A performance prediction model training method, performance prediction method, structural design method, and related products of a bismaleimide resin

Abstract

The application discloses a bismaleimide resin performance prediction model training method, a performance prediction method, a structure design method and related products, relates to the technical field of bismaleimide resin, and the performance prediction model training method comprises the following steps: taking a molecular structure as input, taking a test condition as a supplementary input feature, taking a dielectric constant as a label, taking a predicted value of the dielectric constant as output, training a supervised learning model to obtain a dielectric constant prediction model; taking the molecular structure as input, taking melting point data as a label, taking a predicted value of the melting point data as output, training the supervised learning model to obtain a melting point prediction model; and taking the molecular structure as input, taking a 5% thermal decomposition temperature as a label, taking a predicted value of the 5% thermal decomposition temperature as output, training the supervised learning model to obtain a thermal decomposition temperature prediction model. The three prediction models are obtained by training the supervised learning model, so that the performance prediction of the bismaleimide resin material can be realized based on the theoretical model.

Description

Technical Field

[0001] This application relates to the field of bismaleimide resin technology, and in particular to a method for training a performance prediction model, a performance prediction method, a structural design method, and related products for bismaleimide resin. Background Technology

[0002] In recent years, with the continuous development of 5G communication technology, automotive electronics, and aerospace engineering, electronic devices in related fields have shown a significant trend towards miniaturization, lightweighting, and high-frequency and high-speed operation. This technological revolution has placed higher demands on the comprehensive performance of electronic packaging materials, among which low-dielectric-constant materials play a particularly prominent role in high-frequency circuit systems. Developing resin-based composite materials that combine low dielectric constant and high thermal stability has become an important research direction for promoting technological upgrading in the electronic information industry. Furthermore, the feasibility of material processing technology is also a core factor determining its engineering applications.

[0003] Bismaleimide resins, as a class of addition-polymer thermosetting polyimide materials with maleimide as the active end group, exhibit no low-molecular-weight byproducts during curing, and have short molecular chains and a high degree of crosslinking between crosslinking points. These materials, with their excellent thermal stability, low hygroscopicity, and good dielectric properties, coupled with their simple monomer synthesis, abundant raw material sources, and commercial cost advantages, show broad application prospects in the field of large-scale electronic device manufacturing. However, due to the high symmetry of molecular chains and the regularity of crystal structures, most bismaleimide resins have high prepolymer melting points, significantly affecting the material's processing and molding performance. Conventional strategies to improve processing performance by introducing asymmetric structures often lead to increased material polarity, decreased dielectric properties, and negatively impacted thermal stability. This interrelationship among multiple properties has become a technical bottleneck restricting the development of high-performance bismaleimide resins. Therefore, it is urgent to conduct research on the optimization design of molecular structures to effectively improve processing performance while maintaining high thermal stability and excellent dielectric properties.

[0004] To meet the comprehensive performance requirements of materials in engineering applications, researchers need to make scientific trade-offs among various performance indicators. The traditional research and development model of resin materials mainly relies on experimental trial and error. This empirical research and development process not only leads to long material development cycles and high economic costs, but also makes it difficult to achieve the goal of improving heat resistance and processing performance while reducing dielectric constant due to the complexity of the material structure-performance relationship. Summary of the Invention

[0005] The purpose of this application is to provide a training method for a performance prediction model of bismaleimide resin, a performance prediction method, a structural design method, and related products. These methods can predict the performance of bismaleimide resin materials based on theoretical models and can simultaneously consider multiple properties such as dielectric constant and heat resistance.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] In a first aspect, this application provides a method for training a performance prediction model for bismaleimide resins, including:

[0008] Acquire bismaleimide data and supplementary data; the bismaleimide data includes: the molecular structure of bismaleimide resin and corresponding experimental data; the experimental data includes: dielectric constant data, melting point, and 5% thermal decomposition temperature; the supplementary data includes: supplementary dielectric constant data and supplementary temperature data; the supplementary dielectric constant data includes: the molecular structure of several types of resin and corresponding dielectric constant data; the supplementary temperature data includes: the molecular structure of polyimide resin and corresponding melting point data and 5% thermal decomposition temperature data; the dielectric constant data includes: dielectric constant and corresponding test conditions;

[0009] Using the molecular structure as input, the test conditions as supplementary input features, the dielectric constant as a label, and the predicted value of the dielectric constant as output, a supervised learning model is trained to obtain a dielectric constant prediction model.

[0010] Using the molecular structure as input, the melting point data as labels, and the predicted value of the melting point data as output, a supervised learning model is trained to obtain a melting point prediction model.

[0011] Using the molecular structure as input, the 5% thermal decomposition temperature as a label, and the predicted value of the 5% thermal decomposition temperature as output, a supervised learning model is trained to obtain a thermal decomposition temperature prediction model.

[0012] The performance prediction model includes: the dielectric constant prediction model, the melting point prediction model, and the thermal decomposition temperature prediction model.

[0013] Secondly, this application provides a method for predicting the performance of bismaleimide resins, including:

[0014] Obtain the molecular structure of the bismaleimide resin to be predicted;

[0015] The molecular structure is input into the dielectric constant prediction model to obtain the predicted value of the dielectric constant of the bismaleimide resin to be predicted; the dielectric constant prediction model is trained by the performance prediction model training method of bismaleimide resin described above.

[0016] The molecular structure is input into the melting point prediction model to obtain the predicted value of the melting point data of the bismaleimide resin to be predicted; the melting point prediction model is trained by the performance prediction model training method of bismaleimide resin described above.

