Optical glass melting process prediction method, system, equipment and medium

Abstract

The present application discloses an optical glass melting process prediction method, system, equipment and medium, and relates to the technical field of optical glass. The method comprises: obtaining melting process data and bubble evaluation data of optical glass, extracting features from image data and ultrasonic reflection signals, and obtaining corresponding bubble features; obtaining a bubble grade prediction model based on a first preset model and according to bubble features and grade data; obtaining a bubble feature prediction model based on a second preset model and according to melting process data and bubble features; obtaining a target bubble grade and restriction conditions, traversing at least one of bubble density, bubble area and bubble depth based on a random network search algorithm, and calling a bubble grade prediction model for prediction to obtain a bubble grade, and using a bubble parameter corresponding to a bubble grade consistent with the target bubble grade as a target bubble feature parameter. The present application has the effect of reducing the cost of glass melting process.

Description

Technical Field

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

[0002] Optical glass, as an important material, is widely used in optical instruments, photographic equipment, laser technology and other fields, and its performance is directly affected by the melting process. The melting process of optical glass is a crucial step in the manufacturing process, and its quality and control are directly related to the optical performance and physical properties of the final product. However, bubbles often appear in the melting process, which can cause light scattering, significantly reduce light transmittance, and affect the imaging quality and performance of optical devices.

[0003] At present, most studies on the bubble problem use trial and error methods for process control and prediction. Although this method can achieve certain optimization, it is often complicated, time-consuming and inefficient. Therefore, how to quickly and accurately design the melting process to effectively reduce the generation of bubbles, reduce the impact of bubbles, and improve the quality of optical glass has become an important issue that the industry needs to solve. In order to optimize the melting process of optical glass, it is necessary to explore more accurate control strategies and methods based on the understanding of the bubble formation mechanism to improve production efficiency and ensure the excellent optical properties of optical glass. Summary of the invention

[0004] In order to reduce the optimization cost of the optical glass melting process, the present application provides an optical glass melting process prediction method, system, equipment and medium.

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

[0006] Optical glass melting process prediction method, including:

[0007] Acquiring melting process data and bubble evaluation data of the optical glass, wherein the bubble evaluation data includes image data, ultrasonic reflection signal and grade data of the optical glass;

[0008] Performing feature extraction on the image data and the ultrasonic reflection signal to obtain corresponding bubble features; the bubble features include at least one of a bubble density feature, a bubble area feature, and a bubble depth feature;

[0009] Based on the first preset model, and according to the bubble characteristics and the grade data, a bubble grade prediction model is obtained;

[0010] Based on the second preset model, and according to the smelting process data and the bubble characteristics, a bubble characteristic prediction model is obtained;

[0011] Obtaining a target bubble level and restriction conditions, traversing at least one of bubble density, bubble area and bubble depth based on a random network search algorithm, and calling the bubble level prediction model to predict, obtaining the bubble level, and using the bubble parameters corresponding to the bubble level consistent with the target bubble level as the target bubble characteristic parameters; the bubble parameters include at least one of bubble density, bubble area and bubble depth, and the target bubble characteristic parameters include at least one of target bubble density, target bubble area and target bubble depth;

[0012] A target vector is generated according to the target bubble characteristic parameters, and the bubble characteristic prediction model is called to set process preconditions and ranges. A group of process parameters is screened out according to the Monte Carlo algorithm and the target vector, and the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters.

[0013] By adopting the above technical scheme, the melting process data and bubble evaluation data of the optical glass are first obtained, the bubble evaluation data includes the image data, ultrasonic reflection signal and grade data of the optical glass, and then the image data and the ultrasonic reflection signal are feature extracted to obtain the corresponding bubble features; the bubble features include at least one of the bubble density features, the bubble area features and the bubble depth features, and then based on the first preset model, a bubble grade prediction model is obtained according to the bubble features and the grade data, and based on the second preset model, a bubble feature prediction model is obtained according to the melting process data and the bubble features, and then the target bubble grade and restriction conditions are obtained, at least one of the bubble density, the bubble area and the bubble depth is traversed based on the random network search algorithm, and the bubble grade prediction model is called for prediction to obtain To the bubble level, the bubble parameters corresponding to the bubble level consistent with the target bubble level are used as target bubble characteristic parameters, the bubble parameters include at least one of bubble density, bubble area and bubble depth, the target bubble characteristic parameters include at least one of target bubble density, target bubble area and target bubble depth, finally, a target vector is generated according to the target bubble characteristic parameters, and the bubble characteristic prediction model is called, the process preconditions and range are set, and a group of process parameters are screened out according to the Monte Carlo algorithm and the target vector, and the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters; compared with the method of optimizing process parameters by trial and error, the method of the present invention improves the optimization efficiency of the optical glass melting process, reduces the optimization cost of the optical glass melting process, and realizes the digital tuning of the optical glass melting process.

