Optimizing Heat Exchanger Design with Machine Learning and Genetic Algorithms

This research investigates the application of machine learning, specifically neural networks (NNs), and genetic algorithms (GAs) for multi-objective optimization of heat exchangers. Focusing on the tube fin heat exchanger (TFHE), the study utilizes inlet air velocity and tube ellipticity as optimization variables. To achieve optimal heat transfer and pressure drop performance, Computational Fluid Dynamics (CFD) simulations were conducted for various Reynolds numbers (150-750) and tube ellipticities (0.2-1). These simulation data were used to train back-propagation NNs, resulting in prediction models for heat transfer coefficient and pressure drop. The non-dominated sorting genetic algorithm II (NSGA-II) with elitist retention strategy was employed to optimize the prediction results of the NNs. The optimization outcomes are presented as a Pareto front, highlighting the trade-off between heat transfer coefficient and pressure drop. The study found that with a Reynolds number of 541 and an ellipticity of 0.34, the TFHE experienced a 20% reduction in pressure drop with minimal impact on the heat transfer coefficient, leading to a j/f value 1.28 times higher than the original design. In conclusion, this research demonstrates the effectiveness of combining machine learning and genetic algorithms for multi-objective optimization of heat exchangers. This approach offers a more efficient and accurate method for determining optimal design parameters, highlighting the importance of inlet air velocity and tube ellipticity in the optimization process. The optimized TFHE achieves a substantial reduction in pressure drop while maintaining high heat transfer performance.