Turbofan Engine Behaviour Forecasting using Flight Data and Machine Learning Methods

• Autor(a) principal: Silva, Fernanda Cavalcante da
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.6/12666
• topic: Condition Monitoring; Forecasting; Machine Learning; Manutenção Preditiva; Motor Turbofan; Neural Networks; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Aeronáutica

The modern gas turbine engine widely used for aircraft propulsion is a complex integrated system which undergoes deterioration during operation due to the degradation of its gas path components. This dissertation outlines the importance of Engine Condition Monitoring (ECM) for a more efficient maintenance planning. Different ML approaches are compared with the application of predicting engine behaviour aiming at finding the optimal time for engine removal. The selected models were OLS, ARIMA, NeuralProphet, and Cond-LSTM. Long operating and maintenance history of two mature CF6-80C2 turbofan engines were used for the analysis, which allowed for the identification of the impact of different factors on engine performance. These factors were also considered when training the ML models, which resulted in models capable of performing prediction under specified operation and flight conditions. The Machine Learning (ML) models provided forecasting of the Exhaust Gas Temperature (EGT) parameter at take-off phase. Cond-LSTM is shown to be a reliable tool for forecasting engine EGT with a Mean Absolute Error (MAE) of 7.64?, allowing for gradual performance deterioration under specific operation type. In addition, forecasting engine performance parameters has shown to be useful for identifying the optimal time for performing important maintenance action, such as engine gas path cleaning. This thesis has shown that engine removal forecast can be more precise by using sophisticated trend monitoring and advanced ML methods.