High-modulus glass fiber formula design method and glass fiber
By optimizing the high-modulus glass fiber formulation design through machine learning models, the problem of time-consuming and labor-intensive traditional methods has been solved, enabling efficient and accurate material research that meets the needs of large wind turbine blades.
Patent Information
- Application Number
- CN202511415909.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-30
- Publication Date
- 2025-12-30
AI Technical Summary
Existing high-modulus glass fibers are insufficient to meet the design requirements of giant wind turbine blades, and traditional R&D methods consume a lot of manpower and resources and are difficult to iterate and upgrade quickly.
Machine learning models are used to collect experimental data on high-modulus glass fibers from a database, establish training and testing sets, select appropriate computing platforms and models, evaluate and optimize the models, predict the performance of high-modulus glass fiber formulations, and select the formulations that meet the requirements.
It achieves efficient and accurate high-modulus glass fiber formulation design, with an elastic modulus of over 95GPa, shortening the development cycle and making it suitable for the production of large wind turbine blades.
Smart Images
Figure CN121237246A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of high modulus glass fibers, and particularly relates to a high modulus glass fiber formula design method and a glass fiber. BACKGROUND
[0002] Under the promotion of the "double carbon" policy, wind power, as a representative of clean and renewable energy, is in a stage of rapid development. With the rapid growth of wind turbine capacity, wind turbine blades have grown from over 30 meters to nearly 100 meters. As the core reinforcing material of wind turbine blades, the elastic modulus of high modulus glass fiber has increased from 75 GPa of E glass fiber to 95 GPa. With the rapid rise of the offshore wind power industry, the length of wind turbine blades exceeds 120 meters, and the weight of the blades increases significantly. The existing high modulus glass fiber is difficult to meet the design requirements of giant wind turbine blades, and therefore it is urgent to develop high modulus glass fiber with higher elastic modulus.
[0003] High modulus glass fiber production technology is difficult, and the total output is low, but its demand growth rate is more than 4 times that of ordinary glass fiber, and its excellent cost performance is currently irreplaceable by other fiber materials.
[0004] The traditional glass fiber research and development method is based on trial and error and experience, and a large amount of manpower, material resources and time cost are required to find the final glass component that meets the design index, which is difficult to meet the current demand of wind power industry for rapid iteration and upgrading of glass fiber performance.
[0005] Therefore, the application provides a high modulus glass fiber formula design method and a glass fiber. SUMMARY
[0006] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background art.
[0007] The technical scheme adopted by the application to solve the technical problem is a high modulus glass fiber formula design method, comprising the following steps:
[0008] S1, data collection: collecting high modulus glass fiber experimental data of experiments completed by a plurality of researchers through a database, and making a data set according to the collected high modulus glass fiber experimental data;
[0009] S2, selecting a computing platform: selecting a reasonable computer language and a development environment to complete the training and verification of the machine learning model;
[0010] S3, establishment of the data set: dividing the collected data set into a training set and a test set according to a certain proportion;
[0011] S4, model establishment and selection: determine the influence factors and important properties, and select the corresponding model for training according to the structural characteristics of the data set;
[0012] S5, model evaluation: input the test set into the machine learning model established by different algorithms, judge whether the evaluation index meets the requirements, if it meets the requirements, select the model with the best evaluation index as the best machine learning model, otherwise, repeat step S4 until the evaluation index meets the requirements;
[0013] S6, based on the best machine learning model, the performance of a series of high modulus glass fiber formulations is predicted, and the performance of all high modulus glass fiber formulations is obtained;
[0014] S7, the high modulus glass fiber formulation with the corresponding performance is preferably selected.
[0015] Preferably, in S1, the database includes one or more of Sci Glass, ScienceDirect and Web of Science.
[0016] Preferably, in S2, the computer language is Python language; and the development environment includes Scikit-learn, NumPy, Pandas, Pytorch and SHAP software packages.
