Performance analysis of a micro-scale integrated hydrogen production system by analytical approach, machine learning, and response surface methodology

Parametric study of micro-scale integrated hydrogen production systems requires great computational efforts due to complex phenomena such as reaction kinetics. In the present study, an innovative combined approach, including machine learning for data generation (pre-processing), analytical techniques (processing), and response surface methodology (post-processing) is developed to investigate an integrated hydrogen production system. In the pre-processing step, appropriate correlations are provided for the species' net rate, mixture properties, and the heat of reactions considering the detailed reaction mechanism of methane steam reforming and combustion, using the decision tree algorithm. A 2D steady-state model for heat and mass transfer is employed to analytically solve the conservation equations in a thermally coupled micro-combustor and catalytic micro-reformer. The post-processing step investigates the effects of seven main operational parameters on CH4 conversion, system efficiency, and quenching distance. It is found that the wall thickness is the most influential parameter in CH4 conversion and system efficiency. Also, the combustor height is the most critical parameter to sustain combustion in the integrated system. The achievements can be employed as guidelines for the initial design of an integrated hydrogen production system. Finally, five optimized designs of the integrated system are suggested for the first time to construct experimental prototypes. (c) 2022 Elsevier Ltd. All rights reserved.

Automated summary

This study presents an innovative combined approach, including machine learning for data generation, analytical techniques, and response surface methodology, to investigate micro-scale integrated hydrogen production systems. The study examines the effects of operational parameters on system performance and provides guidelines for initial design and optimized design suggestions.