Model Adaptation of a Mixed Flow Turbofan Engine

Gas turbine performance models are usually created in an object oriented manner, where different standard components are connected to form the complete model. The characteristics of these components are often represented by component maps and empirical correlations. However, engine specific component characteristics are seldom available to anyone outside of the manufacturers. It is therefore very common for researchers to use publicly accessible or generic component maps instead. But in order to reduce prediction errors the maps have to be modified to fit any specific engine. This thesis work investigates the process of adapting a parametric turbofan engine model to a limited amount of test-data using the propulsion program EVA. Steady state test-data was generated using an initial reference model with SLS operating conditions. Another engine model with different fan, compressor and turbine maps was then used in the adaptation. An initial on-design model was adapted to the highest power test-data point. This model is based on aerothermodynamic equations and is used as a reference to scale the generic component maps to. A sensitivity analysis was done at this point in order to find dependencies between unknown component parameters and test data. These were then included in the cycle solver which employs a version of the Newton-Raphson method. After the fan and compressor maps had been scaled to the design point they were adapted to test-data by adjusting the mass flow parameters in a direct search optimizer. Finally, speed lines in the fan and compressor maps were relabeled to reduce rotor speed errors. The adapted performance model was then validated against the reference model at a few flying conditions. The performance model results demonstrate that it is possible to greatly reduce prediction errors by only adjusting the corrected mass flow in fan and compressor maps. Additionally, rotor speed errors could successfully be corrected as a final step in the adaptation by relabeling speed lines in the component maps. When validated, the adapted model had a maximum parameter error of 1.5%