Synergistic ANN-GA-CFD framework for high-performance Savonius wind turbine optimization with experimental validation

Savonius wind turbine (SWT) optimization via machine learning and optimization techniques has attracted increasing attention; however, most existing studies rely on limited datasets that cover only specific geometric parameters or operating conditions. This limitation constrains the comprehensive exploration of the Savonius wind turbine design space. Therefore, the present study constructs a comprehensive multisource dataset covering the key geometric parameters and operating conditions of SWT. Accordingly, an iterative optimization framework integrating artificial neural networks (ANN), genetic algorithms (GA), and computational fluid dynamics (CFD) is developed. The performed CFD simulations are employed to enrich the dataset by filling critical data gaps. Consequently, two high-accuracy ANN surrogate models are established for straight and twisted SWTs, achieving correlation coefficients of up to 0.98. Accordingly, optimizing the developed models results in optimal designs with maximum power coefficients of 0.1856 and 0.1927 for straight and twisted SWTs, respectively. Employing the developed ANN models with Monte Carlo-based sensitivity analysis enables the quantification of influence percentage of each design parameter and operating condition on SWT performance. Furthermore, the optimal designs are fabricated and experimentally tested under different operating conditions. The experimental measurements show good agreement with the ANN model predictions, ensuring the accuracy of the developed ANN-GA-CFD framework.