ACDP is built upon the Accelerated Materials Development for Manufacturing (AMDM) research program to apply the concept of high throughput experimentation and automated machine learning optimization to accelerating catalyst development. It is a joined effort between IMRE, ICES and IHPC. Industrial catalysts, such as for the Fischer-Tropsch process that converts syngas to hydrocarbons, are typically optimized over a period of decades involving intensive trial and error process. ACDP aims to accelerated this process by a factor of 10X or more, with a focus on catalysts for CO2 conversion with the aim of mitigating high carbon emissions.
Capabilities
We have built a high-throughput syringe pump flow setup that is capable of studying catalytic reactions in flow, with automated in-line Raman diagnostic tool for monitoring of reaction products. An in-line benchtop NMR system for more precise reaction monitoring will also be added in the near future. Reactions will be optimized in an automated in-the-loop fashion using machine learning optimization algorithms via a combination of exploitation and exploration strategies. Further, we are building flow electrochemical cells for CO2 conversion that will be integrated with electrochemical potentiostats and the flow reactor setup, for discovery and optimization of CO2 conversion catalysts.
Achievements
A key focus of ACDP is to make use of machine learning to accelerate catalyst development. We have developed detailed machine learning workflows that encompasses data feature engineering and dimensionality reduction, and hyperparameter tuning of machine learning algorithms. In addition, we have established best practices for machine learning optimization, such as optimized surrogate functions for Bayesian Optimization and using Particle Swarm Optimization to converge quickly to the global maximum.
We have been working with an MNC to deploy machine learning into their catalyst optimization workflow, with the aim of discovering new catalysts with higher selectivity for commercial deployment.