Machine learning-driven gerotor profile synthesis and optimization using Conditional Generative Adversarial Networks

The gerotor tooth profile plays a critical role in the performance of hydraulic systems, particularly in applications such as engine lubrication and automatic transmission. Traditional design methods often rely on predefined mathematical curves and iterative adjustments, limiting optimization flexibility. This study proposes a Conditional Generative Adversarial Network-based approach for automatic gerotor profile generation. The model was trained on a dataset consisting of high-performance profiles developed through analytical base curves, enabling the generation of new designs optimized to maximize flow rate and minimize irregularity. To facilitate practical implementation, a coordinate extraction process was used to convert generated images into usable Computer-Aided Design geometry. An area-based evaluation method was developed to assess flow rate and irregularity performance. Compared to results from Camus theory, the proposed method achieved high accuracy, with a signed error of +0.16% for flow rate and $-0.18$% for irregularity. Validation using Computational Fluid Dynamics (CFD) was performed with dynamic mesh modeling at 5000 RPM, confirming a 32.3% increase in average flow rate and a 74.7% reduction in flow irregularity compared to traditional ovoid profiles. Additionally, outlet pressure fluctuation was reduced by 53.6%, with a 10.1% improvement in average pressure. These results highlight CGAN’s effectiveness as a data-driven design tool for enhancing gerotor performance and suggest a promising direction for future multi-objective optimization in hydraulic components.