CN-121257339-B - Fiber glass modulus prediction method based on optimized MAKISHIMA-Mackenzie formula

CN121257339BCN 121257339 BCN121257339 BCN 121257339BCN-121257339-B

Abstract

The invention discloses a fiber glass modulus prediction method based on an optimization MAKISHIMA-Mackenzie formula. The invention establishes an oxide dissociation energy optimization model based on a differential weight coefficient for the first time, can optimally characterize the difference of contribution of different oxides to the model in a glass network structure by introducing independent optimization coefficients, greatly reduces the prediction error on the high-modulus glass modulus compared with the traditional MM formula, simultaneously reserves clear physical images based on dissociation energy and components, can reveal the internal connection between the glass microstructure and macroscopic modulus, and avoids the 'black box' problem of a pure machine learning method. In addition, the invention can complete the modulus prediction by only using the easily obtained basic parameters such as the dissociation energy, the molar volume and the like, improves the calculation efficiency by more than hundred times, does not need high-performance calculation resources, and obviously reduces the time cost and the technical threshold for developing the high-performance glass fiber.

Inventors

ZHAO ZIYU
ZHAO MING
LANG YUDONG
LIU XIN
Lv Shiwu
ZHAO QIAN

Assignees

南京玻璃纤维研究设计院有限公司

Dates

Publication Date: 20260512
Application Date: 20251204

Claims (9)