[0017] The molecular structure is input into the thermal decomposition temperature prediction model to obtain the predicted value of the 5% thermal decomposition temperature of the bismaleimide resin to be predicted; the thermal decomposition temperature prediction model is trained by the performance prediction model training method of bismaleimide resin described above.

[0018] The predicted values ​​of the dielectric constant, melting point, and 5% thermal decomposition temperature of the bismaleimide resin to be predicted are the performance prediction data.

[0019] Thirdly, this application provides a structural design method for bismaleimide resins, including:

[0020] Construct several virtual molecular structures of bismaleimide resins;

[0021] Calculate the synthetic accessibility score for each of the virtual molecular structures;

[0022] The virtual molecular structure is input into the dielectric constant prediction model to obtain the corresponding dielectric constant prediction values; the dielectric constant prediction model is trained by the performance prediction model training method of bismaleimide resin described above.

[0023] The virtual molecular structure is input into the melting point prediction model to obtain the corresponding predicted melting point data values; the melting point prediction model is trained by the performance prediction model training method of bismaleimide resin described above.

[0024] The virtual molecular structure is input into the thermal decomposition temperature prediction model to obtain the corresponding 5% thermal decomposition temperature prediction value; the thermal decomposition temperature prediction model is trained by the performance prediction model training method of bismaleimide resin described above.

[0025] Based on the predicted dielectric constant, predicted melting point, predicted 5% thermal decomposition temperature, and synthesis accessibility score, a comprehensive score is calculated for each of the virtual molecular structures.

[0026] The design structure of the bismaleimide resin was determined based on the comprehensive score.

[0027] Fourthly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the performance prediction model training method for bismaleimide resin described above, or the performance prediction method for bismaleimide resin described above, or the structural design method for bismaleimide resin described above.

[0028] Fifthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for training a performance prediction model of bismaleimide resin, the above-described method for predicting the performance of bismaleimide resin, or the above-described method for structural design of bismaleimide resin.

[0029] Sixthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for training a performance prediction model of bismaleimide resin, the above-described method for predicting the performance of bismaleimide resin, or the above-described method for structural design of bismaleimide resin.

[0030] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0031] This application provides a performance prediction model training method, performance prediction method, structural design method, and related products for bismaleimide resin. The performance prediction model training method includes: acquiring bismaleimide data and supplementary data; the bismaleimide data includes: the molecular structure of bismaleimide resin and corresponding experimental data; the experimental data includes: dielectric constant data, melting point, and 5% thermal decomposition temperature; the supplementary data includes: dielectric constant supplementary data and temperature supplementary data; the dielectric constant supplementary data includes: the molecular structure of several types of resin and corresponding dielectric constant data; the temperature supplementary data includes: the molecular structure of polyimide resin and corresponding melting point data and 5% thermal decomposition temperature data; the dielectric constant data... The data includes: dielectric constant and corresponding test conditions; using the molecular structure as input, the test conditions as supplementary input features, the dielectric constant as a label, and the predicted value of the dielectric constant as the output, a supervised learning model is trained to obtain a dielectric constant prediction model; using the molecular structure as input, the melting point data as a label, and the predicted value of the melting point data as the output, a supervised learning model is trained to obtain a melting point prediction model; using the molecular structure as input, the 5% thermal decomposition temperature as a label, and the predicted value of the 5% thermal decomposition temperature as the output, a supervised learning model is trained to obtain a thermal decomposition temperature prediction model; the performance prediction model includes: the dielectric constant prediction model, the melting point prediction model, and the thermal decomposition temperature prediction model. This application obtains three prediction models by training a supervised learning model. These three prediction models can simultaneously consider multiple properties such as dielectric constant and heat resistance, thereby enabling the prediction of various properties of bismaleimide resin materials based on theoretical models. At the same time, this application improves the accuracy of machine learning models in predicting the dielectric constant, melting point and 5% thermal decomposition temperature of bismaleimide resin by introducing experimental data of various types of resins as a supplement, thus solving the problem of small data for materials. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart illustrating a method for training a performance prediction model for bismaleimide resin according to an embodiment of this application.

[0034] Figure 2 A schematic diagram showing the data distribution of a dataset of dielectric constant, melting point, and 5% thermal decomposition temperature provided in an embodiment of this application;

[0035] Figure 3 This is a schematic diagram of the process for constructing a performance prediction model of a graph convolutional neural network and a multilayer perceptron according to an embodiment of this application;

[0036] Figure 4 A schematic diagram illustrating the accuracy of a bismaleimide resin performance prediction model provided in an embodiment of this application;

[0037] Figure 5 A schematic diagram of a virtual bismaleimide resin structure combination provided in an embodiment of this application;

[0038] Figure 6 This is a schematic diagram illustrating the results of verifying the reliability of the model provided in one embodiment of this application;

[0039] Figure 7 A schematic diagram comparing the model performance of Embodiment 4 and Comparative Examples 1 and 2, which are embodiments of this application;

[0040] Figure 8 This is a performance space diagram of a predicted candidate bismaleimide resin structure provided in an embodiment of this application.

[0041] Figure 9 A schematic diagram of the preferred bismaleimide resin structure obtained by screening according to an embodiment of this application;

[0042] Figure 10 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0044] Developing a material property prediction method based on theoretical models to enable rapid screening and design of bismaleimide resin molecular structures that combine low dielectric constant, high heat resistance, and easy processability has become a key problem that urgently needs to be solved in this field.

