Inherently Interpretable Physics-Informed Neural Network for Battery Modeling and Prognosis.

Area of Science:

• Battery Technology
• Artificial Intelligence
• Electrochemical Engineering

Background:

• Lithium-ion batteries are crucial for modern applications, necessitating accurate modeling and prognosis for reliable operation.
• Existing data-driven models suffer from large parameters and lack interpretability, while model-based methods require extensive design parameters and are computationally complex.
• Predicting the end-of-discharge (EOD) is vital for critical mission operations.

Purpose of the Study:

• To develop a novel physics-informed neural network (PINN) for lithium-ion battery modeling and prognosis.
• To address the limitations of existing data-driven and model-based approaches by creating an interpretable and data-efficient model.
• To accurately predict the end-of-discharge (EOD) for enhanced battery operational decision-making.

Main Methods:

• Proposes a physics-informed neural network (PINN) named BattNN, designed based on the equivalent circuit model (ECM).
• Ensures the BattNN is constrained by physical laws, with its forward propagation adhering to physical principles for inherent interpretability.
• Validates the method through discharge experiments under random loading profiles, creating a new dataset.

Main Results:

• The BattNN requires minimal training data (approx. 30 samples) and has a short average training time (21.5 s).
• Achieves high prediction accuracy with few learnable parameters across three datasets.
• Demonstrates significant reductions in prediction Mean Absolute Errors (MAEs) compared to other neural networks: 77.1%, 67.4%, and 75.0%.

Conclusions:

• The proposed BattNN offers a highly accurate, interpretable, and data-efficient solution for lithium-ion battery modeling and prognosis.
• BattNN effectively bridges the gap between data-driven and model-based methods, enhancing battery reliability for critical applications.
• The study provides open-access data and code, facilitating further research and development in battery intelligence.