Accurately predicting the useful life of lithium-ion batteries before their capacities start to wane could allow producers to supply longer-living batteries to automakers, saving ones with shorter expected lifespans for other applications.

Research published in Nature Energy from scientists at the Massachusetts Institute of Technology (MIT), Stanford University, and the Toyota Research Institute (TRI) discovered that combining comprehensive experimental data and artificial intelligence revealed the key for accurately predicting useful battery life.

After researchers trained the machine-learning model with a few hundred million data points, the algorithm predicted how many more cycles each battery would last, based on voltage declines and other factors. The predictions were within 9% of the actual cycle life. Separately, the algorithm categorized batteries as either long- or short-life expectancy, based on the first five charge/discharge cycles. Here, the predictions were correct 95% of the time.

“The standard way to test new battery designs is to charge and discharge the cells until they die. Since batteries have a long lifetime, this process can take many months and even years,” says Peter Attia, co-lead author and Stanford doctoral candidate in Materials Science and Engineering. “It’s an expensive bottleneck in battery research.”

The work was carried out at the Center for Data-Driven Design of Batteries, an academic-industrial collaboration that integrates theory, experiments, and data science.

Study co-authors Muratahan Aykol and Patrick Herring brought TRI’s experience with Big Data to the project and their own expertise on battery development.

One project focus was to find a better way to charge batteries in 10 minutes. To generate the training data set, the team charged and discharged the batteries until each one reached the end of its useful life, which they defined as a 20% capacity loss.

“Advances in computational power and data generation have recently enabled machine learning to accelerate progress for a variety of tasks. These include prediction of material properties,” MIT Chemical Engineering Professor Richard Braatz says. His team led the machine-learning portion of the research. “Our results show how we can predict the behavior of complex systems far into the future.”

Lithium-ion batteries tend to be stable for a while, then take a sharp turn downward. Plummet points vary widely – batteries lasted from 150 to 2,300 cycles in project testing – partly the result of testing different methods of fast charging and normal differences that emerge in commercially produced devices that depend on molecular interfaces.

“For all of the time and money that gets spent on battery development, progress is still measured in decades,” Herring says. “We are reducing one of the most time-consuming steps by an order of magnitude.”

The research method could shorten the time for validating batteries with new chemistries, manufacturers could use the sorting technique to grade batteries, and recyclers could find cells in used EV battery packs that have enough life in them for secondary uses. The research could also speed battery manufacturing, lower production costs, and optimize charging procedures.

Massachusetts Institute of Technology Center for Computational Engineering 

Stanford University Materials Science & Engineering

Toyota Research Institute