Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating Physics-Based Modeling with Machine Learning for Lithium-Ion Batteries

Hao Tu, Scott Moura, Yebin Wang, Huazhen Fang

TL;DR

This work tackles the challenge of accurate LiB voltage prediction across broad C-rates and aging by merging physics-based models with data-driven learning. It introduces two state-informed hybrid frameworks, HYBRID-1 and HYBRID-2, and applies them to integrate the SPMT electrochemical/thermal model and the NDC ECM with feedforward neural networks, yielding four hybrid models: SPMTNet-1, SPMTNet-2, NDCNet-1, and NDCNet-2. An aging-aware extension AA-NDCNet-1 further augments the FNN with state-of-health information to maintain accuracy through cycle life. Across extensive simulations and experiments, these hybrids achieve high voltage predictive accuracy at low to very high C-rates with improved computational efficiency, outperforming pure ML and standard physics-based models, and demonstrating robustness to aging effects. These results support practical deployment in battery management tasks, including high-power applications and aging-aware monitoring, with avenues for further refinement via architecture search and recurrent variants.

Abstract

Mathematical modeling of lithium-ion batteries (LiBs) is a primary challenge in advanced battery management. This paper proposes two new frameworks to integrate physics-based models with machine learning to achieve high-precision modeling for LiBs. The frameworks are characterized by informing the machine learning model of the state information of the physical model, enabling a deep integration between physics and machine learning. Based on the frameworks, a series of hybrid models are constructed, through combining an electrochemical model and an equivalent circuit model, respectively, with a feedforward neural network. The hybrid models are relatively parsimonious in structure and can provide considerable voltage predictive accuracy under a broad range of C-rates, as shown by extensive simulations and experiments. The study further expands to conduct aging-aware hybrid modeling, leading to the design of a hybrid model conscious of the state-of-health to make prediction. The experiments show that the model has high voltage predictive accuracy throughout a LiB's cycle life.

Integrating Physics-Based Modeling with Machine Learning for Lithium-Ion Batteries

TL;DR

This work tackles the challenge of accurate LiB voltage prediction across broad C-rates and aging by merging physics-based models with data-driven learning. It introduces two state-informed hybrid frameworks, HYBRID-1 and HYBRID-2, and applies them to integrate the SPMT electrochemical/thermal model and the NDC ECM with feedforward neural networks, yielding four hybrid models: SPMTNet-1, SPMTNet-2, NDCNet-1, and NDCNet-2. An aging-aware extension AA-NDCNet-1 further augments the FNN with state-of-health information to maintain accuracy through cycle life. Across extensive simulations and experiments, these hybrids achieve high voltage predictive accuracy at low to very high C-rates with improved computational efficiency, outperforming pure ML and standard physics-based models, and demonstrating robustness to aging effects. These results support practical deployment in battery management tasks, including high-power applications and aging-aware monitoring, with avenues for further refinement via architecture search and recurrent variants.

Abstract

Mathematical modeling of lithium-ion batteries (LiBs) is a primary challenge in advanced battery management. This paper proposes two new frameworks to integrate physics-based models with machine learning to achieve high-precision modeling for LiBs. The frameworks are characterized by informing the machine learning model of the state information of the physical model, enabling a deep integration between physics and machine learning. Based on the frameworks, a series of hybrid models are constructed, through combining an electrochemical model and an equivalent circuit model, respectively, with a feedforward neural network. The hybrid models are relatively parsimonious in structure and can provide considerable voltage predictive accuracy under a broad range of C-rates, as shown by extensive simulations and experiments. The study further expands to conduct aging-aware hybrid modeling, leading to the design of a hybrid model conscious of the state-of-health to make prediction. The experiments show that the model has high voltage predictive accuracy throughout a LiB's cycle life.
Paper Structure (16 sections, 26 equations, 10 figures, 2 tables)

This paper contains 16 sections, 26 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Contributions
  4. Organization
  5. Hybrid Physics-ML Modeling for LiBs
  6. The Proposed Hybrid Modeling Frameworks
  7. The FNN Model
  8. Hybrid Modeling via SPMT+FNN
  9. The SPMTNet-1 and SPMTNet-2 Models
  10. Simulation Validation
  11. Hybrid Modeling via NDC+FNN
  12. The NDCNet-1 and NDCNet-2 Models
  13. Experimental Validation
  14. Aging-Aware Hybrid Modeling
  15. Discussion
  16. ...and 1 more sections

Figures (10)

  • Figure 1: Comparison of physics-based models for LiB and their applicable current range.
  • Figure 2: Block diagrams of (a) the HYBRID-1 framework and (b) the HYBRID-2 framework.
  • Figure 3: FNN architecture with two fully connected hidden layers.
  • Figure 4: Block diagrams of (a) the SPMTNet-1 model and (b) the SPMTNet-2 model.
  • Figure 5: Testing results of the SPMTNet-1 and SPMTNet-2 models.
  • ...and 5 more figures

Theorems & Definitions (4)

  • Remark 1
  • Remark 2
  • Remark 3
  • Remark 4