Research progress of calculation method combining machine learning model and finite element

Recently, the Institute of Mechanics, Chinese Academy of Sciences and Beijing Information Science and Technology University have cooperated to make important progress in the construction of data-driven material constitutive models and finite element combined research.
This research is the first in the world to propose and implement a calculation method based on a physical mechanism-driven machine learning model combined with finite element.
The calculation method is applied to the lithium metal that has received wide attention, and it is realized at different temperatures and Accurate description of mechanical behavior in deformed scenes.

Because of its high theoretical capacity (3860 mAh/g), low density and low potential (approximately -3.
04 V), the lithium metal electrode is an ideal anode material for lithium batteries.
Accurately understanding and characterizing the temperature, stress and rate-related deformation behavior of lithium metal anodes is the key to improving the life and reliability of lithium metal batteries.
However, due to the interaction between multiple physical fields and factors such as temperature field, force field, rate effect, and limited experimental data, there is still a lack of reliable physical models to describe the temperature-stress-rate-deformation behavior of lithium metal.
In this study, the researchers constructed a new data-driven constitutive model by combining machine learning methods with physical mechanisms.
This model can not only accurately reproduce the results of stress-strain experiments of lithium metal at different temperatures and strain rates, but also can predict the temperature-stress-rate-deformation behavior of lithium metal in a larger temperature and strain rate range.
In addition, the machine learning constitutive model can also be effectively combined with finite element calculation methods, making full use of the advantages of traditional finite element calculation in multi-physics, complex boundary and deformation system numerical simulation.
This research provides new ideas for solving the precise description of temperature, stress, rate, deformation and other behaviors of engineering materials and the development of efficient numerical methods.

Related research results were published on J.

　　Physics-driven-machine learning constitutive model predicts the deformation rate of lithium metal under different temperature and stress control, including the full service temperature and