Dec 3, 2024
8:00pm - 10:00pm
Hynes, Level 1, Hall A
Omar Allam1,Tyler Sours1,Shivang Agarwal1,Jordan Crivelli-Decker1,Marc Cormier2,Steffen Ridderbusch1,Dan Zhao1,Yunyun Wang1,Stephen Glazier2,Ang Xiao1
SandboxAQ1,Novonix2
Omar Allam1,Tyler Sours1,Shivang Agarwal1,Jordan Crivelli-Decker1,Marc Cormier2,Steffen Ridderbusch1,Dan Zhao1,Yunyun Wang1,Stephen Glazier2,Ang Xiao1
SandboxAQ1,Novonix2
Predicting battery cycle life with high accuracy (e.g., within 5% of actual capacity) is critical for speeding up the iteration loop during battery materials optimization. Traditional long-term cycling methods, while effective at capturing overall performance across a cell’s lifetime, often fail to provide detailed insights from the early cycles due to lower resolution data. Ultra-High Precision Coulometry (UHPC) offers transformative potential for predictive machine learning (ML) models by providing high-resolution early-life cycling data that captures subtle electrochemical changes not readily observed by conventional long-term cyclers. By leveraging UHPC data and advanced feature engineering, our prediction framework trains machine learning models that predict battery degradation with high fidelity using only a few initial cycles measured on UHPC cycling systems. These engineered features enable our models to generalize well across different cell chemistries, suggesting broader applicability than traditional methods. Our models leveraging UHPC can quickly and reliably estimate the cycle number at which significant capacity loss occurs, drastically reducing the time to accurate predictions compared to traditional long-term cycling tests. This approach not only improves our understanding of degradation mechanisms but also offers a scalable and efficient solution for battery management, thereby accelerating advancements in battery technology.