Optical glass chemical strengthening process prediction method, system, equipment and medium

Abstract

The present application discloses a method, system, device and medium for predicting the chemical strengthening process of optical glass, and relates to the technical field of process simulation. The method comprises: obtaining the original experimental data of the target optical glass drop; calculating the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio, and adding the estimated refractive index to the process parameter set to obtain the current process parameter set; based on the decision tree algorithm, regression modeling is performed according to the current process parameter set and the performance parameter set to obtain an initial decision tree model; preprocessing the data of the current process parameter set and the performance parameter set to obtain model training data; training and testing the initial decision tree model according to the model training data to obtain a process performance prediction model; calling the process performance prediction model. The present application has the effect of reducing the design cost of the glass chemical strengthening process.

Description

Technical Field

[0001] The present application relates to the technical field of process simulation, and in particular to a method, system, equipment and medium for predicting the chemical strengthening process of optical glass. Background Art

[0002] Optical glass is an indispensable material in modern optical instruments and is widely used in many fields such as electronic information, automobiles, aerospace, and medical treatment. With the advancement of technology, optical products such as mobile phone screens, tablet computer displays, and camera lenses are moving towards thinness and lightness, which puts higher requirements on the strength, durability, scratch resistance, and impact resistance of optical glass. Traditional glass materials often cannot meet the dual requirements of performance and quality.

[0003] In order to solve this problem, chemical strengthening technology came into being. Compared with physical tempering, through ion exchange and other means, it can significantly improve the performance of optical glass without increasing its weight. This process can effectively improve the toughness and strength of glass, meeting the urgent demand for high-performance optical glass in today's market. However, traditional chemical strengthening process design usually requires repeated adjustments to multiple preparation parameters, multiple experiments and tests to obtain target performance. This method is not only time-consuming and costly, but also inefficient. The existing technology lacks a simulation prediction method for the chemical strengthening process of optical glass, which is costly when designing the strengthening process and has low utilization of data assets obtained from historical experiments. Summary of the invention

[0004] The present invention provides a method, system, equipment and medium for predicting the chemical strengthening process of optical glass, which at least solves the problems in the prior art of lacking a simulation prediction method for the chemical strengthening process of optical glass, high cost when designing the strengthening process, and low utilization rate of data assets obtained from historical experiments.

[0005] In the first aspect, the present application provides a method for predicting the chemical strengthening process of optical glass, which adopts the following technical solution:

[0006] Optical glass chemical strengthening process prediction method, including:

[0007] Obtaining original experimental data of the target optical glass drop; the original experimental data includes process parameter data and performance parameter data, and the process parameter data corresponds to a process parameter set Including glass width, tempering temperature, molten salt ion concentration ratio, content and tempering furnace stirring rate, the performance parameter data corresponding to the performance parameter set Including surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of tempered glass;

[0008] Calculate the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio, and add the estimated refractive index to the process parameter set. In the current process parameter set ;

[0009] Based on the decision tree algorithm and according to the current process parameter set and the performance parameter set Perform regression modeling to obtain an initial decision tree model;

[0010] For the current process parameter set and the performance parameter set Preprocess the data in to obtain model training data;

[0011] Training and testing the initial decision tree model according to the model training data to obtain a process performance prediction model;

[0012] calling the process performance prediction model;

[0013] Set process preconditions and process parameter ranges, and search for optimal process parameters based on the conjugate gradient algorithm so that the corresponding predicted performance meets the user's required performance;

[0014] According to the optimal process parameters, recommended process parameters are output.

[0015] By adopting the above technical solution, the original experimental data of the target optical glass drop material is first obtained, and the original experimental data includes process parameter data and performance parameter data. The process parameter set corresponding to the process parameter data Including glass width, tempering temperature, molten salt ion concentration ratio, optical glass Content and tempering furnace stirring rate, performance parameter data corresponding to the performance parameter set The surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of the tempered glass are included. Then the estimated refractive index of the tempered glass is calculated according to the molten salt ion concentration ratio, and the estimated refractive index is added to the process parameter set. In the current process parameter set , and then based on the decision tree algorithm and according to the current process parameter set and performance parameter sets Perform regression modeling to obtain the initial decision tree model, and then and performance parameter sets The data in is preprocessed to obtain model training data, and then the initial decision tree model is trained and tested according to the model training data to obtain a process performance prediction model, and then the process performance prediction model is called, the process preconditions and process parameter ranges are set, and the optimal process parameters are searched according to the conjugate gradient algorithm, so that the corresponding predicted performance meets the user's required performance, and finally the recommended process parameters are output according to the optimal process parameters. Compared with the situation of adjusting and designing the optical glass chemical strengthening process by relying on experience or experimental trial and error, the present invention applies the decision tree algorithm to the optical glass chemical strengthening process regulation, and obtains a more accurate process performance prediction model through efficient and complete data training, and applies the conjugate gradient algorithm to the optical glass chemical strengthening process regulation, thereby realizing the digitalization of the optical glass chemical strengthening process tuning; at the same time, the empirical formula for estimating the refractive index is incorporated into the process performance prediction model for the first time, which enhances the generalization ability and interpretability of the model, thereby reducing the complexity of the operation, improving the efficiency of the process performance prediction model, and reducing the design cost of the chemical strengthening process.