[0014] Optionally, the bubble features include bubble density features, bubble area features and bubble depth features, and the step of extracting features from the image data and the ultrasonic reflection signal to obtain corresponding bubble features includes:

[0015] Performing feature extraction on the image data according to a pre-generated convolutional neural network (CNN) to obtain the bubble density feature and the bubble area feature;

[0016] The ultrasonic reflection signal is subjected to feature extraction according to a pre-generated recurrent neural network (RNN) to obtain the bubble depth feature.

[0017] By adopting the above technical solution, in order to obtain the corresponding bubble features, the image data is feature extracted according to the pre-generated convolutional neural network CNN to obtain the bubble density features and bubble area features, and the ultrasonic reflection signal is feature extracted according to the pre-generated recurrent neural network RNN ​​to obtain the bubble depth features.

[0018] Optionally, the first preset model is a SVM model.

[0019] Optionally, the second preset model is a random forest model, and the step of obtaining the bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics includes:

[0020] preprocessing the smelting process data according to the smelting process data to obtain preprocessed data;

[0021] Dividing the preprocessed data and the bubble features into a training set and a test set according to a preset ratio, and setting and optimizing the hyperparameters of the random forest model; the hyperparameters include the number of random forest decision trees, the maximum number of features, the maximum depth, and the minimum number of sample splits;

[0022] Training the random forest model according to the training set to obtain a trained random forest model;

[0023] The trained random forest model is tested according to the test set to obtain a random forest model that passes the test, and the random forest model that passes the test is used as a bubble feature prediction model.

[0024] By adopting the above technical scheme, in order to obtain a bubble feature prediction model, the smelting process data is first preprocessed according to the smelting process data to obtain preprocessed data, and then the preprocessed data and the bubble features are divided into a training set and a test set according to a preset ratio, and the hyperparameters of the random forest model are set and optimized, and the hyperparameters include the number of random forest decision trees, the maximum number of features, the maximum depth and the minimum number of sample splits, and then the random forest model is trained according to the training set to obtain a trained random forest model, and finally the trained random forest model is tested according to the test set to obtain a tested random forest model, and the tested random forest model is used as the bubble feature prediction model.

[0025] Optionally, the step of setting the process preconditions and ranges, selecting a set of process parameters according to the Monte Carlo algorithm and the target vector, wherein the bubble characteristic parameters corresponding to the set of process parameters are closest to the target bubble characteristic parameters, comprises:

[0026] In the initial state, the parameters in the process parameter set Φ corresponding to the smelting process data are set to a random state, and annealing is performed for 500 iterations;

[0027] After the annealing is completed, the first vector product is calculated according to the corresponding bubble characteristic vector and the target vector in the current state, and it is determined whether the difference between the modulus length of the first vector product and the modulus length of the target vector is within a preset range, and whether the modulus length of the first vector product meets a preset condition. If yes, the corresponding process parameter value is used as a set of process parameters;

[0028] Otherwise, according to the original data magnitude Q i In [c*Q i ,d*Q i ] randomly select 500 parameter points, c is less than d, and c and d are both constants greater than zero, for each parameter point, determine whether the difference between the modulus of the corresponding bubble feature vector and the modulus of the target vector is within the preset range, if so, calculate the corresponding second vector product according to the corresponding bubble feature vector and the target vector, and determine whether the modulus of the second vector product is less than the modulus of the second vector product calculated last time;

[0029] If the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time, the process parameters are updated; otherwise, whether to accept the current parameter point is determined according to the difference between the modulus length of the second vector product and the modulus length of the second vector product calculated last time;

[0030] Iterate 1000 times or meet the iteration endpoint to obtain the final process parameters, and use the final process parameters as a set of process parameters.

