas path sensor fault. An exceptional situation is that the power-setting sensor fault may not be distinguished from a component fault if both faults have the same fault signature. In addition, the measurement noise has small impact on prediction accuracy. As the GA-based method is computationally slow, it is only recommended for off-line applications. The introduced GA-based diagnostic method is generic so it can be applied to different gas turbine engines. This paper will be presented at the ISABE 2017 Conference, 5-8 September 2017, Manchester, UK