Method, system and storage medium for assisting material synthesis using multimodal large models

Abstract

The present invention discloses a method, system and storage medium for assisting material synthesis using a multimodal large model. The method establishes a corresponding relationship between the embedding of material raw materials and processing processes and the embedding of material performance by constructing a multimodal large model of material raw materials and processing-performance. In the application stage, the method can realize three reasoning functions: reasoning about material performance based on material raw materials and processing processes, inferring the required material formula based on target performance, and judging the credibility of a given formula. The method and system provided by the present invention significantly improve the efficiency of material formula development while ensuring the reliability of the formula. The present invention is applicable to the formula development of various materials such as rubber, plastic, and coatings, and has important practical value.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence-assisted material development, and in particular to a method, system and storage medium for assisted material synthesis using a multimodal large model. Background Art

[0002] At present, the development of material formula mainly relies on the experience and trial and error of engineers. Engineers need to select the type and ratio of raw materials based on experience, design processing parameters, and then conduct experimental verification. If the experimental results do not meet the expected performance indicators, it is necessary to adjust the formula and process parameters and conduct the experiment again. This development model has the following problems:

[0003] Long material formula development cycle: It takes an engineer an average of 60 days to develop a material formula that meets the requirements, which seriously affects the R&D efficiency and market response speed of new products.

[0004] Difficulty in predicting material properties: Before determining the formula, it is difficult for engineers to accurately predict the performance indicators of the final product and they can only confirm it through experimental verification, which leads to a large waste of manpower and material resources.

[0005] The credibility of material formulas is difficult to assess: When a new material formula is obtained, it is difficult for engineers to determine whether the formula can achieve the claimed performance indicators before experimental verification, which increases the trial and error costs in the R&D process.

[0006] Existing technologies lack effective tools to assist engineers in rapidly developing material formulations, predicting material properties, and evaluating formulation credibility. This makes the material R&D process inefficient, limits innovation, and makes it difficult to meet the market's demand for rapid development of new materials. Summary of the invention

[0007] The purpose of the present invention is to provide a method, system and storage medium for assisting material synthesis using a multimodal large model, so as to improve the development efficiency of material formulations such as rubber, plastic, and coatings, as well as the credibility of material formulations and performance.

[0008] In order to solve the above technical problems, the present invention provides a method for using a multimodal large model to assist material synthesis, comprising the following steps:

[0009] Constructing a material raw material and processing-performance multimodal macromodel, wherein the multimodal macromodel includes embedding of material raw materials and processing process and embedding of material performance;

[0010] Using known material formula data to train the multimodal large model;

[0011] Execute reasoning applications based on trained multi-modal large models;

[0012] The steps of the reasoning application include: inputting material formula and performance, and judging the credibility of the material formula and performance;

[0013] The steps of judging the material formulation and performance credibility include:

[0014] Calculate the embedding of raw materials and processing processes in material formulas and establish mapping relationships with material properties;

[0015] Conduct multi-dimensional assessment:

[0016] Theoretical evaluation based on physicochemical properties;

[0017] Statistical evaluation based on historical data;

[0018] Performance evaluation based on multi-scale simulations; and

[0019] Dynamically adjust the weights based on the importance of performance indicators to perform weighted fusion and obtain the material performance probability vector;

[0020] Calculating the distance between the probability vector and the performance vector claimed by the formulation;

[0021] The credibility of the recipe is judged based on the distance, and key factors affecting the credibility are analyzed.

[0022] Optionally, the step of training the multimodal large model using known material formula data includes: taking material raw materials and processing processes as input, taking performance indicators of the synthesized material as output, and establishing a corresponding relationship between the embedding of the material raw materials and processing processes and the embedding of material properties.

[0023] Optionally, the material includes at least one of rubber, plastic or paint.

