Carnegie Mellon University
October 14, 2019

Chemistry Professor Receives Grant to Develop Algorithm for Designing Better Concrete

By Ben Panko

Associate Professor of Chemistry Newell Washburn has received a more than half-million-dollar grant from the Advanced Research Projects Agency-Energy (ARPA-E) to develop a machine learning algorithm that could help design more durable and energy-efficient concrete.

Concrete is the most widely used engineered material on earth and the second-most used substance after water. Its production poses a significant burden to the environment by comprising over 5% of global carbon dioxide emissions. Creating more environmentally-friendly concrete has proven to be difficult due to the composite’s complex composition. Through this project Washburn and his collaborators plan to create a method that uses machine learning to assess new formulations for concrete in order to identify products that can be created in a more sustainable manner while maintaining the high strength required for bridges, roads and other infrastructure

ARPA-E, a branch of the Department of Energy focused on advancing energy technologies, has set aggressive goals for doubling the durability of concrete and halving the embodied CO2 while maintaining cost and other performance metrics, said Washburn. His team, which includes Barnabás Poczos of the Carnegie Mellon University Machine Learning Department and Kimberly Kurtis from the School of Civil and Environmental Engineering at Georgia Tech, will address ARPA-E’s goals by developing a machine learning algorithm to predict the properties of blends of traditional and alternative cements with limestone and minimally processed clays. Their work is especially novel because the development of a machine learning tool for accelerated materials discovery for infrastructure use has been largely overlooked.

“It has been demonstrated that incorporating minimally processed clays and other minerals can significantly improve both the performance and sustainability of cementer binders,” Washburn said. “However, our co-principal investigator Professor Kurtis has shown how complex overcoming the variability in material properties is, depending on the feedstocks and how they were processed. Our goal is to use advanced machine learning algorithms to identify the critical hidden variables and develop design tools for the broad deployment of next-generation concrete.”

Currently concrete mixes are developed by conducting numerous trial experiments. As concrete producers need to be certain that the mix they are using is sufficiently strong, they tend to create designs based on mixes that have been previously proven effective. This prevents concrete producers from adopting minimally-processed alternative materials that have improved sustainability but greater variability.

Additionally, calculating which formulations to use to maximize strength and minimize cost and energy, and in what ratios, is extremely complex because of how many aggregates can be found in concrete and how varied the formulas can be. Washburn and his team plan to harness the power of Bayesian hierarchical machine learning algorithms to predict the composition of potential alternative formulations of concrete that have enhanced durability and reduced embodied carbon dioxide.

The ultimate goal is to incorporate these algorithms into an artificial intelligence tool that can be used by structural engineers and construction companies to design concrete with improved durability and sustainability within two years.

The research program will also involve an advisory committee led by Karen Scrivener, director of the Laboratory of Construction Materials at École Polytechnique Fédérale de Lausanne in Switzerland, and composed of representatives from leading companies in the concrete industry.