As nations work to eliminate carbon emissions, batteries will play a huge role.
Electric vehicles powered by batteries seem likely to dominate the future of commercial and consumer transportation.
Likewise, large stationary batteries will augment renewables like wind and solar by storing energy when production exceeds demand on the electrical grid, then sending that energy back to the grid when needed.
But even today’s most advanced lithium-ion batteries don’t yet have the combination of economics, durability and energy density necessary to meet all future energy goals.
“By 2035, most major industrialized countries have said they want to be mostly carbon-free or have non-emitting vehicles,” said Eric Dufek, department manager for Idaho National Laboratory’s Energy Storage & Advanced Transportation Department. “We need a lot of really good batteries, and we need to do all this in less than 13 years. There’s a real urgency because these things don’t happen overnight.”
To tackle this challenge, battery scientists from the U.S., the United Kingdom and Germany have proposed the Battery Data Genome, a central repository for battery test information that could be accessed by researchers around the world. The proposal was written by Dufek and 26 colleagues representing more than a dozen research institutions including INL, the National Renewable Energy Laboratory and Argonne National Laboratory. The proposal appears in the journal Joule.
Finding data for machine learning and AI
Several industries—including genomics, pharmacology and cybersecurity—have relied on advanced data science methods such as machine learning and artificial intelligence to make great strides in research and development.
But unlike other industries, battery researchers don’t have the extensive data sets necessary to make full use of advanced data science methods.
The Battery Data Genome is envisioned as a data hub designed to provide the large volume of information necessary for researchers to use these same methods to accelerate the pace of battery technology development.
To make tomorrow’s stationary and electric vehicle batteries, researchers are exploring novel battery chemistries, materials, cell designs and other technologies. Machine learning and artificial intelligence can help develop computer models to test these technologies before expensive and time-consuming laboratory tests.
The more data, the closer these models can reflect reality. Machine learning and artificial intelligence can help integrate all that data into the model, in essence, making sense of data sets that might otherwise be too large to manage.
“With advanced data science methods there’s a lot of opportunity for technology development, advancement toward deployment and validation,” Dufek said. “There are a lot of material properties that are extremely important. You can use data sets to identify a lot of useful materials and cell design during the early steps of R&D.”
The information that researchers need to develop new battery types includes electrochemical data, thermal data, materials data and characterization data, as well as meta data—the data that describes the data.
“As the lead test lab for the [Department of Energy’s] Vehicle Technologies Office, we have a lot more data than most people, and we have a lot of really high-quality data as well,” Dufek said. “But we don’t have enough to be as inclusive as we would like.”
The availability of open-source information on batteries is especially limited. “If you look at the current experimental data sources, there’s not much available for open use across the community. And, no single entity has sufficient bandwidth or resources to gather all the data that’s necessary to do build the models we need.”
More data for high fidelity models
The Battery Data Genome proposal envisions a digital environment where researchers can choose whether the information they submit is open source or restricted access.
“All the data that we’re capturing from materials discovery is going to build on itself,” Dufek said. “You’re able to get high fidelity models that tell you, for instance, more about how battery technologies fail. That, in turn, tells you what you need to improve to make a better product.”
The authors also suggest standardized data collection practices that would make it easier for researchers to incorporate other institutions’ information into their models. Standardizing data will go a long way toward making research accessible to the wider battery research community, said Tanvir Tanim, an R&D Engineer and the Group Lead for Energy Storage Technology at INL.
“Sharing the information between institutions is difficult, in part because the datasets are so large,” he said. “Right now people collect data in different ways, so a lot of that information is unusable.”
A broad base of support
While the paper does not describe a specific project underway for developing the Battery Data Genome—a cloud-based application is one option—it does provide a roadmap for organizing the data hub.
The idea has a lot of support among battery researchers, and efforts are underway to make the idea a reality.
“This is an urgent need if we really want to see energy storage advancing to where we need it for transportation and stationary applications,” Dufek said.