[0031] By adopting the above technical scheme, in order to screen the process parameters, in the initial state, the parameters in the process parameter set Φ corresponding to the smelting process data are set to a random state, and annealing is performed for 500 iterations; after the annealing is completed, the corresponding first vector product is calculated according to the corresponding bubble feature vector and the target vector in the current state, and it is determined whether the difference between the modulus length of the first vector product and the modulus length of the target vector is within a preset range, and whether the modulus length of the first vector product meets the preset conditions. If both are true, the corresponding process parameter values ​​are taken as a set of process parameters; otherwise, according to the original data magnitude Q i In [c*Q i ,d*Q i ] randomly select 500 parameter points, c is less than d, and c and d are both constants greater than zero. For each parameter point, determine whether the difference between the modulus length of the corresponding bubble feature vector and the modulus length of the target vector is within the preset range. If so, calculate the corresponding second vector product according to the corresponding bubble feature vector and the target vector, and determine whether the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time. If the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time, update the process parameters. Otherwise, update the process parameters according to the formula Determine whether to accept the current parameter point; after 1000 iterations or when the iteration end point is met, obtain the final process parameters, and use the final process parameters as a set of process parameters.

[0032] Optionally, the image data is images of the upper and lower surfaces of the optical glass, the ultrasonic reflection signal is the ultrasonic reflection signal of the upper and lower surfaces of the optical glass, and the smelting process data includes at least one of the smelting feeding time t, the feeding weight G, the temperature T1~TN at various locations in the smelting furnace, and the stirring rate v.

[0033] Optionally, after the step of obtaining the bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics, the method further includes:

[0034] Obtaining target smelting process data; the target smelting process data includes target smelting charging time, target charging weight, target temperature T1~TN at each part of the smelting furnace, and target stirring rate;

[0035] Inputting the target smelting process data into the bubble characteristic prediction model to obtain predicted characteristic data of the bubble; the predicted characteristic data includes at least one of predicted density data, predicted area data and predicted depth data;

[0036] The predicted characteristic data is input into the bubble level prediction model to obtain the corresponding predicted bubble level.

[0037] By adopting the above technical scheme, in order to predict the bubble grade according to the smelting process data, the target smelting process data is first obtained, and the target smelting process data includes the target smelting feeding time, the target feeding weight, the target temperature T1~TN at each part of the smelting furnace, and the target stirring rate. Then, the target smelting process data is input into the bubble characteristic prediction model to obtain the predicted characteristic data of the bubbles, and the predicted characteristic data includes the predicted density data, the predicted area data and the predicted depth data. Then, the predicted characteristic data is input into the bubble grade prediction model to obtain the corresponding predicted bubble grade.

[0038] In the second aspect, the present application also provides an optical glass melting process prediction system, which adopts the following technical solution:

[0039] Optical glass melting process prediction system, including:

[0040] A data acquisition module, used to acquire melting process data and bubble evaluation data of the optical glass, wherein the bubble evaluation data includes image data, ultrasonic reflection signal and grade data of the optical glass;

[0041] A feature extraction module, used to extract features from the image data and the ultrasonic reflection signal to obtain corresponding bubble features; the bubble features include bubble density features, bubble area features and bubble depth features;

[0042] A bubble level prediction model generation module, used to obtain a bubble level prediction model based on a first preset model and according to the bubble characteristics and the level data;

[0043] A bubble characteristic prediction model generation module, used for obtaining a bubble characteristic prediction model based on a second preset model and according to the smelting process data and the bubble characteristics;

[0044] A target bubble characteristic parameter generation module is used to obtain a target bubble level and restriction conditions, traverse the bubble density, bubble area and bubble depth based on a random network search algorithm, and call the bubble level prediction model to predict, obtain the bubble level, and use the bubble density, bubble area and bubble depth corresponding to the bubble level consistent with the target bubble level as target bubble characteristic parameters; the target bubble characteristic parameters include target bubble density, target bubble area and target bubble depth;

[0045] A process parameter screening module is used to generate a target vector according to the target bubble density, the target bubble area and the target bubble depth, and call the bubble characteristic prediction model, set process preconditions and ranges, and screen out a group of process parameters according to the Monte Carlo algorithm and the target vector, wherein the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters.

[0046] In a third aspect, the present application also provides a computer device, which adopts the following technical solution:

[0047] 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.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0049] 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.

