Statistical Modeling of Aircraft Engine Fuel Flow Rate

Knowledge of the fuel flow rate is important for accurate fuel burn and emissions inventory generation and for accurate engine performance modeling. In this paper, machine learning techniques are applied to aircraft flight recorder data to model the fuel flow rate as a function of the aircraft altitude, ground speed, vertical speed, and takeoff mass in the airborne phases of flight. Models are built using the Classification and Regression Trees (CART) and the Least Squares Boosting algorithms. The models are found to be interpretable. They are also shown to predict fuel flow rate and its variability more accurately than other frequently-used methods for fuel flow rate estimation. A sensitivity analysis of the predicted fuel flow rate to takeoff mass is also conducted. The proposed nonparametric models can help set a benchmark for developing more sophisticated data-driven statistical models of engine fuel flow rate.