Polymers are everywhere in modern life, from cars to mobile phones, but their ubiquity comes at a steep cost. Only 9% of all polymers are being recycled, a figure that urgently needs to improve, says TU Delft Associate Professor Sid Kumar. He is leveraging machine learning to find and design polymers that are better recyclable, paving the way for a more sustainable future.
Most existing polymers are difficult to recycle due to their chemical structure. But what if we could tweak their structure to maintain essential properties-like strength and stiffness-while improving their recyclability? The problem, as Kumar explains, is the sheer number of possibilities.
-Polymers are very large molecules composed of repeating units. These repeating units can range anywhere from 10-100s of atoms. The uniqueness of arrangement determines the polymer properties. Even for a small repeating unit, with just 20 atoms, arranging them in different ways already creates 1018 possible combinations,- he says. Testing each option experimentally would take lifetimes, and even simulations cannot screen them all’efficiently.
The solution: machine learning
This is where artificial intelligence comes in. Kumar and his team use machine learning to identify polymers with desired properties from the ocean of possibilities. -AI helps us find the needle in a haystack,- he explains.
The properties of metals depend heavily on how they are processed-such as heating, cooling, and deformation. We are using machine learning to optimise processing conditions for desired properties, enhancing both the sustainability and recyclability of metals.
Sid Kumar
Their recent development is an advanced machine learning algorithm that requires only a small amount of input data to discover new polymers. Additionally, the algorithm is designed to be interpretable and explainable, ensuring collaboration between AI and human scientists.
-Molecules are unintelligible to computers. While we understand what bisphenol A diglycidyl ether is, a computer doesn-t-it doesn’t have a numerical value to assign to it,- Kumar says. -We-ve embedded interpretability into the AI, which not only increases its capacity to handle complexity but also improves trust, usability, and collaboration for researchers.-
Healable polymers for a circular economy
The team tested their algorithm on vitrimers-a new class of healable polymers. Vitrimers combine durability with end-of-life recyclability, offering a promising solution to plastic waste. These polymers can repair themselves when heated, thanks to their unique molecular bonds.
However, commercially available vitrimers are limited by the scarcity of suitable molecular building blocks, which restricts their self-healing properties and broader applications. Kumar-s team set a target temperature for self-healing and used their algorithm to identify promising molecular candidates. What once could have taken years-or remained impossible-was achieved in just days.
This study, conducted in collaboration with researchers from the University of Washington and Microsoft, has already shown practical applications. For instance, these vitrimers can be incorporated into recyclable circuit boards , demonstrating their potential in high-tech industries. By extending the lifecycle of plastic products and enabling on-site repairs, recyclable polymers could potentially reduce the 430 million tons of plastic produced annually and mitigate economic losses from replacing damaged parts.
A sustainable future for plastics and metals
Kumar-s machine learning approach is also being applied to metals. -The properties of metals depend heavily on how they are processed-such as heating, cooling, and deformation,- he says. -We are using machine learning to optimise processing conditions for desired properties, enhancing both the sustainability and recyclability of metals.- Keeping humans in the loop remains a core principle. -Factories need practical solutions that are interpretable for operators,- Kumar notes. -AI shouldn-t be a black box-it should empower human decision-making. And I think it has a huge potential to improve circularity of materials.-
