Accelerated battery life predictions through synergistic combination of physics-based models and machine learning

Abstract

There are tremendous economic and technical benefits to shortening battery test periods through robust predictive methods. Accurate long-term forecasting of battery life enables proactive planning of battery management (e.g., cell replacements) and preemptive actions to modify operating conditions to improve safety and life. The ever-evolving landscape of battery materials and applications ensure an abiding need for early capture of aging mechanisms. Herein we report on accelerated determination of battery aging mechanisms together with prediction of future capacity loss. Sigmoidal rate expressions (SREs) are used as diagnostic and predictive engines to evaluate aging mechanisms at play. We demonstrate three methods by which SRE parameters are early assessed. Overall results indicate that for cases dominated by loss of lithium inventory we can predict end-of-test capacity loss using less than three weeks of data. In many cases, predictions are within 5%–10% relative error and to within 1%–2% absolute error of observed performance.