1. A method for predicting fiber glass modulus based on an optimized MAKISHIMA-Mackenzie formula, the method comprising the steps of: Step 1, collecting components and Young modulus data of glass fibers of different systems from a disclosed glass performance database to form a dataset, wherein the dataset comprises data of a SiO 2 -Al 2 O 3 -MgO ternary system and an expansion system thereof; step 2, calculating the total number of atoms and the total number of different cations in each glass molecular model; Step3, constructing a correlation model of the components of the glass fiber and Young modulus thereof based on the total number of atoms, the total number of different cations, different oxide dissociation energies and molar volumes; Step 4, optimizing the association relation model by utilizing a genetic algorithm based on the data set so as to minimize root mean square error between the predicted modulus and the experimental modulus; step 5, based on the data set, selecting data of different Young modulus intervals to verify generalization of the association relation model, and adjusting according to a verification result to obtain a final association relation model; step 6, inputting the components of the glass fiber to be predicted into a final association relation model, and outputting a corresponding Young modulus result; the association relation model of the components of the glass fiber and the Young modulus of the glass fiber constructed in the step 3 is specifically as follows: ; wherein F is a modulus-related parameter, N is the total number of atoms in the model, i is an oxide species, Is the component mole ratio of the i-th oxide, Is the dissociation energy of the i-th oxide, Optimizing the coefficient for the dissociation energy of the ith oxide, The molar volume of the i-th oxide, Indicating the total number of oxide species.
2. The method for predicting the modulus of fiber glass based on the optimized MAKISHIMA-Mackenzie formula according to claim 1, wherein the correlation model is optimized by adopting NGAS-II algorithm in step 4.
3. The method for predicting the fiber glass modulus based on the optimized MAKISHIMA-Mackenzie formula according to claim 2, wherein the correlation model is optimized by NGAS-II algorithm in step 4, and specifically comprises: step 4-1, dividing the data set in the step1 into a training set, a testing set and a verification set; step 4-2, constructing an optimization objective function: ; Wherein n is the number of training samples, E pred,j is the predicted Young's modulus of the jth sample, E exp,j is the corresponding experimental Young's modulus measurement value, and the optimization objective is to minimize the RMSE value; Step 4-3, setting an initial value range of dissociation energy coefficients delta i of each oxide, and generating an initial solution set with a population scale of Q, wherein each individual represents a complete group of dissociation energy coefficient combinations; step 4-4, for each individual in the population, i.e. a group of dissociation energy coefficient combinations, performing the following operations: ① Substituting the group of dissociation energy coefficient combinations into the association relation model established in the step 3; ② Calculating the related parameters F of the modulus of all glass fibers in the training set; ③ Establishing a linear regression relationship between the modulus-related parameter F and the experimental Young's modulus E exp : E exp = a·F + b Determining coefficients a and b by a least square method; ④ Calculating an RMSE value corresponding to the dissociation energy coefficient combination of the group as a fitness index, wherein the smaller the RMSE is, the higher the fitness is; step 4-5, genetic manipulation is performed, including: Selecting m individuals at random each time by adopting a tournament selection method, selecting the individuals with the highest fitness to enter the next generation, and repeating the process until the individuals with the same number as the population scale are selected; Performing crossover operation, pairing selected individuals with probability P c , generating offspring by adopting a simulated binary crossover method, and performing cross-over on each pair of father and father , ) The offspring coefficient calculation formula is: ; Wherein, the The i-th oxide dissociation energy coefficient of the first parent, The i-th oxide dissociation energy coefficient of the second parent, The i-th oxide dissociation energy coefficient for the first offspring, The i-th oxide dissociation energy coefficient of the second filial generation, and beta is a cross distribution parameter; Performing mutation operation, mutating the individual with probability P m , and using polynomial mutation method to obtain dissociation energy coefficient of the selected individual The perturbation is performed according to the following formula: ; In the formula, The dissociation energy coefficient of the i-th oxide after mutation, Is the dissociation energy coefficient of the ith oxide before mutation, Is the maximum value of the dissociation energy coefficient of the i oxides, Is the minimum value of the dissociation energy coefficient of the i-th oxide, Random numbers in the range of [ -0.1,0.1 ]; And 4-6, performing iterative optimization, namely taking a new population generated through selection, crossing and mutation operations as a next generation population, and recording the optimal fitness value and the corresponding dissociation energy coefficient combination in each generation population, wherein the termination condition of the iteration is that the iteration is stopped when any one of the following conditions is met: ; wherein the units of the second preset threshold value and the third preset threshold value are respectively ; Step 4-7, outputting the dissociation energy coefficient of each oxide finally obtained by optimization And outputting an optimal set of dissociation energy coefficient combinations, thereby obtaining an optimized association relation model.
4. The method for predicting the modulus of fiber glass based on the optimized MAKISHIMA-Mackenzie formula according to claim 3, wherein the initial value range in the step 4-2 is set to be [0.5, 2.0].
5. The method for predicting fiber glass modulus based on optimized MAKISHIMA-Mackenzie formula as claimed in claim 3, wherein the first preset threshold value is 500 and the second preset threshold value is 0.01 in the steps 4-6 , The third preset threshold value is 0.5 。
6. A method for predicting fiber glass modulus based on the optimized MAKISHIMA-Mackenzie formula as defined in claim 3, wherein the verification process in step 5 includes: carrying out Young modulus prediction on verification set data outside the training set by using the optimized association relation model in the step 4, and calculating a prediction error of the verification set; And if the prediction error of the verification set is larger than the error of the training set and the difference exceeds a preset threshold, adjusting the genetic algorithm parameters or expanding the training set, and then returning to the execution step 4 to re-optimize the association relation model.
7. A fiber glass modulus prediction system based on the method of any one of claims 1 to 6, wherein the system comprises: the first module is used for acquiring the components and Young modulus data of glass fibers of different systems from the disclosed glass performance database to form a data set, wherein the data set comprises data of a SiO 2 -Al 2 O 3 -MgO ternary system and an expansion system thereof; A second module for calculating the total number of atoms and the total number of different cations in each glass molecular model; The third module is used for constructing a correlation model of the components of the glass fiber and Young modulus of the glass fiber based on the total number of atoms, the total number of different cations, different oxide dissociation energies and molar volumes; A fourth module for optimizing the association model using a genetic algorithm based on the dataset to minimize a root mean square error between the predicted modulus and the experimental modulus; a fifth module, configured to verify generalization of the association model by selecting data of different young modulus intervals based on the data set, and adjust the generalization according to a verification result to obtain a final association model; and a sixth module for inputting the components of the glass fiber to be predicted into a final association relation model and outputting a corresponding Young modulus result.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of claims 1 to 6 when executing the computer program.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of any one of claims 1 to 6.

