A mini review on the applications of artificial intelligence (AI) in surface chemistry and catalysis

Abstract

This review critically analyzes the incorporation of artificial intelligence (AI) in surface chemistry and catalysis to emphasize the revolutionary impact of AI techniques in this field. The current review examines various studies that using AI techniques, including machine learning (ML), deep learning (DL), and neural networks (NNs), in surface chemistry and catalysis. It reviews the literature on the application of AI models in predicting adsorption behaviours, analyzing spectroscopic data, and improving catalyst screening processes. It combines both theoretical and empirical studies to provide a comprehensive synthesis of the findings. It demonstrates that AI applications have made remarkable progress in predicting the properties of nanostructured catalysts, discovering new materials for energy conversion, and developing efficient bimetallic catalysts for CO2 reduction. AI-based analyses, particularly using advanced NNs, have provided significant insights into the mechanisms and dynamics of catalytic reactions. It will be shown that AI plays a crucial role in surface chemistry and catalysis by significantly accelerating discovery and enhancing process optimization, resulting in enhanced efficiency and selectivity. This mini-review highlights the challenges of data quality, model interpretability, scalability, and ethical, and environmental concerns in AI-driven research. It highlights the importance of continued methodological advancements and responsible implementation of artificial intelligence in catalysis research.

Intelligent chemometric modelling of Al2O3 supported mixed metal oxide catalysts for oxidative dehydrogenation of n-butane using simple features

“…1 gives an illustration of the workflow for developing supervised learning models to predict catalyst performance using simple experimentally obtained features. This workflow, commonly employed by researchers, 39,45,46 underpins the approach taken in this work. The following sections provide more details about each stage of the process.…”

“…20 These characteristics make unsupervised learning techniques more valuable in feature selection and dimensionality reduction applications compared to catalyst performance prediction. [37][38][39] Supervised learning is generally preferred for performance prediction due to its ability to learn from labelled data.…”