A method for predicting thermal stability and screening formulations of fluorophosphate glasses

Abstract

This invention provides a method for predicting the thermal stability of fluorophosphate glasses and screening their formulations, relating to the field of materials design technology. The method includes: obtaining multiple candidate characteristic parameters for different fluorophosphate glasses, where each candidate characteristic parameter characterizes the composition of the fluorophosphate glass; predicting the thermal stability index of a target glass based on the candidate characteristic parameters, where the thermal stability index characterizes the strength of the target glass's thermal stability, and the target glass characterizes the glass prepared using a single candidate characteristic parameter as a component; and determining a target parameter from the multiple candidate characteristic parameters based on the thermal stability index, as the formulation screening result. This method can improve upon existing technologies where the fluorophosphate glass system has a complex compositional space and significant nonlinear interactions between components, and where traditional manual methods for screening glass formulations suffer from low efficiency and insufficient accuracy.

Description

Technical Field

[0001] This invention relates to the field of materials design technology, specifically to a method for predicting the thermal stability of fluorophosphate glass and screening its formulation. Background Technology

[0002] High-energy laser systems place higher demands on the service stability of optical windows, lenses, prisms, and other optical components. While traditional fused silica materials possess good optical transmittance and chemical stability, their laser damage threshold, nonlinear optical response, and overall service performance still fall short of fully meeting the development requirements of next-generation high-power laser systems in some high-energy laser applications. Therefore, researchers have begun exploring the development of novel optical materials with high laser damage resistance, aiming to replace or partially replace fused silica materials.

[0003] Fluorophosphate glasses possess advantages such as ease of fabrication, low nonlinear refractive index, high ultraviolet transmittance, and a wide range of compositional control, demonstrating promising development potential in applications such as high-energy lasers, ultraviolet optics, infrared windows, and special functional glasses. Improving their thermal stability is a key issue in further enhancing their engineering application value and industrial development potential. Fluorophosphate glasses with poor thermal stability are prone to crystallization during melting, forming, annealing, and subsequent processing, which affects the material's optical homogeneity, mechanical properties, and long-term service stability.

[0004] Traditional fluorophosphate glass formulation design relies primarily on empirical rules, repeated melting trials, and manual trial and error. These methods typically consume significant experimental time and raw material resources. Furthermore, due to the complex compositional space of fluorophosphate glass systems and the significant nonlinear interactions between components, traditional trial-and-error methods often struggle to quickly obtain formulations with excellent thermal stability. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method for predicting the thermal stability of fluorophosphate glass and screening its formulations. This method aims to improve upon the existing technologies where the fluorophosphate glass system has a complex compositional space and significant nonlinear interactions between components, and where traditional manual screening of glass formulations suffers from low efficiency and insufficient accuracy.

[0006] The technical solution adopted in this invention is a method for predicting the thermal stability of fluorophosphate glass and screening its formulation, comprising: Multiple candidate characteristic parameters for different fluorophosphate glasses are obtained, wherein the candidate characteristic parameters characterize the composition of the fluorophosphate glass; Based on the candidate feature parameters, the thermal stability index of the target glass is predicted. The thermal stability index characterizes the strength of the thermal stability of the target glass. The target glass characterizes the glass prepared using a single candidate feature parameter as a component. Based on the thermal stability index, a target parameter is determined from a plurality of candidate feature parameters as the formulation screening result.

[0007] In one optional implementation, the thermal stability index of the target glass is predicted based on the candidate feature parameters through a prediction model, the training process of which is as follows: Obtain an initial dataset, which includes multiple parameter pairs. Each parameter pair includes a first parameter and a second parameter. The first parameter characterizes the composition of the fluorophosphate glass, and the second parameter characterizes the thermal properties of the glass with the first parameter as its composition during the preparation process. Based on the second parameter, determine the target variable corresponding to the first parameter; The first parameter and the target variable are used as training data. The training data is preprocessed to obtain preprocessed training data, which is used as the training set. The pre-built initial model is trained using the training set to obtain the trained initial model, which is then used as the prediction model.

