Leveraging genetic programming to analyze heat transfer and pressure drop characteristics in open-cell metal foams

A numerical analysis of pressure drops and heat transfer characteristics in open-cell metal foams is presented, and predictive correlations relating thermo-fluid-dynamic performance to foam topology and flow conditions are proposed. The new possibilities offered by additive manufacturing in terms of cell geometry customization have in fact expanded the design parameter space for metal foams. Consequently, surrogate models able to account for specific geometrical features are required. For this purpose, a large dataset of cells was generated from a broad variety of configurations, resulting in a total of 540 simulations. The numerical simulations were conducted in laminar flow conditions, using air as the heat transfer fluid. The flow patterns within the cells were analyzed, to uncover the mechanisms driving the observed heat transfer and pressure drop behavior. From this extensive dataset, predictive correlations were derived for key performance indicators, including the friction factor, Nusselt number, and area goodness factor. Correlations were developed through symbolic regression, implemented using genetic programming. Unlike conventional regression approaches based on predefined functional forms, this approach derives thermo-hydraulic correlations directly from the data. The proposed correlations were then assessed against selected experimental and numerical datasets from the literature, including porous structures and cell geometries not represented in the regression dataset. The proposed protocol is intended to provide both a better understanding of how geometric and fluid dynamics parameters influence the performance of 3D printable metal foams, and how Machine Learning techniques can be helpful towards design of improved heat exchangers.