Towards a condition monitoring scheme for thermoacoustic instability detection and fuel blend performance classification using machine learning

The investigation and improvement in fuel performance and combustion is necessary in order to minimize emissions and operation costs in various engineering applications e.g. aerospace. Among these factors, nevertheless, ensuring safe operation is a priority: undesired phenomena, such as thermoacoustic instabilities, can have detrimental effects on jet engines, gas turbines and combustors, in general, due to excessive vibrations. It is for this reason that monitoring and design schemes should be able to identify the potential of occurrence of such events. This is a difficult task due to the complexity of the nature of these events. This paper is a preliminary investigation into the performance and characterization of various fuel blends and the examination of the vibration levels expected for different combustion states of a gas turbine engine. We tackle the issue from the perspective of modifying the input to the system (i.e. the fuel composition) in order to investigate nonlinear behavior of the gas turbine engine through the development of a multi-class classification algorithm. Features from a vibration channel for each of the fuel blends were extracted for both classification modelling and cluster analysis.