Sustainable hydrogen production from NaBH4 using Co@CHE Catalyst: Experimental and MLP-Based modeling for autonomous fuel systems

Abstract

In this study, a catalytic hydrogen production process was investigated by integrating experimental results with Multi-Layer Perceptron (MLP) neural network models to address the energy needs of future autonomous systems powered by hydrogen. Sodium borohydride (SBH, NaBH₄) hydrolysis was conducted using a cobalt-based catalyst (Co@CHE) synthesized with coffee hydrochar extract, a biomass-derived waste material. While solution medium and temperature were kept constant, the catalyst dosage and NaBH₄ concentration were varied, yielding 844 experimental data points. Additional experiments were carried out to fill gaps in the dataset, which was then expanded to 23,355 data points using polynomial regression. The Co@CHE catalyst exhibited good activity, achieving hydrogen generation rate (HGR) of 65791 mL g⁻¹ min⁻¹ at 313 K. Reusability tests further confirmed its stability, showing 60 % retention of catalytic activity after six consecutive cycles, while maintaining 100 % hydrogen yield. Two MLP models were developed: one to predict the amount of hydrogen produced and the other to estimate reaction time. The first model used NaBH₄ amount, catalyst dosage, and reaction time as inputs, while the second used NaBH₄ amount, catalyst dosage, and hydrogen volume. Both models demonstrated excellent prediction performance, with R² values of 0.99376 for hydrogen yield and 0.99577 for reaction time, respectively. These results confirm that the proposed MLP models are highly effective for accurately modeling hydrogen production and can support the design of efficient, catalyst-based hydrogen systems.