Machine Learning-Based State of Health and Aging-Aware Power Capability Estimation for Battery Systems

Abstract

The rapid shift toward transportation electrification underscores the urgent need for intelligent battery management strategies that ensure the safety, reliability, and longevity of lithium-ion batteries. This dissertation presents a comprehensive framework for machine learning–based battery state-of-health (SOH) estimation and aging-aware power capability (state-of-power, SOP) assessment, addressing critical challenges posed by battery degradation and diverse operating conditions.

First, a novel SOP measurement methodology is developed and validated across multiple commercial cell chemistries, pulse durations, temperatures, and aging levels, offering significant accuracy improvements over conventional hybrid pulse power characterization (HPPC) techniques. Data generated from this work reveals key challenges in SOH and SOP estimation, particularly under extreme operating conditions and during long-term aging.

To address the limitations of conventional health indicators (HIs) used in data-driven SOH estimation, this work then proposes a multi-modal fusion approach that combines features from partial voltage curves and histogram data. By integrating these complementary sources of health information, the proposed model improves estimation accuracy and robustness while reducing data requirements. This approach also enhances the understanding of battery degradation and aging behaviors, supporting the development of more effective aging-aware battery management system (BMS) algorithms.

Building on these findings, a machine learning–based battery voltage estimation model is further introduced and embedded into a numerical search-based SOP estimation framework. This model exhibits strong generalization under varying temperature and state-of-charge (SOC) conditions and is validated via hardware-in-the-loop (HIL) testing. To ensure sustained accuracy over the battery's lifecycle, a self-updating, cloud-integrated framework is proposed. This system leverages both lab and real-world operational data to adapt to aging-induced changes and cell-to-cell variations.

Collectively, this work advances the development of intelligent, aging-aware battery management systems, delivering scalable, adaptive solutions for electric vehicles and energy storage applications.