[0017] Preferably, in S3, the data set is randomly divided into a training set and a test set, wherein the training set accounts for 80%, and the test set accounts for 20%.
[0018] Preferably, in S3, the data set needs to be preprocessed and feature engineered, including the following steps:
[0019] S301, the preprocessing of the data set mainly includes data deduplication, elimination of empty label samples and removal of samples with low content;
[0020] S302, the feature engineering of the data set is used to enhance the accuracy of model prediction and reduce the noise interference of training data, and is used to screen samples of chemical elements with extremely low frequency, insufficient abundance and significant influence on oxygen balance.
[0021] Preferably, in S4, the model establishment and selection specifically includes the following steps:
[0022] S401, the influence factors are determined as various oxides constituting high modulus glass fiber;
[0023] S402, the important properties include density, elastic modulus, liquidus temperature and glass fiber forming temperature;
[0024] S403, rotating the machine learning model includes random forest regression and artificial neural network.
[0025] Preferably, in S5, the evaluation index includes the determination coefficient R 2 and the root mean square error RMSE, the calculation formula is as follows:
[0026]
[0027]
[0028] In the formula, represents the number of data in the data set, represents the measured value, is the predicted value, represents the average value of the measured value;
[0029] The evaluation index determination coefficient R 2 ≥0.88 is a demand value, R 2 ≥0.9 is a preferred value.
[0030] Preferably, the high modulus glass fiber adopts the formula oxide and mass percentage as follows:
[0031] SiO2 53.21%~68.86%
[0032] Al2O3 5.56%~21.51%
[0033] CaO 0%~11.67%
[0034] MgO 11.31%~27.81%
[0035] Na2O 0%~1%
[0036] K2O 0%~0.8%
[0037] Li2O 0%~3%
[0038] TiO2 0%~1%
[0039] Y2O3 0%~6.5%
[0040] La2O3 0%~2%
[0041] CeO2 0%~1%
[0042] ZnO 0%~1%
[0043] Fe2O3 0%~1%
[0044] B2O3 0%~1%
[0045] MnO 0%~1%
[0046] ZrO2 0%~1%.
[0047] Preferably, the mass percentage of oxides in the high-modulus glass fiber satisfies the following conditions: Li2O+CaO≥0.5%, Na2O+K2O≤1.2%, Al2O3+MgO≥16.9%, and SiO2+Al2O3≥50.8%.
[0048] Preferably, the formulation concentration gradient of the high modulus glass fiber is 0.5 wt%.
[0049] The beneficial effects of this invention are as follows:
[0050] 1. The high-modulus glass fiber formulation design method and glass fiber of the present invention construct the best machine learning model through training data to predict materials with target properties, realize efficient and accurate new material research technology, and promote the application of machine learning in interdisciplinary research; the high-modulus glass fiber of the present invention has an elastic modulus of up to 95 GPa and contains a small amount of rare metal oxides, which can be used in the production of large wind turbine blades.
[0051] 2. The high-modulus glass fiber formulation design method and glass fiber described in this invention shorten the development cycle of high-modulus glass fiber formulations by applying efficient machine learning computing and prediction capabilities to research in the field of high-modulus glass fibers and utilizing existing research results and data.
[0052] 3. The high-modulus glass fiber formulation design method and glass fiber described in this invention can directly output the relevant performance results of high-modulus glass fiber by extracting the mass percentage of each element from the high-modulus glass fiber formulation and designing the relationship between the element combination and coefficient ratio of the input formulation, thereby achieving the effect of efficient verification of new materials. Attached Figure Description
[0053] The invention will now be further described with reference to the accompanying drawings.
