Formula optimization method, medium and system for flame-retardant epoxy resin for electric reactor

By acquiring and utilizing multimodal fingerprint data of flame retardant epoxy resin raw materials for reactors, combining chemical models and deep learning to optimize the formula, the problem of low optimization efficiency of flame retardant epoxy resin formula for reactors in the prior art is solved, and the problem of low optimization efficiency of flame retardant epoxy resin formula for reactors in the prior art is realized. The comprehensive performance optimization of reactors.

CN119007871APending Publication Date: 2024-11-22ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID NINGXIA ELECTRIC POWER COMPANY +6
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202410987131.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing flame retardant epoxy resin formulation optimization methods for reactors rely on a large number of trial and error experiments, are inefficient and difficult to fully meet the demanding performance requirements of reactors.

Method used

By obtaining multimodal fingerprint data of raw materials, selecting alternative raw materials with high similarity, using quantitative structure-effect relationship chemical model and pre-trained epoxy resin flame retardant index prediction model, generate and optimize formulas, and combine with deep learning models for performance prediction , choose the best formula.

Benefits of technology

The efficiency and reliability of formula optimization have been greatly improved, and the performance of flame-retardant epoxy resin for reactors has been comprehensively optimized, meeting the strict performance requirements of reactors.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007871A_ABST
    Figure CN119007871A_ABST
Patent Text Reader

Abstract

The invention provides a formula optimization method, medium and system for flame-retardant epoxy resin for electric reactors, and belongs to the technical field of flame-retardant epoxy resin for electric reactors, and the method comprises the following steps: collecting multi-modal fingerprint data of each raw material, and screening out alternative raw materials with the fingerprint similarity of the raw materials higher than a threshold value; generating a formula set containing the alternative raw materials by using a traversal circulation method; calculating the raw material proportion of each formula by applying a quantitative structure-activity relationship model; evaluating the flame retardant property of each formula by utilizing a pre-trained flame retardant index prediction model; selecting a plurality of formulas with the optimal flame retardant index, and performing a small-scale experiment to obtain a material performance vector; according to the performance requirements of the flame-retardant epoxy resin for the electric reactor, selecting a formula closest to the flame-retardant epoxy resin as an optimized final formula; according to the method, the multi-mode fingerprint data, the quantitative structure-activity model, the performance prediction model and other means are comprehensively utilized, and the technical problems that an existing method depends on a large number of trial and error experiments, a large amount of time and cost are input, and efficiency is low are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of flame-retardant epoxy resin for reactors, and in particular, relates to a method, medium and system for optimizing the formulation of flame-retardant epoxy resin for reactors. Background Art

[0002] Reactors are one of the indispensable and important components in power systems. Their main functions are to limit current, regulate voltage, absorb surges, etc. Reactors are widely used in various types of power equipment and electronic products, playing a key role in the stable operation of power systems and the improvement of power quality.

[0003] In order to meet the various performance requirements of reactor applications, the main core material of the reactor is usually flame-retardant epoxy resin. Compared with metal materials, epoxy resin has good insulation, corrosion resistance and processability, and can better meet the requirements of lightweight and complex structure of reactors. However, the flame retardancy of pure epoxy resin itself is poor and cannot meet the strict fire protection standards of reactors. Therefore, when preparing epoxy resin composite materials for reactors, auxiliary materials such as flame retardants are usually added to improve the overall flame retardant performance.

[0004] At present, the commonly used flame-retardant epoxy resin formula for reactors includes epoxy resin, curing agent, inorganic flame retardant (such as aluminum hydroxide, phosphate, etc.) and some additives (such as dispersant, defoamer, etc.). This formula can effectively improve the flame retardant properties of epoxy resin by introducing inorganic flame retardants to meet the fire protection requirements of reactor applications. However, it will also reduce other key properties of epoxy resin, such as electrical properties, thermal properties and mechanical properties, thereby limiting its further application in reactors.

[0005] In order to further improve the comprehensive performance of flame-retardant epoxy resin for reactors, the industry has taken the following measures:

[0006] 1) Develop new high-performance flame retardants, such as magnesium-aluminum-lithium composite metal hydroxides, phosphorus- and nitrogen-containing organic flame retardants, etc., in order to improve the flame retardant properties while minimizing the impact on other properties.

[0007] 2) Optimize the ratio of epoxy resin and flame retardant, and find the best formula that can balance various performances through exploration of different formulas.

[0008] 3) Use a mixed flame retardant system with multiple flame retardant mechanisms, such as inorganic-organic composite flame retardants, in order to achieve a synergistic effect of 1+1>2.

[0009] 4) Using modification technology, such as introducing modified epoxy resin monomers, modified flame retardants, etc., to improve the compatibility and interfacial bonding strength between the epoxy resin matrix and the flame retardant.

[0010] But in general, the existing formula optimization methods mainly rely on a large number of trial and error experiments, which requires a lot of time and cost and has low efficiency. At the same time, the optimal formula obtained is difficult to fully meet the stringent performance requirements of the reactor and has certain limitations. Summary of the invention

[0011] In view of this, the present invention provides a flame retardant epoxy resin formulation optimization method, medium and system for reactors, which can solve the technical problems that the existing formulation optimization method mainly relies on a large number of trial and error experiments, requires a lot of time and cost, and has low efficiency.

[0012] The present invention is achieved in that:

[0013] A first aspect of the present invention provides a method for optimizing a flame retardant epoxy resin formulation for a reactor, comprising the following steps:

[0014] S10, obtaining multimodal fingerprint data of each raw material in the flame-retardant epoxy resin formula for the reactor;

[0015] S20, for each raw material, selecting a material whose multimodal fingerprint data similarity with the raw material is greater than a preset fingerprint similarity threshold as a substitute raw material, and generating a recipe set by using a traversal loop method;

[0016] S30, calculating the raw material ratio of each formula in the formula set using a quantitative structure-activity relationship chemical model;

[0017] S40, using a pre-trained epoxy resin flame retardant index prediction model, inputting the multimodal fingerprint and raw material ratio of each formula in the formula set to obtain a flame retardant index of each formula;

[0018] S50, selecting multiple formulas with the best flame retardant index to form a first formula set, and obtaining multiple necessary properties of the epoxy resin material obtained by conducting a small test on each formula in the first formula set, and generating a performance vector;

[0019] S60. According to the performance vector of each formula, a formula that is closest to the preset required performance vector of the flame-retardant epoxy resin for the reactor is selected as the optimized formula.

[0020] The specific implementation method of step S10 for obtaining the multimodal fingerprint data of each raw material in the flame-retardant epoxy resin formula for the reactor is as follows:

[0021] First, a large number of various raw material samples used in the formulation of flame-retardant epoxy resin for reactors are collected. For each raw material sample, a variety of analytical methods are used to obtain its characteristic data to form multimodal fingerprint data of the raw material. The multimodal fingerprint data includes spectral data, morphological data, thermal analysis data and other characterization data. Spectral data include infrared spectrum, Raman spectrum, ultraviolet spectrum and visible spectrum, etc., which are used to reflect the existence and strength of chemical bonds in the raw materials; morphological data include scanning electron microscope images, atomic force microscope images, transmission electron microscope images and optical microscope images, etc., which are used to reflect the microscopic morphology and size distribution of the raw materials; thermal analysis data include thermogravimetric analysis data, differential scanning calorimetry analysis data, thermomechanical analysis data and dynamic mechanical thermal analysis data, etc., which are used to reflect the thermal stability and phase change behavior of the raw materials; other characterization data include X-ray diffraction data, particle size analysis data, specific surface area data and density data, etc., which are used to reflect the structure, composition and particle size information of the raw materials. By performing the above multimodal analysis on a large number of different raw material samples, a data set containing rich raw material characteristic information can be obtained.

