Physics-Informed Neural Network with Transfer Learning for State Estimation in Lithium-Ion Batteries using the Single Particle Model with Electrolyte

TL;DR

A transfer learning framework for PINNs is proposed to efficiently solve the Single Particle Model with Electrolyte (SPMe) for lithium-ion batteries, improving training speed and generalization.

Problem

The SPMe battery model consists of nonlinear PDEs that are computationally expensive for traditional numerical methods. While PINNs can solve them, training from scratch for each new battery chemistry or condition is slow and inefficient.

Approach

A PINN incorporating SPMe physics (diffusion, transport, reactions) into its loss function was designed. A transfer learning framework was implemented where the model is first pre-trained on general electrochemical dynamics, then adapted to a target battery by transferring weights, freezing selected layers, and fine-tuning the remaining parameters.

Results & Contribution

Validation using PyBaMM showed that the proposed approach significantly reduces training time compared to training from scratch while maintaining accurate voltage prediction and electrochemical consistency. This enables efficient generalization across different battery conditions.