[0054] Figure 1 This is a flowchart of a high-modulus glass fiber formulation design method according to the present invention;
[0055] Figure 2 This is a schematic diagram of a random forest structure in a high-modulus glass fiber formulation design method of the present invention;
[0056] Figure 3 This is a schematic diagram of the neural network structure of a high-modulus glass fiber formulation design method according to the present invention;
[0057] Figure 4A comparison chart of evaluation results for multiple model training sets;
[0058] Figure 5 A comparison chart of evaluation results for multiple model test sets;
[0059] Figure 6 This is a density error diagram of the high-modulus glass fiber of the present invention;
[0060] Figure 7 Error diagram of elastic modulus of high-modulus glass fiber in this invention;
[0061] Figure 8 This invention provides a liquidus temperature error diagram for high-modulus glass fibers.
[0062] Figure 9 Error diagram of high modulus glass fiber molding temperature in this invention; Detailed Implementation
[0063] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0064] like Figure 1 As shown in the figure, a high-modulus glass fiber formulation design method according to an embodiment of the present invention includes the following steps:
[0065] S1. Data Collection: Collect experimental data on high modulus glass fibers from multiple researchers through a database, and create a dataset based on the collected high modulus glass fiber experimental data.
[0066] S2. Selecting a computing platform: Selecting a suitable computer language and development environment to complete the training and validation of machine learning models;
[0067] S3. Dataset creation: Divide the collected dataset into training set and test set according to a certain ratio;
[0068] S4. Model Establishment and Selection: Determine the influencing factors and important performance characteristics, and select the appropriate model for training based on the structural characteristics of the dataset;
[0069] S5. Model Evaluation: Input the test set into the machine learning models built by different algorithms, and determine whether the evaluation index meets the requirements. If it does, select the model with the best evaluation index as the best machine learning model. Otherwise, repeat step S4 until the evaluation index meets the requirements.
[0070] S6. Based on the best machine learning model, the performance of a series of high modulus glass fiber formulations is predicted to obtain the performance of all high modulus glass fiber formulations.
[0071] S7. Select the high-modulus glass fiber formulation with corresponding performance.
[0072] Furthermore, in S1, the database includes one or more of Sci Glass, ScienceDirect, and Web of Science.
[0073] Furthermore, in step S2, the computer language is Python; the development environment includes multiple software packages such as Scikit-learn, NumPy, Pandas, PyTorch, and SHAP.
[0074] Furthermore, in step S3, the dataset is randomly divided into a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%.
[0075] Furthermore, in step S3, the dataset needs to undergo corresponding preprocessing and feature engineering, including the following steps:
[0076] S301. The preprocessing of the dataset mainly includes data deduplication, removal of empty label samples, and removal of samples with too low content.
[0077] S302. Feature engineering of the dataset is used to enhance the accuracy of model predictions and reduce noise interference in the training data, and to screen chemical element samples with extremely low frequency, insufficient abundance and significant impact on oxygen balance.
[0078] Furthermore, in step S4, the establishment and selection of the model specifically includes the following steps:
[0079] S401. The influencing factors are the various oxides that make up high-modulus glass fibers.
[0080] S402. Determine important properties including density, elastic modulus, liquidus temperature and glass fiber forming temperature;
[0081] S403, Rotational machine learning models include random forest regression and artificial neural networks.
[0082] Furthermore, in step S5, the evaluation index includes the coefficient of determination R. 2 The root mean square error (RMSE) is calculated using the following formula:
[0083]
[0084]
[0085] In the formula, Represents the number of data points in the dataset. Indicates the measured value. For predicted values, This represents the average of the measured values;
[0086] The evaluation index, the coefficient of determination R, is mentioned above. 2 A value ≥0.88 is required to meet the demand. 2 ≥0.9 is the preferred value.
[0087] Furthermore, the mass percentage of oxides in the high-modulus glass fiber satisfies the following conditions: Li2O+CaO≥0.5%, Na2O+K2O≤1.2%, Al2O3+MgO≥16.9%, and SiO2+Al2O3≥50.8%.
[0088] Furthermore, the formulation concentration gradient of the high-modulus glass fiber is 0.5wt%.