[0022] Wherein, the step S20 selects a material whose multimodal fingerprint data similarity with the raw material is greater than a preset fingerprint similarity threshold as a substitute raw material, and a specific implementation method of generating a recipe set by using a traversal loop method is as follows:

[0023] First, for each raw material in the flame-retardant epoxy resin formula for reactors, the fingerprint similarity between its multimodal fingerprint data and all other raw materials in the data set is calculated. Then, the raw materials whose fingerprint similarity with the raw material is greater than a preset threshold are selected as alternative raw materials, and the threshold is usually between 0.85 and 0.95. Next, the traversal combination method is used to combine the alternative raw materials in pairs to form a new formula, and the raw material ratio of each new formula is calculated. Through the above steps, a formula set containing a large number of potential formulas can be generated, laying the foundation for subsequent performance optimization.

[0024] The specific implementation of step S30 for calculating the raw material ratio of each formula in the formula set using the quantitative structure-activity relationship chemical model is as follows:

[0025] First, a quantitative structure-activity relationship (QSPR) chemical model is established to predict the optimal usage ratio of each raw material in the formula. The QSPR model needs to be trained and optimized based on a large amount of experimental data. Then, the raw material information of each formula in the formula set generated in step S20 is input into the trained QSPR model to calculate the optimal usage ratio of each raw material. The QSPR model is a statistical learning model based on the chemical structure-performance relationship, which can be used to predict various performance indicators of compounds. In this patent, the QSPR model is used to predict the optimal usage ratio of each raw material in the formula, providing an important basis for subsequent performance optimization.

[0026] The specific implementation method of step S40 using the pre-trained epoxy resin flame retardant index prediction model, inputting the multimodal fingerprint and raw material ratio of each formula in the formula set, and obtaining the flame retardant index of each formula is as follows:

[0027] The epoxy resin flame retardant index prediction model is a bidirectional pyramid neural network model based on multimodal fusion. The model is divided into multiple pyramid layers, each of which is responsible for predicting one or more performance indicators. Each pyramid layer contains 5 sub-network modules: a spectral data prediction network module, a morphology data prediction network module, a thermal analysis data prediction network module, other characterization data prediction network modules, and a multimodal fusion network module. The multimodal fusion network adopts an attention mechanism or a gating mechanism to fuse the features of different modes, that is, to fuse the outputs of each module, and finally predict all performance indicators. The multimodal fingerprint data and raw material ratio of each formula in the formula set generated in step S20 are input into the trained prediction model to obtain the flame retardant performance index of each formula.

[0028] The specific implementation of step S50 is as follows: selecting multiple formulations with the best flame retardant index to form a first formulation set, obtaining multiple necessary properties of epoxy resin materials obtained by conducting a small test on each formulation in the first formulation set, and generating a performance vector:

[0029] First, according to the flame retardant index of each formula obtained in step S40, multiple formulas with the best flame retardant index are selected to form a first formula set. Then, for each formula in the first formula set, an experimental small test is carried out to prepare epoxy resin materials, and its flame retardant properties, electrical properties, thermal properties, mechanical properties and processing properties are tested. Multiple performance indicators corresponding to each formula are composed of a performance vector, where the flame retardant properties include vertical combustion grade, oxygen index, heat release rate, smoke density, etc.; electrical properties include volume resistivity, dielectric constant, insulation strength, arc tracking index, etc.; thermal properties include thermal shrinkage, glass transition temperature, thermal expansion coefficient, thermal stability, etc.; mechanical properties include tensile strength, bending strength, impact strength, hardness, etc.; processing properties include fluidity, formability, processable temperature range, etc. The detailed performance data obtained through the experimental small test can more accurately describe the comprehensive performance characteristics of each formula.

[0030] Wherein, the step S60 selects the formula closest to the preset required performance vector of the flame-retardant epoxy resin for the reactor according to the performance vector of each formula, and the specific implementation method as the optimized formula is as follows:

[0031] First, the target performance requirements of the flame-retardant epoxy resin for reactors are determined in advance to form an ideal performance vector, which includes target values ​​of indicators such as flame retardant performance, electrical performance, thermal performance, mechanical performance and processing performance. Then, the Euclidean distance between the performance vector of each formula in the first formula set in step S50 and the target performance vector is calculated. Finally, the formula with the closest distance to the target performance vector is selected as the final optimized formula. Through this step, the optimal formula with the most comprehensive performance close to the target requirements can be selected according to the actual application requirements of the flame-retardant epoxy resin for reactors.

[0032] The multimodal fingerprint data refers to characteristic data of raw materials obtained through a variety of analysis methods, including spectral data, morphological data, thermal analysis data and other characterization data.

[0033] The spectral data includes one or more of infrared spectrum, Raman spectrum, ultraviolet spectrum and visible spectrum, which are used to reflect the existence and strength of chemical bonds;

[0034] The morphological data includes one or more of a scanning electron microscope image, an atomic force microscope image, a transmission electron microscope image, and an optical microscope image, which are used to reflect the microscopic morphology and size distribution of the raw material;

[0035] The thermal analysis data includes one or more of thermogravimetric analysis data, differential scanning calorimetry analysis data, thermomechanical analysis data, and dynamic mechanical thermal analysis data, which are used to reflect the thermal stability and phase change behavior of the raw materials;

[0036] The other characterization data include one or more of X-ray diffraction data, particle size analysis data, specific surface area data, and density data, which are used to reflect the structure, composition, and particle size information of the raw materials.

[0037] The performance vector is a vector composed of flame retardant properties, electrical properties, thermal properties, mechanical properties, and processing properties.

[0038] The performance vector is a vector composed of flame retardant properties, electrical properties, thermal properties, mechanical properties, and processing properties;

[0039] The flame retardant properties include one or more of vertical burning grade, oxygen index, heat release rate, and smoke density;

[0040] The mechanical properties include one or more of tensile strength, flexural strength, impact strength, and hardness;

[0041] The thermal properties include one or more of thermal shrinkage, glass transition temperature, thermal expansion coefficient, and thermal stability;

[0042] The electrical properties include one or more of volume resistivity, dielectric constant, dielectric strength, and arc tracking index;

[0043] The processing performance includes one or more of fluidity, formability, and processable temperature range.

[0044] Among them, the structure of the epoxy resin flame retardant index prediction model is a bidirectional pyramid neural network model based on multimodal fusion, which is composed of multiple pyramid layers, and each pyramid layer is responsible for predicting one or more types of performance indicators.

