Search

CN-121999943-A - Material formula design and standardization method and system based on artificial intelligence

CN121999943ACN 121999943 ACN121999943 ACN 121999943ACN-121999943-A

Abstract

The invention relates to the technical field of research and development of artificial intelligence auxiliary materials, and discloses a material formula design and standardization method and system based on artificial intelligence, wherein the method comprises the following steps of collecting multisource basic data, and constructing a four-level standardized data set structure after standardization processing; the method comprises the steps of constructing a material composition-structure-performance association constraint physical model based on a domain mechanism rule, constructing a formula-performance association model by adopting a deep neural network and combining a graph neural network, optimizing model parameters based on a variation inference method, storing results by adopting a standardized data format, and constructing a material database by combining a knowledge graph for data sharing and multiplexing. Through intelligent data acquisition, preprocessing, model construction under physical constraint and knowledge mapping association, standardized management, accurate association analysis and efficient recommendation of high polymer material formula data are realized, and research and development period is shortened and material product quality consistency and market competitiveness are improved for assisting material research and development enterprises.

Inventors

  • ZHANG XIN
  • JIANG ZHIYING
  • FANG ZHAOHUA

Assignees

  • 房兆华
  • 张鑫

Dates

Publication Date
20260508
Application Date
20260206

Claims (10)

  1. 1. The material formula design and standardization method based on artificial intelligence is characterized by comprising the following steps: s1, collecting multisource basic data, carrying out standardized processing on the data, and then constructing a four-level standardized data set structure; S2, constructing a material composition-structure-performance association constraint physical model based on a domain mechanism rule; S3, constructing a formula-performance association model by adopting a deep neural network and combining a graph neural network, and introducing a physical constraint loss function in the training process; s4, optimizing model parameters based on a variation inference method, improving generalization capability in a low data environment, and adopting a regularization strategy to reduce overfitting; And S5, storing the result by adopting a standardized data format based on the optimized association model, and constructing a material database by combining the knowledge graph for data sharing and multiplexing.
  2. 2. The artificial intelligence based material formulation and normalization method of claim 1, wherein the multi-source base data comprises: Public data collected by web crawler technology; unstructured data extracted using NLP techniques; And (3) acquiring and forming enterprise internal development and retest verification self-contained data through standardized input specifications, and client enterprise private data which is in butt joint through an API (application program interface).
  3. 3. The artificial intelligence based material formulation and normalization method according to claim 1, wherein the four-level normalization data set structure comprises: first-stage classification, namely classifying the multisource basic data subjected to standardized processing according to purposes; The second class classification is to divide the first class classification into different main performance directions on the basis of the first class classification; three-stage classification, namely dividing the process into different process categories on the basis of two-stage classification; And four-stage classification, namely dividing the three-stage classification into different auxiliary performance categories.
  4. 4. The artificial intelligence based material formulation and normalization method according to claim 1, wherein the normalization process in step S1 comprises the steps of: S11, converting data units, wherein the units of all material experimental data and calculation simulation data are unified into a preset unit system; S12, format normalization, adopting a standardized data storage format; s13, data cleaning, namely filling the missing data, removing the abnormal data and correcting measurement errors; and S14, feature normalization, and maximum and minimum normalization or Z score normalization processing is adopted for the logarithmic variable.
  5. 5. The artificial intelligence based material formulation and normalization method according to claim 1, wherein the composition-structure-performance association constraint physical model comprises: The component proportion constraint model is used for constructing a proportion constraint relation among the raw material components based on chemical reactivity, interaction relation or stability conditions among the components in the material reaction or formation process, and is used for limiting the content range of each component in the formula to be in a region where a material system can stably exist or can effectively react; The process-structure association model is used for constructing a mapping relation between process parameters and material structural characteristics based on the influence rule of the process parameters on the microstructure, molecular structure or aggregation state structure of the material in the preparation or processing process of the material, and is used for restraining the process parameters to fall into an achievable interval capable of forming target structural characteristics; And the performance prediction constraint model is used for constructing a theoretical value interval or an achievable range of the target performance based on physical or empirical association relation among material composition, structural characteristics and macroscopic performance and is used for limiting the prediction result of the model not to exceed the reasonable performance boundary of the material system under the component proportion and the process condition.
  6. 6. The artificial intelligence based material formulation and normalization method of claim 1, wherein the physical constraint loss function comprises: constraint loss based on laws of physics, for ensuring that the recipe predictions meet conservation of mass; Constraint loss is based on material characteristics, and the constraint loss is used for limiting the prediction performance of the model not to exceed the allowable range of the material theory; based on the constraint of process compatibility, the method is used for ensuring that the recommended formula and the preparation parameters can be stably realized by the existing industrial equipment.
  7. 7. The artificial intelligence-based material formulation design and standardization method according to claim 1 is characterized in that the deep neural network adopts a multi-layer fully-connected network structure for learning nonlinear performance relationship of materials, and the graph neural network is used for graph representation learning of formulation components and interaction relationship thereof and capturing mapping relationship between material structure and performance.
  8. 8. The artificial intelligence based material formulation and normalization method according to claim 1, wherein the knowledge graph comprises: Defining a knowledge graph entity, wherein the knowledge graph entity comprises a raw material entity, a formula entity, a process entity, a main performance entity, an auxiliary performance entity and an application field Jing Shiti; Establishing an entity association relationship, wherein the entity association relationship comprises a raw material-formula composition relationship, a formula-process adaptation relationship, a formula-performance mapping relationship, a performance-application scene matching relationship and a raw material-raw material synergy/antagonism relationship; the hidden association relation is mined, the similar formula recommended relation, the alternative raw material matching relation and the technological parameter optimization direction relation are mined based on the association rule mining or graph embedding algorithm, and the semantic connection of the knowledge graph is enriched.
  9. 9. The method for designing and standardizing material formulation based on artificial intelligence according to claim 1, wherein the standardized data format in step S5 adopts a data storage mode conforming to international material standards, and builds the association relationship of material properties based on knowledge graph, so as to improve the reusability and interoperability of data.
  10. 10. An artificial intelligence based material formulation and normalization system for use in an artificial intelligence based material formulation and normalization method according to any one of claims 1 to 9, comprising: The data acquisition module and the classification module are used for acquiring multi-source basic data, and after standardized processing is carried out on the data, a four-level standardized data set structure is constructed; the physical modeling module is used for constructing a material composition-structure-performance association constraint physical model based on a domain mechanism rule and establishing a mathematical mapping relation; the artificial intelligent training module is used for constructing a material performance prediction model by adopting a deep neural network and combining a graph neural network, and introducing a physical constraint loss function in the training process; The model optimization module is used for optimizing a material performance prediction model based on a variation inference method and is used for predicting generalization capability and robustness; The knowledge graph construction module is used for defining the relationship between the entity and the association, mining hidden association and constructing a high polymer material formula knowledge graph; The prediction and standardization module is used for storing the optimized prediction result, and constructing a material database by combining the knowledge graph for data sharing and multiplexing; and the user interaction module is used for receiving target parameters and constraint conditions input by a user and displaying candidate recipe results and recommendation basis.