[0089] When using the design method provided by this invention,
[0090] During data collection, experimental data on high-modulus glass fibers were collected from databases such as Sci Glass, ScienceDirect, and Web of Science. The experimental data on high-modulus glass fibers included the oxide composition of the high-modulus glass fiber formulation, as well as properties such as density, elastic modulus, liquidus temperature, and fiber forming temperature. The data was allocated and calculated according to the molar ratio of the high-modulus glass fiber formulation to obtain the molar ratio of each element, and the data was collected one by one to build a dataset.
[0091] When building the dataset, the original data is divided into training set and test set. The dataset is divided according to a set ratio: the training set accounts for 80% and the test set accounts for 20%. Data deduplication, empty label samples are removed and samples with low content are removed from the dataset. Corresponding feature engineering is also performed, including feature selection, feature extraction and feature transformation.
[0092] When selecting a computing platform, Python is used as the computer language, and the development environment includes PyCharm and third-party libraries such as Scikit-learn, NumPy, Pandas, PyTorch, and SHAP available in the Python language community.
[0093] When establishing and selecting models, a machine learning model was used to establish objective functions for high-modulus glass fiber formulation and properties such as density, elastic modulus, liquidus temperature and fiber forming temperature, and common models such as random forest regression and artificial neural networks were adopted.
[0094] During model evaluation, the test set is input into machine learning models built with different algorithms to determine whether the evaluation metrics meet the requirements. If they do, the model with the best evaluation metrics is selected as the best machine learning model; otherwise, step S4 is repeated until the evaluation metrics meet the requirements. The formulas for calculating the coefficient of determination R² and root mean square error RMSE are as follows:
[0095]
[0096]
[0097] In the formula, Represents the number of data points in the dataset. Indicates the measured value. For predicted values, This represents the average of the measured values;
[0098] The requirement is that the coefficient of determination R² ≥ 0.88 is necessary to meet the evaluation criteria for machine learning models.
[0099] Finally, based on the optimal machine learning model, the performance of a series of gradient high-modulus glass fiber formulations was predicted, yielding the performance of all high-modulus glass fiber formulations. The content of each oxide is a percentage of the constituent components, ranging from 0% to 100%. The oxides in the high-modulus glass fiber formulations include SiO2, Al2O3, CaO, MgO, Na2O, K2O, Li2O, TiO2, Y2O3, CeO2, La2O3, ZnO, Fe2O3, B2O3, MnO, and ZrO2. The mass percentage of oxides in the high-modulus glass fiber satisfies the following conditions: Li2O+CaO≥0.5%, Na2O+K2O≤1.2%, Al2O3+MgO≥16.9%, and SiO2+Al2O3≥50.8%. The high-modulus glass fiber formulations with the corresponding performance were then selected.
[0100] The high-modulus glass fiber of this invention has an elastic modulus of up to 95 GPa and contains a small amount of rare metal oxides, which can be used in the production of large wind turbine blades.
[0101] The aforementioned high-modulus glass fibers can be prepared using the following method:
[0102] A1: Calculate the required mass of various raw materials based on the high modulus glass fiber formulation;
[0103] A2: Weigh all the raw materials and send them to the mixing tank for mixing. After mixing evenly, the mixture is sent to the kiln head silo of the pool kiln to obtain the batch material.
[0104] A3: The mixture in the kiln head hopper is fed into the pool kiln. In the pool kiln, the mixture is clarified and homogenized under high temperature conditions of 1580-1650℃. High-quality glass melt enters the drawing operation channel.
[0105] A4: The molten glass in the drawing process channel is cooled to 1265-1390℃, flows out through the platinum spindle, and is quickly drawn into glass fibers by the drawing machine.
[0106] As a specific example, the density and elastic modulus properties of the high-modulus glass fiber were tested.
[0107] The experimental performance data of the high-modulus glass fibers designed using machine learning models in Examples 1-10 are shown in Table 1.
