How DOE Researchers Are Using AI And Robots To Design The Successor To Lithium-Ion Batteries

Building the better battery may be too big a job for humans alone, according to researchers at the Joint Center for Energy Storage Research, so the researchers are enlisting the help of artificial intelligence and robots.
In 2016, JCESR Director George Crabtree announced the group had settled on an organic flow battery to provide energy storage on the grid longer and cheaper than lithium-ion can. Such a battery would rely on a combination of molecules made from earth-abundant, low-cost elements like carbon, hydrogen and oxygen.
The question is, which combination?
“There are thousands, maybe hundreds of thousands of candidates out there,” Crabtree said this month during Stanford University’s StorageX Symposium. “We just haven't found the right one yet.”
That search could take years if researchers did it the traditional way. During most of human history, researchers employed direct design, Crabtree said, by thinking of candidate molecules, making them, putting them in a battery and seeing if it works.
“That was dramatically changed when we started to do computational screening.”
Using computational screening, researchers could simulate hundreds or thousands of molecules on a computer much faster than synthesizing and testing them in a laboratory.
But JCESR researchers need the process to be quicker still, so they have turned to what Crabtree calls a self-driving laboratory, using artificial intelligence to find molecular structures that might satisfy the multitude of performance requirements needed for a battery, things like solubility, stability, voltage crossover, or multi-electron transfer.
“This is a new feature,” he said. “So with AI, instead of simulating hundreds or thousands of molecules you go immediately to the most promising five or ten and consider only those.”
That’s when the robots take over.
“You could then have automatic synthesis. So the laboratory, using a robot, would make the material—automatic characterization, run it through lots of machines, send that information back to the artificial-intelligence brain here to score the material. Did it actually work? If it failed I wonder why it failed; let's try something else. So it actively learns from every cycle of this synthesis route. And this is what's coming to the fore.”
JCESR is in the ninth year of what was originally a five-year project to develop much better batteries for grid storage and transportation.
“If you want to have a completely carbon-free grid, you're going to have to have somewhere between 500 and 1,000 hours of continuous storage discharge,” Crabtree said, “and this, of course, is well beyond what lithium-ion can do.”
The organic redox-flow candidate for grid storage would contain liquids infused with organic molecules that carry a charge. The organic molecules should be inexpensive, recyclable, and harmless to the environment.
"The big advantage of this so-called flow battery is that it's scalable, so if I make that tank of active ions 10 times larger I can store 10 times the energy density," Crabtree said at an earlier talk. "You can't do that with a lithium-ion battery."
Watch George Crabtree describe JCESR’s update at Stanford’s StorageX Symposium: