Poly(ML): Machine Learning for Improved Sustainable Polymers

The environmental impacts of polymer waste are well-publicised, and there is an increasing demand from policymakers and the public to reduce polymer use and improve recycling or degradability. In addition, there is an opportunity to reduce environmental impact by producing the polymers from bio-derived inputs, which are often themselves waste products from other processes. Whilst the polymer industry is fully engaged with these topics, there is still significant basic research to be done: to produce bio-derived polymers with excellent engineering properties, and to understand the different opportunities
presented by the very wide range of possible input materials.

It is also well known that in many fields, machine learning has played a vital and important role in making better use of very large data sets resulting from complex physical processes; examples include use of data in healthcare and weather prediction. Importantly, machine learning enables us to build on our current best understanding of the subject whilst efficiently exploring new possibilities, for example in synthesis or processing of polymers, in an efficient and well-guided manner. This will potentially allow us to make advances in polymer design much more quickly than can currently be done
through processes of trial and error.

Unfortunately, the application of machine learning to polymers is challenging because of their very complex structures and behaviours. In this project, we will bring together experts in polymer synthesis, polymer engineering and machine learning in order to solve this challenge. Specifically, we will be able to produce polymer systems in which we can carefully vary and measure the relevant properties, such as stiffness or strength, in order to produce high quality training data from which the machines can learn. We will then use the machine learning to show us how to optimise the polymer properties.

During this project, we will produce open-source computational tools which will be made available on the internet. Industrial polymer producers will be able to use these toolboxes to develop new polymers, and will also be able to expand
and adapt them for future needs. The toolboxes will also support 'distributed manufacturing', allowing small-scale manufacturers worldwide to obtain locally produced polymers designed to have properties that meet their needs.
Combined with new production methods such as 3D printing, this will help to deliver a low-waste, localised 'circulareconomy',meeting specific local manufacturing needs. Hence, this project will play a key role in global polymer development over a very wide range of economic scales.

The key societal benefit of this project will be the production of high performance bio- and waste-derived polymers with bespoke properties. This will greatly help to reduce the environmental impact of polymer production and use, and help
move closer to a net zero-waste economy. Further, the production of polymers with bespoke properties will improve manufacturability, allow more effective application, and improve end of life recycling, by enabling optimisation of the
polymers for these processes. We will ensure these benefits by working with industrial users, the academic community, the public and policymakers.

More specifically, we will work with industrial partners to ensure that the machine learning algorithms developed are usable by the global polymer industry. This will be achieved through the production and exploitation of a flexible open-source machine learning platform, whose capabilities can be further enhanced with industry in the future. Whilst developing the machine learning capability, we will use monomers that are or could be bio-derived, or produced from waste products, and in which it is feasible to enable recycling and degradation by introducing functionalisation into the backbone of the polymer.

At all times, we will ensure that that the materials are well defined, with predictable compositions, so that they can be delivered at scale reproducibly and the machine learning has meaning and use for industry. We will also work carefully on generation and storage of characterisation data to ensure that this is compatible with industrial processes, lack of suitable data being one of the key impediments to the use of machine learning for polymer synthesis in the past.

The outcomes of the project will also be of benefit to the global distributed circular economy. Here, we will work with distributed manufacturing and testing networks, including the Fab-Lab network. We will make our open-source platform
available so that local manufacturers can add to it, and use it to support their design, synthesis and manufacturing needs, again through the choice and optimisation of well-chosen bio- and waste-derived polymer systems with appropriate
properties.

We anticipate significant academic outputs from the project. This will be an exemplar project for the use of machine learning in materials development, with wider applications outside polymer synthesis. We will use, for the first time in this context, a combination of machine learning techniques that will allow us to incorporate current and future physically based models but also to look for additional higher order relationships in the data, giving insights that will allow development of other models in the future. The machine learning will also allow more targeted experimental design, allowing more rapid model development. These opportunities will be exploited through research collaborations with globally leading polymer scientists.

Finally, we will build on the high profile of these subjects in the public perception. Through media, science fairs and outreach events, we will show how the outcomes of this research can allow us to continue to benefit from the use of
polymers in many applications to which they are ideally suited, whilst at the same time overcoming concerns about the long-term effects of this use.