Development of Machine Learning based Inverse Design for 2D Aerofoil

This research exemplifies initial stages of the development of an inverse design process for an ultra-large horizontal axis wind turbine (HAWT) blade. The full-scale HAWT blades are complex as twist angles and chord lengths at different radial sections are varied along the length of the blade. Therefore, the design procedure has been reorganized into a modular evolution. The present stage focuses on the development of 2D simulation data that is subsequently applied to train predictive models of inverse design. Simple 2D cases of NACA 2412 aerofoil were considered for controlled validation, data generation and refinement of the Machine Learning (ML) framework. A step-by-step strategy has been adopted to ensure detailed design complexities are considered with proven data and design principles to form trusted baselines for 3D analysis leading to the development of future complex aerofoil designs. In this research, the generated data from 2D simulations was used to train the Levenberg-Marquardt (ML) learning algorithm model to develop a baseline inverse design methodology to estimate geometric parameters of the aerofoil from desired lift (Cl) to drag (Cd) coefficient ratios. The prediction level of the ML algorithm showed 0.998 of R2 followed by minimal errors of the ratio being 0.00955 at highest frequency of occurrences and mean squared error of 0.1197 that confirms the modelling approach while establishes trust for its future scalability.