[0008] In an optional embodiment, the first parameter includes the P2O5 content, LiF content, BaO content, Al2O3 content, SiO2 content, MgO content, K2O content, CaF2 content, TiO2 content, CaO content, Na2O content, PbF2 content, GeO2 content, AlF3 content, ZnF2 content, BaF2 content, PbO content, TeO2 content, CdF2 content, MnO2 content, PbBr2 content, and B2O3 content of the fluorophosphate glass; The second parameter includes the glass transition temperature Tg, crystallization initiation temperature Tx, and melting temperature Tm during the preparation of the fluorophosphate glass.

[0009] In one optional implementation, determining the target variable corresponding to the first parameter based on the second parameter includes: The target variable is calculated based on the second parameter, as follows: In the formula, Kgl represents the target variable.

[0010] In one optional implementation, the first parameter and the target variable are used as training data, and the training data is preprocessed to obtain preprocessed training data, including: The training data is initially preprocessed to obtain preprocessed training data, which is used as the first training data. Using the Spearman analysis strategy, feature filtering is performed on the first training data to obtain the filtered first training data, which is then used as the second training data. The second training data is augmented using the WGAN data augmentation strategy to obtain augmented second training data, which is then used as the preprocessed training data.

[0011] In one optional implementation, the training data undergoes initial preprocessing to obtain preprocessed training data, including: The missing samples in the training data are removed to obtain the third training data; The abnormal samples in the third training data are removed to obtain the fourth training data; The first parameter in the fourth training data is normalized to obtain the training data after initial preprocessing.

[0012] In one optional implementation, the Spearman analysis strategy is used to perform feature filtering on the first training data to obtain filtered first training data, including: Determine the correlation coefficient between the first parameter and the target variable in the first training data, wherein the correlation coefficient characterizes the degree of influence of any component data of the first parameter on the target variable; Target training data with a correlation coefficient greater than or equal to a preset threshold are selected from the first training data and used as the first training data after selection.

[0013] In one optional implementation, the second training data is augmented using the WGAN data augmentation strategy to obtain augmented second training data, including: Based on the second training data, multiple simulated data samples are generated using the generator in WGAN; The discriminator in the WGAN is used to select at least one target simulated data sample from the plurality of simulated data samples as augmented training data; The enhanced training data is merged into the second training data to obtain the enhanced second training data.

[0014] In one optional implementation, determining a target parameter from a plurality of candidate feature parameters based on the thermal stability index includes: The candidate feature parameter that has the largest thermal stability index and whose total component amount is within a preset total range is determined as the target parameter.

[0015] As can be seen from the above technical solution, the beneficial technical effects of the present invention are as follows: The technical solution provided in this application first obtains multiple candidate characteristic parameters for different fluorophosphate glasses. Then, based on the candidate characteristic parameters, the thermal stability index of the target glass is predicted. The thermal stability index characterizes the strength of the thermal stability of the target glass, and the target glass characterizes the glass prepared using a single candidate characteristic parameter as a component. Finally, based on the thermal stability index, the target parameter is determined from multiple candidate characteristic parameters as the formulation screening result. In this way, by predicting the thermal stability index corresponding to different fluorophosphate glass combination formulations and judging the quality of the formulation based on the thermal stability index, the initial screening of glass formulations is achieved, thereby reducing the cost of manual experiments and improving the problems of low efficiency and insufficient accuracy in the traditional manual screening of glass formulations due to the complex composition space and significant nonlinear interactions between components in the fluorophosphate glass system in existing technologies. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 A flowchart illustrating a method for predicting the thermal stability of fluorophosphate glass and screening its formulation, provided for an embodiment of this application; Figure 2 A schematic diagram illustrating the experimental verification results of the model provided in this application embodiment. Detailed Implementation

[0018] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0019] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0020] This embodiment provides a method for predicting the thermal stability of fluorophosphate glass and screening its formulation, which may include: Step 110: Obtain multiple candidate characteristic parameters for different fluorophosphate glasses, wherein the candidate characteristic parameters characterize the composition of the fluorophosphate glass; Step 120: Based on the candidate feature parameters, predict the thermal stability index of the target glass. The thermal stability index characterizes the strength of the thermal stability of the target glass. The target glass characterizes the glass prepared using a single candidate feature parameter as a component. Step 130: Based on the thermal stability index, determine the target parameter from the plurality of candidate feature parameters as the formulation screening result.

