Approach Based on Machine Learning for the Sensitivity Assessment of an Aerodynamic Shape

This study presents an innovative approach to the aerodynamic optimization of wing-in-ground (WIG) effect vehicles, with a focus on enhancing operational efficiency and promoting environmental sustainability. A coupled framework was developed in which Reynolds-averaged Navier–Stokes (RANS) computational-fluid-dynamics simulations were linked to an adaptive neuro-fuzzy inference system (ANFIS) surrogate, after which a genetic algorithm (GA) was employed for design exploration. Within this framework, the four-dimensional (4D) design space, defined by camber, thickness, angle of attack, and height-to-chord ratio (h/c), was traversed efficiently, and response-surface methodology was applied so that in-depth parameter sensitivities could be quantified. It was found that camber and ground clearance accounted for 36% and 31% of the variance in the lift-to-drag ratio, whereas angle of attack and thickness contributed 22% and 11%, respectively. By means of GA optimization guided by the ANFIS surrogate, a lift-to-drag (L/D) improvement of 55.9% over the baseline configuration was achieved. Numerical validation through three-dimensional (3D) WIG simulations confirmed that modifications to camber and ground clearance altered the velocity field in a manner consistent with induced-drag reduction, thereby substantiating the predicted performance gains. Owing to these enhancements, fuel consumption and associated emissions would be lowered, aligning the design with emerging environmental standards. These findings were shown to highlight the paramount importance of camber, ground clearance, angle of attack, and thickness in the aerodynamic sensitivity assessment and optimization of WIG vehicles, thereby affirming the capacity of machine learning–assisted approaches to revolutionize the practice of aerodynamic design.