Description

Material formula design and standardization method and system based on artificial intelligence Technical Field The invention relates to the technical field of development of artificial intelligence auxiliary materials, in particular to a material formula design and standardization method and system based on artificial intelligence. Background In the research and development and industrialization process of the polymer material, the formula design is a core link for determining the material performance, the production cost and the application suitability, and a reasonable formula is required to meet the basic performance requirements of mechanical strength, thermal stability, chemical inertness and the like, adapt to the individuation requirements of specific application scenes (such as the dielectric performance of an electronic packaging material and the biocompatibility of a medical material), and simultaneously consider the feasibility of a preparation process and the economy of large-scale production. On one hand, the traditional high polymer material formula design mainly depends on experience accumulation and trial-and-error experiments of research personnel, has obvious limitations that on the one hand, the high polymer material system is complex, the types, proportions and preparation process parameters (temperature, shear rate, pressure and reaction duration) of raw material components (such as monomers, cross-linking agents and auxiliary agents) have strong nonlinear coupling relations, the internal association of component-structure-performance is difficult to accurately capture only by experience, the formula research period is as long as months to years, the research and development cost is high, on the other hand, the formula data of different research institutions and production enterprises are stored in the form of isolated documents and local databases, the data formats are not uniform (such as disordered raw material dosage units and standard difference of performance detection), critical information is absent (such as the influence of unrecorded process fluctuation on the performance), the cross-platform compatibility is poor, the data is difficult to share and reuse, and a large amount of valuable experimental data is idle, so that effective support cannot be provided for the follow-up formula optimization. Disclosure of Invention In order to solve the problems that the traditional polymer formula design depends on experience trial and error and data are difficult to disperse and reuse, so that the research and development period is long, the cost is high and the optimization efficiency is low in the prior art, the invention discloses a material formula design and standardization method and system based on artificial intelligence. In order to achieve the above purpose, the invention adopts the following technical scheme: the material formula design and standardization method based on artificial intelligence comprises the following steps: s1, collecting multisource basic data, carrying out standardized processing on the data, and then constructing a four-level standardized data set structure; S2, constructing a material composition-structure-performance association constraint physical model based on a domain mechanism rule; S3, constructing a formula-performance association model by adopting a deep neural network and combining a graph neural network, and introducing a physical constraint loss function in the training process; s4, optimizing model parameters based on a variation inference method, improving generalization capability in a low data environment, and adopting a regularization strategy to reduce overfitting; And S5, storing the result by adopting a standardized data format based on the optimized association model, and constructing a material database by combining the knowledge graph for data sharing and multiplexing. Optionally, the multi-source base data includes: Public data collected by web crawler technology; unstructured data extracted using NLP techniques; And (3) acquiring and forming enterprise internal development and retest verification self-contained data through standardized input specifications, and client enterprise private data which is in butt joint through an API (application program interface). Optionally, the four-level standardized data set structure includes: first-stage classification, namely classifying the multisource basic data subjected to standardized processing according to purposes; The second class classification is to divide the first class classification into different main performance directions on the basis of the first class classification; three-stage classification, namely dividing the process into different process categories on the basis of two-stage classification; And four-stage classification, namely dividing the three-stage classification into different auxiliary performance categories. Optionally, the normalization process in step S1 i