Machine learning can lead to cleaner water
A Louisiana State University-Penn State research team is using machine learning (ML) to developer a smarter approach to ionic separations which could improve water treatment, resource recovery and energy production.
The U.S. Department of Energy has awarded a $1.5 million to a Louisiana State University-Penn State research team to develop a smarter approach to ionic separations, critical chemical reactions needed for water treatment, resource recovery and energy production. Using machine learning (ML), the researchers are building a platform to accelerate the discovery phase of determining better materials and processes by combining continuously refined and cross-linked molecular, materials and device data. The platform bridges transport, thermodynamic and kinetic phenomena from nano-level to plant-scale equipment.
The team comprises Chris Arges, associate professor in the Penn State Department of Chemical Engineering and adjunct professor in the LSU Department of Chemical Engineering; principal investigator Revati Kumar, associate professor in the LSU Department of Chemistry with a joint appointment in the LSU Center for Computation & Technology; and Cain Endowed Chair José Romagnoli, professor in the LSU Department of Chemical Engineering.
The researchers aim to optimize common and often costly processes in the chemical manufacturing and energy industries, such as purifying water or extracting and recycling valuable metals, including lithium and copper.
“A challenge in the development of new materials for separation science is the scarce availability of data that limits the implementation and effectiveness of machine learning methods,” Romagnoli said.
He explained in this project, the available data will help train machine learning algorithms, which can then be incorporated into compositional, physics-informed models. Those models will generate synthetic data to train machine learning surrogate models, which predict outcomes based on the statistical likelihood of occurrence with the supplied data, rather than playing out an entire simulation. While the researchers still conduct some physical experiments to validate the accuracy and usefulness of the machine learning results, Romagnoli said the process alleviates the cost and time required to generate large datasets solely from physical experiments.
“Machine learning is beautiful because you don’t have to have a priori knowledge of the relationships between things,” Kumar said. “That’s what the machine learning does — it takes large data sets and helps you see patterns. Normally, if you look at all that data, you have no chance of making sense of it, so this method is a big improvement and there are many validation steps.”
In addition to identifying and validating more efficient approaches to separate ions to obtain freshwater from salt water, the machine learning platform may be able to help recover metals from wastewater, including the fermentation broths used in biofuel production.
“There are many value-added chemicals in that mixture that would be good building blocks for commodity chemicals and new products, such as biodegradable plastics,” said Arges, who is also affiliated with the Center of Integrated Energy Systems in the Institutes of Energy and the Environment at Penn State.
Arges explained how more efficient ionic separations have important impacts ranging from sourcing lithium for lithium-ion batteries to ensuring reliable freshwater to create steam for electric power plants.
“There are many uses of this technology, and we look forward to sharing our playbook with the wider scientific community so they can use the tools we’re developing to advance their own research objectives in the area of ionic separations,” Arges said.