A physics-informed digital twin reveals the mechanistic threshold for knee-point onset in lithium-ion batteries

Abstract

Lithium-ion batteries can degrade abruptly after extended stable operation through a failure transition known as the knee point, yet the internal electrochemical states governing this transition are invisible to standard monitoring. Here we develop a physics-informed digital twin based on deep operator networks that captures lithium-ion diffusion dynamics across electrode particles and electrolyte with stable recursive simulation over full charge–discharge cycles. Deploying this digital twin across 16 cells under diverse conditions and temperatures (10°C, 25°C, 40°C), we track degradation mode evolution — distinguishing loss of lithium inventory from loss of active material on each electrode — with root-mean-square errors below 3% from routine 0.3C charging data alone, without requiring specialized diagnostic tests. Real-time monitoring of anode overpotentials reveals a quantitative threshold in the relationship between anode active material loss and lithium plating onset that separates cells developing knee points from those that do not. This threshold, which quantifies the critical coupling strength of the plating-degradation feedback loop, is detectable at early life stages when conventional capacity measurements show no difference between cells, and provides a quantitative criterion for both predicting and potentially preventing knee-point onset.