[0021] In this embodiment, several candidate characteristic parameters for different fluorophosphate glasses are first obtained. Then, based on the candidate characteristic parameters, the thermal stability index of the target glass is predicted. The thermal stability index characterizes the strength of the thermal stability of the target glass, and the target glass represents the glass prepared using a single candidate characteristic parameter as a component. Finally, based on the thermal stability index, the target parameter is determined from the multiple candidate characteristic parameters as the formulation screening result. In this way, by predicting the thermal stability index corresponding to different fluorophosphate glass combination formulations and judging the quality of the formulation based on the thermal stability index, the initial screening of glass formulations is achieved, thereby reducing the cost of manual experiments and improving the problems of low efficiency and insufficient accuracy in the traditional manual screening of glass formulations due to the complex composition space and significant nonlinear interactions between components in the fluorophosphate glass system in the prior art.

[0022] The following is a detailed explanation of each step in the method for predicting the thermal stability of fluorophosphate glass and screening its formulation: In step 110, the candidate feature parameters can be a set of data containing the content of various components of fluorophosphate glass (hereinafter referred to as glass), that is, a fluorophosphate glass formulation (hereinafter referred to as formulation). In this embodiment, the user can randomly generate multiple sets of component content data within a pre-set range of glass component content, thereby constructing multiple different candidate feature parameters for glass, that is, multiple different formulations, as candidate options for subsequent formulation screening.

[0023] In step 120, the thermal stability index of the target glass can be predicted based on the candidate feature parameters. This prediction can be achieved through a prediction model, and the training process of the prediction model can be as follows: Obtain an initial dataset, which includes multiple parameter pairs. Each parameter pair includes a first parameter and a second parameter. The first parameter characterizes the composition of the fluorophosphate glass, and the second parameter characterizes the thermal properties of the glass with the first parameter as its composition during the preparation process. Based on the second parameter, determine the target variable corresponding to the first parameter; The first parameter and the target variable are used as training data. The training data is preprocessed to obtain preprocessed training data, which is used as the training set. The pre-built initial model is trained using the training set to obtain the trained initial model, which is then used as the prediction model.

[0024] In this embodiment, the first parameter may include the P2O5 content, LiF content, BaO content, Al2O3 content, SiO2 content, MgO content, K2O content, CaF2 content, TiO2 content, CaO content, Na2O content, PbF2 content, GeO2 content, AlF3 content, ZnF2 content, BaF2 content, PbO content, TeO2 content, CdF2 content, MnO2 content, PbBr2 content, and B2O3 content of the fluorophosphate glass. Understandably, the first parameter refers to multiple different glass formulations pre-stored by the user, used as input data for training the initial model. The first parameter in this embodiment can be collected from a pre-built glass formulation database (e.g., the SciGlass glass database) or from relevant technical literature on fluorophosphate glasses.

[0025] The second parameter may include the glass transition temperature Tg, crystallization initiation temperature Tx, and melting temperature Tm during the preparation of fluorophosphate glass. Understandably, the second parameter is the user's calculation of various temperature parameters (i.e., thermal properties) during glass preparation based on the first parameter, used to calculate the target variable corresponding to the first parameter. This target variable is used to form a training sample pair with the first parameter to guide the model's training, learning, and prediction.

[0026] In this embodiment, determining the target variable corresponding to the first parameter based on the second parameter may include: The target variable is calculated based on the second parameter, as follows: In the formula, Kgl represents the target variable. The larger the Kgl value, the stronger the resistance to crystallization and the higher the thermal stability of the fluorophosphate glass throughout the glass-forming space. Compared with the traditional method of characterizing glass thermal stability using only Tx-Tg, the target variable proposed in this embodiment further introduces Tm as the denominator. When both Tx-Tg and Tm are large, the potential crystallization driving force of the glass may be large, and there may still be a risk of crystallization during actual preparation. Therefore, this embodiment selects Kgl as the target variable to improve the rationality of the thermal stability evaluation.

