Sangwook Kim,¹ Zonggen Yi,¹ Ross R. Kunz,¹ Eric J. Dufek,¹ Tanvir Tanim,¹ Kevin L. Gering¹, Bor-Rong Chen,¹ Peter Weddle, ² Kandler Smith, ²

¹ Energy and Environmental Science and Technology, Idaho National Laboratory, Idaho Falls, Idaho 83415 USA

² Center for Energy Conversion & Storage Systems, National Renewable Energy Laboratory, Golden, CO 80401, USA

242^nd ECS Meeting, Atlanta, GA, Oct. 9 - 13, 2022

Symposium: A03 – Lithium Ion Batteries

Lithium-ion batteries play a central role in powering electric vehicles, stationary energy storage systems, and consumer electronics. Lately, machine learning and deep learning (DL) has been successfully used to gain insights into battery degradation during battery operation and predict lifetime in battery research and development community. Fast and robust classification and quantification of battery aging (e.g., Loss of Lithium Inventory (LLI) and Loss of active Material (LAM)) and accurate long-term forecasting of battery life enable more proactive planning of battery management and preemptive actions of modified operating conditions to achieve safe operations and prolong battery life.

Here, we present the development of Incremental Capacity (IC)-DL framework for fast charging conditions as a diagnostic tool for aging mode classification and quantification (Figure 1). The classification and quantification of dominant battery aging modes is conducted using a synthetic-data-based DL modeling framework. Over 6000 initial conditions and 26,000 different aging conditions are generated using IC model to train DL. We applied trained DL classification and quantification algorithm to 22 Gr/NMC532 pouch cells with a different loading and charging protocol tested up to 600 cycles. Besides the analysis of reference performance test data at C/20, cycle-by-cycle data at higher C-rate is analyzed. This IC-DL framework as a diagnostic tool is also used in different battery chemistries, such as Gr/NMC811and LTO/LMO. IC-DL framework enables unique, rapid identification and quantification of the dominant aging modes at different C-rates in different battery chemistries. Additionally, a prognostic tool, Sigmoidal Rate Expression (SRE)-type mathematics are employed to evaluate the capacity loss and aging mode (i.e., LLI). SREs are robust engines that contain three variables that capture the thermodynamic and kinetic “thumbprint” of the mechanism progression within the context of a batch reactor scenario [1]. We show two different methods by which SRE parameters can be early assessed based on quantified values from IC-DL framework; (1) extrapolative techniques using specialized functions to determine SRE parameter convergence and (2) a technique based on deep learning and Monte Carlo framework. Overall results from both methods confirm that we can predict capacity loss and LLI at the end of test (i.e., 600 cycles) within 1-2% absolute error using three weeks of testing data (or 125 cycles). We believe the IC-DL framework combined with SRE-base prognostics will reduce lithium-ion batteries development cycle as well as shorten the turnover time for validation, fulfilling the demands of rapidly growing battery market.

References

1. K. Gering, Electrochimica Acta, 228(20), 636-651 (2017)