Chemical reaction yield depends on several factors, ranging from catalyst materials selection and preparation method, surface science and reaction conditions. The complex interplay between these factors leads to laborious optimisation of the reaction parameters requiring expensive trial runs. For certain industrially-important reaction classes, many catalysts have been tried and documented over the years; these problems are well-suited for data-driven approaches.
Using the water-gas-shift reaction as a used-case, we develop an AI platform for reaction optimisation and catalyst discovery. We incorporate fundamental physics- and chemistry-based features to construct rationalisable and reproducible AI models. The AI platform (Fig 1) includes automated machine-learning AutoML and Intuitive Graphical User Interface (GUI) for usability and ML workflow templates for reusability, allowing broader user adoption of AI tools in materials discovery.
IHPC envisions the scalable platform to serve as a bedrock for future lab digitalisation effort with public or private research laboratories where researchers progressively contribute to an expanding dataset, create new ML and optimisation templates for sustainable chemical process development with speed, efficiency and lower costs.
Fig 1. AI Platform for Reaction Optimisation