A new kind of polymer membrane created by Georgia Tech researchers could reshape how refineries process crude oil, dramatically reducing the energy and water required while extracting useful materials.
The researchers have also created artificial intelligence tools to predict the performance of the DUCKY polymer membranes, which could accelerate the development of new ones.
The implications are stark: the initial separation of crude oil components is responsible for roughly 1% of energy used across the globe. The membrane separation technology the researchers are developing could have several uses, from biofuels and biodegradable plastics to pulp and paper products.
According to M.G. Finn, professor and James A. Carlos Family Chair in the School of Chemistry and Biochemistry, the researchers are establishing concepts that can be used with different molecules or polymers, but they have been applied to crude oil because that’s the most challenging current target.
In its raw state, crude oil includes thousands of compounds that have to be processed and refined to produce useful materials, such as gas and fuels, plastics, textile, food additives, medical products and more. There are many steps in squeezing out the valuable substances, but it starts with distillation, a water- and energy-intensive process.
Researchers have been trying to develop membranes to do that work instead, filtering out the desirable molecules while skipping the boiling and cooling.
According to Ryan Lively, Thomas C. DeLoach Jr. Professor in the School of Chemical and Biomolecular Engineering, crude oil is an important feedstock, but most people don’t think about how it is processed.
“These distillation systems are massive water consumers, and the membranes simply are not. They’re not using heat or combustion. They just use electricity. You could ostensibly run it off of a wind turbine, if you wanted. It’s just a fundamentally different way of doing a separation,” Lively said.
A new family of polymers is what makes the team’s membrane formula so powerful. The researchers used building blocks called spirocyclic monomers that assemble together in chains with lots of 90-degree turns, forming a kinky material that doesn’t easily compress and forms pores that selectively bind and permit desirable molecules to pass through. The polymers are not rigid, meaning they are easier to make in large quantities. They also have a well-controlled flexibility or mobility that allows pores of the right filtering structure to come and go over time.
The DUCKY polymers are created through a chemical reaction that’s easy to produce at a scale that would be useful for industrial purposes. It comes from a Nobel Prize-winning family of reactions called click chemistry, which gives the polymers their name. The reaction is called copper-catalysed azide-alkyne cycloaddition — abbreviated to CuAAC and pronounced “quack”. Thus: DUCKY polymers.
In isolation, the three key characteristics of the polymer membranes aren’t new; it is their unique combination that makes them a novelty and effective.
The research team included ExxonMobil scientists, who discovered how effective the membranes could be. The scientists took the crudest of the crude oil components — the sludge left at the bottom after the distillation process — and pushed it through one of the membranes. The process extracted even more valuable materials.
“That’s actually the business case for a lot of the people who process crude oils. They want to know what they can do that’s new. Can a membrane make something new that the distillation column can’t?” Lively said. “Of course, our secret motivation is to reduce energy, carbon, and water footprints, but if we can help them make new products at the same time, that’s a win-win.”
The team’s AI models can come into play to predict such outcomes. In a related study, Lively, Finn and researchers in Rampi Ramprasad’s Georgia Tech lab described using machine learning algorithms and mass transport simulations to predict the performance of polymer membranes in complex separations.
“This entire pipeline, I think, is a significant development. And it’s also the first step toward actual materials design,” said Ramprasad, professor and Michael E. Tennenbaum Family Chair in the School of Materials Science and Engineering. “We call this a ‘forward problem’, meaning you have a material and a mixture that goes in — what comes out? That’s a prediction problem. What we want to do eventually is to design new polymers that achieve a certain target permeation performance.”
Complex mixtures like crude oil might have hundreds or thousands of components, so it may be difficult to accurately describe each compound in mathematical terms, explain how it interacts with the membrane and extrapolate the outcome.
Training the algorithms involved combing through experimental literature on solvent diffusion through polymers to build an enormous dataset. But, according to Ramprasad, knowing ahead of time how a proposed polymer membrane might work would accelerate a material’s design process, which is basically trial-and-error currently.
The default approach of making the material and testing it takes time. Using a data-driven or machine learning-based approach uses past knowledge in an efficient manner.
“It’s a digital partner: You’re not guaranteed an exact prediction, because the model is limited by the space spanned by the data you use to train it. But it can extrapolate a little bit and it can take you in new directions, potentially. You can do an initial screening by searching through vast chemical spaces and make go, no-go decisions up front,” Ramprasad said.
Lively said that prior to meeting Ramprasad, he had long been a sceptic about the ability of machine learning tools to tackle the kinds of complex separations he works with.
Developing the AI tools also involved comparing the algorithms’ predictions to actual results, including with the DUCKY polymer membranes. The experiments showed the AI model’s predictions were within 6 to 7% of actual measurements.
“It’s astonishing,” Finn said. “My career has been spent trying to predict what molecules are going to do. The machine learning approach, and Rampi’s execution of it, is just completely revolutionary.”
