Image Courtesy: Society for Science
An 18-year-old California student has developed a model that could make it easier to determine how much life remains in an aging electric vehicle battery. Colin Jie Chu, a student at The Nueva School in Palo Alto, developed a system that estimates the health of lithium-ion batteries with a reported prediction error of just 2.36 percent.
Chu’s research earned him a place as a finalist in the 2026 Regeneron Science Talent Search, one of the United States’ leading science and mathematics competitions. His work tackles a growing challenge for the EV industry: accurately determining how batteries degrade over time.
Battery degradation happens gradually as cells are repeatedly charged and discharged and exposed to different temperatures and operating conditions. Because the degradation cannot be directly measured while a battery is operating, engineers instead have to infer its condition from electrical data. Chu analyzed data from 22 batteries that were deliberately aged using electrical signals designed to replicate changing driving conditions, according to the Society for Science.
He then combined an equivalent circuit model, which represents a battery’s electrical behavior using simplified mathematical components, with machine learning. The hybrid approach was designed to estimate battery state of health while accounting for changing operating conditions.
Chu conducted the research through Stanford University’s Young Investigators Program, working at Professor Simona Onori’s Stanford Energy Control Lab with researchers and industry partners. The work was later presented at the Modeling, Estimation, and Control Conference in Chicago and published in the Journal of The Electrochemical Society.
Knowing the condition of an EV battery has practical consequences. As lithium-ion cells age, their usable capacity gradually falls, reducing driving range and potentially affecting decisions about maintenance, reuse and recycling. More accurate health estimates could also help battery management systems make better decisions about charging and remaining useful life.
The reported 2.36 percent error, however, does not mean the model is ready to predict the lifespan of every EV battery. The research was conducted using specific experimental data, and broader validation would be needed across different battery chemistries, ages, vehicle platforms and real-world driving conditions.
Chu began the project in 2024 while still in high school and has since expanded his research into battery “kneepoints,” periods when degradation can begin accelerating, as well as lifetime estimation.
For an EV industry increasingly focused on battery longevity, research like Chu’s could eventually make battery health easier to measure and understand. The technology is not yet ready for widespread deployment, but the project shows how combining physics-based models with machine learning could provide a more accurate picture of one of an EV’s most important and expensive components.
