Predicting hydrogen production in porous foams for steam methane reforming: A combined approach using computational fluid dynamics and machine learning regression models

Abstract

This study investigates the steam methane reforming (SMR) process within catalytic porous media through numerical simulations, emphasizing the influence of structural parameters in open-cell foams. The foams were created utilizing the Laguerre-Voronoi tessellation technique, which was chosen for its ability to yield high specific surface area and superior heat transfer capabilities. The key parameters studied include inlet velocity (0.5–4 m/s), foam porosity (between 0.7 and 0.9), and pore diameter (1.65–2 mm). To explore the interactions between reactive flow and heat transfer, computational fluid dynamics (CFD) simulations were employed. The results revealed that the structural properties of the catalytic foams significantly influence fluid-solid interactions, pressure drop, heat transfer efficiency, and hydrogen production. To address the computational challenges associated with complex foam geometries, four machine learning (ML) regression models were introduced to create continuous predictive frameworks based on intrinsic foam properties. Among these, Ordinary Least Squares (OLS) emerged as a reliable model for estimating hydrogen production across various foam configurations, achieving an impressive accuracy of 99 % on both training and test datasets. Additionally, the increase in foam length in samples with larger pore diameters and higher porosity, which typically shows reduced methane conversion, lower hydrogen output, and decreased pressure drops, demonstrated potential performance improvements. This study's pore-scale analysis offers a nuanced understanding of the interplay between foam structure and SMR efficiency, emphasizing the transformative role of porous catalytic media in advancing clean energy technologies.