Share
Export Citation
Application of Machine Learning for Enhancing the Efficiency of Aircraft Model Testing in Wind Tunnels
Purwadi
International Review of Aerospace Engineering
Q3Abstract
Wind tunnel testing remains a fundamental approach for obtaining high fidelity aerodynamic data in aircraft design. However, it is often constrained by high operational cost and time-consuming experimental procedures. This challenge becomes more significant for small passenger aircraft, where extensive parameter sweeps across multiple control surface configurations generate large and highly structured datasets. In order to address these limitations, this study proposes a machine learning-based framework for efficient prediction and optimization of aerodynamic characteristics derived from wind tunnel experiments. The framework incorporates a leakage-free validation strategy tailored for structured datasets, including GroupKFold cross-validation and a prospective holdout scheme to evaluate rigorously model generalization. Particular emphasis is placed on the accurate prediction of aerodynamic coefficients, especially those associated with complex lateral directional dynamics. Several machine learning models, including Artificial Neural Networks (ANN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), are systematically compared and optimized by using a Genetic Algorithm (GA). The result demonstrates that the optimized Random Forest model achieves an average coefficient of determination (R2) of 84.6% under extrapolation scenarios, indicating strong predictive capability beyond trained conditions. Furthermore, the proposed framework shows the potential to reduce wind tunnel testing requirements by up to 49% significantly improving experimental efficiency without compromising accuracy. This study contributes a robust and practically applicable methodology for integrating machine learning into aerodynamic data analysis, bridging the gap between experimental testing and data-driven modeling. The findings highlight the transformative potential of adaptive, data-driven approaches in modern aerodynamic experimentation.