[0045] Furthermore, the first pyramid layer is responsible for flame retardant performance indicators, the second pyramid layer is responsible for the comprehensive indicators of flame retardant performance and electrical properties, the third pyramid layer is responsible for the comprehensive indicators of flame retardant performance, electrical properties and thermal properties, the fourth pyramid layer is responsible for the comprehensive indicators of flame retardant performance, electrical properties, thermal properties and mechanical properties, and the fifth pyramid layer is responsible for the comprehensive indicators of flame retardant performance, electrical properties, thermal properties, mechanical properties and processing performance.

[0046] Furthermore, each pyramid layer includes five network modules: spectral data prediction network module, morphological data prediction network module, thermal analysis data prediction network module, other characterization data prediction network module and multimodal fusion network module.

[0047] The spectral data prediction network module is used to extract the characteristics of spectral data using a convolutional neural network or a recurrent neural network structure, and predict the performance indicators corresponding to the characteristics of the spectral data.

[0048] The morphology data prediction network module is used to extract the morphology features of the morphology data using a convolutional neural network, and predict the performance indicators corresponding to the morphology features.

[0049] The thermal analysis data prediction network module is used to extract the thermal analysis features of the thermal analysis data using a fully connected neural network or an LSTM structure, and predict the performance indicators corresponding to the thermal analysis features.

[0050] The other characterization data prediction network module is used to extract other characterization features of other characterization data using a fully connected neural network or an LSTM structure, and predict performance indicators corresponding to the other characterization features.

[0051] Furthermore, the multimodal fusion network module is used to adopt an attention mechanism or a gating mechanism to fuse the different spectral data prediction network modules, morphological data prediction network modules, thermal analysis data prediction network modules, and other characterization data prediction network modules, and ultimately predict all performance indicators corresponding to all multimodal fingerprint data.

[0052] Furthermore, in the training process of the epoxy resin flame retardant index prediction model, four prediction network modules are first trained separately, and then the output features of the four networks are input into the multimodal fusion network for end-to-end training, and cross-validation and other methods are used throughout the training process to ensure the generalization ability of the model.

[0053] Specifically, in the training process of the epoxy resin flame retardant index prediction model, the four prediction network modules are first trained separately, and supervised learning is performed using the corresponding modal data and their performance indicators. Then, the output features of the four network modules are input into the multimodal fusion network module for end-to-end training. During the entire training process, methods such as cross-validation can be used to ensure the generalization ability of the model.

[0054] Through this bidirectional pyramid neural network model, the multimodal characterization data (spectrum, morphology, thermal analysis and others) of raw materials can be effectively converted into key performance indicators of flame-retardant epoxy resin for reactors, providing a reliable prediction basis for subsequent formula optimization.

[0055] 1. Acquisition of training data set: Collect a large number of epoxy resin samples with multimodal characterization data and performance test data as training samples. These samples should cover different raw material formulas, production processes, etc. to ensure the generalization ability of the model.

[0056] For each sample, obtain its multimodal fingerprint data, including:

[0057] 1) Spectral data: infrared spectrum, Raman spectrum, ultraviolet spectrum, visible spectrum, etc.

[0058] 2) Morphological data: scanning electron microscope images, atomic force microscope images, transmission electron microscope images, optical microscope images, etc.

[0059] 3) Thermal analysis data: thermogravimetric analysis data, differential scanning calorimetry analysis data, thermomechanical analysis data, dynamic mechanical thermal analysis data, etc.

[0060] 4) Other characterization data: X-ray diffraction data, particle size analysis data, specific surface area data, density data, etc.

[0061] At the same time, each sample is experimentally tested to obtain its flame retardant properties, electrical properties, thermal properties, mechanical properties and processing properties as label data.

[0062] The above multimodal fingerprint data and performance indicator data are standardized to form the final training data set.

[0063] 2. Model structure and training: The aforementioned bidirectional pyramid neural network model based on multimodal fusion is adopted.

[0064] Each pyramid layer contains five network modules: spectral data prediction network, morphological data prediction network, thermal analysis data prediction network, other characterization data prediction network and multimodal fusion network.

[0065] First, train the four prediction network modules separately, namely:

[0066] 1) Spectral data prediction network: The input is spectral data, and the output is the corresponding performance index. It uses CNN or RNN structure.

[0067] 2) Shape data prediction network: The input is shape data, and the output is the corresponding performance index. CNN structure is used.

[0068] 3) Thermal analysis data prediction network: The input is thermal analysis data, and the output is the corresponding performance index. A fully connected network or LSTM structure is used.

[0069] 4) Other representation data prediction network: The input is other representation data, and the output is the corresponding performance index. It also uses a fully connected network or LSTM structure.

[0070] The output features of the four prediction networks are then fed into a multimodal fusion network for end-to-end training. The fusion network uses methods such as attention mechanisms or gating mechanisms, and the output is all performance indicators.

[0071] During the training process, you can use cross-validation to evaluate model performance and use techniques such as early stopping to prevent overfitting. At the same time, you can try transfer learning to initialize the main network using pre-trained sub-network parameters to speed up model convergence.

[0072] Finally, the entire bidirectional pyramid neural network model is end-to-end fine-tuned to further optimize the performance prediction accuracy.

[0073] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for optimizing the formulation of a flame-retardant epoxy resin for a reactor.

[0074] A third aspect of the present invention provides a flame retardant epoxy resin formulation optimization system for reactors, which includes the above-mentioned computer readable storage medium.

[0075] Compared with the prior art, the flame retardant epoxy resin formulation optimization method, medium and system for reactor provided by the present invention have the following beneficial effects:

[0076] 1. Make full use of the multimodal characteristic information of raw materials to improve the accuracy of formula performance prediction

[0077] Compared with the traditional formula development method that only relies on single performance test data, the method of the present invention makes full use of multimodal characterization data such as spectrum, morphology, thermal analysis, etc. of raw materials. These rich raw material characteristic information can more comprehensively reflect the chemical and physical properties of raw materials, providing a more reliable data basis for subsequent performance prediction and formula optimization.

[0078] The deep learning prediction model based on multimodal fusion can effectively transform the multimodal feature data of raw materials into key performance indicators of flame-retardant epoxy resin for reactors, such as flame retardancy, electrical properties, thermal properties, etc. The model has a unique structure. Through the cascade of multiple pyramid layers, it can gradually and comprehensively consider the coupling relationship between different performance indicators and finally output a comprehensive performance prediction result. Compared with the traditional single performance prediction model, this method can more accurately evaluate the comprehensive performance of the formula.

[0079] 2. Significantly improve the efficiency and reliability of recipe optimization

[0080] Existing formulation optimization methods mainly rely on a large number of trial and error experiments, which require a lot of time and cost. The method of the present invention uses the QSPR model to predict the optimal usage ratio of raw materials in the formula, and then combines the deep learning model to predict performance, which greatly reduces the demand for experiments. At the same time, by conducting a small test to verify the candidate formula with the best flame retardant performance, it is further ensured that the obtained optimal formula can meet the actual application needs.

[0081] Compared with traditional empirical formula development, the method of the present invention is more scientific, systematic and reliable. First, by making full use of the multimodal characteristic data of raw materials, the properties of raw materials can be described more accurately; second, by combining the prediction capabilities of QSPR and deep learning models, the formula candidates with the best performance can be quickly screened out; finally, further optimization is performed through experimental verification to ensure that the final formula obtained can meet the stringent performance requirements of the reactor. Therefore, this method greatly improves the efficiency and reliability of formula optimization.