[0027] In this embodiment, the first parameter and the target variable are used as training data. Preprocessing the training data to obtain preprocessed training data may include: The training data is initially preprocessed to obtain preprocessed training data, which is used as the first training data. Using the Spearman analysis strategy, feature filtering is performed on the first training data to obtain the filtered first training data, which is then used as the second training data. The second training data is augmented using the WGAN data augmentation strategy to obtain augmented second training data, which is then used as the preprocessed training data.

[0028] In this embodiment, the training data undergoes initial preprocessing to obtain preprocessed training data, which may include: The missing samples in the training data are removed to obtain the third training data; The abnormal samples in the third training data are removed to obtain the fourth training data; The first parameter in the fourth training data is normalized to obtain the training data after initial preprocessing.

[0029] In this embodiment, samples with missing Tg, Tx, and Tm values ​​in the training data are first removed. Then, samples with abnormal component parameters (i.e., the first parameter) (e.g., exceeding a preset content range) or abnormal component sums are removed. Finally, each component parameter is normalized to ensure that different components are within a comparable data scale. This standardization provides the foundation for subsequent correlation coefficient screening and data augmentation, improving the robustness of subsequent model training and thermal stability prediction.

[0030] In this embodiment, the Spearman analysis strategy is used to perform feature filtering on the first training data to obtain the filtered first training data, which may include: Determine the correlation coefficient between the first parameter and the target variable in the first training data, wherein the correlation coefficient characterizes the degree of influence of any component data of the first parameter on the target variable; Target training data with a correlation coefficient greater than or equal to a preset threshold are selected from the first training data and used as the first training data after selection.

[0031] In this embodiment, Spearman analysis is first used to calculate the Spearman correlation coefficient between the component parameters of the glass (the first parameter) and the target variable Kgl. This correlation coefficient can reflect the monotonic correlation between variables, does not require the data to follow a normal distribution, and is more suitable for small sample and skewed distribution data in this embodiment.

[0032] For example, the Spearman correlation coefficients between the aforementioned 22 first parameters and the target variable Kgl are calculated respectively. The results show that PbF2, GeO2, PbO, and PbBr2 have a strong positive influence on Kgl; MnO2, SiO2, BaO, and P2O5 have a strong negative influence on Kgl. To remove redundant features, this embodiment sets a preset threshold of 0.05. When the absolute value of the Spearman correlation coefficient between a feature and Kgl is greater than or equal to 0.05, the feature is retained; when the absolute value of the Spearman correlation coefficient is less than 0.05, the feature is removed from the dataset. In this way, the dimensionality of the input features can be reduced, the adverse effects of weakly correlated features and noisy features on model training can be weakened, and the generalization ability of the subsequent model can be improved. It is understood that the output of the Spearman feature selection is not only used to reduce dimensionality, but also serves as the input boundary for subsequent WGAN data augmentation. Because fluorophosphate glass data is characterized by small sample sizes and high dimensionality and sparsity, directly performing generative augmentation on all 22 component parameters would require the generator to simultaneously learn a large number of weakly correlated or severely missing features, easily leading to sample distribution drift. This embodiment first uses Spearman filtering under Kgl constraints to determine the target training data, and then uses this target training data as the generative space for WGAN data augmentation, thereby establishing a synergistic relationship of dependency between feature selection and data augmentation.

[0033] In this embodiment, the second training data is augmented using the WGAN data augmentation strategy to obtain augmented second training data, which may include: Based on the second training data, multiple simulated data samples are generated using the generator in WGAN; The discriminator in the WGAN is used to select at least one target simulated data sample from the plurality of simulated data samples as augmented training data; The enhanced training data is merged into the second training data to obtain the enhanced second training data.

[0034] In this embodiment, to address the issues of small sample size and sparse distribution of thermal stability data for fluorophosphate glass, WGAN is introduced for data augmentation. WGAN (Wasserstein Generative Adversarial Networks) can include a generator and a discriminator. The generator receives random noise and generates simulated fluorophosphate glass samples (i.e., simulated data samples). The discriminator determines whether the simulated data samples meet retention criteria. For example, the discriminator generates a corresponding score for each simulated data sample and then filters the simulated data samples based on this score; samples with scores higher than a preset score are retained, while those with scores lower than a preset score are discarded. Compared to traditional GANs, WGAN uses Wasserstein distance to measure the difference between the distribution of real samples and the distribution of generated samples, which can alleviate the problems of training instability and mode collapse, making it more suitable for data augmentation tasks involving small sample materials.