[0045] The prediction method of this application includes the following steps: inputting the molecular graph and adjacency matrix of the bismaleimide resin material into the prediction model to obtain the predicted performance values ​​of the bismaleimide resin; the prediction model is trained from a dataset; wherein, the dataset includes the SMILES and performance parameters of the bismaleimide resin. This application utilizes data-driven and materials genome technology to guide the rational design of resin structures, which can significantly improve the efficiency and accuracy of material research and development, efficiently develop low-cost and highly practical bismaleimide resins, and open up new paths for the development and application of low-dielectric, easily processed, and high-temperature resistant bismaleimide resins.

[0046] This application addresses the technical bottlenecks of traditional experimental trial-and-error methods in designing bismaleimide resins with conflicting properties, including low R&D efficiency and high economic costs. By constructing a high-precision predictive model, the rationality and efficiency of bismaleimide resin molecular structure design can be significantly improved, while reducing R&D costs.

[0047] However, bismaleimide resin systems suffer from a lack of experimental data, which is mostly derived from multi-component systems or actual product-level tests. This makes it difficult for existing experimental data to accurately reflect the intrinsic structure-property relationship of the material. Directly using such data to build machine learning models would significantly reduce the model's predictive accuracy and generalization ability.

[0048] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1:

[0050] In one exemplary embodiment, such as Figure 1 As shown, a method for training a performance prediction model for bismaleimide resin is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using a server as an example, and includes the following steps S1 to S4. Wherein:

[0051] S1. Obtain bismaleimide data and supplementary data; the bismaleimide data includes: the molecular structure of bismaleimide resin and corresponding experimental data; the experimental data includes: dielectric constant data, melting point, and 5% thermal decomposition temperature; the supplementary data includes: supplementary dielectric constant data and supplementary temperature data; the supplementary dielectric constant data includes: the molecular structure of several types of resin and corresponding dielectric constant data; the supplementary temperature data includes: the molecular structure of polyimide resin and corresponding melting point data and 5% thermal decomposition temperature data; the dielectric constant data includes: dielectric constant and corresponding test conditions. In this embodiment, the test conditions refer to the test frequency and test temperature.

[0052] The experimental data for bismaleimide resin materials is relatively limited, and most of it involves multi-component measurements, lacking more accurate experimental-grade structural and performance data. Models derived directly from this limited data have low accuracy and weak generalization ability. To improve model accuracy, this embodiment incorporates supplementary data from various resin structures during data collection, based on the structural characteristics of bismaleimide resin. The dielectric constant of a polymer is primarily determined by its structure and typically exhibits temperature and frequency dependence. For the dielectric constant model, this embodiment supplements it with dielectric constant data from various types of resins, including thermosetting and thermoplastic resins. It also collects test frequency and temperature data (dielectric properties are related to polymer structure and test conditions; using dielectric constant data from different types of polymers (including thermoplastic and thermosetting) for data augmentation helps the model reasonably learn the relationship between structure and dielectric constant, improving model reliability and generalization ability). For melting point and 5% thermal decomposition temperature, this embodiment supplements it with melting point and 5% thermal decomposition temperature data from polyimide resins, which also contain imides (for melting point and 5% thermal decomposition temperature, polyimide and bismaleimide share the same imide groups; this structural similarity leads to similar properties and common patterns in thermal properties. Therefore, introducing polyimide data for data augmentation helps improve model generalization ability).

[0053] Then, the dataset is cleaned to improve its quality.

[0054] The preferred method for cleaning the dataset is to cover the dielectric constant test frequency from 1*10. -5 Hz to 1.4*10 8The test data ranged from -196℃ to 380℃, with a wide data distribution that conformed to a normal distribution. For dielectric constant test frequencies with excessively large distribution skewness, the original data was replaced with the logarithm (base e), ensuring that the model better learned the relationship between the dielectric constant and the material structure and test conditions. When the same structure had different melting point data, the structure was discarded if the difference between the maximum and minimum values ​​was greater than 50℃, and the average melting point was used as the target performance for the remaining structures. The test atmosphere for the 5% thermal decomposition temperature of the resin material was set to a nitrogen atmosphere. When the same structure had different 5% thermal decomposition temperatures, the structure was discarded if the difference between the maximum and minimum values ​​was greater than 50℃, and the average 5% thermal decomposition temperature was used as the target performance for the remaining structures.

[0055] The dataset typically contains over 100 data entries. Preferably, the experimental data for the dielectric constant, melting point, and 5% thermal decomposition temperature of the bismaleimide resin are 30-60 entries each, for example, 60 entries for the dielectric constant, 51 entries for the melting point, and 30 entries for the 5% thermal decomposition temperature. The dataset also includes 100-2500 supplementary performance data entries besides the bismaleimide resin experimental data, for example, 2344 entries for the dielectric constant of various resin types (polyolefins, polyurethanes, etc.), 475 entries for the 5% thermal decomposition temperature of polyimide resin, and 141 entries for the melting point of polyimide. The introduction of this supplementary data expands the chemical space for model learning and enhances the model's generalization ability. Those skilled in the art will understand that the performance data in this dataset can generally be obtained by collecting data from publicly available literature and / or obtaining it from public databases.

[0056] S2. Using the molecular structure as input, the test conditions as supplementary input features, the dielectric constant as a label, and the predicted value of the dielectric constant as output, train the supervised learning model to obtain the dielectric constant prediction model.

[0057] S3. Using the molecular structure as input, the melting point data as a label, and the predicted value of the melting point data as output, train a supervised learning model to obtain a melting point prediction model.

