Project 1-2

In this project, we aim to develop a hybrid AI-based modeling approach for lithium-ion batteries. The potential of machine-learning methods in a wide range of areas has motivated its recent use in the context of computational physics. In particular, deep learning provides efficient and novel modeling approaches based on learning certain tasks from examples. This is of interest, especially in the context of complex multiscale systems, where the underlying governing physics is not completely known, or the computational cost required for simulation through conventional numerical methods is high. Another aspect that promotes the usage of deep-learning approaches is the need for fast solvers that can be implemented in iterative tasks such as optimization and control.

The recent advancement in physics-informed machine learning led to a set of computational frameworks well suited for the solution of forward and inverse problems related to several different types of partial differential equations (PDEs). This project aims to develop models based on data and the available physical knowledge for lithium-ion batteries through physics-informed machine learning, providing a hybrid AI-based modeling approach.

Specific objectives:

• Development of efficient computational methods based on deep learning for fast and accurate modeling of lithium-ion batteries
• Including physical laws into the deep-learning-based model to enhance accuracy, generalizability, and robustness
• Preparation of the dataset required for model development and designing evaluation (in close collaboration with other projects)
• Comparison of the deep-learning-based model with the conventional numerical methods regarding efficiency and accuracy
• Implementation of the developed model for optimization of lithium-ion batteries

Expected outcome:

• Physics-informed machine learning tools for modeling lithium-ion batteries
• A computational framework for the development of hybrid-AI-based techniques
• A concrete set of examples and datasets for the development and evaluation of deep-learning-based models for lithium-ion batteries