EvoSim: AI System for Advanced Battery Modeling Solutions

Paper: EvoSim: Learning to Model, Modeling to Learn

Authors: Yun-Wei Song, Jinkai Tao, Jun-Dong Zhang, Rui Zhang, Yi-Min Wu, Qiang Zhang

Executive Summary

EvoSim is a new AI system designed to enhance the process of creating physics-based models, particularly in the field of battery technology. Traditional methods for developing these models involve selecting appropriate physical processes, defining the states and equations governing these processes, and determining relevant parameters through experiments. However, existing AI technologies struggle to make these complex decisions on their own.

EvoSim addresses this limitation by autonomously evolving based on experimental results. It utilizes discrepancies found in experimental data to refine the mechanisms and equations it employs. This means that when the model predicts outcomes that don't match actual data, EvoSim learns from these errors and adjusts its approach. Additionally, it tests the plausibility of its physical models against separate sets of experimental data to ensure reliability.

The research evaluated EvoSim’s performance on two specific tasks related to battery modeling. First, it was tasked with predicting the onset of lithium metal plating at varying temperatures and charge rates. The model achieved a mean absolute error of 1.79% in state of charge. Second, it performed dynamic voltage predictions under conditions simulating vehicle operation, resulting in a root mean square error of just 7.62 mV. Notably, these results exceeded the accuracy of models previously developed by human experts.

The self-evolution capability of EvoSim led to a significant reduction in both model and physics-related errors, improving accuracy by approximately 36% compared to baseline measures. This advancement is crucial for industries relying on precise modeling for product development and research, as it translates experimental observations into validated models more effectively than traditional methods.

EvoSim demonstrates potential applications in various sectors where accurate modeling of physical processes is essential. In particular, the battery industry could benefit from its capabilities in optimizing battery performance and safety, which are critical factors in electric vehicle development and energy storage solutions. By automating the model refinement process, EvoSim may accelerate innovation and improve the reliability of scientific predictions, ultimately enhancing product development timelines and reducing costs.

This research, while presented in a controlled evaluation scenario, highlights a significant step forward in integrating AI into scientific modeling, potentially transforming how industries approach complex physical systems.

Academic Abstract

Physics-based models connect scientific explanation with quantitative prediction. Constructing them requires selecting physical processes, defining states and governing equations, specifying couplings, and identifying parameters from experiments. Existing AI systems remain limited in making these model structure decisions autonomously. We introduce EvoSim, a self-evolving AI scientist for physical modeling. It uses experimental discrepancies to drive mechanism and equation revisions and held-out experimental data to test physical plausibility. Exploration traces make updates to knowledge, skills, and multi-agent orchestration. This co-evolution improves physics-based models and EvoSim's ability to select mechanisms, diagnose failures, and coordinate research. We evaluate EvoSim on two industrial battery modeling tasks. It predicts lithium-metal-plating onset from 25 to 45 degrees Celsius and 2 C to 6 C with a mean absolute error of 1.79% in state of charge. Dynamic voltage prediction under vehicle driving conditions achieves a root mean square error of 7.62 mV, surpassing the reported accuracy of models developed by human experts. Self-evolution reduces model and physics errors by approximately 36% relative to baseline, demonstrating improved scientific modeling capability. EvoSim turns experimental observations into validated models and cumulative research expertise.