[0024] Optionally, the theoretical evaluation based on physicochemical properties includes: physical constraint verification based on physicochemical properties; the statistical evaluation based on historical data includes: calculating the confidence interval of the current formula in historical experimental data and cross-validating using industry standard databases and historical successful formula libraries; the performance evaluation based on multi-scale simulation includes: multi-scale analysis combining microscopic material simulation with macroscopic material performance prediction.

[0025] Optionally, the reasoning application further includes:

[0026] Input the raw materials and processing process, and infer the properties of the synthesized material;

[0027] Input the target material properties and infer the required raw materials and processing procedures.

[0028] Optionally, the step of inferring the performance of the synthesized material includes:

[0029] Inputting raw materials and processing processes into the multimodal macromodel;

[0030] Obtaining multiple possible material performance results output by the multi-modal large model;

[0031] The multiple possible material performance results are screened to obtain the material performance that best meets the actual conditions.

[0032] Optionally, the step of inferring the required materials and processing procedures includes:

[0033] Building a vector database based on known material formulas, wherein each vector in the vector database corresponds to a raw material and a processing process of a material formula;

[0034] Calculate the embedding vector of the target material according to its input properties;

[0035] Using an orthogonal method in the vector database to search for multiple material formulas closest to the embedded vector;

[0036] The final raw materials and processing procedures are determined from the multiple material recipes.

[0037] The present invention also provides a system for implementing the above-mentioned method of using a multimodal large model to assist material synthesis, comprising:

[0038] A model building module is used to build a material raw material and processing-performance multi-modal large model, wherein the multi-modal large model includes the embedding of material raw materials and processing process and the embedding of material performance;

[0039] A training module is used for training using a known material formula, wherein the material raw material and the processing process are used as input, and the performance index of the synthesized material is used as output, and a corresponding relationship between the embedding of the material raw material and the processing process and the embedding of the material performance is established;

[0040] The inference module is used to perform inference applications based on the trained multimodal large model.

[0041] Optionally, the reasoning module includes:

[0042] The performance inference unit is used to infer the performance of the synthesized material based on the input material raw materials and processing process;

[0043] The recipe reasoning unit is used to infer the required raw materials and processing procedures based on the input target material properties;

[0044] The credibility judgment unit is used to judge the credibility of the input material formula and performance.

[0045] Optionally, also include:

[0046] Vector database, used to store vectors corresponding to known material formulas;

[0047] The human-computer interaction module is used to receive manual screening results and confirm operations.

[0048] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method for using a multimodal large model to assist in material synthesis as described above are implemented.

[0049] Compared with the prior art, the present invention has at least the following beneficial effects:

[0050] By adopting the method provided by the present invention, the multimodal large model technology is applied to the field of material synthesis, realizing the intelligent development of material formula. By establishing the correlation between the raw materials and processing process of the material and the material performance, the method can effectively predict the material performance, or reversely infer the required material formula according to the target performance, which can significantly improve the efficiency of material research and development.

[0051] Furthermore, the present invention improves the accuracy of material formula reasoning by introducing a vector database and an orthogonal search method; by setting up a human-computer interaction link and having professional engineers participate in screening and confirmation, the hallucination problem that may be caused by a large model is effectively avoided, thereby ensuring the reliability of the reasoning results; at the same time, the present invention also provides a judgment mechanism for the credibility of the formula, which provides an important reference basis for the verification of the material formula. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic flow chart of a method for using a multi-modal large model to assist material synthesis in Embodiment 1 of the present invention;

[0053] Figure 2 This is a system module diagram for utilizing a multi-modal large model to assist material synthesis in Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0054] The following is a more detailed description of a method, system and storage medium for assisting material synthesis using a multimodal large model of the present invention in conjunction with a schematic diagram, wherein preferred embodiments of the present invention are shown, and it should be understood that those skilled in the art can modify the present invention described herein while still achieving the advantageous effects of the present invention. Therefore, the following description should be understood as being widely known to those skilled in the art, and not as a limitation of the present invention.

