Multi-objective Optimization of Staggered Tube Banks in Cross-flow Using Machine Learning and Genetic Algorithm

This paper presents the numerical multi-objective optimization of staggered tube banks in cross-flow using neural networks and genetic algorithm. The objective is to determine the optimal dimensionless transverse and longitudinal pitches that establish a proper compromise between heat transfer enhancement and pressure drop minimization across a wide range of inlet Reynolds numbers (۱,۰۰۰–۵۰,۰۰۰). Tube banks simulations are performed for randomly selected pairs of design points to generate data on Nusselt number and friction factor. This dataset is used to train neural networks, which predict heat transfer and pressure drop characteristics as functions of dimensionless pitches. Appropriate objective functions are defined using trained neural networks and integrated into Genetic Algorithm to efficiently identify Pareto-optimal solutions. Results indicate that Reynolds number has a negligible effect on the Pareto front, as the optimal trade-offs between heat transfer and pressure drop remain consistent across different flow regimes. The best point on the Pareto front, defined as the solution with the minimum distance to the utopia point, exhibits dimensionless longitudinal and transverse pitches of approximately ۰.۹۰ and ۱.۳۰, respectively, regardless of the Reynolds number. Additionally, the study confirms that compact tube banks with dimensionless longitudinal pitches smaller than ۱.۰, often excluded in experimental and numerical studies, can be successfully simulated and optimized using the proposed framework. The findings provide practical guidelines for designing high-efficiency staggered tube banks and demonstrate a computationally efficient approach to optimize heat exchanger configurations without relying on empirical correlations.

کلیدواژه ها:

Cross-flow staggered tube banks ، Multi-objective optimization ، Heat transfer ، Genetic Algorithm ، Machine learning