A CFD-Informed, Experimentally Validated ANN Digital Twin for Charge and Discharge of an AB5 Metal Hydride Reactor

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Solid-state hydrogen storage in metal hydride reactors offers a safe, low-pressure alternative to conventional hydrogen storage, but the batch nature of the system and the prohibitive cost of solid-state hydrogen concentration sensors present a significant challenge for effective process control. This study presents an integrated modelling framework for an AB5 metal hydride reactor, progressing from a validated computational fluid dynamics model through operational sensitivity analysis to an artificial neural network digital twin encompassing both the charge and discharge states. An experimentally validated finite element porous flow and reaction model achieving regression coefficient values of 0.9 or higher was used to perform a sensitivity analysis and generate training data. The sensitivity analysis found a desorption threshold of around 90 °C for the heating fluid below which no significant desorption takes place. This threshold is absent from the charge state, which likely contributed to the differences in artificial neural network architecture performance asymmetry. This was then used to perform artificial neural network-based regression to generate digital twin models for the two respective operational states of the reactor. The best-performing models achieved a regression coefficient of 0.99 or higher and a mean squared error of 10⁻⁴ or lower. This paper serves as not only a report of the findings but also as an outline of the modelling framework implemented.

This publication has 39 references indexed in Scilit: