AI工具可凭几天测试数据预估电池寿命

Core Insights - A new AI tool based on "discovery learning" has been developed by scientists at the University of Michigan, capable of accurately predicting the cycle life of new batteries using only a few days of experimental data [1][2] - Traditional testing methods require hundreds to thousands of charge-discharge cycles, taking months or even years to determine when a battery's capacity falls below 90% of its design value, while the new AI system can estimate battery life using data from just the first 50 cycles, saving approximately 98% of time and 95% of energy consumption [1][2] Group 1 - The AI system consists of three core modules: a "learner" that proposes questions and determines which battery prototypes to build, an "interpreter" that analyzes historical data and simulates internal reactions, and a "think tank" that integrates experimental results and past experiences to predict battery cycle life [1] - The AI captures degradation trends from early data and identifies key influencing factors, such as how high temperatures affect chemical mechanisms that may be negligible in low-temperature environments [2] Group 2 - The model was validated using data from Farasis Energy's pouch batteries, demonstrating its ability to predict the performance of more complex battery structures despite being trained on simpler cylindrical batteries, indicating strong generalization capabilities [2] - The technology has potential for expansion into other dimensions of battery performance, such as safety and fast charging, and the "discovery learning" paradigm may be applicable in other fields like chemistry and materials science, which are often constrained by high costs and long research cycles [2]