Lithium-ion batteries are the most common batteries used widely in everything from consumer electronics to scooters and cars. However, such batteries are highly flammable. A team from Johns Hopkins University and the Toyota Research Institute are aiming to develop better batteries with new AI tools and high-throughput automated experiments, increasing the speed of discovery and revealing why certain combinations work well.
"Battery chemistry is extremely complicated. A change in one property can influence several others, so understanding those relationships is very difficult," says study leader Yayuan Liu, assistant professor of chemical and biomolecular engineering and an associate researcher with the Ralph O'Connor Sustainable Energy Institute.
Image caption: Yayuan Li
Image credit: Will Kirk / Johns Hopkins University
New types of advanced electrochemical materials, substances that store, transmit, or generate energy, are difficult to discover. Scientists often labor over a trial-and-error process, spending years manually testing countless combinations of materials and chemical formulations.
The research team focused specifically on using electrolytes for fluoride-ion batteries, an emerging alternative to lithium-ion batteries. Such electrolytes for batteries are not the type found in sports drinks, but the concept is similar. They're both liquids containing charged particles, but while sports drinks include minerals natural to the body, battery electrolytes contain chemicals dissolved in a solvent that allows electrical charges to move through a battery.
"Fluoride-ion batteries can potentially provide two to three times the energy density of lithium-ion batteries," says Andong Liu, a member of the team and a doctoral student in chemical and biomolecular engineering. "Another advantage is that these batteries can use aqueous electrolytes, which are much safer because they aren't flammable. Fluoride is also far more abundant than lithium, so long-term supply chain concerns are less relevant. These improvements could eventually make fluoride-ion batteries especially useful for larger-scale technologies, such as drones."
Despite their promise, fluoride-ion batteries have been difficult to develop because researchers have struggled to identify electrolytes that balance conductivity, stability, and ion transport. Some electrolytes offer excellent conductivity but poor stability, while others are stable but transport ions too slowly.
To address this challenge, the team is building a high-throughput platform that uses software to rapidly test large numbers of electrolyte combinations. Traditional experiments involve humans doing the manual labor and can take around 10 minutes for each, according to Dian-Zhao Lin, a member of the team and a postdoctoral fellow in chemical and biomolecular engineering. The group's automated experimentation system reduces that timeline to approximately three to four minutes per experiment, including preparation and cleaning.
"When you scale that across hundreds or even thousands of experiments, the time savings become enormous," Lin says. "It allows us to generate much more data and move much faster."
The data generated through those experiments is then paired with causal machine learning models developed by scientists at the Toyota Research Institute.
"Most machine learning models can predict how a particular electrolyte formulation may affect performance, and explainability tools can aid in interpreting those predictions," says project collaborator Amanda Volk, a senior research scientist at the Toyota Research Institute. "Our goal is to move beyond predicting which formulations work to understanding why they work by identifying causal relationships and mechanistic pathways that connect electrolyte chemistry to performance."
Image caption: A zoomed-in view of battery experiments.
Image credit: Will Kirk / Johns Hopkins University
Reasoning is a rising topic in the field of AI that fills a critical role in physical sciences.
"We all know correlation is not causation. But what if we could break down this barrier? Our team is trying to see how far we can push causal reasoning to close the gap between materials research and device development," says Kevin Tran, senior manager at the Toyota Research Institute.
Liu says that deeper understanding could have impacts far beyond fluoride-ion batteries alone.
"If we understand the real factors that impact performance, then we gain much stronger predictive power for future designs," Liu says. "The insights we develop here could eventually transfer to many other battery systems and energy technologies. It could help researchers across the energy field discover better materials much more quickly and efficiently."
Funding for the project is provided by the Toyota Research Institute's University Research Program.
