Session: 02-01 Intelligent Modeling and Predictive Analytics for Energy Systems

Paper Number: 172151

172151 - Leveraging Machine Learning Techniques for Lithium-Ion Batteries Advanced Design Review 

Accurately capturing lithium-ion batteries thermal and mechanical dynamics is an essential requirement for design review, and battery performance assessment and improvement. One of the main safety-related events, that is central to the design review task, is the thermal runaway (TR), where an increase in battery temperature, due to internal and/or external factors, causes further heat generation, leading to uncontrollable temperature rise, often resulting in system failure or damage.

Accurate modeling of TR dynamics enables the identification of critical safety risks, and supports the design of safer high-voltage battery systems. In the presented work, experimental data and advanced physics-based simulations were utilized to build an extensive database capturing key TR measurements, including temperature, cell mass, voltage, and total released heat, among other collected and simulated data. To enhance the developed dataset quality, data augmentation techniques were applied, addressing inconsistencies and improving the robustness of the dataset.

Analysis of the experimental and simulated data revealed significant variability when examined individually, which is driven by factors such as cell chemistry and testing conditions. To overcome this challenge, a machine learning (ML) model was developed, trained, and tested to predict key TR metrics. Furthermore, data grouping techniques were deployed to reduce variability and improve the model accuracy. The ML model demonstrated strong performance, achieving a mean absolute percentage error (MAPE) of approximately 9% when validated against a dataset comprising 600 healthy tests and 103 TR tests.

This work provides a robust framework for understanding TR dynamics and predicting critical safety metrics, such as total released heat, with high accuracy. By integrating experimental data, physics-based simulations, and ML techniques, this study helps advance the development of reliable TR models and supports more robust battery system design.

Presenting Author: Ala Eddine Omrani American Bureau of Shipping

Presenting Author Biography: Ala Eddine Omrani holds a Ph.D. in Mechanical Engineering from the University of Houston, and has over 7 years of experience in data-driven and physics-based modeling and simulation. His work spans a range of industries, including oil and gas, automotive, advertising, and offshore vessels classification and certification. Currently, Ala is a Senior Engineer II – Modeling and Simulation at the American Bureau of Shipping (ABS), where he applies advanced modeling techniques to drive innovation in offshore vessel classification, certification, and related engineering solutions.