Machine learning approach to balance heat transfer and pressure loss in a dimpled tube: Generative adversarial networks in computational fluid dynamics

• GANs optimize dimple depth and pitch in heat pipes, balancing heat transfer and pressure loss. • Three distinct GANs-CFD datasets are created, each mixing GAN-generated and original points. • PEC shows deviations up to 15.49% compared to results from CFD simulations alone. • Average errors for dimple depth and pitch in GANs-CFD datasets are 6.1 % and ∼0 %, respectively. • GANs significantly reduce computational time and costs in optimizing heat pipe designs. In heat tubes, dimples enhance heat transfer but also introduce significant pressure loss. This study aims to optimize dimple depth and longitudinal pitch to balance heat transfer efficiency with pressure loss. An innovative machine learning approach using generative adversarial networks (GANs) is applied to identify optimal dimple configurations. GANs were trained on a foundational dataset derived from computational fluid dynamics (CFD) simulations. Initially, CFD simulations generated twenty sample data points to form the base dataset. By employing an ensemble learning method, GANs produced ten additional data points. Combining these with ten randomly chosen points from the original dataset resulted in a hybrid dataset called GANs-CFD, containing twenty points. This procedure was repeated three times to create three distinct GANs-CFD datasets. Performance evaluation criteria (PEC) optimization using these datasets showed that the results deviated by a maximum of 15.49 % from those obtained solely through CFD simulations. The average errors for optimized dimple depth and pitch across the three GANs-CFD datasets were 6.1 % and ∼0 %, respectively. These findings demonstrate the potential of GANs to significantly reduce computational time and cost in optimizing pipes for transporting heat in a fluid designs involving complex trade-offs between heat transfer and pressure loss.