[0035] During the training of WGAN, real fluorophosphate glass data filtered by Spearman features are used as WGAN training samples, enabling the generator to learn the composition distribution and Kgl distribution of real samples. After training, the generator generates simulated data samples, and the target simulated data samples that meet the retention criteria are merged with the second training data to obtain the enhanced second training data.

[0036] Understandably, the simulated data samples generated by WGAN in this embodiment maintain the same feature dimensions as the second training data obtained through Spearman filtering, and retain the thermally stable label distribution corresponding to Kgl. The simulated data samples are not randomly constructed independently of the feature filtering results, but rather supplement the sparse regions of the second training data within the effective composition space after Spearman denoising. Thus, Spearman feature filtering addresses the question of "on which features to enhance," while WGAN data augmentation addresses the question of "how to supplement samples within the effective feature space." Together, they provide more stable training data for subsequent prediction models.

[0037] Thus, after initial preprocessing, Spearman filtering, and WGAN data augmentation, preprocessed training data is obtained, serving as the training set. Finally, the pre-built initial model is trained using the training set to obtain the trained initial model, which serves as the prediction model.

[0038] In this embodiment, the initial model can be any one of the following: random forest model, extreme random tree model, XGBoost model, support vector regression model, gradient boosting decision tree model, neural network model, or stacked ensemble model. During the training of the initial model, the selected fluorophosphate glass composition parameters (i.e., the first parameter in the preprocessed training data) are used as input features, and the corresponding Kgl is used as the output target variable to establish a nonlinear mapping relationship between the composition and thermal stability. After the model training is completed, the predictive performance of the model is evaluated using a test set. Evaluation metrics may include at least one of the following: coefficient of determination R², mean squared error (MSE), root mean square error (RMSE), and mean absolute error (MAE).

[0039] After preprocessing and training in this embodiment, experimental results show that, compared with the conventional benchmark model without Spearman feature selection and WGAN data augmentation, the technical solution proposed in this embodiment can improve the model's generalization ability by 22.5% to 53.7%. This demonstrates that the combined strategy of Spearman feature selection and WGAN data augmentation can effectively improve the generalization ability and mean square error performance of the fluorophosphate glass thermal stability prediction model.

[0040] The aforementioned performance improvement stems from the synergistic effect between the various steps: Kgl unifies the thermal stability evaluation objective, Spearman uses Kgl to remove weakly correlated components and determine the effective feature space, WGAN supplements sparse regions of samples in this effective feature space, and the initial machine learning model learns the mapping relationship between the components and Kgl based on the enhanced effective samples. Removing any step would disrupt this synergistic chain. For example, lacking Spearman's screening would cause WGAN to amplify redundant noise features; lacking WGAN enhancement would leave the model still limited by small sample sizes; and lacking Kgl's unified objective would make it difficult for feature selection and the prediction model to optimize around the same thermal stability criterion.

[0041] Model Performance Validation: To verify the synergistic gain between Spearman feature selection and WGAN data augmentation described in this application embodiment, four model validation scenarios were set up, and 20 independent repeated experiments were conducted under the same data partitioning principle to determine the coefficient of determination R. 2 As a performance evaluation metric, a performance box plot is drawn, such as... Figure 2 As shown. Here, model-1 represents the baseline case (including cases such as...). Figure 2The models shown are: ET-Baseline (Extreme Random Tree Baseline), GDBT-Baseline (Gradient Boosting Decision Tree Baseline), XGBoost-Baseline (XGBoost Baseline), and RF-Baseline (Random Forest Baseline), which do not employ feature selection or data augmentation; model-2, which employs feature selection but not data augmentation, with Spearman correlation coefficients used for feature selection; model-3, which employs data augmentation but not feature selection, with WGAN used for data augmentation; and model-4, which employs both feature selection and data augmentation, with Spearman correlation coefficients used for feature selection and WGAN used for data augmentation.