[0016] Optionally, the decision tree algorithm is based on the current process parameter set and the performance parameter set The steps for performing regression modeling and obtaining an initial decision tree model include:

[0017] Evaluate the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the formula (a) determines the first mutually exclusive parameter group;

[0018] Performing a linear transformation on the first mutually exclusive parameter group to obtain a corresponding first merging parameter;

[0019] Based on the decision tree algorithm and according to the first merging parameter, the current process parameter set and the performance parameter set The model is constructed to obtain the initial decision tree model.

[0020] By adopting the above technical solution, in order to generate the initial decision tree model, the current process parameter set is first evaluated according to the Bayesian conditional probability. The mutual exclusivity of the parameters in the first mutually exclusive parameter group is determined, and then the first mutually exclusive parameter group is linearly transformed to obtain the corresponding first merged parameter. Finally, based on the decision tree algorithm and according to the first merged parameter and the current process parameter set and performance parameter sets A model is constructed to obtain an initial decision tree model; the decision tree algorithm of the present application is an algorithm improved by Bayesian conditional probability. By using Bayesian conditional probability to calculate the mutual exclusivity of optical glass parameters, the complexity of the operation is reduced, the efficiency of the process performance prediction model is improved, and thus the design cost of the chemical strengthening process is reduced.

[0021] Optionally, the step of training and testing the initial decision tree model according to the model training data to obtain a process performance prediction model includes:

[0022] Dividing the model training data into a training set and a test set according to a preset ratio;

[0023] The hyperparameters of the initial decision tree model are set according to the random network search algorithm, and the root mean square error RMSE and the determination coefficient R 2 As an evaluation indicator;

[0024] Training the initial decision tree model according to the training set to obtain a trained decision tree model;

[0025] The trained decision tree model is tested according to the test set, and it is determined whether the error is within a preset range according to the evaluation index. If so, the trained decision tree model is used as a process performance prediction model.

[0026] By adopting the above technical solution, in order to obtain the process performance prediction model, the model training data is first divided into a training set and a test set according to a preset ratio, and then the hyperparameters of the initial decision tree model are set according to the random network search algorithm, and the root mean square error RMSE and the determination coefficient R 2 As an evaluation indicator, the initial decision tree model is then trained according to the training set to obtain a trained decision tree model. Finally, the trained decision tree model is tested according to the test set, and the evaluation indicator is used to determine whether the error is within the preset range. If it is within the preset range, the trained decision tree model is used as the process performance prediction model.

[0027] Optionally, the step of setting process preconditions and process parameter ranges, and searching for optimal process parameters according to a conjugate gradient algorithm so that the corresponding prediction performance meets the performance requirements of the user includes:

[0028] Evaluate the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the formula (a) determines the second mutually exclusive parameter group;

[0029] For the current process parameter set The data in the first embodiment are normalized to obtain corresponding normalized data, and a parameter space is defined according to the second mutually exclusive parameter group and the normalized data;

[0030] Performing search space mapping for each process parameter combination according to the standardized data and the second combined parameter, and defining parameter space distance according to Minkowski distance;

[0031] Select the initial value and initial search direction, and set the initial search;

[0032] According to the iteration point and search direction Iterate For the k Iteration points, For the k +1 iteration point, The search step length is For the k The search direction of the iteration point, For the k+1 The search direction of the iteration point, For the k The update factor for the iteration point, is the gradient in Minkowski space;

[0033] The process performance is predicted according to the process performance prediction model to obtain a prediction result, and it is determined whether a termination condition is met according to the prediction result. If so, the optimal process parameters are searched.

[0034] By adopting the above technical solution, in order to realize the search for the optimal process parameters, the current process parameter set is first evaluated according to the Bayesian conditional probability. The mutual exclusivity of the parameters in the process is determined, and the second mutually exclusive parameter group is determined. Then, the current process parameter set The data in is standardized to obtain the corresponding standardized data, and the parameter space is defined according to the second mutually exclusive parameter group and the standardized data, and then the search space mapping is performed for each process parameter combination according to the standardized data and the second merged parameter, and the parameter space distance is defined according to the Minkowski distance, and then the initial value and the initial search direction are selected, and the initial search is set, and then according to the iteration point and search direction Iterate, It is the gradient in the Minkowski space, and finally the process performance is predicted according to the process performance prediction model to obtain the prediction result, and whether the termination condition is met is judged according to the prediction result. If the termination condition is met, the optimal process parameters are searched; the conjugate gradient algorithm of the present application is a conjugate gradient algorithm improved by using the Minkowski distance, which realizes hyperparameter tuning and process parameter prediction by using the conjugate gradient algorithm improved by using the Minkowski distance, realizes the limitation and optimization of discrete eigenvalues, and also reduces the complexity of calculation, which provides support for the reliability of model prediction.