[0082] 3. Achieve comprehensive optimization of the performance of flame-retardant epoxy resin for reactors

[0083] Existing formulation optimization methods usually focus on improving flame retardancy, while ignoring other key performance indicators, such as electrical, thermal and mechanical properties. This approach often fails to meet the stringent requirements of reactors for comprehensive material performance.

[0084] The method of the present invention can truly achieve comprehensive optimization of the performance of flame-retardant epoxy resin for reactors by constructing a comprehensive performance vector covering indicators such as flame retardant performance, electrical performance, thermal performance, mechanical performance and processing performance, and taking this as the optimization target. When selecting the final optimized formula, not only a single flame retardant index is considered, but more attention is paid to the coordinated balance of various performance indicators to ensure that the obtained formula can fully meet the strict performance requirements of the reactor.

[0085] In summary, the flame-retardant epoxy resin formulation optimization method for reactors proposed in the present invention makes full use of the multimodal characteristic data of raw materials and combines advanced machine learning technology to achieve rapid and accurate prediction and comprehensive optimization of formulation performance. Compared with the existing technology, this method greatly improves the efficiency and reliability of formulation development and solves the technical problem that the existing formulation optimization method mainly relies on a large number of trial and error experiments, requires a lot of time and cost, and has low efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0087] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0089] like Figure 1 FIG. 1 is a flow chart of a method for optimizing a flame retardant epoxy resin formulation for a reactor provided by the present invention. The method comprises the following steps:

[0090] S10, obtaining multimodal fingerprint data of each raw material in the flame-retardant epoxy resin formula for the reactor;

[0091] S20, for each raw material, selecting a material whose multimodal fingerprint data similarity with the raw material is greater than a preset fingerprint similarity threshold as a substitute raw material, and generating a recipe set by using a traversal loop method;

[0092] S30, using a quantitative structure-activity relationship chemical model, calculating the raw material ratio of each formula in the formula set;

[0093] S40, using a pre-trained epoxy resin flame retardant index prediction model, inputting a multimodal fingerprint and a raw material ratio of each formula in the formula set to obtain a flame retardant index of each formula;

[0094] S50, selecting multiple formulas with the best flame retardant index to form a first formula set, and obtaining multiple necessary properties of the epoxy resin material obtained by conducting a small test on each formula in the first formula set, and generating a performance vector;

[0095] S60. According to the performance vector of each formula, a formula that is closest to the preset required performance vector of the flame-retardant epoxy resin for the reactor is selected as the optimized formula.

[0096] The specific implementation methods of the above steps are described in detail below:

[0097] Step S10: Obtaining multimodal fingerprint data of each raw material in the flame retardant epoxy resin formula for reactor

[0098] The specific implementation method of this step is as follows: First, a large number of raw material samples used in the formulation of flame-retardant epoxy resin for reactors are collected. For each raw material sample, multiple analysis methods are used to obtain its characteristic data to form multimodal fingerprint data of the raw material.

[0099] The so-called multimodal fingerprint data refers to the raw material characteristic data obtained through spectral analysis, morphological analysis, thermal analysis and other characterization analysis methods. Specifically including:

[0100] 1) Spectral data: including infrared spectrum, Raman spectrum, ultraviolet spectrum and visible spectrum, etc., which are used to reflect the existence and strength of chemical bonds in raw materials.

[0101] 2) Morphological data: including scanning electron microscope images, atomic force microscope images, transmission electron microscope images and optical microscope images, etc., which are used to reflect the microscopic morphology and size distribution of the raw materials.

[0102] 3) Thermal analysis data: including thermogravimetric analysis data, differential scanning calorimetry analysis data, thermomechanical analysis data and dynamic mechanical thermal analysis data, etc., which are used to reflect the thermal stability and phase change behavior of raw materials.

[0103] 4) Other characterization data: including X-ray diffraction data, particle size analysis data, specific surface area data and density data, etc., which are used to reflect the structure, composition and particle size information of the raw materials.

[0104] By performing the above multimodal analysis on a large number of different raw material samples, a data set containing rich raw material characteristic information can be obtained, laying a solid foundation for subsequent formula optimization.

[0105] Step S20: For each raw material, select a material whose multimodal fingerprint data similarity with the raw material is greater than a preset fingerprint similarity threshold as a substitute raw material, and generate a recipe set by traversing the loop method.

[0106] The specific implementation method of this step is as follows:

[0107] 1) For each raw material in the flame-retardant epoxy resin formula for reactors, calculate the fingerprint similarity between its multimodal fingerprint data and all other raw materials in the dataset.

[0108] 2) Select the raw material whose fingerprint similarity with the raw material is greater than the preset threshold as the alternative raw material. The threshold can be set according to the specific situation, usually between 0.85 and 0.95.

[0109] 3) Use the traversal combination method to combine the replaceable raw materials in pairs to form a new formula. For each new formula, calculate its raw material ratio.

[0110] 4) Through the above steps, a recipe set containing a large number of potential recipes can be generated. This recipe set covers the original recipe and the new recipe based on raw material substitution, laying the foundation for subsequent performance optimization.

[0111] The purpose of this is to enrich the formulation space by replacing raw materials and provide more options for finding the optimal formulation. At the same time, the setting of the similarity threshold can ensure that the substitute raw materials have a high chemical and physical similarity with the raw materials, thereby ensuring the feasibility of the formulation.

[0112] Step S30: Calculate the raw material ratio of each formula in the formula set using the quantitative structure-activity relationship chemical model

[0113] The specific implementation method of this step is as follows:

[0114] 1) Establish a quantitative structure-activity relationship (QSPR) chemical model to predict the optimal ratio of each raw material in the formula. The QSPR model needs to be trained and optimized based on a large amount of experimental data.

[0115] 2) Input the raw material information of each recipe in the recipe set generated in step S20, and use the trained QSPR model to calculate the optimal usage ratio of each raw material.

[0116] The QSPR model is a statistical learning model based on the chemical structure-performance relationship, which can be used to predict various performance indicators of compounds, such as melting point, boiling point, solubility, biological activity, etc. In this patent, the QSPR model is used to predict the optimal usage ratio of each raw material in the formula, providing an important basis for subsequent performance optimization.

[0117] By inputting the recipe information generated in step S20 into the QSPR model, the optimal usage ratio of each raw material in each recipe can be obtained, which provides necessary data support for the next step of performance prediction and optimization.

[0118] Step S40: Using the pre-trained epoxy resin flame retardant index prediction model, input the multimodal fingerprint and raw material ratio of each formula in the formula set to obtain the flame retardant index of each formula

[0119] The specific implementation method of this step is as follows:

[0120] 1) A bidirectional pyramid neural network model based on multimodal fusion is constructed to predict the flame retardant performance indicators of epoxy resin formulas. The structure of the model is divided into multiple pyramid layers, each layer is responsible for predicting one or more types of performance indicators.

[0121] 2) Each pyramid layer contains 5 sub-network modules: spectral data prediction network, morphological data prediction network, thermal analysis data prediction network, other characterization data prediction network, and multimodal fusion network.

