Ensemble Machine Learning Methods for Unsteady Aerodynamics Modeling from Flight Data

In this paper, two ensemble machine learning-based methods, Bagging and Boosting, are applied to model the unsteady aerodynamics of an aircraft from flight test data. Bagging generates a predictive model by combining multiple decision trees that have been trained using the bootstrap on different input-output datasets. Boosting generates a predictive model by growing trees sequentially based on previously grown trees, with each decision tree fitting on a modified version of the data. The effectiveness of these two data-driven methods is investigated and validated by estimating the standard research aircraft's force and moment coefficients. The proposed methods' estimated results are statistically analysed and found to be highly correlated with measured data and to have a significantly lower root mean squared error (RMSE). Furthermore, these estimated aerodynamic force and moment coefficients are compared to the estimated coefficient model from the most commonly used maximum likelihood estimation method (MLE). The estimated results were found to be on par with the MLE predicted aerodynamic models. Moreover, Bagging and Boosting-based methods do not require the solution of the equation of motion, which is advantageous for generalised nonlinear modelling applications such as load estimation, aeroelasticity, and fault diagnosis, detection, and identification.