[0055] The present invention is described in more detail in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the present invention will become more apparent from the following description. It should be noted that the accompanying drawings are in very simplified form and are not in exact proportions, and are only used to facilitate and clearly assist in illustrating the purpose of the embodiments of the present invention.

[0056] Please refer to Figure 1 The present invention provides a method for using a multi-modal large model to assist material synthesis, comprising the steps of:

[0057] S1. Constructing a material raw material and processing-performance multimodal large model, wherein the multimodal large model includes the embedding of material raw materials and processing process and the embedding of material performance;

[0058] S2. Training the multimodal large model using known material formula data;

[0059] S3. Execute reasoning applications based on the trained multimodal large model.

[0060] The steps of reasoning application in step S3 include: inputting material formula and performance, and judging the credibility of material formula and performance;

[0061] The steps of judging the material formulation and performance credibility include:

[0062] Calculate the embedding of raw materials and processing processes in material formulas and establish mapping relationships with material properties;

[0063] Conduct multi-dimensional assessment:

[0064] Theoretical evaluation based on physicochemical properties;

[0065] Statistical evaluation based on historical data;

[0066] Performance evaluation based on multi-scale simulations; and

[0067] Dynamically adjust the weights based on the importance of performance indicators to perform weighted fusion and obtain the material performance probability vector;

[0068] Calculating the distance between the probability vector and the performance vector claimed by the formulation;

[0069] The credibility of the recipe is judged based on the distance, and key factors affecting the credibility are analyzed.

[0070] The method for using a multimodal large model to assist in material synthesis provided by the embodiment of the present invention is particularly suitable for formula development of materials such as rubber, plastic and coating.

[0071] In this embodiment, the multimodal big model used is an artificial intelligence model that can simultaneously process and understand multiple modal data such as text, images, and sounds, and can achieve cross-modal information understanding and knowledge expression. In the field of material synthesis, the multimodal big model can simultaneously process multiple types of information about materials, thereby performing a more comprehensive analysis and prediction of material properties and synthesis processes.

[0072] In a specific example, step S1 can choose to refer to the architecture design of the image-text large model: build a multimodal large model that can generate two embedded representations: one is the embedded representation of material raw materials and processing, and the other is the embedded representation of material properties. The architecture of this model refers to the image-text multimodal large model, in which the embedding method of material raw materials and processing is similar to the embedding method of images, and the embedding method of material properties is similar to the embedding method of text. A mapping relationship is established between these two embeddings, so that the model can understand the correspondence between material formula and performance.

[0073] In step S2, the material raw material and the processing process are taken as input, and the performance index of the synthesized material is taken as output, and a corresponding relationship between the embedding of the material raw material and the processing process and the embedding of the material performance is established.

[0074] During the training process, the model gradually establishes the mapping relationship between material components, processing technology and final performance by learning a large amount of data on known formulas.

[0075] In a specific example, the training process includes: using a large amount of known material formula data to train the model, taking the material raw materials and processing process as the input sequence, and taking the corresponding material performance indicators as the output sequence. Through training, the model gradually learns how to convert material formulas into embedded representations, and how to convert material properties into embedded representations, and finally establishes an accurate correspondence between the two embeddings.

[0076] In a specific example, the input data of the multimodal large model includes: text description information of the material, molecular structure diagram, spectral data, experimental data charts, etc. Through the comprehensive analysis of multimodal input data, the accuracy and reliability of the material synthesis scheme are improved.

[0077] In step S3, based on the trained multimodal large model, the following reasoning applications can also be performed:

[0078] Input the raw materials and processing process, and infer the properties of the synthesized material;

[0079] Input the target material properties and infer the required raw materials and processing procedures.

[0080] Therefore, the three types of reasoning applications include:

[0081] S3.1: Material Performance Reasoning

[0082] The reasoning process includes:

[0083] a) inputting raw materials and processing processes into the multimodal macromodel;

[0084] b) Obtain multiple possible material performance results output by a multi-modal large model;

[0085] c) Screening the multiple possible material performance results to obtain the material performance that best meets the actual conditions.