[0122] 3) The spectral data prediction network, morphological data prediction network, thermal analysis data prediction network and other characterization data prediction networks respectively use convolutional neural networks, recurrent neural networks or fully connected network structures to extract the features of the corresponding modal data and predict the corresponding performance indicators.

[0123] 4) The multimodal fusion network uses an attention mechanism or a gating mechanism to fuse the features of different modalities and ultimately predict all performance indicators, including flame retardant properties, electrical properties, thermal properties, mechanical properties, and processing properties.

[0124] 5) The multimodal fingerprint data and raw material ratio of each formula in the formula set generated in step S20 are input into the trained prediction model to obtain the flame retardant performance index of each formula.

[0125] The key to this prediction model is that it can effectively convert the multimodal characterization data of raw materials into key performance indicators of flame-retardant epoxy resin for reactors, providing a reliable basis for subsequent formula optimization.

[0126] Step S50: Select multiple formulations with the best flame retardant index to form a first formulation set, and obtain multiple necessary properties of epoxy resin materials obtained by conducting a small test on each formulation in the first formulation set, and generate a performance vector

[0127] The specific implementation method of this step is as follows:

[0128] 1) According to the flame retardant index of each formulation obtained in step S40, multiple formulations with the best flame retardant index are selected to form a first formulation set. The optimal value of the flame retardant index can be determined according to actual needs.

[0129] 2) For each formula in the first formula set, conduct a small-scale experiment to prepare epoxy resin materials and test their flame retardant properties, electrical properties, thermal properties, mechanical properties and processing properties.

[0130] 3) Multiple performance indicators corresponding to each formula are combined into a performance vector, including:

[0131] -Flame retardant properties: such as vertical burning grade, oxygen index, heat release rate, smoke density, etc.

[0132] -Electrical properties: such as volume resistivity, dielectric constant, insulation strength, arc tracking index, etc.

[0133] - Thermal properties: such as thermal shrinkage, glass transition temperature, thermal expansion coefficient, thermal stability, etc.

[0134] - Mechanical properties: such as tensile strength, flexural strength, impact strength, hardness, etc.

[0135] -Processing properties: such as fluidity, formability, processing temperature range, etc.

[0136] The detailed performance data obtained through small-scale experimental tests can more accurately describe the comprehensive performance characteristics of each formula, providing an important basis for the subsequent selection of the optimal formula.

[0137] Step S60: According to the performance vector of each formula, select the formula closest to the preset required performance vector of the flame retardant epoxy resin for the reactor as the optimized formula

[0138] The specific implementation method of this step is as follows:

[0139] 1) Determine the target performance requirements of flame-retardant epoxy resin for reactor in advance to form an ideal performance vector. This vector includes the target values ​​of flame retardant performance, electrical performance, thermal performance, mechanical performance and processing performance.

[0140] 2) Calculate the Euclidean distance between the performance vector of each recipe in the first recipe set and the target performance vector in step S50.

[0141] 3) Select the formula that is closest to the target performance vector as the final optimized formula.

[0142] Through this step, the optimal formula with the most comprehensive performance close to the target requirements can be selected according to the actual application requirements of flame-retardant epoxy resin for reactors. This optimization method based on multi-index performance vector can better balance and take into account various key performance indicators and obtain the optimal formula that meets the actual application requirements.

[0143] In summary, the flame retardant epoxy resin formulation optimization method for reactors involved in this patent mainly includes the following steps:

[0144] 1. Obtain multimodal fingerprint data of each raw material in the flame-retardant epoxy resin formula for reactors, including spectrum, morphology, thermal analysis and other characterization data.

[0145] 2. Generate a large number of potential formulas by using alternative raw materials, and calculate the optimal proportion of raw materials in each formula.

[0146] 3. Use the pre-trained neural network model to accurately predict the flame retardant performance indicators of each formula.

[0147] 4. Select multiple formulations with the best flame retardant performance for small-scale tests to obtain detailed performance vector data.

[0148] 5. According to the performance requirements of the reactor application, select the formula closest to the target performance vector as the final optimization result.

[0149] This optimization method makes full use of the multimodal characterization data of raw materials and realizes the rapid optimization of the flame-retardant epoxy resin formula for reactors through the prediction of machine learning models and the optimization of performance vectors.

[0150] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for optimizing the formulation of a flame-retardant epoxy resin for a reactor.

[0151] A third aspect of the present invention provides a flame retardant epoxy resin formulation optimization system for reactors, which includes the above-mentioned computer readable storage medium.

[0152] In order to better understand and implement the present invention, a specific embodiment 1 of the present invention in a computer readable storage medium or a computer system for a computer program is provided below. The specific steps of this embodiment 1 are as follows:

[0153] Step S10: Obtaining multimodal fingerprint data of each raw material in the flame retardant epoxy resin formula for reactor

[0154] For each raw material sample i, its multimodal fingerprint data can be obtained, including spectral data Morphological data Thermal analysis data and other characterization data These data can be obtained by:

[0155] 1. Spectral data Including infrared spectrum, Raman spectrum, ultraviolet spectrum and visible spectrum, etc., which are used to reflect the existence and strength of chemical bonds in raw materials. They can be obtained using testing instruments such as Fourier transform infrared spectrometer (FTIR) and Raman spectrometer.

[0156] 2. Morphological data Including scanning electron microscope (SEM) images, atomic force microscope (AFM) images, transmission electron microscope (TEM) images and optical microscope images, which are used to reflect the microscopic morphology and size distribution of raw materials.

[0157] 3. Thermal analysis data Including thermogravimetric analysis (TGA) data, differential scanning calorimetry (DSC) data, thermomechanical analysis (TMA) data and dynamic mechanical thermal analysis (DMA) data, which are used to reflect the thermal stability and phase change behavior of raw materials.

[0158] 4. Other characterization data Including X-ray diffraction (XRD) data, particle size analysis data, specific surface area data and density data, etc., which are used to reflect the structure, composition and particle size information of the raw materials.

[0159] By performing the above multimodal analysis on a large number of different raw material samples, a data set containing rich raw material characteristic information can be obtained. in represents the multimodal feature vector of the i-th raw material, y (i) The actual application performance of the raw material in flame-retardant epoxy resin for reactors is shown. These data lay the foundation for subsequent formulation optimization.

[0160] Step S20: For each raw material, select a material whose multimodal fingerprint data similarity with the raw material is greater than a preset fingerprint similarity threshold as a substitute raw material, and generate a formula set by using a traversal loop method.

[0161] For each raw material i, its multimodal fingerprint data x needs to be calculated (i) With the dataset The fingerprint similarity of all other raw materials in . Here, cosine similarity is used as the similarity metric:

[0162]

[0163] Where ‖x‖ represents the L2 norm of vector x.

[0164] Set a similarity threshold τ, usually 0.85≤τ≤0.95. For raw material i, select all raw materials with similarity greater than τ as alternative raw materials to form a set

[0165] Then, the set is traversed and combined. The raw materials in the formula are combined in pairs to form a new formula. For each new formula j, its raw material ratio can be calculated Where M is the total number of raw materials Indicates the usage ratio of the mth raw material in recipe j.

[0166] Through the above steps, a recipe set containing a large number of potential recipes can be generated. in Represents the multimodal fingerprint data of the jth recipe. This recipe set lays the foundation for subsequent performance optimization.