[0035] Optionally, the hyperparameters of the initial decision tree model include regression type, booster, maximum depth of tree model, feature and data sampling ratio and maximum number of feature value buckets. The regression type is L2 regularization, the booster is gbdt, the maximum depth of tree model is 7, the feature and data sampling ratio is 80%, and the maximum number of feature value buckets is 300. By adopting the above technical solution,

[0036] Optionally, the calculation formula for estimating the refractive index is:

[0037] =

[0038] in, To estimate the refractive index, is the refractive index of the optical glass before chemical strengthening, is the molten salt ion concentration ratio, and is the addition coefficient.

[0039] Optionally, after the step of training and testing the initial decision tree model according to the model training data to obtain the process performance prediction model, the method further includes:

[0040] Obtain target process parameter data; the target process parameter data includes target glass width, target tempering temperature, target molten salt ion concentration ratio, target content, target stirring rate, and target estimated refractive index;

[0041] The target process data is input into the process performance prediction model to obtain corresponding predicted performance parameters; the predicted performance parameters include predicted surface compressive stress, predicted central tensile stress, predicted stress layer depth, predicted weight gain and predicted refractive index.

[0042] By adopting the above technical solution, in order to predict the performance parameters of tempered glass under specific process parameters, the target process parameter data is first obtained, and the target process parameter data includes target glass width, target tempering temperature, target molten salt ion concentration ratio, target content, target stirring rate and target estimated refractive index, and then input the target process data into the process performance prediction model to obtain the corresponding predicted performance parameters. The predicted performance parameters include predicted surface compressive stress, predicted central tensile stress, predicted stress layer depth, predicted weight gain and predicted refractive index.

[0043] In the second aspect, the present application also provides an optical glass chemical strengthening process prediction system, which adopts the following technical solutions:

[0044] Optical glass chemical strengthening process prediction system, including:

[0045] The original experimental data acquisition module is used to acquire the original experimental data of the target optical glass drop material; the original experimental data includes process parameter data and performance parameter data, and the process parameter data corresponds to the process parameter set Including glass width, tempering temperature, molten salt ion concentration ratio, the optical glass content and tempering furnace stirring rate, the performance parameter data corresponding to the performance parameter set Including surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of tempered glass;

[0046] An estimated refractive index processing module is used to calculate the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio, and add the estimated refractive index to the process parameter set. In the current process parameter set ;

[0047] Modeling module, based on decision tree algorithm and according to the current process parameter set and the performance parameter set Perform regression modeling to obtain an initial decision tree model;

[0048] A preprocessing module is used to process the current process parameter set and the performance parameter set Preprocess the data in to obtain model training data;

[0049] A model generation module, used to train and test the initial decision tree model according to the model training data to obtain a process performance prediction model;

[0050] A model calling module, used for calling the process performance prediction model;

[0051] The optimal process parameter prediction module is used to set the process preconditions and process parameter ranges, and search for the optimal process parameters according to the conjugate gradient algorithm so that the corresponding prediction performance meets the user's required performance;

[0052] The recommended process parameter output module is used to output recommended process parameters according to the optimal process parameters.

[0053] In a third aspect, the present application also provides a computer device, which adopts the following technical solution:

[0054] A computer device comprises a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the method described in the first aspect when executing the computer program.

[0055] In a fourth aspect, the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0056] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute the method described in the first aspect.

[0057] In summary, the present application at least includes the following beneficial technical effects: firstly, the original experimental data of the target optical glass drop material is obtained, the original experimental data includes process parameter data and performance parameter data, and the process parameter set corresponding to the process parameter data is obtained. Including glass width, tempering temperature, molten salt ion concentration ratio, optical glass Content and tempering furnace stirring rate, performance parameter data corresponding to the performance parameter set The surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of the tempered glass are included. Then the estimated refractive index of the tempered glass is calculated according to the molten salt ion concentration ratio, and the estimated refractive index is added to the process parameter set. In the current process parameter set , and then based on the decision tree algorithm and according to the current process parameter set and performance parameter sets Perform regression modeling to obtain the initial decision tree model, and then and performance parameter sets The data in is preprocessed to obtain model training data, and then the initial decision tree model is trained and tested according to the model training data to obtain a process performance prediction model, and then the process performance prediction model is called, the process preconditions and process parameter ranges are set, and the optimal process parameters are searched according to the conjugate gradient algorithm so that the corresponding prediction performance meets the user's performance requirements, and finally the recommended process parameters are output according to the optimal process parameters. Compared with the situation of adjusting and designing the optical glass chemical strengthening process by relying on experience or experimental trial and error, the present invention applies the optimized decision tree algorithm to the optical glass chemical strengthening process regulation, and obtains a more accurate process performance prediction model through efficient and complete data training, realizing the digitalization of the optical glass chemical strengthening process tuning; at the same time, the empirical formula for estimating the refractive index is incorporated into the process performance prediction model for the first time, enhancing the generalization ability and interpretability of the model; in addition, the Bayesian conditional probability is used for the first time to calculate the mutual exclusivity of optical glass parameters, reducing the complexity of the operation, improving the efficiency of the process performance prediction model, and thus reducing the design cost of the chemical strengthening process. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present application.

