CN-121565342-B - Reverse design method, device, equipment and storage medium of material structure

CN121565342BCN 121565342 BCN121565342 BCN 121565342BCN-121565342-B

Abstract

The application provides a reverse design method, device and equipment of a material structure and a storage medium, and relates to the field of material design. And constructing a three-dimensional feature vector matrix of the sample material, wherein the three-dimensional feature vector in the matrix is used for representing the three-dimensional microstructure features of the sample material. Extracting macroscopic performance parameters of the sample material, and taking the macroscopic performance parameters as model condition information. And forming training data pairs by the three-dimensional feature vectors and the model condition information. And carrying out iterative training on the conditional diffusion model by using the training data pair until reaching the training ending condition to obtain the target conditional diffusion model. And inputting the target macroscopic performance parameters into a target condition diffusion model to obtain a target three-dimensional microstructure meeting the target macroscopic performance parameter conditions. The three-dimensional microstructure of the material can be reversely designed according to the required performance, the material development efficiency is improved, the development cost is reduced, and the designed structure can more accurately meet the required performance requirement and be more fit with a real process.

Inventors

SHEN PAN
MA LIMING
LI WEI
WANG HAN
CHEN LIPENG

Assignees

之江实验室

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (9)

1. A method of reverse engineering a material structure, the method comprising: Constructing a three-dimensional feature vector matrix of the sample material, wherein three-dimensional feature vectors in the three-dimensional feature vector matrix are used for representing three-dimensional microstructure features of the sample material; extracting macroscopic performance parameters of the sample material, and taking the macroscopic performance parameters as model condition information; Forming training data pairs by the three-dimensional feature vector matrix and the model condition information; Performing iterative training on a pre-constructed conditional diffusion model by using the training data pair until a training ending condition is reached to obtain a target conditional diffusion model; inputting target macroscopic performance parameters into the target conditional diffusion model to obtain a target three-dimensional microstructure meeting the requirements of the target macroscopic performance parameters; Constructing a three-dimensional eigenvector matrix of the sample material, comprising: Extracting structural parameters of the sample material from the target document, wherein the structural parameters comprise at least one of grain characteristic parameters and phase characteristic parameters; Discretizing the representative volume units to obtain a three-dimensional voxel structure formed by three-dimensional voxel units; Generating a three-dimensional feature vector corresponding to the three-dimensional voxel unit according to the structural parameters; Determining a target position of a corresponding three-dimensional feature vector in the three-dimensional feature vector matrix according to the position of the three-dimensional voxel unit in the three-dimensional voxel structure; and according to each target position, forming each three-dimensional feature vector into the three-dimensional feature vector matrix.
2. The method of reverse engineering a material structure according to claim 1, wherein the structural parameters include the grain characteristic parameters and the phase characteristic parameters, the grain characteristic parameters include at least a grain direction, the phase characteristic parameters include at least a phase type, and generating the three-dimensional characteristic vector corresponding to the three-dimensional voxel unit according to the structural parameters includes: Determining Euler angles corresponding to the grain directions of the three-dimensional voxel units, and converting the Euler angles into quaternions; obtaining a pre-constructed mapping relation, wherein the mapping relation is the corresponding relation between the phase type and the number; Determining a target number corresponding to the three-dimensional voxel unit according to the mapping relation; and combining the quaternion and the target number to obtain the three-dimensional feature vector.
3. The method of reverse engineering a material structure according to claim 1, wherein extracting macroscopic performance parameters of the sample material comprises: Extracting a stress-strain curve of the sample material from the target document; performing parameter fitting on the stress-strain curve data to determine a constitutive equation of the sample material; Determining mechanical property parameters of the sample material based on the constitutive equation; and taking the mechanical property parameter as the macroscopic property parameter.
4. The method of reverse engineering a material structure according to claim 1, wherein generating a representative volume element corresponding to the sample material based on the structural parameters comprises: Taking the structural parameters as simulation constraint conditions; based on the simulation constraint conditions, generating representative volume units with three-dimensional microstructures meeting the conditions corresponding to the structural parameters through a microstructure generating tool, wherein the number of the representative volume units is multiple, and the three-dimensional microstructures of the representative volume units are different from each other.
5. The method of reverse engineering a material structure according to claim 1, wherein training the pre-constructed conditional diffusion model using the training data pair until reaching a training end condition to obtain a target conditional diffusion model, comprises: Mapping the three-dimensional feature vector matrix to a potential space through a pre-constructed encoder to obtain a corresponding potential space representation; performing conditional forward diffusion to add real noise to the potential spatial representation to obtain the first A noise potential representation of the step; Inputting the model condition information corresponding to the three-dimensional feature vector matrix into a coding network to obtain a condition embedded vector; Potential representation of the noise, number of steps of diffusion The conditional embedded vector is input into a denoising network to perform reverse denoising, so as to obtain predicted noise; constructing a loss function according to the predicted noise and the real noise; Updating parameters of the encoder and the denoising network with the aim of minimizing the loss function to optimize the conditional diffusion model.
6. The method of reverse engineering a material structure according to claim 1, wherein when the number of target three-dimensional microstructures is plural, after determining a target three-dimensional microstructure that meets the target macroscopic performance parameter, further comprising: Performing macroscopic performance simulation on the target three-dimensional microstructure, and determining actual macroscopic performance parameters; Comparing the target macroscopic performance parameter with the actual macroscopic performance parameter to determine a performance deviation value of the target three-dimensional microstructure, wherein the performance deviation value is used for representing the degree of difference between the target macroscopic performance parameter and the actual macroscopic performance parameter; and taking the target three-dimensional microstructure with the minimum performance deviation value as a final target three-dimensional microstructure.
7. A reverse engineering apparatus for a material structure, the apparatus comprising: the matrix construction module is used for constructing a three-dimensional feature vector matrix of the sample material, wherein the three-dimensional feature vector in the three-dimensional feature vector matrix is used for representing the three-dimensional microstructure features of the sample material; the condition construction module is used for extracting macroscopic performance parameters of the sample material and taking the macroscopic performance parameters as model condition information; The training pair construction module is used for constructing training data pairs by the three-dimensional feature vector matrix and the model condition information; The model training module is used for carrying out iterative training on a pre-constructed conditional diffusion model by utilizing the training data pair until reaching a training end condition to obtain a target conditional diffusion model; The structural design module is used for inputting target macroscopic performance parameters into the target conditional diffusion model to obtain a target three-dimensional microstructure meeting the requirements of the target macroscopic performance parameters; the matrix construction module comprises: The document acquisition module is used for acquiring a target document in the field of alloy materials; the structure parameter extraction module is used for extracting the structure parameters of the sample material from the target document, wherein the structure parameters comprise at least one of grain characteristic parameters and phase characteristic parameters; The volume unit generation module is used for generating a representative volume unit corresponding to the sample material based on the structural parameters; the discrete module is used for discretizing the representative volume units to obtain a three-dimensional voxel structure formed by three-dimensional voxel units; the vector generation module is used for generating a three-dimensional feature vector corresponding to the three-dimensional voxel unit according to the structural parameters; The target position determining module is used for determining the target position of the corresponding three-dimensional feature vector in the three-dimensional feature vector matrix according to the position of the three-dimensional voxel unit in the three-dimensional voxel structure; the matrix generation module is used for forming a three-dimensional feature vector matrix from the three-dimensional feature vectors according to each target position.
8. An apparatus, comprising: A memory for storing a computer program; A processor for implementing the steps of the method for reverse engineering a material structure according to any one of claims 1 to 6 when executing said computer program.
9. A storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method of reverse engineering a material structure according to any one of claims 1 to 6.

