Diagnostic-free onboard battery health assessment

Abstract

Diverse usage patterns induce complex and variable aging behaviors in lithium-ion batteries, complicating accurate health diagnosis and prognosis. In this work, we leverage portions of operational measurements from charging or dynamic discharging in combination with an interpretable machine learning model to enable rapid, onboard battery health diagnostics and prognostics without offline diagnostic testing and access to historical data. We integrate mechanistic constraints derived from differential voltage analysis within an encoder-decoder to extract electrode health states in a physically interpretable latent space, which enables improved reconstruction of the degradation path with onboard aging mechanisms tracking. The diagnosis model can be flexibly applied across diverse applications with slight fine-tuning. We demonstrate the model’s versatility by applying it to three battery-cycling datasets consisting of 422 cells under different operating conditions, with a mean absolute error of less than 2% for health diagnosis under varying conditions, highlighting the utility of an interpretable, diagnostic-free model.