[0059] Figure 2 It is a schematic diagram of the structure of the system of this application.

[0060] Figure 3 It is a structural block diagram of the computer device of the present application. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-3 It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0062] The embodiments of the present application disclose a method for predicting the chemical strengthening process of optical glass.

[0063] Reference Figure 1 , optical glass chemical strengthening process prediction method, including:

[0064] Step S11, obtaining original experimental data of the target optical glass drop.

[0065] The original experimental data includes process parameter data and performance parameter data. The process parameter set corresponding to the process parameter data is Including glass width, tempering temperature, molten salt ion concentration ratio, optical glass Content and tempering furnace stirring rate, performance parameter data corresponding to the performance parameter set Including the surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of tempered glass.

[0066] It should be further explained that the process parameter data are various parameters related to the optical glass production process, such as the glass width (D), tempering temperature (T), molten salt ion concentration ratio (S), content (c), tempering furnace stirring rate (v), glass width (D) refers to the thickness or width of optical glass, tempering temperature (T) refers to the heating temperature required during the manufacturing process, molten salt ion concentration ratio (S) refers to the ion concentration of molten salt during chemical strengthening, Content (c) is the content of optical glass The content of the content, the stirring rate of the tempering furnace (v) refers to the stirring speed of the glass in the furnace during the tempering process. The performance data are parameters related to the performance of the finished optical glass (chemically strengthened optical glass, i.e. tempered glass), such as the surface compressive stress (CS), central tensile stress (CT), stress layer depth (DOL), weight gain (M) and refractive index (n) of the tempered product in step S11. d ), the surface compressive stress (CS) of the tempered product reflects the pressure condition on the glass surface and is an important indicator for evaluating its strength. The central tensile stress (CT) indicates the tension at the center of the glass. The depth of the stress layer (DOL) characterizes the depth of stress distribution. The weight gain (M) refers to the weight added during the chemical strengthening process. The refractive index (n d ) indicates the optical properties of optical glass after chemical strengthening.

[0067] Step S12, calculating the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio, and adding the estimated refractive index to the process parameter set In the current process parameter set .

[0068] The calculation formula for estimating the refractive index is:

[0069] =

[0070] in, To estimate the refractive index, is the refractive index of optical glass before chemical strengthening, is the molten salt ion concentration ratio, and is the addition coefficient.

[0071] It should be noted that in the optical glass industry, there is an estimated relationship between salt ion concentration and refractive index: , Can be used as an estimate of the refractive index , is the sum coefficient of each component, which is substituted into the above formula and is represented as and , is the molar ratio of each component, which is substituted into the above formula and is characterized as and , this application incorporates the above relationship into the process parameter data, thereby incorporating the empirical formula into the prediction model to enhance the accuracy and interpretability of the model.

[0072] Step S13, based on the decision tree algorithm and according to the current process parameter set and performance parameter sets Perform regression modeling to obtain the initial decision tree model.

[0073] It should be noted that the decision tree algorithm of this application is the LightGBM algorithm optimized based on Bayesian theorem.

[0074] Step S14, the current process parameter set and performance parameter sets The data in is preprocessed to obtain model training data.

[0075] It should be noted that the preprocessing operations of this application include missing value checking and KNN nearest neighbor filling, 3σ principle to eliminate outliers, data standardization, etc.

[0076] Step S15, training and testing the initial decision tree model according to the model training data to obtain a process performance prediction model.

[0077] Step S16, calling the process performance prediction model.

[0078] Step S17, setting process preconditions and process parameter ranges, and searching for optimal process parameters according to the conjugate gradient algorithm so that the corresponding prediction performance meets the user's required performance.

[0079] It should be noted that the conjugate gradient algorithm of the present application is an improved conjugate gradient algorithm based on the Minkowski distance.

[0080] Step S18: outputting recommended process parameters according to the optimal process parameters.

[0081] It can be understood that, in step S18, the optimal process parameters are output as recommended process parameters.

[0082] In the above embodiment, the original experimental data of the target optical glass drop material is first obtained, and the original experimental data includes process parameter data and performance parameter data. The process parameter set corresponding to the process parameter data is Including glass width, tempering temperature, molten salt ion concentration ratio, optical glass Content and tempering furnace stirring rate, performance parameter data corresponding to the performance parameter set The surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of the tempered glass are included. Then the estimated refractive index of the tempered glass is calculated according to the molten salt ion concentration ratio, and the estimated refractive index is added to the process parameter set. In the current process parameter set , and then based on the decision tree algorithm and according to the current process parameter set and performance parameter sets Perform regression modeling to obtain the initial decision tree model, and then and performance parameter sets The data in is preprocessed to obtain model training data, and then the initial decision tree model is trained and tested according to the model training data to obtain a process performance prediction model, and then the process performance prediction model is called, the process preconditions and process parameter ranges are set, and the optimal process parameters are searched according to the conjugate gradient algorithm, so that the corresponding predicted performance meets the user's required performance, and finally the recommended process parameters are output according to the optimal process parameters. Compared with the situation of adjusting and designing the optical glass chemical strengthening process by relying on experience or experimental trial and error, the present invention applies the decision tree algorithm to the optical glass chemical strengthening process regulation, and obtains a more accurate process performance prediction model through efficient and complete data training, and applies the conjugate gradient algorithm to the optical glass chemical strengthening process regulation, thereby realizing the digitalization of the optical glass chemical strengthening process tuning; at the same time, the empirical formula for estimating the refractive index is incorporated into the process performance prediction model for the first time, which enhances the generalization ability and interpretability of the model, thereby reducing the complexity of the operation, improving the efficiency of the process performance prediction model, and reducing the design cost of the chemical strengthening process.