[0058] S4. Using the molecular structure as input, the 5% thermal decomposition temperature as a label, and the predicted value of the 5% thermal decomposition temperature as output, train a supervised learning model to obtain a thermal decomposition temperature prediction model. The performance prediction model includes: the dielectric constant prediction model, the melting point prediction model, and the thermal decomposition temperature prediction model.

[0059] It is worth noting that there are no specific restrictions on the order of steps S2-S4.

[0060] The supervised learning model selected in this embodiment includes an input layer, an intermediate layer, and a fully connected layer. The input layer is used to encode the molecular structure in the form of a molecular graph. The intermediate layer is used to update the embedding vector of the target node by aggregating neighbor node information through continuous convolution and batch normalization operations. The fully connected layer is used to output the model with the experimental performance data corresponding to the molecular structure.

[0061] The intermediate layers can include graph convolutional layers, batch normalization layers, and pooling layers. Graph convolutional layers aggregate the feature relationships between adjacent nodes and update the node feature representations. Batch normalization layers normalize the features to accelerate the model convergence speed. Pooling layers sum the features of all nodes to obtain the entire graph feature representation.

[0062] The model used in this embodiment is a learning framework based on graph convolutional neural networks and multilayer perceptrons. The multilayer perceptron (MLP) layer consists of stacked fully connected layers and activation functions. This model incorporates two MLP layers to perform non-linear transformations on features, acquiring more complex information and improving the model's expressive power. Finally, a fully connected layer is used for the output.

[0063] Specifically, this application combines the cleaned dataset to construct a machine learning model that combines a graph convolutional neural network with a multilayer perceptron, automatically learning the relationship between the structure and properties of bismaleimide resin, and obtaining a performance prediction model for bismaleimide resin. Compared with the machine learning model based on Gaussian process regression and the XGBoost regression model based on gradient boosting tree, its prediction accuracy is higher.

[0064] In this embodiment, a learning framework based on graph convolutional neural networks and multilayer perceptrons is used. The bismaleimide resin structure is encoded in the form of a molecular graph as the input to the model. The embedding vector of the target node is updated by aggregating neighbor node information through continuous convolution and batch normalization operations. Finally, two fully connected layers are connected, and the experimental performance data corresponding to the structure is used as the model output. The dataset is divided into training and testing ratios of 8:2, and the potential relationship between the bismaleimide resin structure and performance is automatically learned.

[0065] Since the dielectric constant of polymers is usually frequency-dependent and temperature-dependent, this embodiment adds the test frequency and test temperature of the dielectric constant as supplementary features to the model input. This helps the model capture the law of change of dielectric constant with environmental conditions, and at the same time improves the model prediction accuracy, it reflects the direct relationship between the dielectric constant of bismaleimide resin polymer and macroscopic behavior.

[0066] The method of encoding the bismaleimide resin structure in the form of a molecular diagram preferably includes: representation of atomic information, representation of bond information, and representation of molecular adjacency matrix.

[0067] The method of updating the embedding vector of the target node by aggregating neighbor node information through continuous convolution and batch normalization operations preferably includes: using batch normalization immediately after each layer of graph convolution to accelerate the convergence speed of the model and improve the stability of the model.

[0068] Preferably, the connection between the two fully connected layers is: after the GCN, two fully connected layers with 300 nodes are connected. This combined use of GCN and MLP can capture local and global features in graph data, thereby improving the expressive power of the model.

[0069] In this embodiment, the dielectric constant, melting point, and 5% thermal decomposition temperature of the bismaleimide resin are expressed in a material property space manner.

[0070] Those skilled in the art should know that the 5% thermal decomposition temperature reflects the high-temperature resistance of the resin material; the melting point reflects the processing performance of the resin material.

[0071] This embodiment improves the accuracy of the machine learning model in predicting the dielectric constant, melting point, and 5% thermal decomposition temperature of bismaleimide resin by introducing experimental data from various types of resins, thus solving the problem of limited material data. Regarding the prediction of the dielectric constant, this embodiment achieves higher accuracy in the R-value of the test set. 2 The R-value is 0.88, and the mean absolute error is 0.19; regarding melting point prediction, the R-value on the test set is... 2 The R-value is 0.88, with a mean absolute error of 28℃; regarding the prediction of the 5% thermal decomposition temperature, the R-value of the test set is... 2 It can be 0.82, with an average absolute error of 22℃.

[0072] Example 2:

[0073] Based on the same inventive concept, embodiments of this application provide a method for predicting the performance of bismaleimide resins, including:

[0074] A1. Obtain the molecular structure of the bismaleimide resin to be predicted.

[0075] A2. Input the molecular structure into the dielectric constant prediction model to obtain the predicted value of the dielectric constant of the bismaleimide resin to be predicted; the dielectric constant prediction model is trained by the performance prediction model training method of bismaleimide resin described in Example 1.

[0076] A3. Input the molecular structure into the melting point prediction model to obtain the predicted value of the melting point data of the bismaleimide resin to be predicted; the melting point prediction model is trained by the performance prediction model training method of bismaleimide resin described in Example 1.

[0077] A4. Input the molecular structure into the thermal decomposition temperature prediction model to obtain the predicted value of the 5% thermal decomposition temperature of the bismaleimide resin to be predicted; the thermal decomposition temperature prediction model is trained by the performance prediction model training method of bismaleimide resin described in Example 1.

[0078] The predicted values ​​of the dielectric constant, melting point, and 5% thermal decomposition temperature of the bismaleimide resin to be predicted are the performance prediction data.