[0108] Table 1 Data Table for Examples 1-10
[0109]
[0110]
[0111] Based on the above figure, the final formulation of the high-modulus glass fiber is Example 2, and the mass percentage of its oxides is:
[0112] The composition of the high-modulus glass fiber of this invention is as follows: SiO2: 60.26%; Al2O3: 19.55%; CaO: 3.35%; MgO: 12.67%; Na2O: 0; K2O: 0; Li2O: 1.86%; TiO2: 0; Y2O3: 2.3%; CeO2: 0; La2O3: 0; ZnO: 0; Fe2O3: 0; B2O3: 0.1%; MnO: 0; ZrO2: 0. The high-modulus glass fiber of this invention has an elastic modulus of up to 95 GPa or higher. The glass fiber composition contains a small amount of rare metal oxides and can be used in the production of large wind turbine blades.
[0113] Working principle: During data collection, experimental data on high-modulus glass fibers are collected from databases such as Sci Glass, ScienceDirect, and Web of Science. The experimental data on high-modulus glass fibers includes the oxide composition of the high-modulus glass fiber formulation, as well as properties such as density, elastic modulus, liquidus temperature, and fiber forming temperature. The data is allocated and calculated according to the molar ratio of the high-modulus glass fiber formulation to obtain the molar ratio of each element, and is collected one by one to build a dataset.
[0114] When building the dataset, the original data is divided into training set and test set. The dataset is divided according to a set ratio: the training set accounts for 80% and the test set accounts for 20%. Data deduplication, empty label samples are removed and samples with low content are removed from the dataset. Corresponding feature engineering is also performed, including feature selection, feature extraction and feature transformation.
[0115] When selecting a computing platform, Python is used as the computer language, and the development environment includes PyCharm and third-party libraries such as Scikit-learn, NumPy, Pandas, PyTorch, and SHAP available in the Python language community.
[0116] When establishing and selecting models, a machine learning model was used to establish objective functions for high-modulus glass fiber formulation and properties such as density, elastic modulus, liquidus temperature and fiber forming temperature, and common models such as random forest regression and artificial neural networks were adopted.
[0117] During model evaluation, the test set is input into machine learning models built with different algorithms to determine whether the evaluation metrics meet the requirements. If they do, the model with the best evaluation metrics is selected as the best machine learning model; otherwise, step S4 is repeated until the evaluation metrics meet the requirements. The formulas for calculating the coefficient of determination R² and root mean square error RMSE are as follows:
[0118]
[0119]
[0120] In the formula, Represents the number of data points in the dataset. Indicates the measured value. For predicted values, This represents the average of the measured values;
[0121] The requirement is that the coefficient of determination R² ≥ 0.88 is necessary to meet the evaluation criteria for machine learning models.
[0122] Finally, based on the optimal machine learning model, the performance of a series of gradient high-modulus glass fiber formulations is predicted, yielding the performance of all high-modulus glass fiber formulations. The content of various oxides is a percentage of the constituent components, ranging from 0% to 100%. The oxides in the high-modulus glass fiber formulations include SiO2, Al2O3, CaO, MgO, Na2O, K2O, Li2O, TiO2, Y2O3, CeO2, La2O3, ZnO, Fe2O3, B2O3, MnO, and ZrO2. The mass percentage of oxides in the high-modulus glass fiber satisfies the following conditions: Li2O+CaO≥0.5%, Na2O+K2O≤1.2%, Al2O3+MgO≥16.9%, SiO2+Al2O3≥50.8%. The high-modulus glass fiber formulations with corresponding performance are then selected. The high-modulus glass fiber of this invention has an elastic modulus of up to 95 GPa, and contains a small amount of rare metal oxides, making it suitable for the production of large wind turbine blades.