[0086] In a specific example, when a rubber formula is input, the model may output multiple sets of possible performance indicators, such as hardness, tensile strength, elongation, etc., and the engineer selects the most reasonable result based on professional knowledge.

[0087] S3.2: Material Recipe Reasoning

[0088] The reasoning process includes:

[0089] a) constructing a vector database based on known material formulas, wherein each vector in the vector database corresponds to a raw material and a processing process of a material formula;

[0090] b) calculating the embedding vector of the target material according to the input target material properties;

[0091] c) searching the vector database for multiple material formulas closest to the embedded vector using an orthogonal method;

[0092] d) Determining the final raw materials and processing process from the multiple material recipes.

[0093] In a specific example, when a rubber material with a specific hardness and tensile strength needs to be developed, the system will find the closest formulas in the vector database for engineers to choose.

[0094] S3.3: Formula credibility judgment

[0095] The steps of judging the material formulation and performance credibility specifically include:

[0096] a) Construct embedded representation: Calculate the embedding of raw materials and processing processes in the material formula and establish a mapping relationship with material properties;

[0097] b) Conduct multi-dimensional assessment:

[0098] Physical constraint verification based on physicochemical properties;

[0099] Calculate the confidence interval of the current formula in historical experimental data;

[0100] Cross-validation using industry standard databases and historical successful formula libraries;

[0101] Multi-scale analysis combining microscopic material simulations with macroscopic material property predictions; and

[0102] c) Perform credibility calculation:

[0103] Dynamically adjust the weights based on the importance of performance indicators to perform weighted fusion and obtain the material performance probability vector;

[0104] Calculating the distance between the probability vector and the performance vector claimed by the formulation;

[0105] The credibility of the recipe is judged based on the distance, and key factors affecting the credibility are analyzed.

[0106] In a specific embodiment, the step of determining the material formulation and performance credibility specifically includes:

[0107] First, the embedded representation is constructed. The system calculates the embedding vectors of the raw materials and processing processes in the material formula, and establishes a mapping relationship with the material properties. This step ensures that the model can accurately capture the relationship between raw materials, processing methods and final performance.

[0108] Secondly, the system evaluates the material formula based on multi-dimensional criteria:

[0109] In the physical constraint verification phase, the system screens based on the physical and chemical properties of known materials. For example, for rubber materials, the system will check the compatibility between components, the rationality of cross-linking process parameters, etc., screen out formulas that do not conform to scientific laws, and ensure the feasibility of material design.

[0110] In the experimental data consistency analysis, the confidence interval of the formula in the historical experimental data set is calculated. Taking the rubber formula as an example, if the predicted tensile strength exceeds the 95% confidence interval of the historical data, the system will increase the uncertainty score accordingly to quantify the degree of deviation.

[0111] In the knowledge base comparison phase, the system calls the industry standard database for verification to determine whether the formula is significantly different from the existing feasible solutions, thereby improving the robustness of the prediction. For example, for rubber formulas for specific purposes, the system will check whether they meet the performance requirements specified in relevant national standards.

[0112] In multi-scale predictive analysis, the system combines microscopic methods such as molecular dynamics simulations and density functional theory calculations with macroscopic performance predictions to ensure that the reasoning process remains consistent at different scales and avoid the error accumulation that may result from single-scale analysis, such as simulating and analyzing the effect of cross-linked network structures on the mechanical properties of materials.

[0113] Furthermore, the reliability is quantitatively calculated:

[0114] Dynamic weights are set according to the importance of different performance indicators, and the evaluation results of each dimension are weighted and fused to obtain a more representative material performance probability vector.

