Integrating Machine Learning and Physics-Based Models for Battery Design Optimization

  • Lu, Wei (University of Michigan)

Please login to view abstract download link

A full engineering analysis of batteries and battery systems is essential for developing products and tools that enable the seamless integration of battery systems into vehicles and other applications. With the implementation of multiscale battery models, we anticipate that cells, modules, and packs will be amenable to predictive performance analysis for both design and manufacturing. In this talk, I will present some of our recent work on developing advanced battery simulation tools and discuss how these tools can facilitate improved battery design and management. Accurate prediction of capacity fade and battery lifetime is critical for cell design, identifying optimal operating conditions and control strategies, and guiding cell maintenance. Since capacity fade arises from multiple mechanisms, an integrated approach that considers all contributing factors is necessary. I will discuss: A comprehensive capacity fade model, including its experimental validation and applications for battery optimization. A multiscale approach that consistently couples mechanics and electrochemistry at both particle and electrode scales, enabling the simulation of various electrode phenomena. A general framework for optimizing battery health while meeting both energy and power requirements. Simulation-based battery design often requires a vast number of simulations to determine optimal design variables, which can be computationally prohibitive. I will introduce a new methodology called Self-directed Online Learning Optimization (SOLO), which integrates dynamic deep neural networks with finite element calculations. SOLO reduces computational time by up to five orders of magnitude compared with conventional heuristic methods and outperforms all state-of-the-art algorithms tested in our experiments. Finally, I will present a pruner-and-sampler approach for efficiently determining model or design parameters, which can reduce battery testing time by approximately 75%. Together, these tools provide a powerful platform for accelerating battery design, optimization, and management.