Towards a circular economy: designing biopolymers with machine learning

Originally published by TNO. Source: Richting een circulaire economie: biopolymeren door machine learning, tno.nl. This article is an English adaptation of the original Dutch text. At the time of writing, polySCOUT was a TNO programme with a planned spin-off.

Imagine 2050: a sustainable chemical industry in which waste is a raw material, and biomass and CO₂ fill the remaining gaps. The question for the coming years is which materials we will use to get there. Do we stay with plastics from fossil sources, or do we make use of what materials from biomass can offer?

Unlocking sustainable materials

There is no silver bullet. For some applications, today's plastics remain the best choice, and sources such as biomass and CO₂ still require efficient production methods. For other applications, new materials outperform existing ones, and the task is to scale up the technology and make it more widely available on the market.

The question is how to find these better-performing materials. Today's portfolio was built over more than a hundred years, largely through empirical research into structures, properties, applications and processes. There is no time for a similar path with CO₂-based and biobased polymers: 2050 is only one investment cycle away.

Data science meets materials science

Thanks to a huge increase in computing power, new machine learning algorithms have been developed over the past ten years, including for materials science. Machine learning has great potential for predicting the properties of new polymers and for designing polymers that meet desired properties.

Promise and challenges of data-driven materials innovation

The promise of machine learning is a faster and cheaper design process. It allows design to focus on functionality, sustainability and what happens to a material at the end of its life, such as recyclability or biodegradation, all at the same time. To deliver on this promise, several challenges must be addressed together:

• Accurate models need large amounts of data, while data on polymers and their properties are scarce and fragmented.
• Accurate models need algorithms that represent polymers adequately. Current algorithms use simplified representations (so-called fingerprints) that leave out essential polymer characteristics, the role of additives and the influence of processing history.
• New materials must be designed to be safe and sustainable, with the right end-of-life options, whereas most current machine learning models design for functionality only.

The polySCOUT programme

In 2022, TNO started the polySCOUT programme to develop sustainable polymers using machine learning.

The vision: to make the power of data-driven materials innovation accessible. By bringing together knowledge from different disciplines, polySCOUT aims to change how new polymer materials are designed and made.

The approach: research groups combine their expertise in data and materials science to build predictive models. polySCOUT captures the real-world complexity of polymers. Advanced analytical techniques and a carefully curated database enable accurate modelling of complex (co)polymer systems, addressing design questions that go beyond technical performance.

The results: polySCOUT does not stop at theory. During the programme it already delivered improved models and new materials, and it enabled collaboration with external partners. Experimentally validated designs point the way to safe and sustainable materials that meet industry needs.

The design approach looks beyond technical performance, taking into account CO₂ impact, toxicity and safety, sustainability, cost-effectiveness and end-of-life.

From innovation to impact

TNO aims to turn this data-driven innovation into impact through a spin-off. TNO Ventures supports technologies like polySCOUT on their way to market, including access to TNO's expertise, technology and facilities.

Example of collaboration: BIOTTEK

In the JTF project BIOTTEK, Senbis aims to develop a biodegradable polyester for use as a textile fibre. This could (partly) remove an important source of microplastics in seas and oceans. In this project, TNO is further developing the machine learning algorithms and models to support Senbis and RUG/NHL-Stenden in their polymer development.

Get involved

We invite industry to help develop and validate this approach for the application areas that matter to them. In this way, you gain experimentally validated candidates for new polymers that fit your needs.

Source: TNO, "Richting een circulaire economie: biopolymeren door machine learning", tno.nl. Adapted and translated with attribution.