[0115] Furthermore, a variety of mathematical methods are used to calculate the performance vector difference, that is, to calculate the distance between the weighted probability vector and the formula's claimed performance vector, in order to quantify the credibility level of the formula. For example, the Euclidean distance is used to evaluate the degree of deviation of the hardness prediction value, and the KL divergence is used to evaluate the difference in the tensile property distribution.

[0116] Use SHAP (Shapley Additive Explanations) or LIME (Local Interpretable Model-Agnostic Explanations) methods to analyze influencing factors, so as to accurately optimize the formula and improve the reliability and scientificity of material design. For example, analyze whether the reason why a formula has low credibility is that the content of a specific component exceeds the reasonable range.

[0117] Finally, evaluation conclusions and optimization suggestions are output based on the calculation results: for formulas with higher credibility, they are directly recommended for use; for formulas with medium credibility, experimental verification is recommended, and optimization suggestions are provided, such as adjusting the proportion of specific ingredients or optimizing processing parameters; for formulas with lower credibility, correction or rejection is recommended, and an optimization path based on historical data is provided to improve R&D efficiency and reduce trial and error costs.

[0118] In a specific example, for formulas with a credibility higher than 0.9, the system directly recommends their use; for formulas with a credibility between 0.7-0.9, the system will give specific optimization suggestions; for formulas with a credibility lower than 0.7, the system will provide an optimization path based on historical successful cases.

[0119] The method provided by the embodiment of the present invention applies multimodal large model technology to the field of material synthesis, realizing intelligent material formula development. By establishing the correlation between material raw materials and processing processes and material properties, the method can effectively predict material properties, or reversely infer the required material formula based on the target performance, significantly improving the efficiency of material research and development.

[0120] Embodiment 2

[0121] This embodiment provides a system for using a multi-modal large model to assist material synthesis. Please refer to Figure 2 , specifically including model building module, training module and reasoning module.

[0122] The model building module is used to build a material raw material and processing-performance multimodal large model, and the multimodal large model includes the embedding of material raw materials and processing process and the embedding of material performance.

[0123] The training module is connected to the model building module and is used to train the multimodal large model using known material formulas.

[0124] Specifically, the material raw material and the processing process are taken as input, and the performance index of the synthesized material is taken as output, and a corresponding relationship between the embedding of the material raw material and the processing process and the embedding of the material performance is established;

[0125] The reasoning module is connected to the training module and specifically includes:

[0126] Performance reasoning unit: used to infer the performance of the synthesized material based on the input material raw materials and processing process.

[0127] Recipe reasoning unit: used to infer the required raw materials and processing procedures based on the input target material properties.

[0128] Credibility judgment unit: used to judge the credibility of input material formula and performance.

[0129] The system for assisting material synthesis using a multimodal large model provided in this embodiment also includes a vector database, which is connected to the reasoning module and is used to store vectors corresponding to known material formulas.

[0130] The system for assisting material synthesis using a multimodal large model provided in this embodiment also includes a human-computer interaction module, which is connected to the reasoning module and is used to receive manual screening results and confirmation operations.

[0131] In a specific example, the system can be deployed on a cloud server and provide services through a web interface. Engineers can access the system through a browser to perform material development work.

[0132] Accordingly, other embodiments of the present application may also provide a computer-readable storage medium in which computer executable instructions are stored, and when the computer executable instructions are executed by the processor, the various method embodiments of the present application are implemented. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be a computer-readable instruction, a data structure, a module of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0133] In summary, the present invention provides a method, system and storage medium for using a multimodal large model to assist in material synthesis. Traditional material formula development requires engineers to spend about 60 days, while the use of the method and system of the present invention can improve development efficiency. By reducing the number of experiments and shortening the development cycle, the cost of material research and development is greatly reduced. By combining the learning of multimodal large models with the professional judgment of engineers, the accuracy and reliability of formula development are improved. The system stores and utilizes historical formula data through a vector database, thereby achieving effective accumulation and inheritance of material development knowledge. The method of the present invention is not only applicable to rubber materials, but can also be extended to the formulation development of other materials such as plastics and coatings.