[0167] Step S30: Calculate the raw material ratio of each formula in the formula set using the quantitative structure-activity relationship chemical model

[0168] In order to predict the optimal ratio of each raw material in the formula, a quantitative structure-activity relationship (QSPR) chemical model was established. This model can be trained and optimized by machine learning using a large amount of experimental data.

[0169] Given a multimodal fingerprint data x of recipe j (j) And raw material ratio p (j) , the QSPR model can predict the performance index y of the formula (j) The QSPR model can be built using methods such as linear regression or neural networks:

[0170] y (j) =f(x (j) ,p (j) ;θ)

[0171] Where θ is the model parameter and f(·) is the model function.

[0172] When training the QSPR model, you can use the dataset The sample (x (i) ,p (i) ,y (i) ), by minimizing the prediction error ‖y (i) -f(x (i) ,p (i) ;θ)‖ 2 To optimize the model parameters θ.

[0173] Then, the recipe set generated in step S20 is The multimodal fingerprint data x for each recipe in(j) By inputting the trained QSPR model, the optimal usage ratio p of raw materials for each formula can be obtained. (j) This provides the necessary data support for subsequent performance prediction and optimization.

[0174] Step S40: Using the pre-trained epoxy resin flame retardant index prediction model, input the multimodal fingerprint and raw material ratio of each formula in the formula set to obtain the flame retardant index of each formula

[0175] In order to predict the flame retardant performance index of epoxy resin formula, a bidirectional pyramid neural network model based on multimodal fusion was constructed. The structure of the model is as follows:

[0176] y=g(x;θ)

[0177] where x = [x spec ,x morph ,x therm ,x other ,p] is the input, including multimodal fingerprint data and raw material ratio, y is the output, including indicators such as flame retardancy, electrical properties, thermal properties, mechanical properties and processing properties, and θ is the model parameter.

[0178] The structure of the model is divided into multiple pyramid layers, each of which is responsible for predicting one or more performance indicators. Each pyramid layer contains 5 sub-network modules:

[0179] 1. Spectral data prediction network f spec (x spec θ spec ), which is used to extract spectral features and predict the corresponding performance.

[0180] 2. Shape data prediction network f morph (x morph θ morph ), which is used to extract morphological features and predict corresponding performance.

[0181] 3. Thermal analysis data prediction network therm (x therm θ therm ), which is used to extract thermal analysis features and predict the corresponding performance.

[0182] 4. Other representation data prediction networks f other (x other θ other ), which is used to extract other features and predict the corresponding performance.

[0183] 5. Multimodal Fusion Network fusion ([h spec ,h morph ,h therm ,hother ,p];θ fusion ), use attention mechanism or gating mechanism to fuse different modal features and predict all performance indicators.

[0184] where h spec =f spec (x spec θ spec ),h morph =f morph (x morph θ morph ),h therm =f therm (x therm θ therm ),h other =f other (x other θ other ) represents different modal features.

[0185] During the training process, the four prediction network modules are first trained separately, and supervised learning is performed using the corresponding modal data and their performance indicators. Then the output features of the four networks are input into the multimodal fusion network for end-to-end training. During the entire training process, methods such as cross-validation can be used to ensure the generalization ability of the model.

[0186] Finally, the recipe set generated in step S20 is The multimodal fingerprint data x for each recipe in (j) And its raw material ratio p (j) Input the trained prediction model g(·;θ) to obtain the flame retardant performance index y of each formula. (j) .

[0187] Step S50: Select multiple formulations with the best flame retardant index to form a first formulation set, and obtain multiple necessary properties of epoxy resin materials obtained by conducting a small test on each formulation in the first formulation set, and generate a performance vector

[0188] The flame retardant performance index y of each formulation obtained in step S40 (j) , select multiple formulas with the best flame retardant performance to form the first formula set

[0189] For the first recipe set For each formula j, a small test was conducted to prepare epoxy resin materials and test their flame retardant properties. Electrical properties Thermal properties Mechanical properties and processing performance

[0190]

[0191] The multiple performance indicators corresponding to each recipe are combined into a performance vector:

[0192]

[0193] in:

[0194] Flame retardant properties Including vertical combustion level, oxygen index, heat release rate and smoke density

[0195] Electrical properties Including indicators such as volume resistivity, dielectric constant, insulation strength and arc tracking index

[0196] Thermal properties Including indicators such as thermal shrinkage, glass transition temperature, thermal expansion coefficient and thermal stability

[0197] Mechanical properties Including tensile strength, flexural strength, impact strength and hardness

[0198] Processing performance Including indicators such as fluidity, formability and processing temperature range

[0199] The detailed performance data obtained through small-scale experimental tests can more accurately describe the comprehensive performance characteristics of each formula, providing an important basis for the subsequent selection of the optimal formula.

[0200] Step S60: According to the performance vector of each formula, select the formula closest to the preset required performance vector of the flame retardant epoxy resin for the reactor as the optimized formula

[0201] First, it is necessary to predetermine the target performance requirements of flame-retardant epoxy resin for reactors and form an ideal performance vector y target The vector contains target values ​​for flame retardancy, electrical properties, thermal properties, mechanical properties, and processing properties.

[0202] Then, calculate the first formula set The performance vector y of each recipe in (j) and the target performance vector y target The Euclidean distance between:

[0203] d (j) =‖y (j) -y target ‖

[0204] Finally, the recipe closest to the target performance vector is selected, i.e. As the final optimized formula.

[0205] Through this step, the optimal formula with the most comprehensive performance close to the target requirements can be selected according to the actual application requirements of flame-retardant epoxy resin for reactors. This optimization method based on multi-index performance vector can better balance and take into account various key performance indicators and obtain the optimal formula that meets the actual application requirements.

[0206] In summary, the flame retardant epoxy resin formulation optimization method for reactors involved in this patent mainly includes the following key steps:

[0207] 1. Obtain multimodal fingerprint data of each raw material in the flame-retardant epoxy resin formula for reactors, including spectrum, morphology, thermal analysis and other characterization data.

[0208] 2. Generate a large number of potential formulas by using alternative raw materials, and calculate the optimal proportion of raw materials in each formula.

[0209] 3. Use the pre-trained neural network model to accurately predict the flame retardant performance indicators of each formula.

[0210] 4. Select multiple formulations with the best flame retardant performance for small-scale tests to obtain detailed performance vector data.

[0211] 5. According to the performance requirements of the reactor application, select the formula closest to the target performance vector as the final optimization result.

[0212] This optimization method makes full use of the multimodal characterization data of raw materials, and realizes the rapid optimization of the flame retardant epoxy resin formula for reactors through the prediction of machine learning models and the optimization of performance vectors. Compared with the traditional empirical formula development method, this method is more scientific, efficient and reliable.

[0213] Specifically, the principle of the present invention is:

[0214] 1. Comprehensive utilization of multimodal feature data

[0215] Existing formulation development methods usually rely only on single performance test data, such as flame retardancy, electrical properties, etc. The method of the present invention makes full use of multimodal characteristic data of raw materials, such as spectrum, morphology, thermal analysis, etc. These multimodal data can more comprehensively and accurately describe the chemical structure, physical form and other characteristics of the raw materials, providing a more reliable basis for subsequent performance prediction and formulation optimization.

