The power of machine learning technology has been harnessed to reverse long-held suppositions about the way lithium-ion batteries charge and discharge. By coupling machine learning with knowledge obtained from experiments and physics equations, new criteria can be developed to help engineers design longer-lasting battery cells.

The application of scientific machine learning to battery cycling and failure mechanisms is expected to accelerate the development of safer fast-charging lithium-ion battery packs for electric vehicles. The machine learning application generates new information on how electrode particles release stored lithium ions during battery charging. Source: Greg Stewart/SLAC National Accelerator LaboratoryResearchers from Stanford University, U.S. Lawrence Livermore National Laboratory, MIT and SLAC National Accelerator Laboratory first observed the actions of cathode particles created from nickel, manganese and cobalt, which soak in lithium ions during discharge and shed them during charging.

Experimental and modeling data were integrated into machine learning algorithms, which revealed that the particles themselves manage the pace of lithium ions leaving cathode particles during charging. A few particles shed many of their ions straight away while others discharge hardly any or no particles at all. Quick-release particles also carry on streaming ions at a quicker rate than slow-release or non-releasing neighboring particles.

An improved understanding of this uneven charging and discharging process offers scope for improving the cost, storage capacity and durability of batteries.

A paper on this development is published in Nature Materials.