Harnessing AI to Battle Battery Challenges

Researchers are using artificial intelligence to speed up exploration to enhance batteries.

They have developed an AI-enabled robotic platform for testing aqueous electrolytes, possibly paving the way for faster innovations that make batteries better. (Image courtesy of Carnegie Mellon University College of Engineering.)

They have developed an AI-enabled robotic platform for testing aqueous electrolytes, possibly paving the way for faster innovations that make batteries better. (Image courtesy of Carnegie Mellon University College of Engineering.)

While the future may be unknown, there is one thing that humans will continue to rely on: batteries. With that in mind, reliance on electric vehicles, smartphones and other devices is driving research into developing safer, cost-effective, faster-charging and more reliable batteries. To aid in that research, scientists at Carnegie Mellon University’s (CMU’s) College of Engineering turned to artificial intelligence (AI).

Faced with the fact that experimentation often takes years, the team decided to take an autonomous approach. In a recently published paper, the team outlined its use of machine learning as a method of discovering potential compounds and materials. The team focused its efforts on aqueous electrolytes due to them being ideal for renewable energy storage. The team’s approach vastly minimizes the time required for exploration, which may potentially lead to faster development of new battery innovations.

“Designing high-performing aqueous batteries is an important process to solve,” said Adarsh Dave, mechanical engineering student. “However, there is a staggering number of possible formulations here to choose from—that’s where our design process comes in.”

The team developed a robotic platform integrated with machine learning, Otto, to discover battery electrolytes through measurement of potential properties that would be ideal for use in batteries. Otto is attached to a Bayesian optimizer and autonomously runs, resulting in real-time machine learning that learns to plan sequential experiments.

Schematic figure of a robotic platform and software architecture. (Image courtesy of Carnegie Mellon University College of Engineering.)

Schematic figure of a robotic platform and software architecture. (Image courtesy of Carnegie Mellon University College of Engineering.)

According to the researchers, Otto mixes feeder solutions and measures pH, ionic conductivity, and electrochemical properties in a symmetric platinum electrode cell. A computer with control software relays the electrolytes that Otto needs to test. Otto then sends back the electrolytes’ properties. The repeated back forth gradually enhances Otto’s machine learning capabilities. That means faster testing is occurring 24/7.

In what would have taken humans much longer, Otto was able to test 140 electrolyte formulas in 40 hours. The results indicated that a mixed-anion sodium electrolyte offered more stability than a benchmark electrolyte. Otto’s precision and repeatability were able to discover a new possibility that could have easily been missed by a human.

“While no robot or algorithm will replace a highly-trained chemist’s intuition for innovation, our system certainly automates and accelerates routine science and design tasks,” said Jay Whitacre, director of the Scott Institute for Energy Innovation and professor of engineering and public policy and material science engineering. “I hope to see my colleagues in other labs automate away the boring stuff and really accelerate the pace of battery innovation.”

Although Otto may not be part of other research yet, CMU College of Engineering’s focus on machine learning and enhancing battery testing caught the eye of Facebook. In September, the two launched the Open Catalyst Project. Its aim is aligned with CMU research to find better ways of storing renewable energy. According to Larry Zitnick, research scientist, large-scale deployment of lithium-ion batteries “for days or weeks or reserve power … is prohibitively expensive.”

AI is being developed to help researchers explore new electrocatalysts that are less expensive and have a smaller environmental impact than platinum. The ongoing project aims to create AI that can more quickly predict atomic interactions in much less time. Similar to discovering electrolytes, instead of testing three or four compositions or materials a year, AI enables the testing of 40,000 simulations a year.

Although Otto and other AI-integrated platforms may still be in the early stages of development, these studies indicate that battery improvements may be coming sooner rather than later.