[0216] By making full use of the multimodal feature data of raw materials, the method of the present invention can better characterize the properties of the raw materials themselves, thereby providing a basis for screening out better alternative raw materials. In the process of generating a recipe set, using the similarity of multimodal fingerprint data to replace raw materials can not only enrich the recipe space, but also ensure that the new recipe obtained has a high similarity with the original recipe in chemical and physical properties, thereby ensuring the feasibility of the recipe.

[0217] 2. Performance prediction based on machine learning

[0218] In order to quickly and accurately predict the comprehensive performance of the formula, the method of the present invention adopts a multimodal fusion prediction model based on deep learning. The model has a unique structure and can gradually and comprehensively consider the coupling relationship between different performance indicators through the cascade of multiple pyramid layers, and finally output a comprehensive performance prediction result.

[0219] Each pyramid layer contains 5 sub-network modules, which use convolutional neural networks, recurrent neural networks and other structures to extract the features of different modal data and predict the corresponding performance indicators. At the same time, the multimodal fusion network uses attention mechanism or gating mechanism to fuse the features of different modalities to obtain the final comprehensive performance prediction. This multimodal fusion method can make full use of the information contained in various feature data of raw materials, and is more accurate and reliable than the prediction of a single modality.

[0220] During the training process, the four prediction network modules are first trained separately, and supervised learning is performed using the corresponding modal data and their performance indicators. The output features of the four networks are then input into the multimodal fusion network for end-to-end training. The generalization ability of the model is ensured through methods such as cross-validation, and techniques such as transfer learning can be tried to further optimize the model performance.

[0221] This multimodal fusion prediction method based on machine learning can efficiently convert the rich characterization data of raw materials into key performance indicators of flame-retardant epoxy resin for reactors, providing a reliable basis for subsequent formula optimization.

[0222] 3. Optimization based on multi-index performance vector

[0223] The method of the present invention not only focuses on single flame retardant performance when optimizing the formula, but also pays more attention to multiple key indicators such as electrical performance, thermal performance, mechanical performance and processing performance. By constructing a comprehensive performance vector covering these performance indicators and taking it as the optimization target, it can ensure that the final formula obtained can meet the stringent performance requirements of the reactor in all aspects.

[0224] When selecting the optimal formula, the method of the present invention calculates the distance between the performance vector of each candidate formula and the target performance vector, and selects the formula closest to the target as the final result. This optimization method based on multi-index performance vectors can better balance various key performances and find the optimal formula that meets actual application needs.

[0225] Compared with the existing optimization methods that only focus on a single performance indicator, the method of the present invention can achieve comprehensive optimization of the performance of flame-retardant epoxy resin for reactors, providing strong support for meeting the demanding application requirements of reactors.

[0226] In summary, the flame retardant epoxy resin formulation optimization method for reactors proposed in the present invention makes full use of the multimodal feature data of raw materials and combines advanced machine learning technology to achieve rapid and accurate prediction and comprehensive optimization of formulation performance. Compared with the existing technology, this method has significant advantages in improving prediction accuracy, optimization efficiency, and achieving comprehensive performance optimization, and provides effective technical support for the development of flame retardant epoxy resin composite materials that meet the demanding application requirements of reactors.

[0227] In order to better implement the present invention, Example 2 of the present invention in a specific application scenario is provided below. In this Example 2, a company needs to develop a high-performance flame-retardant epoxy resin composite material as a core to meet the strict fire resistance requirements of the reactor to meet various key indicators of the product.

[0228] Firstly, a large number of flame-retardant epoxy resin samples for reactors with different raw material compositions were collected, and multimodal characterization analysis was performed on these samples to obtain rich raw material characteristic data. Specifically, they include:

[0229] 1. Spectral data: Fourier transform infrared spectrometer (FTIR) and Raman spectrometer were used to test the infrared spectrum and Raman spectrum of each sample. Infrared spectrum can reflect the existence and strength of chemical bonds, while Raman spectrum can obtain more detailed molecular vibration information.

[0230] 2. Morphological data: The microscopic morphology of the samples was characterized by scanning electron microscopy (SEM) and atomic force microscopy (AFM), and information such as particle size, shape, and distribution was obtained.

[0231] 3. Thermal analysis data: The thermal stability and phase change behavior of the samples were analyzed using thermogravimetric analyzer (TGA) and differential scanning calorimetry (DSC).

[0232] 4. Other characterization data: The structure and particle size of the samples were tested using X-ray diffractometer (XRD) and laser particle size analyzer.

[0233] Through the above multimodal characterization, a training data set containing 100 samples was obtained, each of which has complete multimodal fingerprint data and corresponding performance test results. The performance test of the samples covers indicators such as flame retardancy, electrical properties, thermal properties, mechanical properties and processing properties.

[0234] Based on this training dataset, a bidirectional pyramid neural network model based on multimodal fusion was established to predict the comprehensive performance of flame-retardant epoxy resin for reactors.

[0235] The model has 5 pyramid layers, each of which is responsible for predicting one or more performance indicators. Each pyramid layer contains 5 sub-network modules:

[0236] 1. Spectral data prediction network: Using convolutional neural network structure, extract the characteristics of spectral data and predict the corresponding performance indicators.

[0237] 2. Morphology data prediction network: Using convolutional neural network structure, extract the features of morphology data and predict the corresponding performance indicators.

[0238] 3. Thermal analysis data prediction network: Use a fully connected network or LSTM structure to extract the features of thermal analysis data and predict the corresponding performance indicators.

[0239] 4. Other representation data prediction network: Use a fully connected network or LSTM structure to extract features of other representation data and predict corresponding performance indicators.

[0240] 5. Multimodal fusion network: Use attention mechanism or gating mechanism to fuse different modal features and predict all performance indicators.

[0241] During the training process, the four prediction network modules are first trained separately, and supervised learning is performed using the corresponding modal data and their performance indicators. Then the output features of the four networks are input into the multimodal fusion network for end-to-end training. During the entire training process, the cross-validation method is used to ensure the generalization ability of the model.

[0242] The performance of the trained prediction model is shown in Table 1:

[0243] Table 1 Performance table of the model

[0244] Performance Indicators Prediction Error Flame retardant properties 5.2% Electrical properties 3.8% Thermal properties 4.6% Mechanical properties 6.1% Processing performance 4.9%

[0245] It can be seen that the prediction accuracy of the prediction model in various performance indicators is above 95%, which can provide a more reliable basis for subsequent formula optimization.

[0246] After completing the model training, a set of 500 formulas was generated for the existing flame-retardant epoxy resin formula for reactors by replacing raw materials. For each new formula, the optimal usage ratio of each raw material was calculated. The specific method is as follows:

[0247] First, a quantitative structure-activity relationship (QSPR) chemical model was established, which can predict the performance indicators of the formulation based on the multimodal fingerprint data of the raw materials. The QSPR model was trained by the least squares method using the samples in the training dataset:

[0248] y=f(x,p;θ)

[0249] Where x represents the multimodal fingerprint data vector of the raw material, p represents the usage ratio vector of the raw material, y represents the performance index vector, and θ is the model parameter.