[0079] Example 3:

[0080] Based on the same inventive concept, embodiments of this application provide a structural design method for bismaleimide resin, including:

[0081] B1. Construct several virtual molecular structures of bismaleimide resins.

[0082] Specifically:

[0083] B11. Determine the gene type for the bismaleimide resin structure.

[0084] In this embodiment, gene types for bismaleimide resin structures are defined; wherein, the gene types preferably include: 1) maleic anhydride gene; 2) diamine gene with only two amino groups at both ends of the main chain; 3) ammonium acid gene containing a benzene ring in the main chain, wherein the benzene ring is connected to an amino group and a formic acid; 4) diphenol gene containing a benzene ring in the main chain, wherein the benzene ring is connected to two hydroxyl groups; 5) maleic acid diamine gene; 6) haloalkane gene containing halogen atoms at both ends of the main chain; 7) amino alcohol gene with one amino group and one hydroxyl group connected to each end of the main chain; 8) diacid gene with only two formic acid groups at both ends of the main chain.

[0085] B12. Based on the gene type, retrieve the corresponding genes from the database.

[0086] Use public databases to search for genes and obtain diverse gene pools.

[0087] The gene pool may include 700-800 diamine genes, 1-100 ammonium acid genes, 100-200 diphenol genes, 1-50 haloalkanes genes, 1-50 amino alcohol genes, and 100-200 diacid genes; for example, 788 diamine genes, 50 ammonium acid genes, 101 diphenol genes, 33 haloalkanes genes, 48 ​​amino alcohol genes, and 111 diacid genes.

[0088] B13. Based on a preset template, several genes are synthesized into several bismaleimide resins to obtain several virtual molecular structures of bismaleimide resins.

[0089] In this embodiment, a virtual molecular structure of bismaleimide resin is obtained based on the gene pool. The combination of the virtual molecular structure is achieved by using the SMARTS (SMilesARbitraryTargetSpecification) module in RDKit to construct a chemical reaction template, which is implemented by Python programming. SMARTS is a language in RDKit for describing molecular structures, providing specific symbols to describe the atoms, bonds and their connections in molecules. In this embodiment, by using the molecular connection method of the SMARTS module in Python, four reaction templates of rules (1)-(4) are constructed, realizing the reaction methods (connection methods) as shown in (1)-(4). Specifically, the SMARTS patterns of reactants and products are defined; reactants and products are combined to create a SMARTS format template for the reaction; and the chemical reaction of the molecule is performed using the reaction template.

[0090] In addition to the common route for synthesizing bismaleimide resins via the reaction of maleic anhydride and diamine, this embodiment summarizes other possible experimental synthetic routes and defines three novel reaction templates. This not only enriches the synthetic methods for bismaleimide resins but also provides more possibilities for the design of subsequent high-performance resin materials. The chemical templates include the following combination rules:

[0091] Rule (1):

[0092]

[0093] In this process, maleimide and diamine genes are used as reactants and react in a 2:1 ratio to generate bismaleimide (a common synthetic route).

[0094] Rule (2):

[0095]

[0096] In this process, maleimide, ammonium acid gene, and diphenol gene are reacted in a ratio of 2:2:1 to generate bismaleimide resin.

[0097] Rule (3):

[0098]

[0099] In this process, maleic diamine and halogenated hydrocarbons react in a 2:1 ratio to generate bismaleimide resin.

[0100] Rule (4):

[0101]

[0102] Among them, maleimide, amino alcohol gene, and diacid gene react in a 2:2:1 ratio to generate bismaleimide resin.

[0103] The virtual molecular structure of the bismaleimide resin can be more than 10,000, preferably 10,000-15,000, for example 10,744.

[0104] B2. Calculate the synthetic accessibility score for each of the virtual molecular structures.

[0105] In this embodiment, the method for evaluating the synthetic difficulty of the virtual molecular structure can be the synthetic accessibility score and the Belstein complexity index, which are conventional in the art, with the synthetic accessibility score (SA) being preferred.

[0106] B3. Input the virtual molecular structure into the dielectric constant prediction model to obtain the corresponding dielectric constant prediction values; the dielectric constant prediction model is trained by the performance prediction model training method of bismaleimide resin described in Example 1.

[0107] B4. Input the virtual molecular structure into the melting point prediction model to obtain the corresponding melting point data prediction values; the melting point prediction model is trained by the performance prediction model training method of bismaleimide resin described in Example 1.

[0108] B5. Input the virtual molecular structure into the thermal decomposition temperature prediction model to obtain the corresponding 5% thermal decomposition temperature prediction value; the thermal decomposition temperature prediction model is trained by the performance prediction model training method of bismaleimide resin described in Example 1.

[0109] B6. Calculate the comprehensive score for each of the virtual molecular structures based on the predicted dielectric constant, predicted melting point, predicted 5% thermal decomposition temperature, and synthesis accessibility score.

[0110] The formula for calculating the comprehensive score is as follows:

[0111]

[0112]

[0113] Among them, Score, ε, T M T d S A These represent the overall score, dielectric constant of the virtual molecular structure, melting point, 5% thermal decomposition temperature, and synthesis accessibility score, respectively; ε min ε max T represents the minimum and maximum values ​​of the dielectric constant of the virtual molecular structure, respectively; Mmin T Mmax T represents the minimum and maximum melting points of the virtual molecular structure, respectively; dmin T dmax S represents the minimum and maximum values ​​of the 5% thermal decomposition temperature of the virtual molecular structure, respectively; Amin S Amax These represent the minimum and maximum values ​​of the synthetic accessibility score for the virtual molecular structure, respectively.

