Science

Artificial intelligence supports the creation of new bespoke plastics – summaries

In the tradition of naming historical – and prehistoric – periods based on the relationships between people and the materials around them, some call the 20th century the plastic age.

Today, the balance of visibility tends to be one of the problems caused by the increasing presence of polymers in human life as they began to replace materials that are less common, more expensive, heavier and more difficult to adapt to new uses. Much of the technological development over the past 100 years, however, has included plastics, rubber, and synthetic fibers that have transformed the automotive, textile, aerospace industries, as well as packaging – like the rightly infamous single-use bags and PET bottles – and equipment under other.

Natural polymers have been used for centuries, but synthetic versions did not emerge until the early decades of the 20th century. The pioneer Bakelite was patented in 1909. The birthmark of polymer science is in 1920 when the German Hermann Staudinger published an article revealing the formation of polymer chains. Staudinger received the Nobel Prize in Chemistry in 1953 for his work.

Polymer means that it (only) consists of many (poly) parts. They are macromolecules made up of long chains of atoms and smaller molecules called monomers. The different sizes of these chains, their spatial structure and the practically infinite chemical compositions also lead to numerous possible properties.

This large number of possible combinations between different elements, in chains of different sizes and in different sequences of atoms poses challenges for the use of artificial intelligence in the search for new polymers in so-called rational material design. Artificial intelligence, and machine learning in particular, have proven to be powerful tools for predicting properties and thus for developing new materials more efficiently, faster and cheaper than the traditional method of trial and error, which is largely based on this process. Important results have been obtained for older materials such as metal and ceramic alloys, but the variety and complexity of polymers associated with the relative youth of the field create additional difficulties.

Research carried out at the University of Chicago and published in Science Advances in late 2020 has brought us closer to this possibility of using algorithms to determine which combination of monomers leads to the polymer with the desired properties for a particular application – such as lightness and durability for new aerospace vehicles – and also for materials with properties that reduce their impact on the environment, such as: B. biodegradability.

The use of artificial intelligence in the field of materials begins with large databases in order to search for relationships between composition, structure and other attributes as well as the properties of different materials. In the traditional approach, materials are synthesized and then analyzed to characterize their properties and assess their suitability for the intended use. With artificial intelligence, it is expected that it will be possible to inform the desired properties and, in response, get some sort of recipe for the most promising materials.

However, the lack of empirical data and the quality of this data have a significant impact on this development. Another problem with polymers has been the number of recordings required to train a neural network (the tool used in this case) with known molecules until the properties of new materials can be predicted.

The University of Chicago group combined artificial intelligence, modeling, and simulation to train a neural network made from just 2,000 hypothetical polymers that were computationally engineered to test the tool. Previously it was thought that up to millions of polymer chains could be required to achieve this result.

The trained network was able to accurately predict the properties associated with different polymer chains and above all showed that this is a possible and promising way for empirically obtained polymer data sets. The expectation is therefore that from now on we will see advances in the use of plastics and other polymer materials, which are indispensable for major challenges such as the energy transition among many others.

Back to top button