[0083] As a further implementation of the method, based on a decision tree algorithm and according to the current process parameter set and performance parameter sets The steps for performing regression modeling and obtaining an initial decision tree model include:

[0084] Step S21, evaluate the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the first mutually exclusive parameter group is determined.

[0085] It should be noted that, in step S21, the calculation formula of Bayesian conditional probability can be used: The mutual exclusivity evaluation is performed on any two process parameters. Specifically, from the current process parameter set Select a process parameter pair (X, Y), where X and Y represent the current process parameter set. Then, the prior probability P(X), P(Y) and joint probability (X, Y) of the two parameters are determined according to the process parameter pair. The prior probabilities P(X) and P(Y) are obtained by calculating the probability of X and Y in the current process parameter set. The respective occurrence frequencies are obtained, and the joint probability (X, Y) is obtained by X, Y in the current process parameter set The frequency of simultaneous occurrence is obtained, and then the conditional probability is calculated according to the calculation formula of prior probability, joint probability and Bayesian conditional probability. Then, according to the prior probability, joint probability and conditional probability, it is determined whether X and Y are mutually exclusive. A mutual exclusivity threshold can be set and the conditional probability can be compared with the mutual exclusivity threshold to determine whether X and Y are mutually exclusive. The following is a detailed description: For the process parameter pair (D, v), if the prior probability P(D=10)=0.3, P(v=5)=0.2, P(v=5, D=10)=0.05, the calculated conditional probability P(D=10|v=5)=0.075, if the mutual exclusivity threshold is 0.1, at this time 0.075<0.1, it is considered that the glass width (D) and the tempering furnace stirring rate (v) are mutually exclusive.

[0086] Step S22: performing a linear transformation on the first mutually exclusive parameter group to obtain a corresponding first merging parameter.

[0087] It should be noted that, based on the above situation that the glass width (D) and the tempering furnace stirring rate (v) are mutually exclusive, it is necessary to perform a linear transformation on the glass width (D) and the tempering furnace stirring rate (v) to merge the glass width (D) and the tempering furnace stirring rate (v) into one parameter or feature, so that their subspaces are mutually exclusive. Through linear transformation, the merging of mutually exclusive parameters is achieved. For example, the feature M generated by the linear transformation of the glass width (D) and the tempering furnace stirring rate (v) is: M=a*D+b*v, where a and b are both constants.

[0088] Step S23, based on the decision tree algorithm and according to the first merged parameter and the current process parameter set and performance parameter sets The model is constructed to obtain the initial decision tree model.

[0089] It should be noted that when performing regression modeling according to the decision tree algorithm, a histogram bucketing operation will be performed on the above-mentioned process parameter data. By merging the parameters in step S21 and step S22, the data dimension is reduced and the calculation efficiency is improved.

[0090] In the above implementation, in order to generate an initial decision tree model, the current process parameter set is first evaluated according to the Bayesian conditional probability. The mutual exclusivity of the parameters in the first mutually exclusive parameter group is determined, and then the first mutually exclusive parameter group is linearly transformed to obtain the corresponding first merged parameter. Finally, based on the decision tree algorithm and according to the first merged parameter and the current process parameter set and performance parameter sets A model is constructed to obtain an initial decision tree model; the decision tree algorithm of the present application is an algorithm improved by Bayesian conditional probability. By using Bayesian conditional probability to calculate the mutual exclusivity of optical glass parameters, the complexity of the operation is reduced, the efficiency of the process performance prediction model is improved, and thus the design cost of the chemical strengthening process is reduced.

[0091] As a further implementation of the method, the steps of training and testing the initial decision tree model according to the model training data to obtain the process performance prediction model include:

[0092] Step S31, dividing the model training data into a training set and a test set according to a preset ratio.

[0093] It is understandable that the preset ratio may be 8:2 or 7:3.

[0094] Step S32: set the hyperparameters of the initial decision tree model according to the random network search algorithm, and calculate the root mean square error RMSE and the determination coefficient R 2 as an evaluation indicator.

[0095] Among them, the hyperparameters of the initial decision tree model include regression type, booster, maximum depth of tree model, feature and data sampling ratio and maximum number of eigenvalue buckets. The regression type is L2 regularization, the booster is gbdt, the maximum depth of tree model is 7, the feature and data sampling ratio is 80%, and the maximum number of eigenvalue buckets is 300.