[0123] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A high modulus glass fiber formulation design method characterized by: The method comprises the following steps: S1, data collection: collecting high-modulus glass fiber experimental data of a plurality of scientific researchers who have completed experiments through a database, and making a data set according to the collected high-modulus glass fiber experimental data; S2, selecting a computing platform: selecting a reasonable computer language and a development environment to complete the training and verification of the machine learning model; S3, establishment of the data set: dividing the collected data set into a training set and a test set according to a set proportion; S4, establishment and selection of the model: determining the influence factors and important performances, and selecting a corresponding model for training according to the structural characteristics of the data set; S5, model evaluation: inputting the test set into different machine learning models to determine whether the evaluation index meets the requirements, if the evaluation index meets the requirements, selecting the model with the best evaluation index as the best machine learning model, otherwise, repeating step S4 until the evaluation index meets the requirements; S6, based on the best machine learning model, the performance of a series of gradient high-modulus glass fiber formulations is predicted to obtain the performance of all high-modulus glass fiber formulations; S7, selecting a high-modulus glass fiber formulation with a corresponding performance.
2. A high modulus glass fiber formulation design method according to claim 1, characterized by: In the S1, the database comprises one or more of Sci Glass, ScienceDirect and Web of Science.
3. A high modulus glass fiber formulation design method according to claim 2, characterized by: In the S2, the computer language is Python language; and the development environment comprises a plurality of software packages such as Scikit-learn, NumPy, Pandas, Pytorch and SHAP.
4. The method of designing a high modulus glass fiber formulation and the glass fiber according to claim 3, wherein: In the S3, the data set is randomly divided into a training set and a test set, wherein the proportion of the training set is 80%, and the proportion of the test set is 20%.
5. A high modulus glass fiber formulation design method according to claim 4, characterized by: In the S3, the data set needs to be preprocessed and feature engineered, comprising the following steps: S301, the preprocessing of the data set comprises data deduplication, rejection of empty label samples and removal of samples with excessively low content; S302, the feature engineering of the data set is used to enhance the accuracy of model prediction and reduce the noise interference of training data, and is used to screen samples of chemical elements with extremely low frequency, insufficient abundance and significant influence on oxygen balance.
6. A high modulus glass fiber formulation design method according to claim 5, characterized by: In the S4, the establishment and selection of the model specifically comprises the following steps: S401, the influence factors are determined as various oxides constituting the high-modulus glass fiber; S402, the important performances include density, elastic modulus, liquidus temperature and glass fiber forming temperature; S403, the rotating machine learning model comprises a random forest regression and an artificial neural network.
7. A high modulus glass fiber formulation design method according to claim 6, characterized by: In the S5, the evaluation index includes a determination coefficient R 2 and a root mean square error RMSE, whose calculation formula is as follows: wherein represents the number of data in the data set, denotes the measured value, is the predicted value, denotes the average value of the measured values; wherein the evaluation indicator is the coefficient of determination R 2 ≥ 0.88 is a value meeting the requirements, R 2 ≥ 0.9 is a preferred value.
8. A high modulus glass fiber suitable for use in a high modulus glass fiber formulation design method according to any one of claims 1-7, characterized by: The formula oxides and mass percentages of the high-modulus glass fiber are as follows: SiO2 53.21%~68.86% Al2O3 5.56%~21.51% CaO 0%~11.67% MgO 11.31%~27.81% Na2O 0%~1% K2O 0%~0.8% Li2O 0%~3% TiO2 0%~1% Y2O3 0%~6.5% La2O3 0%~2% CeO2 0%~1% ZnO 0%~1% Fe2O3 0%~1% B2O3 0%~1% MnO 0%~1% ZrO2 0%~1%.
9. A high modulus glass fiber according to claim 8, characterized by: The mass percentage of the oxides in the high modulus glass fiber satisfies Li2O+CaO≥0.5%, Na2O+K2O≤1.2%, Al2O3+MgO≥16.9%, SiO2+Al2O3≥50.8%.
10. A high modulus glass fiber according to claim 9, characterized by: The high modulus glass fiber has a formula concentration gradient of 0.5wt%.