[0114] B7. Determine the design structure of the bismaleimide resin based on the comprehensive score.

[0115] In this embodiment, the design structure of the bismaleimide resin material is determined by sorting the materials according to their comprehensive scores.

[0116] The process includes filtering the material property space after obtaining it, using a weighted scalar function.

[0117] Those skilled in the art will know that the structural information of bismaleimide resins is generally obtained by converting the structure of the bismaleimide resin to be analyzed into the corresponding SMILES input. SMILES, short for Simplified Molecular Input Line Entry Specification, is a specification that explicitly describes the molecular structure using ASCII strings.

[0118] In this embodiment, the bismaleimide resin gene types, in addition to the traditional diamine gene, also include ammonium acid gene, diphenol gene, maleic acid diamine gene, amino alcohol gene, and diacid gene, which greatly expands the chemical space of the virtual resin structure. The selected preferred structures are expected to be widely used in the aerospace field.

[0119] Example 4:

[0120] This embodiment demonstrates the performance prediction method and reliability verification of bismaleimide resin:

[0121] C1. Collect performance datasets of dielectric constant, melting point, and 5% thermal decomposition temperature from the literature. The performance distribution is as follows: Figure 2 As shown, where, Figure 2 (a) represents the dielectric constant property. Figure 2 (b) represents melting point properties. Figure 2 (c) Performance at 5% thermal decomposition temperature, where dielectric constant data includes test frequencies from 1*10⁻⁶. -5 Hz to 1.4*10 8 The data includes experimental data for 60 bismaleimide resins and 2344 various types of resins (polyolefins, polyurethanes, etc.) at test temperatures ranging from -196℃ to 380℃. The 5% thermal decomposition temperature data includes experimental data for 30 bismaleimide resins and 475 polyimide resins. The melting point data includes experimental data for 51 bismaleimide resins and 141 polyimide resins.

[0122] C2. Clean the collected dataset. For the dielectric constant test frequency, replace the original data with the data after taking the lg value; when the same structure has different melting point data, discard the structure if the difference between the maximum and minimum values ​​is greater than 50°C, and use the average melting point value as the target performance for the remaining structures; set the test atmosphere for the 5% thermal decomposition temperature of the resin material to a nitrogen atmosphere; when the same structure has different 5% thermal decomposition temperatures, discard the structure if the difference between the maximum and minimum values ​​is greater than 50°C, and use the average 5% thermal decomposition temperature value as the target performance for the remaining structures.

[0123] C3. Based on the cleaned dataset, construct a graph convolutional neural network combined with a multilayer perceptron performance prediction model, as follows: Figure 3 ,include:

[0124] (1) Divide the dataset into training set and test set in a ratio of 8:2.

[0125] (2) Molecular structure is represented in the form of molecular diagrams, specifically by the molecular adjacency matrix and the atomic number, aromaticity, number of hydrogen atoms and implicit valence of each atom in the molecule, as well as the type, spatial configuration, direction and aromaticity of bonds in the molecule.

[0126] (3) The model input is the feature vector encoded by the molecular graph and the test conditions of the dielectric constant. The corresponding performance is used as the model output. Batch normalization is used after each graph convolutional layer in the model. Finally, two fully connected layers are connected. The performance prediction model of bismaleimide resin is obtained through forward derivation and backpropagation training. The accuracy of the performance prediction model in this embodiment is as follows: Figure 4 As shown, where, Figure 4 (a) represents the accuracy of the dielectric constant prediction. Figure 4 (b) is the accuracy of melting point prediction. Figure 4(c) is the 5% accuracy of thermal decomposition temperature prediction.

[0127] C4. Input the structural information of the bismaleimide resin to be analyzed into the performance prediction model to obtain the dielectric constant, melting point and 5% thermal decomposition temperature performance data of the bismaleimide resin to be analyzed.

[0128] The bismaleimide resin to be analyzed includes existing bismaleimide resins and virtual bismaleimide resins, wherein the method for obtaining virtual bismaleimide resins includes the following steps:

[0129] (1) Define the gene types of bismaleimide resin, including maleic anhydride gene, diamine gene, ammonium acid gene, diphenol gene, maleic acid diamine gene, halohydrocarbon gene, amino alcohol gene and diacid gene.

[0130] (2) Gene pools were obtained by searching for genes in the public databases SciFinder, PubChem, and ChemSpider, including 788 diamine genes, 50 ammonium acid genes, 101 diphenol genes, 33 halohydrocarbon genes, 48 ​​amino alcohol genes, and 111 diacid genes.

[0131] (3) By combining maleic anhydride genes, diamine genes, ammonium acid genes, diphenol genes, maleic diamine genes, haloalkanes genes, amino alcohol genes, and diacid genes, 10,744 virtual bismaleimide resin structures were obtained, with the following combinations: Figure 5 As shown.

[0132] S5. Collect experimental data of existing bismaleimide resins not included in the model and compare them with the model's predicted values, such as... Figure 6 As shown. For dielectric constant data, since radar operating frequencies are high-frequency (MHz) and operating temperatures are mostly at room temperature, data at room temperature and 1MHz were used for comparison. A structure not used for model training was collected, in which the experimental and predicted dielectric constant values ​​were 2.94 and 2.96, respectively; the experimental and predicted melting point values ​​were 160℃ and 145℃, respectively; and the experimental and predicted 5% thermal decomposition temperature values ​​were 476℃ and 440℃, respectively. The predicted values ​​of the three models were quite close to the actual values, verifying the reliability of the method.