Description

Reverse design method, device, equipment and storage medium of material structure Technical Field The application relates to the field of material design, in particular to a reverse design method, a reverse design device, reverse design equipment and a storage medium of a material structure. Background With the rapid development of the fields of aerospace, automobile industry, energy equipment and the like, the performance requirements on various materials are also becoming severe. The traditional material development follows a trial-and-error method, namely, a large number of candidate materials are prepared through experiments, the performance of each candidate material is determined through experimental tests or mathematical models, and materials meeting the required performance are selected from the candidate materials. However, the method only predicts the performance of the prepared material, and can not reversely customize the material according to the required performance, and has the defects of low development efficiency and high cost. Therefore, it is necessary to provide a reverse design method of a material structure, which improves the material development efficiency and reduces the development cost. Disclosure of Invention In view of this, the present application provides a method for reverse engineering a material structure, the method comprising: Constructing a three-dimensional feature vector matrix of the sample material, wherein three-dimensional feature vectors in the three-dimensional feature vector matrix are used for representing three-dimensional microstructure features of the sample material; extracting macroscopic performance parameters of the sample material, and taking the macroscopic performance parameters as model condition information; Forming training data pairs by the three-dimensional feature vector matrix and the model condition information; Performing iterative training on a pre-constructed conditional diffusion model by using the training data pair until a training ending condition is reached to obtain a target conditional diffusion model; And inputting the target macroscopic performance parameters into the target conditional diffusion model to obtain the target three-dimensional microstructure meeting the requirements of the target macroscopic performance parameters. Optionally, constructing a three-dimensional eigenvector matrix of the sample material includes: Acquiring a target document in the field of alloy materials; extracting structural parameters of the sample material from the target document, wherein the structural parameters comprise at least one of grain characteristic parameters and phase characteristic parameters; generating a representative volume element corresponding to the sample material based on the structural parameter; discretizing the representative volume units to obtain a three-dimensional voxel structure formed by three-dimensional voxel units; Generating a three-dimensional feature vector corresponding to the three-dimensional voxel unit according to the structural parameters; Determining a target position of a corresponding three-dimensional feature vector in the three-dimensional feature vector matrix according to the position of the three-dimensional voxel unit in the three-dimensional voxel structure; and according to each target position, forming each three-dimensional feature vector into the three-dimensional feature vector matrix. Optionally, the structural parameter includes the grain characteristic parameter and the phase characteristic parameter, the grain characteristic parameter includes at least a grain direction, the phase characteristic parameter includes at least a phase type, and the generating, according to the structural parameter, a three-dimensional characteristic vector corresponding to the three-dimensional voxel unit includes: Determining Euler angles corresponding to the grain directions of the three-dimensional voxel units, and converting the Euler angles into quaternions; obtaining a pre-constructed mapping relation, wherein the mapping relation is the corresponding relation between the phase type and the number; Determining a target number corresponding to the three-dimensional voxel unit according to the mapping relation; and combining the quaternion and the target number to obtain the three-dimensional feature vector. Optionally, extracting macroscopic performance parameters of the sample material includes: Extracting a stress-strain curve of the sample material from the target document; performing parameter fitting on the stress-strain curve data to determine a constitutive equation of the sample material; Determining mechanical property parameters of the sample material based on the constitutive equation; and taking the mechanical property parameter as the macroscopic property parameter. Optionally, generating a representative volume unit corresponding to the sample material based on the structural parameter inc