[0042] In the above validation, Random Forest (RF), Extremely Random Tree (ET), XGBoost (XGB), and Gradient Boosting Decision Tree (GDBT) models were selected as predictors to compare and test the four scenarios. The following table of model validation results lists the R values ​​obtained from 20 repeated experiments. 2 The mean and standard deviation, Figure 2 The box plots further reflect the R-values ​​of each model in 20 repeated experiments. 2 Distribution. The model validation results are shown in the table below: From the above table and Figure 2 As can be seen, among the four predictors—RF, ET, XGB, and GDBT—case 4 achieves the highest Ri. 2 The mean values ​​reached 0.6666, 0.5878, 0.6364, and 0.6014, respectively. Compared to Case 1 without feature selection and data augmentation, Case 4 showed higher R values ​​on RF, ET, XGB, and GDBT. 2The improvements were approximately 38.8%, 44.5%, 56.3%, and 28.3%, respectively. Compared to Case 2, which only used Spearman feature selection, and Case 3, which only used WGAN data augmentation, Case 4 still maintained a significant overall advantage. This indicates that implementing feature selection or data augmentation alone can only partially improve the modeling effect for small samples, while the combined effect of both can achieve more stable and higher model performance. This result further demonstrates that Spearman feature selection and WGAN data augmentation in this invention are not independent parallel steps, but rather have an inherent synergistic relationship for sparse small dataset problems: Spearman correlation coefficient selection filters out weakly correlated and noisy features based on the target variable Kgl to limit the effective feature space; WGAN data augmentation fills in the sparse regions of the real sample distribution within this effective feature space, avoiding blindly expanding samples in a high-dimensional weakly correlated space. Therefore, the two work together in the subsequent initial model training process, enabling the model to improve sample distribution coverage while reducing the impact of redundant noise, thereby significantly improving model stability, generalization ability, and the reliability of thermal stability formulation selection.

[0043] Thus, based on the initial model trained above, a prediction model is obtained. Then, the compositional parameters (i.e., candidate feature parameters) of the candidate fluorophosphate glass formulations are acquired and input into the trained thermal stability prediction model (i.e., the prediction model) to obtain the Kgl prediction value corresponding to the candidate formulation, which serves as a thermal stability index. The candidate formulations are ranked according to the Kgl prediction values, facilitating subsequent screening in step 130 based on glass formation range, total composition constraints, raw material availability, and experimental synthesability conditions.

[0044] In step 130, determining the target parameter from a plurality of candidate feature parameters based on the thermal stability index may include: The candidate feature parameter that has the largest thermal stability index and whose total component amount is within a preset total range is determined as the target parameter.

[0045] In this embodiment, a high thermal stability fluorophosphate glass formulation was selected through model prediction and experimental verification: 59.1P₂O₅ - 36.2BaO - 4.7AlF₃. This formulation has a Kgl value close to 0.7, exhibiting excellent thermal stability. Further experimental verification of the accuracy of the model's predicted trend and the synthesizability of the predicted formulation demonstrates that the technical solution of this application can effectively guide the design of high thermal stability fluorophosphate glass formulations.

[0046] In summary, this application provides a method for predicting the thermal stability of fluorophosphate glasses and screening formulations for sparse, small datasets. This technical solution first obtains the composition and thermal properties data of fluorophosphate glasses from databases and literature. Then, it uses Kgl, calculated from Tg, Tx, and Tm, as a unified target variable. Addressing the non-normal distribution of the data, Spearman correlation coefficients are used to screen target training data that have a valid monotonic correlation with Kgl. To address the issues of small samples and sparse data, the target training data is used to limit the WGAN generation space and perform data augmentation. Finally, the augmented data is used to train the initial model, and candidate formulations are quickly predicted and screened. This technical solution, through the inherent connection between Kgl target construction, Spearman feature screening, WGAN data augmentation, and model prediction screening, transforms each step from an arbitrarily separable independent operation into a collaborative technical solution for sparse, small datasets. This significantly improves the generalization ability of the fluorophosphate glass thermal stability prediction model, increases the efficiency of formulation decision-making and screening, reduces the cost of traditional trial-and-error experiments, and provides an effective technical solution for the rapid development of high-thermal-stability fluorophosphate glasses in high-energy laser application scenarios.