[0133] Comparative Example 1:

[0134] Modeling is performed using the same dataset as in step C3 of Example 4. The difference from Example 4 is that the Mordred descriptor of SMILES is used to build a machine learning model through XGBoost regression based on gradient boosting trees.

[0135] Comparative Example 2:

[0136] The same dataset as step C3 in Example 4 is used for modeling. The difference from Example 4 is that the Mordred descriptor of SMILES is used to build the machine learning model through Gaussian process regression (GPR).

[0137] Application Example 4

[0138] The models of Example 4 and Comparative Examples 1 and 2 were constructed and run randomly 100 times. The R-value of the resulting model test set was... 2 (Coefficient of determination) and MAE (Mean Absolute Error) are as follows: Figure 7 As shown, where, Figure 7 (a) Comparison of dielectric constant properties Figure 7 (b) Comparison of melting point properties Figure 7 (c) Comparison of performance at 5% thermal decomposition temperature. Figure 7 It can be seen that, for the dielectric constant prediction model, the average R-value of the GCN model on the test set is... 2 MAE and R are 0.80 and 0.22, respectively. 2 The average R-value of the GCN model on the test set is higher than the average of the XGBoost model in Comparative Example 1 (0.79) and higher than the average of the GPR model in Comparative Example 2 (0.57). However, the MAE is lower than the average of the XGBoost model in Comparative Example 1 (0.23) and lower than the average of the GPR model in Comparative Example 2 (0.40). For the melting point prediction model, the average R-value of the GCN model on the test set is... 2 The R-value is 0.88, the average MAE is 29.5℃, and the average R-value of the XGBoost model on the test set in Comparative Example 1 is... 2 The value is 0.61, the average MAE is 48.8℃, and the average R value of the GPR model on the test set in Comparative Example 2 is... 2 The mean value is 0.59, and the average MAE is 52.4℃; the average R-value of the GCN model is... 2 Higher average MAE; for the 5% thermal decomposition temperature prediction model, the average R-value of the GCN model on the test set is lower. 2 It is 0.84, the average MAE is 22.3℃, and the average R... 2 Higher than the average R-value of the XGBoost model test set in Comparative Example 1 2 0.60, higher than the average R-value of the GPR model test set in Comparative Example 2. 2 The average MAE is 0.66, which is lower than the average MAE of 28°C for the XGBoost model test set in Comparative Example 1 and lower than the average MAE of 24.5°C for the GPR model test set in Comparative Example 2. Compared with XGBoost and GPR, the GCN model has higher accuracy, smaller error, and stronger generalization ability, demonstrating the advantages of the method proposed in this embodiment.

[0139] Example 5:

[0140] High-performance bismaleimide resin structure design based on materials genome approach

[0141] In this embodiment, based on the framework of the materials genome approach, the performance prediction model constructed in Example 4 was applied to rapidly screen and obtain a novel bismaleimide resin structure that combines low dielectric constant, low melting point, and high 5% thermal decomposition temperature. The steps include:

[0142] D1. The performance prediction model constructed in Example 4 was used to predict the 10,744 virtual bismaleimide resin candidate structures in Example 4, and their performance data were obtained.

[0143] D2. A weighted scalar function is used to represent the three properties as a comprehensive score, where the weight ratio of dielectric constant, melting point, 5% thermal decomposition temperature, and SASocre is 3:3:3:1. Figure 8 As shown, the spheres from smallest to largest represent overall performance from lowest to highest.

[0144]

[0145] Among them, Score, ε, T M T d SA represents the overall score, dielectric constant of the virtual structure, melting point, 5% thermal decomposition temperature, and synthesis accessibility score, respectively; ε min ε max T represents the minimum and maximum values ​​of the dielectric constant of the virtual structure, respectively; Mmin T Mmax These represent the minimum and maximum melting points in the virtual structure, respectively; T dmin T dmax SA represents the minimum and maximum values ​​of the 5% thermal decomposition temperature in the virtual structure, respectively; min SA max These represent the minimum and maximum values ​​of the synthesis accessibility score in the virtual structure, respectively.

[0146] D3. Based on the comprehensive score, 10 novel bismaleimide resin structures were selected from the top 5% of preferred structures. The predicted values ​​of their structures, dielectric constants at room temperature and 1MHz, melting points, and 5% thermal decomposition temperatures are as follows: Figure 9 As shown, its excellent dielectric properties, high temperature resistance, and easy processing characteristics make it a promising candidate for applications in electronic communications and aerospace.

[0147] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 10As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for training a performance prediction model of bismaleimide resin, a performance prediction method for bismaleimide resin, or a structural design method for bismaleimide resin.

