Machine Learning for Polymer Development


Machine learning is well established as a way through which complicated relationships, such as those present in polymerization systems, can be quantitatively established, but requires sufficient data for efficient training. Unfortunately, data is scarce in many physical sciences and especially so in the world of (dispersed) polymers where characterization of many of the most important features is time-consuming. The overarching scientific objective of this research line is to move away from exclusively data-driven machine learning approaches, which have little practical scope for many scientific applications, and demonstrate the potential of knowledge-guided machine learning in the world of polymer chemistry. In this kind of “chemistry informed” machine learning, the underlying science is embedded into machine learning models, and thus the abilities of conventional machine learning to assimilate data can be utilized whilst eliminating the need for prohibitively large datasets. Key topics of this program are the use of machine learning for reactor control, recipe design and design of polymer materials.

Coordinator: Nicholas Ballard