Intellegens Blog – Stephen Warde, May 2022
Discussing applied machine learning for chemicals, materials and manufacturing – see all blog posts.
What does ‘The Age of Plastic’ mean to you? Perhaps the 1980 Album from the British new wave group, The Buggles? More likely, for scientists and engineers, it’s recognisable as a term used to describe the remarkable rise of this class of material since the development of the first synthetic plastic, Bakelite, in 1907. The fear is that, for future historians, it will define the era when, blinded by their benefits, we lost control of the environmental impact of plastic and its main source, oil, with disastrous consequences.
What has this got to do with machine learning?
Before we get to that, let’s acknowledge the genuine environmental challenge. Plastic production is coupled to oil-centric economic models that drive climate change. Plastic pollution is blighting our seas, with catastrophic consequences for marine life and vital eco systems. We cannot simply innovate our way out of these problems. They require fundamental political decisions on where society invests, major upgrades to recycling infrastructures, and the will to reduce single-use plastic items and unnecessary packaging.
But we should also be aware of the benefits of plastics. It is a remarkable scientific achievement to have generated, in just a few generations, tens of thousands of grades of lightweight, strong, flexible new materials that are now intrinsic to almost everything we use. Can we apply the same genius to making better, more sustainable plastics? Because, while innovation is not a ‘magic bullet’ to solve the plastics challenge, plastics R&D must be part of the solution. We can, for example, increase the percentage of biopolymers derived from non-petrochemical sources. We can make better use of recycled feedstock in manufacturing. We can reduce the usage of harmful additives. We can make plastics more biodegradable for applications where we cannot avoid disposal. And we can make them less likely to degrade in applications where this will significantly increase their usable life.
Many scientists working on plastics would agree that their research retains an element of ‘art’ alongside the science. This is partly because, at the molecular level, the polymers that make up plastics consist of long chain molecules, often tangled in arrangements that are much less ordered than, for example, metals or ceramics. Understanding and precise control of properties can be difficult, and subtle changes in how polymers are blended, processed, and the additives used can make big differences. Not for nothing are cookery terms such as ‘recipe’ used in relation to polymer formulation.
Here is where machine learning comes in. The strong element of empirical research in polymer R&D requires costly investments in both experiment and simulation to identify beneficial tweaks in chemistry, formulation, or process parameters. At Intellegens, we are finding increasing numbers of polymer R&D organisations that want to use machine learning to more effectively mine the data built up through this work. They seek relationships in the data that provide clues on how to change their materials or processes. They use machine learning to identify where it would be most beneficial to focus experiment or simulation, saving time and cost. They are even building models of their systems that might be used to manage polymer production – for example, to change process parameters in response to the inevitable variability in the composition of incoming recycled feedstock.
Machine learning models also provide a great way to capture understanding of the system so that it can be used predictively on future projects. In doing this, we can retain and more effectively re-use knowledge (some of that ‘art’) that otherwise is locked-up in the head of experts who may leave or retire.
“Rewritten by machine on new technology, and now I understand the problems you can see,” sang The Buggles on that 1980 album. They were talking about the impact of video on the radio star. But maybe they were also onto something more positive for the real Age of Plastic?