[0096] Step S33, training the initial decision tree model according to the training set to obtain a trained decision tree model.

[0097] Step S34, testing the trained decision tree model according to the test set, and judging whether the error is within a preset range according to the evaluation index, if so, using the trained decision tree model as a process performance prediction model.

[0098] It should be noted that in step S34, the model performance is evaluated by the evaluation indicators R2 and RMSE, and the surface compressive stress (CS), central tensile stress (CT), stress layer depth (DOL), weight gain (M), refractive index (n d ) and other RMSE errors do not exceed 5% of the mean, and the model is stored in the chemical enhancement process prediction system.

[0099] In the above implementation, in order to obtain the process performance prediction model, the model training data is first divided into a training set and a test set according to a preset ratio, and then the hyperparameters of the initial decision tree model are set according to the random network search algorithm, and the root mean square error RMSE and the determination coefficient R 2 As an evaluation indicator, the initial decision tree model is then trained according to the training set to obtain a trained decision tree model. Finally, the trained decision tree model is tested according to the test set, and the evaluation indicator is used to determine whether the error is within the preset range. If it is within the preset range, the trained decision tree model is used as the process performance prediction model.

[0100] As a further implementation of the method, the steps of setting process preconditions and process parameter ranges, and searching for optimal process parameters according to a conjugate gradient algorithm so that the corresponding predicted performance meets the user's required performance include:

[0101] Step S41, evaluating the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the second mutually exclusive parameter group is determined.

[0102] It should be noted that the merging logic of mutually exclusive parameters in step S41 is basically the same as that in step S22, but steps S21 and S22 are to reduce feature dimensions, focus on feature engineering, reduce the risk of overfitting, and improve the model's ability to learn features. Step S41 is to optimize a specific process parameter combination and does not generate new features. Suppose through steps S41 and S42 we find that the glass width (D) and tempering temperature (T) are closely related to the tempering results, and we do not create new features, but use them in combination. We directly use these features for correlation analysis and optimization without making any changes or combinations to them.

[0103] Step S42, the current process parameter set The data in is standardized to obtain corresponding standardized data, and a parameter space is defined according to the second mutually exclusive parameter group and the standardized data.

[0104] Step S43 , performing search space mapping for each process parameter combination, and defining parameter space distance according to Minkowski distance.

[0105] It should be noted that in step S43, the search space mapping is performed for each process parameter combination. However, due to the particularity of the brand and specification, the glass width (D) and the glass The parameter space of content (c) has only a few discrete values, so the Minkowski distance is used to define the parameter space distance. , where for the parameters glass width (D) and glass Content (c) using p =1 to calculate Manhattan distance, p =2 to calculate the Euclidean distance.

[0106] Step S46, select the initial value and initial search direction, and set the initial search.

[0107] It should be noted that in step S46, the initial value and the initial search direction are selected, the initial search is set according to the historical data, the termination accuracy is set to 0.01, and the step length is Using the line search method, the search target is , which means finding a chemical strengthening process parameter whose corresponding performance values ​​are closest to the target values.

[0108] Step S46, according to the iteration point and search direction Iterate.

[0109] in, For the k Iteration points, For the k +1 iteration point, The search step length is For the k The search direction of the iteration point, For the k+1 The search direction of the iteration point, For the k The update factor for the iteration point, is the gradient in Minkowski space;

[0110] It should be noted that in step S46, for the glass width (D) and the glass The Minkowski distance is used to calculate the gradient components of the two dimensions (c).

[0111] Step S47, predicting the process performance according to the process performance prediction model, obtaining the prediction result, and judging whether the termination condition is met according to the prediction result, if so, searching for the optimal process parameters.

[0112] It is understandable that in step S47, the decision tree model (i.e., process performance prediction model) is called and the performance is predicted, and when the termination condition (i.e., termination accuracy) is achieved, the optimal parameter combination is found. After step S47 (i.e., after the optimal parameter combination is output), the process prediction log is saved.

[0113] In the above implementation, in order to realize the search for the optimal process parameters, the current process parameter set is first evaluated according to the Bayesian conditional probability. The mutual exclusivity of the parameters in the process is determined, and the second mutually exclusive parameter group is determined. Then, the current process parameter set The data in is standardized to obtain the corresponding standardized data, and the parameter space is defined according to the second mutually exclusive parameter group and the standardized data, and then the search space mapping is performed for each process parameter combination according to the standardized data and the second combined parameter, and the parameter space distance is defined according to the Minkowski distance, and then the initial value and the initial search direction are selected, and the initial search is set, and then according to the iteration point and search direction Iterate, It is the gradient in the Minkowski space, and finally the process performance is predicted according to the process performance prediction model to obtain the prediction result, and whether the termination condition is met is judged according to the prediction result. If the termination condition is met, the optimal process parameters are searched; the conjugate gradient algorithm of the present application is a conjugate gradient algorithm improved by using the Minkowski distance, which realizes hyperparameter tuning and process parameter prediction by using the conjugate gradient algorithm improved by using the Minkowski distance, realizes the limitation and optimization of discrete eigenvalues, and also reduces the complexity of calculation, which provides support for the reliability of model prediction.