[0134] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for using a multimodal large model to assist material synthesis, characterized in that: The following steps are involved: Constructing a material raw material and processing-performance multimodal macromodel, wherein the multimodal macromodel includes embedding of material raw materials and processing process and embedding of material performance; Using known material formula data to train the multimodal large model; Execute reasoning applications based on trained multi-modal large models; The steps of the inference application include: Input the raw materials and processing process, and infer the properties of the synthesized material; Input the target material properties and infer the required raw materials and processing process; Input the material formula or performance derived by inference, and judge the credibility of the material formula or performance; The step of inferring the performance of the synthesized material includes: Inputting raw materials and processing processes into the multimodal macromodel; Obtaining multiple possible material performance results output by the multi-modal large model; Screening the multiple possible material performance results to obtain the material performance that best meets the actual situation; The steps of inferring the required materials and processing procedures include: Building a vector database based on known material formulas, wherein each vector in the vector database corresponds to a raw material and a processing process of a material formula; Calculate the embedding vector of the target material according to its input properties; Using an orthogonal method in the vector database to search for multiple material formulas closest to the embedded vector; Determining final material raw materials and processing procedures from the multiple material recipes; Wherein, the step of judging the credibility of the material formula includes: Calculate the embedding of raw materials and processing processes in material formulas and establish mapping relationships with material properties; Conduct multi-dimensional assessment: Theoretical evaluation based on physicochemical properties; Statistical evaluation based on historical data; Performance evaluation based on multi-scale simulations; and Dynamically adjust the weights based on the importance of performance indicators to perform weighted fusion and obtain the material performance probability vector; Calculating the distance between the probability vector and the performance vector claimed by the formulation; The credibility of the recipe is judged based on the distance, and key factors affecting the credibility are analyzed.

2. The method for assisted material synthesis using a multimodal large model according to claim 1, characterized in that: The step of training the multimodal large model using known material formula data includes: taking material raw materials and processing as input, taking performance indicators of synthesized materials as output, and establishing a corresponding relationship between the embedding of the material raw materials and processing and the embedding of material performance.

3. The method for assisted material synthesis using a multimodal large model according to claim 1, characterized in that: The material includes at least one of rubber, plastic or paint.

4. The method for assisted material synthesis using a multimodal large model according to claim 1, characterized in that: The theoretical evaluation based on physicochemical properties includes: physical constraint verification based on physicochemical properties; the statistical evaluation based on historical data includes: calculating the confidence interval of the current formula in historical experimental data and cross-validating using industry standard databases and historical successful formula libraries; the performance evaluation based on multi-scale simulation includes: multi-scale analysis combining microscopic material simulation with macroscopic material performance prediction.

5. A system for implementing the method for using a multimodal large model to assist material synthesis as claimed in any one of claims 1 to 4, characterized in that: include: A model building module is used to build a material raw material and processing-performance multi-modal large model, wherein the multi-modal large model includes the embedding of material raw materials and processing process and the embedding of material performance; A training module is used for training using a known material formula, wherein the material raw material and the processing process are used as input, and the performance index of the synthesized material is used as output, and a corresponding relationship between the embedding of the material raw material and the processing process and the embedding of the material performance is established; The inference module is used to perform inference applications based on the trained multimodal large model to infer the properties of the synthesized material.

6. The system according to claim 5, characterized in that The reasoning module comprises: The performance inference unit is used to infer the performance of the synthesized material based on the input material raw materials and processing process; The recipe reasoning unit is used to infer the required raw materials and processing procedures based on the input target material properties; The credibility judgment unit is used to judge the credibility of the input material formula and performance.

7. The system according to claim 5, characterized in that Also includes: Vector database, used to store vectors corresponding to known material formulas; The human-computer interaction module is used to receive manual screening results and confirm operations.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for assisting material synthesis using a multimodal large model are implemented as described in any one of claims 1 to 4.

Images