[0047] Based on the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.

[0048] In the embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or part of code, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0049] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting the thermal stability of fluorophosphate glass and screening its formulation, characterized in that, include: Multiple candidate characteristic parameters for different fluorophosphate glasses are obtained, wherein the candidate characteristic parameters characterize the composition of the fluorophosphate glass; Based on the candidate feature parameters, the thermal stability index of the target glass is predicted. The thermal stability index characterizes the strength of the thermal stability of the target glass. The target glass characterizes the glass prepared using a single candidate feature parameter as a component. Based on the thermal stability index, a target parameter is determined from a plurality of candidate feature parameters as the formulation screening result.

2. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 1, characterized in that, Based on the candidate feature parameters, the thermal stability index of the target glass is predicted through a prediction model, the training process of which is as follows: Obtain an initial dataset, which includes multiple parameter pairs. Each parameter pair includes a first parameter and a second parameter. The first parameter characterizes the composition of the fluorophosphate glass, and the second parameter characterizes the thermal properties of the glass with the first parameter as its composition during the preparation process. Based on the second parameter, determine the target variable corresponding to the first parameter; The first parameter and the target variable are used as training data. The training data is preprocessed to obtain preprocessed training data, which is used as the training set. The pre-built initial model is trained using the training set to obtain the trained initial model, which is then used as the prediction model.

3. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 2, characterized in that, The first parameter includes the P2O5 content, LiF content, BaO content, Al2O3 content, SiO2 content, MgO content, K2O content, CaF2 content, TiO2 content, CaO content, Na2O content, PbF2 content, GeO2 content, AlF3 content, ZnF2 content, BaF2 content, PbO content, TeO2 content, CdF2 content, MnO2 content, PbBr2 content, and B2O3 content of the fluorophosphate glass; The second parameter includes the glass transition temperature Tg, crystallization initiation temperature Tx, and melting temperature Tm during the preparation of the fluorophosphate glass.

4. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 3, characterized in that, Based on the second parameter, the target variable corresponding to the first parameter is determined, including: The target variable is calculated based on the second parameter, as follows: In the formula, Kgl represents the target variable.

5. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 2, characterized in that, Using the first parameter and the target variable as training data, the training data is preprocessed to obtain preprocessed training data, including: The training data is initially preprocessed to obtain preprocessed training data, which is used as the first training data. Using the Spearman analysis strategy, feature filtering is performed on the first training data to obtain the filtered first training data, which is then used as the second training data. The second training data is augmented using the WGAN data augmentation strategy to obtain augmented second training data, which is then used as the preprocessed training data.

6. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 5, characterized in that, The training data is initially preprocessed to obtain preprocessed training data, including: The missing samples in the training data are removed to obtain the third training data; The abnormal samples in the third training data are removed to obtain the fourth training data; The first parameter in the fourth training data is normalized to obtain the training data after initial preprocessing.

7. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 5, characterized in that, Using the Spearman analysis strategy, feature filtering is performed on the first training data to obtain the filtered first training data, including: Determine the correlation coefficient between the first parameter and the target variable in the first training data, wherein the correlation coefficient characterizes the degree of influence of any component data of the first parameter on the target variable; Target training data with a correlation coefficient greater than or equal to a preset threshold are selected from the first training data and used as the first training data after selection.

8. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 5, characterized in that, Using the WGAN data augmentation strategy, the second training data is augmented to obtain the augmented second training data, including: Based on the second training data, multiple simulated data samples are generated using the generator in WGAN; The discriminator in the WGAN is used to select at least one target simulated data sample from the plurality of simulated data samples as augmented training data; The enhanced training data is merged into the second training data to obtain the enhanced second training data.

9. The method for predicting the thermal stability of fluorophosphate glass and screening its formulation according to claim 1, characterized in that, Based on the thermal stability index, a target parameter is determined from a plurality of candidate feature parameters, including: The candidate feature parameter that has the largest thermal stability index and whose total component amount is within a preset total range is determined as the target parameter.

Images