Author Identifier

Amir Razmjou: https://orcid.org/0000-0002-3554-5129

Document Type

Journal Article

Publication Title

EES Catalysis

Volume

2

Issue

2

First Page

612

Last Page

623

Publisher

Royal Society of Chemistry

School

Mineral Recovery Research Centre / School of Engineering

Funders

Ministry of Higher Education Malaysia (FRGS/1/2022/TK08/UM/02/43) / Australian Research Council / Southeast Asia-European Joint Funding Scheme (JFS21-123 HYPERMIS) / EPRO Adv Tech (IF044-2021) / UM Matching Grant (MG002-2022)

Grant Number

ARC Number : DP200102121

Comments

Haghshenas, Y., Wong, W. P., Gunawan, D., Khataee, A., Keyikoğlu, R., Razmjou, A., ... & Teoh, W. Y. (2024). Predicting the rates of photocatalytic hydrogen evolution over cocatalyst-deposited TiO 2 using machine learning with active photon flux as a unifying feature. EES Catalysis, 2(2), 612-623. https://doi.org/10.1039/d3ey00246b

Abstract

An accurate model for predicting TiO2 photocatalytic hydrogen evolution reaction (HER) rates is hereby presented. The model was constructed from a database of 971 entries extracted predominantly from the open literature. A key step that enabled high accuracy lies in the use of active photon flux (AcP, photons with energy equal to and greater than the bandgap energy of the photocatalyst) as the input feature describing the irradiation. The quantification of AcP, besides being a more direct feature describing the photocatalyst excitation, circumvents the use of lamp power ratings and light intensities as ambiguous inputs as they encompass varying degrees of AcP depending on the irradiation spectra. The AcP unifies four other key performing features (out of 46 initially screened), i.e., cocatalyst work functions, loadings of cocatalyst, alcohol type and concentrations, to afford a physically-intuitive model that can be generalized to a wide range of experimental conditions. The inclusion of AcP as an input to the machine learning model for HER prediction leads to a mean absolute error of 7 μmol h, which is a 90% reduction when compared to a model that does not use AcP. Verification of untested conditions with high HER rates, identified through Bayesian optimization, saw less than 9% deviation from the physically-measured kinetics, thus confirming the validity of the model.

DOI

10.1039/d3ey00246b

Creative Commons License

Creative Commons Attribution-Noncommercial 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 License

Share

 
COinS