Experimental investigation, development of machine learning model and optimization studies of a metal hydride reactor with embedded helical cooling tube

Abstract

Metal hydride hydrogen storage systems are gaining popularity due to their advantages and suitability for various applications in variety of fields. In this study, a metal hydride reactor using LaNi5 as metal hydride and using water as heat transfer fluid flowing through embedded helical tube was fabricated and experimentally studied. The aim of this study is to experimentally investigate the effect of internal heat transfer arrangement on the absorption and desorption of hydrogen in a metal hydride system. The effect of mass flow rate and temperature of heat transfer fluid, effect of hydrogen supply pressure on the rate of hydrogen absorption and desorption was investigated. It was found that the increase in the hydrogen supply pressure, increase in mass flow rate and decrease in the temperature of heat transfer fluid, results in faster hydrogen absorption rate. Similarly, the increase in mass flow rate and the temperature of heat transfer fluid results in faster hydrogen desorption rate. It was also found out that maximum hydrogen absorbed inside metal hydride reactor was around 34.5 g which corresponds to a gravimetric capacity of 1.34 wt%. Further based on the experimental data various machine learning models were developed and tested and then these models were used for performance prediction. It was found that tree-based regression machine learning models are better compared to other models. Such models can thus be used for prediction of rate of hydrogen absorbed or desorbed and metal hydride bed temperature for different parametric values without doing detailed experimental and numerical trial. Finally, the brute force approach is used to optimize the system for three different applications. The optimization techniques help to fix the value of parameters for particular application and hence save multiple experimental run of absorption and desorption. This is a novel work in the field of using machine learning for modelling and optimizing the metal hydride-based reactor. Use of such tools can save lot of computational time, efforts and experimental resources.