[0114] As a further implementation of the method, after the step of training and testing the initial decision tree model according to the model training data to obtain the process performance prediction model, the method further includes:

[0115] Step S56, obtaining target process parameter data.

[0116] Among them, the target process parameter data include target glass width, target tempering temperature, target molten salt ion concentration ratio, target content, target stirring rate, and target estimated refractive index.

[0117] It can be understood that the target process parameter data is the process parameter data for the intended experiment.

[0118] Step S57, inputting the target process data into the process performance prediction model to obtain corresponding predicted performance parameters.

[0119] Among them, the predicted performance parameters include predicted surface compressive stress, predicted central tensile stress, predicted stress layer depth, predicted weight gain and predicted refractive index.

[0120] It can be understood that the predicted performance parameters are the performance parameters of the optical glass after chemical strengthening under the target process parameter data, that is, the performance parameters of the tempered glass.

[0121] In the above embodiment, in order to predict the performance parameters of tempered glass under specific process parameters, the target process parameter data is first obtained, and the target process parameter data includes the target glass width, the target tempering temperature, the target molten salt ion concentration ratio, the target content, target stirring rate and target estimated refractive index, and then input the target process data into the process performance prediction model to obtain the corresponding predicted performance parameters. The predicted performance parameters include predicted surface compressive stress, predicted central tensile stress, predicted stress layer depth, predicted weight gain and predicted refractive index.

[0122] The embodiments of the present application also disclose an optical glass chemical strengthening process prediction system.

[0123] refer to Figure 2 , optical glass chemical strengthening process prediction system, including:

[0124] Optical glass chemical strengthening process prediction system, including:

[0125] The original experimental data acquisition module is used to obtain the original experimental data of the target optical glass drop; the original experimental data includes process parameter data and performance parameter data, and the process parameter data corresponds to the process parameter set Including glass width, tempering temperature, molten salt ion concentration ratio, optical glass Content and tempering furnace stirring rate, performance parameter data corresponding to the performance parameter set Including surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of tempered glass;

[0126] The estimated refractive index processing module is used to calculate the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio and add the estimated refractive index to the process parameter set In the current process parameter set ;

[0127] Modeling module, based on decision tree algorithm and according to the current process parameter set and performance parameter sets Perform regression modeling to obtain an initial decision tree model;

[0128] Preprocessing module, used to process the current process parameter set and performance parameter sets Preprocess the data in to obtain model training data;

[0129] A model generation module is used to train and test the initial decision tree model according to the model training data to obtain a process performance prediction model;

[0130] A model calling module is used to call a process performance prediction model;

[0131] The optimal process parameter prediction module is used to set the process preconditions and process parameter ranges, and search for the optimal process parameters according to the conjugate gradient algorithm so that the corresponding prediction performance meets the user's required performance;

[0132] The recommended process parameter output module is used to output recommended process parameters based on the optimal process parameters.

[0133] The optical glass chemical strengthening process prediction system of the present invention can implement any one of the optical glass chemical strengthening process prediction methods, and the specific working process of the optical glass chemical strengthening process prediction system of the present invention can refer to the corresponding process in the above-mentioned optical glass chemical strengthening process prediction method.

[0134] The embodiment of the present application also discloses a computer device.

[0135] refer to Figure 3 A computer device includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, any one of the above-mentioned optical glass chemical strengthening process prediction methods is implemented.

[0136] The embodiment of the present application also discloses a computer-readable storage medium.

[0137] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any one of the optical glass chemical strengthening process prediction methods.

[0138] Among them, computer-readable storage media can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0139] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. A method for predicting the chemical strengthening process of optical glass, characterized in that: include: Obtain the original experimental data of the target optical glass drop; The original experimental data includes process parameter data and performance parameter data. The process parameter set corresponding to the process parameter data Including glass width, tempering temperature, molten salt ion concentration ratio, content and tempering furnace stirring rate, the performance parameter data corresponding to the performance parameter set Including surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of tempered glass; Calculate the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio, and add the estimated refractive index to the process parameter set. In the current process parameter set ; Based on the decision tree algorithm and according to the current process parameter set and the performance parameter set Perform regression modeling to obtain an initial decision tree model; For the current process parameter set and the performance parameter set Preprocess the data in to obtain model training data; Training and testing the initial decision tree model according to the model training data to obtain a process performance prediction model; calling the process performance prediction model; Set process preconditions and process parameter ranges, and search for optimal process parameters based on the conjugate gradient algorithm so that the corresponding predicted performance meets the user's required performance; Outputting recommended process parameters according to the optimal process parameters; The decision tree algorithm is based on the current process parameter set and the performance parameter set The steps for performing regression modeling and obtaining an initial decision tree model include: Evaluate the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the formula (a) determines the first mutually exclusive parameter group; Performing a linear transformation on the first mutually exclusive parameter group to obtain a corresponding first merging parameter; Based on the decision tree algorithm and according to the first merging parameter, the current process parameter set and the performance parameter set The model is constructed to obtain the initial decision tree model.

