Materials scientists at MIT are asking and answering this very question. Elsa Olivetti PhD ’07, the Esther and Harold E. Edgerton Associate Professor in materials science and engineering, and Rafael Gomez-Bombarelli, the Jeffrey Cheah Career Development Professor and assistant professor of materials science and engineering, are leading a collaboration that pairs cutting-edge computational design techniques with machine learning to assess the properties of materials and to determine how they can be redesigned and improved, or if entirely new materials could be synthesized to do a job better.
“We aspire as people that work on matter and atoms to use computational tools in the same way as engineers in other specialties,” says Gomez-Bombarelli. Mechanical engineers, for example, use programs such as AutoCAD and Ansys to predict how various components will perform in different environments, and chemical engineers use Aspen to understand processes flows.
Now, Olivetti and Gomez-Bombarelli are bringing similar design tools to the field of materials science and applying them at a broad scale. “We can think about what elements to include in a material and do so with a set of tools that inform design across its life cycle, from manufacturing to recycling,” says Olivetti. “That accelerates the screening of materials that might be more sustainable and directs efforts experimentally.”
Olivetti, a MacVicar Faculty Fellow, and Gomez-Bombarelli have worked with their students to assemble a suite of machine learning-based software tools, ranging from natural language processing tools to custom neural networks adapted to use molecular structures as inputs. This suite of tools automatically collates information from published literature and uses volumes of data to develop algorithms for materials synthesis and optimized performance.
The team has been using this process to build better zeolites, a class of materials commonly used in catalysts, chemical filters, and the catalytic converters used to clean vehicle emissions. “We use our tools to extract massive amounts of data from the literature around zeolites,” says Olivetti. “Then we use our predictive modeling algorithm to determine potential subsequent ingredients to add to make the final zeolite.”
Using this system, the researchers were able to work with colleagues to design a new zeolite recipe optimized for removing nitrogen oxide, a major pollutant, from diesel engine exhaust. “We were able to use all this computation to support our collaborators in the lab and hit a narrow, really exciting piece of innovation that would have been really hard to find with traditional trial and error,” says Gomez-Bombarelli.
More sustainable concrete
Predictive synthesis works well in cases such as zeolites, in which there are far too many options to sift through experimentally. It’s also useful when optimizing a mixture of materials is needed to make a product more sustainable.
Consider cement, an essential ingredient of concrete. Thirty billion tons of concrete is used every year, accounting for 8% of global carbon dioxide emissions due to the intense heat needed to create cement from raw materials such as lime, clay, and silica. Developing a more sustainable process requires a clear understanding of how possible replacement materials might mix.
Because zeolites and cement have a similar chemistry, critical aspects of Olivetti and Gomez-Bombarelli’s predictive zeolite work could be applied to the world of cement. The researchers plan to use their techniques to predict how potential concrete ingredients will behave on a molecular level, with the aim of adjusting the recipe to employ, for example, industrial waste materials.
“We use these computational tools to search the space for how to make the best mixture,” says Olivetti. “The way I think about it is, how early in the design of new materials can we think about their environmental implications from extraction to end of life?” Her answer? “The earlier, the better.”