Deciphering the thermal-hydraulic synergy in wavy geometries through neural-computing and Pareto optimal frontiers

Abstract

An advanced Artificial Neural Network (ANN) model was developed to predict and optimize the thermalhydraulic performance of a wavy-tube heat exchanger. Based on experimental data, a multi-layer feed-forward network [2020] was trained using the Levenberg-Marquardt algorithm. With R2 values surpassing 0.977 for both parameters, the model exhibited high predictive precision, maintaining the Mean Absolute Percentage Error (MAPE) under 6.5% for Nu and f. The verified ANN framework was further employed as a reliable surrogate tool for multi-objective optimization. Benchmarked against the experimental Wilson Plot correlation, the optimization identified a theoretical global optimum at a Reynolds number of 8800 and a wavy geometry diameter of 4.99 mm. At this specific configuration, the Thermal Performance Factor (eta) reached a peak value of 3.072, demonstrating a significant enhancement over previously reported discrete experimental results. This confirms that the ANN successfully identified a 'hidden peak' of efficiency between the experimentally tested discrete diameters. Furthermore, a sensitivity analysis based on Garson's algorithm revealed that the wavy geometry diameter is the most influential parameter (36%) on performance, surpassing the influence of the flow arrangement and Reynolds number. Finally, a Pareto Frontier was established to provide a decision-making envelope for the trade-off between heat transfer enhancement and pressure drop penalty. The findings provide a robust computational framework and a ready-to-use mathematical matrix for the efficient design of nextgeneration wavy-tube heat exchangers.