Physics-Informed Neural Networks for Modeling Li-ion Batteries: Solving the Single Particle Model Without Labeled Data

Abstract

Abstract Li-ion batteries are garnering significant attention due to the electrification of critical sectors. High-fidelity battery cell models have proven effective in assessing performance and optimizing design, alleviating the financial burden associated with extensive experimental procedures. However, the computational costs associated with such simulations can become prohibitive, particularly when numerous iterations are required or when integration into small devices, such as battery management systems, is necessary. To address these challenges and provide an alternative to traditional methods such as finite element and finite volume solvers, we propose the development of an algorithm that utilizes Physics-Informed Neural Networks (PINNs) to solve the Single Particle Model across multiple parameter ranges. A notable advantage of this machine learning approach is its capacity to generate competitive solutions post-training by relying solely on the governing equations, without the necessity for experimental or simulation data. Additionally, the lightweight nature of the model indicates its potential for embedding within small-scale devices.