[0148] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0149] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0150] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0151] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0152] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0153] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for training a performance prediction model for bismaleimide resin, characterized in that, include: Obtain bismaleimide data and supplementary data; The bismaleimide data includes: the molecular structure of bismaleimide resin and corresponding experimental data; the experimental data includes: dielectric constant data, melting point, and 5% thermal decomposition temperature; the supplementary data includes: supplementary dielectric constant data and supplementary temperature data; the supplementary dielectric constant data includes: the molecular structure of several types of resin and corresponding dielectric constant data; the supplementary temperature data includes: the molecular structure of polyimide resin and corresponding melting point data and 5% thermal decomposition temperature data; the dielectric constant data includes: dielectric constant and corresponding test conditions; Using the molecular structure as input, the test conditions as supplementary input features, the dielectric constant as a label, and the predicted value of the dielectric constant as output, a supervised learning model is trained to obtain a dielectric constant prediction model. Using the molecular structure as input, the melting point data as labels, and the predicted value of the melting point data as output, a supervised learning model is trained to obtain a melting point prediction model. Using the molecular structure as input, the 5% thermal decomposition temperature as a label, and the predicted value of the 5% thermal decomposition temperature as output, a supervised learning model is trained to obtain a thermal decomposition temperature prediction model. The performance prediction model includes: the dielectric constant prediction model, the melting point prediction model, and the thermal decomposition temperature prediction model; The supervised learning model includes: an input layer, an intermediate layer, and a fully connected layer; The input layer is used to encode the molecular structure in the form of a molecular diagram; The intermediate layer is used to update the embedding vector of the target node by aggregating neighbor node information through continuous convolution and batch normalization operations. The fully connected layer is used to output the model using experimental performance data corresponding to the molecular structure.

2. A method for predicting the performance of bismaleimide resin, characterized in that, include: Obtain the molecular structure of the bismaleimide resin to be predicted; The molecular structure is input into the dielectric constant prediction model to obtain the predicted value of the dielectric constant of the bismaleimide resin to be predicted; the dielectric constant prediction model is trained by the performance prediction model training method of the bismaleimide resin according to claim 1. The molecular structure is input into the melting point prediction model to obtain the predicted value of the melting point data of the bismaleimide resin to be predicted; the melting point prediction model is trained by the performance prediction model training method of the bismaleimide resin according to claim 1. The molecular structure is input into the thermal decomposition temperature prediction model to obtain the predicted value of the 5% thermal decomposition temperature of the bismaleimide resin to be predicted; the thermal decomposition temperature prediction model is trained by the performance prediction model training method of the bismaleimide resin according to claim 1. The predicted values ​​of the dielectric constant, melting point, and 5% thermal decomposition temperature of the bismaleimide resin to be predicted are the performance prediction data.

3. A structural design method for a bismaleimide resin, characterized in that, include: Construct several virtual molecular structures of bismaleimide resins; Calculate the synthetic accessibility score for each of the virtual molecular structures; The virtual molecular structure is input into the dielectric constant prediction model to obtain the corresponding dielectric constant prediction values; the dielectric constant prediction model is trained by the performance prediction model training method of bismaleimide resin as described in claim 1. The virtual molecular structure is input into the melting point prediction model to obtain the corresponding predicted melting point data values; the melting point prediction model is trained by the performance prediction model training method of bismaleimide resin as described in claim 1. The virtual molecular structure is input into the thermal decomposition temperature prediction model to obtain the corresponding 5% thermal decomposition temperature prediction value; the thermal decomposition temperature prediction model is trained by the performance prediction model training method of bismaleimide resin as described in claim 1. Based on the predicted dielectric constant, predicted melting point, predicted 5% thermal decomposition temperature, and synthesis accessibility score, a comprehensive score is calculated for each virtual molecular structure. The design structure of the bismaleimide resin was determined based on the comprehensive score.

4. The structural design method for bismaleimide resin according to claim 3, characterized in that, The construction of several virtual molecular structures of bismaleimide resins specifically includes: Determine the gene type for the bismaleimide resin structure; Based on the gene type, retrieve the corresponding genes from the database; According to a preset template, several genes are synthesized into several bismaleimide resins, resulting in several virtual molecular structures of bismaleimide resins.

5. The structural design method for bismaleimide resin according to claim 4, characterized in that, The gene types include: Maleic anhydride gene; diamine gene with only two amino groups at both ends of the main chain; ammonium acid gene containing a benzene ring in the main chain, with an amino group and a formic acid attached to the benzene ring; diphenol gene containing a benzene ring in the main chain, with two hydroxyl groups attached to the benzene ring; maleic acid diamine gene; haloalkanes with halogen atoms at both ends of the main chain; amino alcohol gene with an amino group and a hydroxyl group attached to each end of the main chain; diacid gene with only two formic acid groups at both ends of the main chain.

6. The structural design method for bismaleimide resin according to claim 3, characterized in that, The formula for calculating the comprehensive score is as follows: ； in, Score , ε , T M , T d , S A These represent the overall score, dielectric constant of the virtual molecular structure, melting point, 5% thermal decomposition temperature, and synthesis accessibility score, respectively; ε min ε max These represent the minimum and maximum values ​​of the dielectric constant of the virtual molecular structure, respectively. T Mmin , T Mmax These represent the minimum and maximum melting points of the virtual molecular structure, respectively. T dmin , T dmax These represent the minimum and maximum values ​​of the 5% thermal decomposition temperature of the virtual molecular structure, respectively. S Amin , S Amax These represent the minimum and maximum values ​​of the synthetic accessibility score for the virtual molecular structure, respectively.

7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the performance prediction model training method for bismaleimide resin according to claim 1, or the performance prediction method for bismaleimide resin according to claim 2, or the structural design method for bismaleimide resin according to any one of claims 3-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the performance prediction model training method for bismaleimide resin as described in claim 1, the performance prediction method for bismaleimide resin as described in claim 2, or the structural design method for bismaleimide resin as described in any one of claims 3-6.

9. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the performance prediction model training method for bismaleimide resin as described in claim 1, the performance prediction method for bismaleimide resin as described in claim 2, or the structural design method for bismaleimide resin as described in any one of claims 3-6.

Images