Hybrid physics-based and data-driven modeling with calibrated uncertainty for lithium-ion battery degradation diagnosis and prognosis

TL;DR

This work addresses the challenge of predicting lithium-ion battery degradation by developing a hybrid physics-based and data-driven framework for online state of health ($SOH$) diagnosis and remaining useful life ($RUL$) prognosis. It builds a physics backbone with statistical learning as an augmentation to achieve better generalizability, interpretable predictions, and well-calibrated uncertainty. Key contributions include evaluating ECM variants (ECM/SPMe/DFN) with parameter-identifiability uncertainty, integrating physics-driven degradation models with residual ML terms, and developing a data-assimilation framework for online state/parameter updates. The approach aims to enable safer, cost-effective battery management, broader deployment, and second-life opportunities through calibrated, robust prognostics.

Abstract

Advancing lithium-ion batteries (LIBs) in both design and usage is key to promoting electrification in the coming decades to mitigate human-caused climate change. Inadequate understanding of LIB degradation is an important bottleneck that limits battery durability and safety. Here, we propose hybrid physics-based and data-driven modeling for online diagnosis and prognosis of battery degradation. Compared to existing battery modeling efforts, we aim to build a model with physics as its backbone and statistical learning techniques as enhancements. Such a hybrid model has better generalizability and interpretability together with a well-calibrated uncertainty associated with its prediction, rendering it more valuable and relevant to safety-critical applications under realistic usage scenarios.

Figures (3)

• Figure 1: Parameter identifiability and sensitivity study based on a cycling dataset. Here, $\boldsymbol{y}=\boldsymbol{\mathcal{H}}[\boldsymbol{u}]+\boldsymbol{e}$ is the observation model, where $\boldsymbol{u}$ is the state vector of the battery model $\mathrm{\partial}_t \boldsymbol{u}=\boldsymbol{\mathcal{L}}[\boldsymbol{u};\boldsymbol{\theta}]$, $\boldsymbol{\mathcal{H}}[\cdot]$ is the observation operator that relates the observed quantity $\boldsymbol{y}$ to state $\boldsymbol{u}$, and $\boldsymbol{e}$ is the observation noise. Moreover, operator $\boldsymbol{\mathcal{L}}[\cdot]$ characterizes the dynamical battery model with $\boldsymbol{\theta}$ as the parameters.
• Figure 2: Fitting full battery degradation dynamics to cycling data. Compared to Figure \ref{['fig:degrade_param']}, we also have a dynamical system $\boldsymbol{\theta}$ that models the degradation explicitly with $\boldsymbol{\phi}$ being its parameters. $\boldsymbol{\mathcal{R}}_{\boldsymbol{u}}$ and $\boldsymbol{\mathcal{R}}_{\boldsymbol{\theta}}$ denote potential residual data-driven models that can be stochastic processes or ML models.
• Figure 3: Integrating the hybrid modeling approach for battery degradation diagnosis and prognosis into an online filtering/data assimilation framework asch_data_2016.