Session: 09-16-03: Energy-Related Multidisciplinary III

Paper Number: 166464

Uncertainty-Aware Surrogate-Accelerated Chemo-Mechanical Phase-Field Fracture Modeling of Fatigue Crack Growth in Lithium-Ion Batteries 

High-fidelity chemo-mechanical phase-field fracture (PFF) models effectively capture fatigue-induced crack growth in lithium-ion batteries (LIBs), where repeated lithiation and de-lithiation cycles accelerate degradation, leading to capacity fade and reduced battery life. However, inherent variability in mechanical and chemical properties—arising from manufacturing inconsistencies, electrode particle size distributions, and processing conditions—introduces significant uncertainty in predicting key quantities of interest (QOIs), such as crack volume plateau regions and degraded elastic energy density of electrode particles over charge-discharge cycles, ultimately hindering accurate lifecycle prediction and reliability assessment of lithium-ion batteries. With the growing demand for efficient and sustainable energy storage, quantifying these uncertainties is essential for improving battery manufacturing processes, ensuring safer designs, and extending operational lifespan. The substantial computational cost of the coupled model poses a major challenge for conducting uncertainty quantification (UQ) and sensitivity analysis (SA). A surrogate model offers an efficient alternative to mitigate this issue. However, conventional one-shot sampling-based surrogates often exhibit lower accuracy than adaptive sampling-based approaches, as commonly used design of experiments (DOEs) rely on predefined training points, leading to the risk of oversampling or under-sampling key regions of the input space. This work presents a data-driven adaptive sparse polynomial chaos expansion (PCE)-based UQ framework for efficient, physics-free simulation of chemo-mechanical PFF evolution in LIB electrode particles using sparse data under mechanical and chemical uncertainties. The framework integrates global SA to systematically identify dominant parameters governing battery degradation across charge-discharge cycles, significantly reducing computational cost while preserving predictive accuracy. Our  adaptive sparse PCE is constructed using an adaptive sampling strategy from a Sobol sequence sampling-based candidate set with an effective convergence criterion to ensure accurate estimation of QOIs. Starting with an initial DOEs, the adaptive sampling methodology iteratively refines the sparse PCE by strategically selecting the most informative sample points, enhancing accuracy of both the surrogate and QOIs while minimizing computational cost. This approach leverages a limited yet optimally selected DOE, requiring only a small number of high-fidelity finite element-based multi-physics simulations implemented efficiently within the FEniCS environment. Preliminary results demonstrate that the proposed framework significantly reduces computational costs while maintaining high prediction accuracy comparable to direct PFF simulations. SA identifies lithium diffusion coefficient and partial molar volume as dominant contributors to variability in crack evolution and elastic energy degradation, underscoring their critical role in fatigue failure. Benchmarking against reference simulations, including statistical metrics and charge-discharge cycle probability density functions, confirms that the proposed framework accurately and efficiently captures uncertainty-aware predictions in LIB degradation metrics. The proposed uncertainty-aware LIB modeling framework improves the scalability and accuracy of battery performance predictions, aiding the development of batteries with optimal energy density, enhanced safety, prolonged lifespan, and reduced resource consumption.

Presenting Author: Avinandan Modak Indian Institute of Technology Roorkee

Presenting Author Biography: Avinandan Modak is currently a doctoral researcher in the Department of Civil Engineering at the Indian Institute of Technology Roorkee, India. He earned his Master of Technology in 2023 from the Department of Civil Engineering at the Indian Institute of Engineering Science and Technology, Shibpur, India. His research focuses on uncertainty quantification, machine learning, and metamodeling, with a strong emphasis on chemo-mechanical phase-field fatigue fracture modeling of Li-ion batteries. Additionally, he works on multi-scale topology optimization for biomechanical applications.

Authors:

Avinandan Modak Indian Institute of Technology Roorkee
Abhinav Gupta Vanderbilt University, USA
Rajib Chowdhury Indian Institute of Technology Roorkee
Ravindra Duddu Vanderbilt University, USA