Machine learning predicts the mechanical properties of porous materials

Machine learning can be used to predict the properties of a group of materials that, according to some, could be as important in the 21st century as plastics in the twentieth century.

The researchers used machine learning techniques to accurately predict the mechanical properties of organometallic structures (MOFs), which could be used to extract water from the desert air, store dangerous gases or fuel cars with hydrogen.

The researchers, led by the University of Cambridge, used their machine learning algorithm to predict the properties of more than 3,000 existing MOFs, as well as those still to be synthesized in the lab.

The results, published in the inaugural edition of the journal Cell Press material, could be used to significantly accelerate the way materials are characterized and designed at the molecular level.

MOFs are self-assembling 3D compounds composed of interconnected metal and organic atoms. Like plastics, they are extremely versatile and can be customized in millions of different combinations. Unlike plastics, which rely on long chains of polymers that only develop in one direction, MOFs have ordered crystalline structures that develop in all directions.

This crystalline structure means that MOFs can be made as building blocks: individual atoms or molecules can be moved in or out of the structure, a level of precision impossible to achieve with plastics.

The structures are very porous and have a massive surface: an MOF the size of a piece of sugar laid flat would cover a surface the size of six football pitches. Perhaps a bit counterintuitive, however, MOFs are extremely efficient storage devices. The pores of any MOF can be customized to form a perfectly formed storage pouch for different molecules, simply by changing the building blocks.

"The fact that MOFs are so porous makes them very adaptable to all kinds of applications, but at the same time, their porous nature makes them extremely fragile," said Dr. David Fairen-Jimenez of the Department of Chemical Engineering and Cambridge Biotechnology, which led the research.

MOFs are synthesized in powder form, but in order to present any practical use, the powder is subjected to pressure and converted into larger and formed pellets. Because of their porosity, many MOFs are crushed in this process, wasting time and money.

To address this problem, Fairen-Jimenez and his Belgian and American collaborators have developed an automatic learning algorithm to predict the mechanical properties of thousands of MOFs, so that only those with the necessary mechanical stability are manufactured.

The researchers used a multi-level computer approach to construct an interactive map of the structural and mechanical landscape of MOFs. First, they used high-throughput molecular simulations for 3,385 MOFs. Secondly, they developed a freely available automatic learning algorithm to automatically predict the mechanical properties of existing MOFs and to synthesize them.

"We are now able to explain the landscape of all materials at the same time," Fairen-Jimenez said. "In this way, we can predict what would be the best material for a given task."

The researchers have launched an interactive website where scientists can design and predict the performance of their own MOFs. Fairen-Jimenez says this tool will help bridge the gap between experimenters and computer scientists working in this field. "This allows researchers to access the tools they need to work with these materials: it simplifies the questions they need to ask," he said.