Cognitive Chemical Manufacturing

(Credit: Dreamstime)

15 June 2020

Researchers based at the STFC Hartree® Centre worked with AstraZeneca – to optimise their chemical manufacturing processes, saving time and costs in large scale product production and manufacturing.

Challenge

Collaborating with The Hartree Centre and IBM has been a mutually beneficial undertaking, providing valuable insight through the incorporation of a Bayesian Optimisation solution

Brian Taylor
AstraZeneca

As a leading pharmaceutical company, AstraZeneca are always looking to meet the significant market demand for their products. To do so, they need to ensure they use the most efficient chemical manufacturing processes to optimise yield of the products’ constituent ingredients. However, this is a complex problem, as each component is composed of a unique combination of chemicals, and each unique combination has a specific set of optimal reactor parameters for maximising yield. Identifying parameters requires large numbers of experiments, with significant financial and resource implications.

Approach

This work has allowed both parties to gain a better understanding of the potential impact of Bayesian Optimisation within a chemical development setting

Brian Taylor
AstraZeneca

The team developed a solution that utilises Bayesian Optimisation to quickly find the most optimal set of reactor parameters in the fewest experiments. Bayesian Optimisation is a highly economical optimisation algorithm that uses Bayesian statistics to determine whether to ‘explore’ - search for novel sets of reactor parameters, or ‘exploit’ - refine a known set of parameters that perform well. By intelligently balancing exploration and exploitation, Bayesian Optimisation can select the optimal set of reactor parameters in far fewer experiments than traditional methods.

Benefits

Using Bayesian Optimisation allowed AstraZeneca to quickly discover optimal reactor parameters for their chemical manufacturing processes. This project - completed as part of the Innovation Return on Research (IROR) programme, a collaboration between STFC and IBM Research - has significant impact in that it reduces the amount of time and money invested to find the desired reactor configuration – reducing the costs to the company, and reducing the time required to scale the production process for commercial manufacturing.

At a glance

Used advanced statistical solution - Bayesian Optimisation - to identify optimal reactor parameters in the fewest experiments 
Accelerates production process for commercial manufacturing
Capable of balancing exploration and exploitation to achieve results economically compared to more traditional lab-based methods
Reduces amount of time and money invested to find desired configurations

Cognitive Chemical Manufacturing

Last updated: 15 June 2020