Description

Fiber glass modulus prediction method based on optimized MAKISHIMA-Mackenzie formula Technical Field The invention belongs to the technical field of glass fibers, and particularly relates to a fiber glass modulus prediction method based on an optimized MAKISHIMA-Mackenzie formula. Background Glass fiber is used as an important reinforcing material and has wide application in the fields of aerospace, wind power generation, constructional engineering and the like. In particular in the field of clean energy equipment, high modulus glass fibers are a key base material for wind turbine blade manufacture. With the rapid development of global energy structure transformation and wind power industry, the size of wind power blades is continuously increased, and higher requirements are put on the mechanical properties of glass fibers, particularly the elastic modulus. The high modulus glass fiber can obviously improve the rigidity and bearing capacity of the blade, prolong the service life and reduce the total life cycle cost, so the development of glass fiber materials with higher modulus has become an urgent need for industry development. However, the traditional glass fiber formula design mainly depends on a trial-and-error method, namely glass samples with different component ratios are prepared through a large number of experiments, and the formula is gradually optimized after the performance of the glass samples is tested. The method has the obvious defects that a large amount of melting experiments are needed, a large amount of time, labor and raw material cost are consumed, theoretical guidance is lacked, the optimization efficiency is low, the component-performance relationship is difficult to systematically explore, and the research and development process of the novel high-modulus glass fiber is severely restricted. In recent years, with the development of artificial intelligence technology, a machine learning method is introduced into the field of glass material performance prediction, and material design is accelerated by establishing a big data driven prediction model. However, the machine learning method also has inherent limitations that the prediction accuracy is highly dependent on the quality and quantity of training data, the generalization capability and reliability of a model are often insufficient for a new system or a specific component range with scarce data, and meanwhile, the machine learning model lacks of physical interpretability, so that the inherent relation between a glass microstructure and macroscopic performance is difficult to reveal, and the modulus strengthening mechanism is not easy to understand deeply. The quantitative structure-performance relationship (QSPR) analysis method based on the theoretical model provides another path for glass modulus prediction. The method associates microstructure parameters of glass with macroscopic properties by establishing descriptors, and has both calculation efficiency and physical significance. For example, patent publication (CN 11623001 a) and document "Yan J, Zhang Y, Wang F, et al. Electronegativity-Based QSPR Analysis for Understanding Structure–Property Relationships of Glass Materials. The Journal of Physical Chemistry B. 2025;129:5033-46." report a model QSPR constructed based on oxide formation energy and coordination number. However, the existing theoretical model has a large error in predicting the modulus of fiber glass, particularly MgO-containing system glass. At present, the main stream system of the high-modulus glass fiber is a SiO 2-Al2O3 -MgO ternary system, and the calculation accuracy of the existing model on the system is insufficient, so that the application of the system in actual formula design is limited. In addition, QSPR models rely on molecular dynamics to simulate and calculate coordination numbers and other structural parameters, and the process is large in calculated amount and long in time consumption, so that the efficiency of glass fiber research and development is reduced. In addition to the QSPR model, the MAKISHIMA-Mackenzie (MM) formula is another classical glass modulus prediction method. The MM formula establishes the relationship between modulus and glass composition and dissociation energy based on the bulk density theory of glass: wherein E is Young's modulus, V t is bulk density factor, C i is mole fraction of the i-th oxide, and G i is dissociation energy. The MM formula has the advantages of clear physical meaning and simple calculation, but the formula also has obvious limitations that on one hand, for a high-modulus fiber glass system, the prediction accuracy of the MM formula is insufficient, the calculated modulus is always obviously smaller than the actual glass modulus, on the other hand, the calculation of the bulk density factor is required to be based on the actual density of glass, and when the glass of an unknown system is designed or predicted, the accurate density of the glass is d