2. The optical glass chemical strengthening process prediction method according to claim 1, characterized in that: The step of training and testing the initial decision tree model according to the model training data to obtain a process performance prediction model comprises: Dividing the model training data into a training set and a test set according to a preset ratio; The hyperparameters of the initial decision tree model are set according to the random network search algorithm, and the root mean square error RMSE and the determination coefficient R 2 As an evaluation indicator; Training the initial decision tree model according to the training set to obtain a trained decision tree model; The trained decision tree model is tested according to the test set, and it is determined whether the error is within a preset range according to the evaluation index. If so, the trained decision tree model is used as a process performance prediction model.

3. The optical glass chemical strengthening process prediction method according to claim 1, characterized in that: The step of setting process preconditions and process parameter ranges, and searching for optimal process parameters according to a conjugate gradient algorithm so that the corresponding prediction performance meets the user's required performance includes: Evaluate the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the formula (a) determines the second mutually exclusive parameter group; For the current process parameter set The data in the first embodiment are normalized to obtain corresponding normalized data, and a parameter space is defined according to the second mutually exclusive parameter group and the normalized data; Performing search space mapping for each process parameter combination according to the standardized data and the second mutually exclusive parameter group, and defining parameter space distance according to Minkowski distance; Select the initial value and initial search direction, and set the initial search; According to the iteration point and search direction Iterate For the k Iteration points, For the k +1 iteration point, The search step length is For the k The search direction of the iteration point, For the k+1 The search direction of the iteration point, For the k The update factor for the iteration point, is the gradient in Minkowski space; The process performance is predicted according to the process performance prediction model to obtain a prediction result, and it is determined whether a termination condition is met according to the prediction result. If so, the optimal process parameters are searched.

4. The optical glass chemical strengthening process prediction method according to claim 1, characterized in that: The hyperparameters of the initial decision tree model include regression type, booster, maximum depth of tree model, feature and data sampling ratio and maximum number of eigenvalue buckets. The regression type is L2 regularization, the booster is gbdt, the maximum depth of tree model is 7, the feature and data sampling ratio is 80%, and the maximum number of eigenvalue buckets is 300.

5. The optical glass chemical strengthening process prediction method according to claim 1, characterized in that: The calculation formula for estimating the refractive index is: = in, To estimate the refractive index, is the refractive index of the optical glass before chemical strengthening, is the molten salt ion concentration ratio, and is the addition coefficient.

6. The optical glass chemical strengthening process prediction method according to claim 1, characterized in that: After the step of training and testing the initial decision tree model according to the model training data to obtain the process performance prediction model, the method further includes: Obtain target process parameter data; the target process parameter data includes target glass width, target tempering temperature, target molten salt ion concentration ratio, target content, target stirring rate, and target estimated refractive index; The target process data is input into the process performance prediction model to obtain corresponding predicted performance parameters; the predicted performance parameters include predicted surface compressive stress, predicted central tensile stress, predicted stress layer depth, predicted weight gain and predicted refractive index.

7. Optical glass chemical strengthening process prediction system, characterized in that: include: The original experimental data acquisition module is used to acquire the original experimental data of the target optical glass drop; The original experimental data includes process parameter data and performance parameter data. The process parameter set corresponding to the process parameter data Including glass width, tempering temperature, molten salt ion concentration ratio, the optical glass content and tempering furnace stirring rate, the performance parameter data corresponding to the performance parameter set Including surface compressive stress, central tensile stress, stress layer depth, weight gain and refractive index of tempered glass; An estimated refractive index processing module is used to calculate the estimated refractive index of the tempered glass according to the molten salt ion concentration ratio, and add the estimated refractive index to the process parameter set. In the current process parameter set ; Modeling module, based on decision tree algorithm and according to the current process parameter set and the performance parameter set Perform regression modeling to obtain an initial decision tree model; A preprocessing module is used to process the current process parameter set and the performance parameter set Preprocess the data in to obtain model training data; A model generation module, used to train and test the initial decision tree model according to the model training data to obtain a process performance prediction model; A model calling module, used for calling the process performance prediction model; The optimal process parameter prediction module is used to set the process preconditions and process parameter ranges, and search for the optimal process parameters according to the conjugate gradient algorithm so that the corresponding prediction performance meets the user's required performance; A suggested process parameter output module, used for outputting suggested process parameters according to the optimal process parameters; The decision tree algorithm is based on the current process parameter set and the performance parameter set The steps for performing regression modeling and obtaining an initial decision tree model include: Evaluate the current process parameter set according to Bayesian conditional probability The mutual exclusivity of the parameters in the formula (a) determines the first mutually exclusive parameter group; Performing a linear transformation on the first mutually exclusive parameter group to obtain a corresponding first merging parameter; Based on the decision tree algorithm and according to the first merging parameter, the current process parameter set and the performance parameter set The model is constructed to obtain the initial decision tree model.

8. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 6.

Images