Material performance prediction and high-throughput screening method based on deep learning force field model

By using a deep learning-based force field model, the problem of traditional empirical force fields being unable to maintain physical consistency under extreme structural distortions and the high computational cost of first-principles calculations is solved. This approach enables efficient and accurate prediction of material properties and high-throughput screening, outputting mechanically stable configurations and improving the engineering practicality of material screening.

CN122290831APending Publication Date: 2026-06-26GUIZHOU NORMAL UNIVERSITY
0 Cites 1 Cited by

Patent Information

Application Number
CN202610425122.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-02
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing high-throughput material screening methods struggle to maintain physical consistency and are prone to force state divergence when faced with large-scale unknown element combinations and extreme structural distortions. Furthermore, the high computational cost of first-principles calculations makes it difficult to support massive data evaluations, resulting in insufficient accuracy in material performance predictions and ineffective convergence of structural relaxation.

Method used

A deep learning-based force field model approach is adopted. By constructing an equivariant graph neural network and combining an automatic differentiation mechanism and physical conservation constraints, material performance prediction and high-throughput screening are performed. This includes data preprocessing, model training, structural relaxation, and performance evaluation. Multi-dimensional physical property calculations and comprehensive evaluations are used to ensure that the model maintains the equivariance of spatial geometry in translation and rotation during training, and to analyze the atomic forces and unit cell stresses that conform to conservative field laws.

Benefits of technology

While reducing computational costs, it improves the accuracy and stability of material performance prediction, can effectively handle unknown crystal configurations, output mechanically stable configurations, and enhances the engineering practical value of high-throughput material screening results.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the technical field of computational material science and material informatics, and discloses a deep learning force field model-based material performance prediction and high-throughput screening method, which comprises the following steps of: inputting material structure data into a first principle calculation module to generate an initial data set, and removing abnormal samples through multiple physical rationality detection; inputting the screened consistency data set into a force field model based on an isovariant graph neural network for training, and constructing a joint loss function by using an automatic differential mechanism and physical conservation constraint to update model parameters; inputting a candidate material into the trained model to obtain energy and stress, and outputting a mechanical stable configuration through structural relaxation; and calculating multi-dimensional physical properties of the configuration, and carrying out weighted evaluation and sequencing output through a multi-index comprehensive scoring function. According to the method, the problems that stress of a traditional force field is easy to diverge and calculation cost of a first principle is too high are solved, and low-cost unknown material high-throughput screening is achieved while physical consistency and precision are guaranteed.
Need to check novelty before this filing date? Find Prior Art