[0250] The trained QSPR model can quickly predict the optimal usage ratio of each raw material in each new formula to ensure that the performance indicators of the formula are optimized.

[0251] Next, the multimodal fingerprint data and raw material ratios of these 500 new formulas were input into the previously trained deep learning prediction model to obtain the comprehensive performance prediction results for each formula.

[0252] According to the performance prediction results, 20 formulations with the best flame retardant performance were selected to form the first formulation set. For these 20 formulations, small-scale tests were carried out to test their detailed performance indicators, including:

[0253] Flame retardant properties: vertical burning rating, oxygen index, heat release rate, smoke density

[0254] Electrical properties: volume resistivity, dielectric constant, insulation strength, arc tracking index

[0255] Thermal properties: thermal shrinkage, glass transition temperature, thermal expansion coefficient, thermal stability

[0256] Mechanical properties: tensile strength, flexural strength, impact strength, hardness

[0257] Processing properties: fluidity, formability, processing temperature range

[0258] The multiple performance indicators of each recipe are combined into a performance vector as follows:

[0259]

[0260] in:

[0261] It represents the flame retardant performance index of the jth formulation;

[0262] represents the electrical performance index vector of the jth formulation;

[0263] represents the thermal performance index vector of the jth recipe;

[0264] Represents the mechanical property index vector of the jth formulation;

[0265] Represents the processing performance index vector of the jth recipe.

[0266] According to the performance target of the reactor product, the ideal performance vector y is determined target . Then the Euclidean distance between the performance vector and the target vector of each recipe in the first recipe set is calculated:

[0267] d (j) =‖y (j) -y target ‖

[0268] Finally, the recipe with the closest distance to the target performance vector is selected, i.e. As the final optimization result.

[0269] Through the above steps, a high-performance flame-retardant epoxy resin composite material that meets the stringent performance requirements of the reactor is developed. The specific performance indicators are shown in Table 2 below:

[0270] Table 2 Performance of high performance flame retardant epoxy resin composite materials

[0271]

[0272]

[0273] It can be seen from the above table that various performance indicators of the flame retardant epoxy resin composite material can meet the strict requirements of reactor products, highlighting the significant advantages of the method of the present invention in achieving comprehensive optimization of flame retardant epoxy resin for reactors.

[0274] Compared with the traditional empirical formula development method, the method of the present invention mainly embodies the following advantages:

[0275] 1. Make full use of the multimodal characteristic data of raw materials to improve the accuracy of formula performance prediction. The deep learning model can effectively convert the multimodal information of raw materials such as spectrum, morphology, thermal analysis, etc. into key performance indicators, providing a reliable basis for subsequent optimization.

[0276] 2. Greatly improved the efficiency and reliability of formula optimization. Combining the prediction of raw material ratio by QSPR model and the prediction of performance by deep learning model, the experimental requirements are greatly reduced and the development cycle is shortened. At the same time, small-scale test verification is used to further ensure that the optimal formula obtained can meet the actual application needs.

[0277] 3. The overall performance optimization of flame-retardant epoxy resin for reactors is achieved. The method of the present invention focuses on multiple indicators such as flame retardant properties, electrical properties, thermal properties, mechanical properties and processing properties to ensure that the optimal formula obtained can meet the stringent requirements of reactors in all aspects.

[0278] In summary, through the method of the present invention, the company has successfully developed a high-performance flame-retardant epoxy resin composite material that meets the strict performance indicators of the reactor, providing strong support for the further upgrading of the reactor products.

[0279] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing the formulation of a flame-retardant epoxy resin for a reactor, characterized in that: The following steps are involved: S10, obtaining multimodal fingerprint data of each raw material in the flame-retardant epoxy resin formula for the reactor; S20, for each raw material, selecting a material whose multimodal fingerprint data similarity with the raw material is greater than a preset fingerprint similarity threshold as a substitute raw material, and generating a recipe set by using a traversal loop method; S30, calculating the raw material ratio of each formula in the formula set using a quantitative structure-activity relationship chemical model; S40, using a pre-trained epoxy resin flame retardant index prediction model, inputting the multimodal fingerprint and raw material ratio of each formula in the formula set to obtain a flame retardant index of each formula; S50, selecting multiple formulas with the best flame retardant index to form a first formula set, and obtaining multiple necessary properties of the epoxy resin material obtained by conducting a small test on each formula in the first formula set, and generating a performance vector; S60. According to the performance vector of each formula, a formula that is closest to the preset required performance vector of the flame-retardant epoxy resin for the reactor is selected as the optimized formula.

2. The method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to claim 1, characterized in that: The multimodal fingerprint data refers to characteristic data of raw materials obtained through various analysis methods, including spectral data, morphological data, thermal analysis data and other characterization data, and the other characterization data include X-ray diffraction data, particle size analysis data, specific surface area data and density data.

3. The method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to claim 1, characterized in that: The performance vector is a vector composed of flame retardant properties, electrical properties, thermal properties, mechanical properties, and processing properties.

4. The method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to claim 1, characterized in that: The structure of the epoxy resin flame retardant index prediction model is a bidirectional pyramid neural network model based on multimodal fusion, which is composed of multiple pyramid layers, and each pyramid layer is responsible for predicting one or more types of performance indicators.

5. The method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to claim 4, characterized in that: The multiple pyramid layers include: the first pyramid layer is responsible for the flame retardant performance index, the second pyramid layer is responsible for the comprehensive index of flame retardant performance and electrical performance, the third pyramid layer is responsible for the comprehensive index of flame retardant performance, electrical performance and thermal performance, the fourth pyramid layer is responsible for the comprehensive index of flame retardant performance, electrical performance, thermal performance and mechanical performance, and the fifth pyramid layer is responsible for the comprehensive index of flame retardant performance, electrical performance, thermal performance, mechanical performance and processing performance.

6. The method for optimizing the formulation of a flame retardant epoxy resin for a reactor according to claim 5, characterized in that: Each pyramid layer includes five network modules: spectral data prediction network module, morphological data prediction network module, thermal analysis data prediction network module, other characterization data prediction network module and multimodal fusion network module.

7. The method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to claim 6, characterized in that: The multimodal fusion network module is used to fuse the outputs of the spectral data prediction network module, the morphological data prediction network module, the thermal analysis data prediction network module, and other characterization data prediction network modules using an attention mechanism or a gating mechanism, and ultimately predict all performance indicators corresponding to all multimodal fingerprint data.

8. The method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to claim 7, characterized in that: The epoxy resin flame retardant index prediction model training process first trains four prediction network modules separately, then inputs the output features of the four networks into the multimodal fusion network module for end-to-end training, and uses a cross-validation method throughout the training process to ensure the generalization ability of the model.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for optimizing the formulation of a flame-retardant epoxy resin for a reactor according to any one of claims 1 to 8.

10. A flame retardant epoxy resin formulation optimization system for reactors, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Citation Information

Cited By

  • Formula optimization method of two-component solvent-free brush coating type transparent waterproof polyurea

    CN119446324A

  • Dry-type air-core reactor prepared based on reactive flame-retardant epoxy resin

    CN120137350A

  • Aluminum alloy component cost optimization method and system

    CN121480886A

  • Epoxy resin formula reverse design system based on iterative optimization and interpretable prediction

    CN121838903A