[0050] In summary, the present application includes at least the following beneficial technical effects: first, the melting process data and bubble evaluation data of the optical glass are obtained, the bubble evaluation data including the image data, ultrasonic reflection signal and grade data of the optical glass, and then the image data and the ultrasonic reflection signal are feature extracted to obtain corresponding bubble features; the bubble features include at least one of bubble density features, bubble area features and bubble depth features, and then based on a first preset model, a bubble grade prediction model is obtained according to the bubble features and the grade data, and based on a second preset model, a bubble feature prediction model is obtained according to the melting process data and the bubble features, and then the target bubble grade and restriction conditions are obtained, at least one of the bubble density, bubble area and bubble depth is traversed based on a random network search algorithm, and the bubble grade prediction model is called to perform A prediction is performed to obtain a bubble level, and bubble parameters corresponding to the bubble level consistent with the target bubble level are used as target bubble characteristic parameters, wherein the bubble parameters include at least one of bubble density, bubble area and bubble depth, and the target bubble characteristic parameters include at least one of target bubble density, target bubble area and target bubble depth. Finally, a target vector is generated according to the target bubble characteristic parameters, and the bubble characteristic prediction model is called to set process preconditions and ranges, and a group of process parameters are screened out according to the Monte Carlo algorithm and the target vector, and the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters. Compared with the method of optimizing process parameters by trial and error, the method of the present invention improves the optimization efficiency of the optical glass melting process, reduces the optimization cost of the optical glass melting process, and realizes the digital tuning of the optical glass melting process. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present application.

[0052] Figure 2 It is a schematic diagram of the structure of the system of this application.

[0053] Figure 3 It is a structural block diagram of the computer device of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. 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.

[0055] The embodiment of the present application discloses a method for predicting an optical glass melting process.

[0056] Reference Figure 1 , optical glass melting process prediction method, including:

[0057] Step S11, obtaining the melting process data and bubble evaluation data of the optical glass.

[0058] The bubble evaluation data includes image data, ultrasonic reflection signals and grade data of the optical glass.

[0059] It should be noted that the image data may be image data of the upper and lower surfaces of the optical glass (V U and V L ), or it can be the image data of one side of the optical glass (V U or V L The ultrasonic reflection signal can be the ultrasonic reflection signal of the upper and lower surfaces of the optical glass (S U and S L ), or it can be the ultrasonic reflection signal of one side of the optical glass (S U or S L , the ultrasonic reflection signal is the reflected wave signal corresponding to the ultrasonic signal; the smelting process data includes at least one of the smelting charging time t, the charging weight G, the temperature T1~TN at each location of the smelting furnace, and the stirring rate v; the grade data includes the grade (L) corresponding to the bubbles in the optical glass.

[0060] Step S12: extracting features from the image data and the ultrasonic reflection signal to obtain corresponding bubble features.

[0061] The bubble characteristics include at least one of a bubble density characteristic, a bubble area characteristic and a bubble depth characteristic.

[0062] Step S13, based on the first preset model, and according to the bubble characteristics and the level data, a bubble level prediction model is obtained.

[0063] Step S14, based on the second preset model and according to the smelting process data and the bubble characteristics, a bubble characteristic prediction model is obtained.

[0064] The first preset model and / or the second preset model may be an SVM model or other artificial intelligence models.

[0065] Step S15, obtain the target bubble level and restriction conditions, traverse at least one of the bubble density, bubble area and bubble depth based on the random network search algorithm, and call the bubble level prediction model to predict to obtain the bubble level, and use the bubble parameters corresponding to the bubble level consistent with the target bubble level as the target bubble characteristic parameters.

[0066] The bubble parameters include at least one of bubble density, bubble area and bubble depth, and the target bubble characteristic parameters include target bubble density, target bubble area and target bubble depth.

[0067] It should be noted that in step S15, when the bubble parameters are bubble density, bubble area and bubble depth, the random grid search algorithm is used according to the target bubble level and the restriction conditions. The acceptable bubble density, bubble area, and bubble depth are traversed, and the SVM model (i.e., bubble grade prediction model) is called to predict the bubble grade, and the corresponding bubble density, area, and depth parameters that are consistent with the target bubble grade are retained.

[0068] It should be further explained that the bubble feature in step S12 (at least one of the bubble density feature, bubble area feature and bubble depth feature) and the traversal object in step S15 (at least one of the bubble density, bubble area and bubble depth) should correspond. For example, if the bubble feature is the bubble density feature and the bubble area feature, the traversal object is the bubble density and the bubble area; if the bubble feature is the bubble density feature, the bubble area feature and the bubble depth feature, the traversal object is the bubble density, the bubble area and the bubble depth.

[0069] Step S16, based on the target bubble characteristic parameter target vector, and calling the bubble characteristic prediction model, setting the process preconditions and range, and selecting a set of process parameters according to the Monte Carlo algorithm and the target vector, the bubble characteristic parameters corresponding to the set of process parameters are closest to the target bubble characteristic parameters.

[0070] In the above embodiment, the melting process data and bubble evaluation data of the optical glass are first obtained, the bubble evaluation data includes the image data, ultrasonic reflection signal and grade data of the optical glass, and then the image data and the ultrasonic reflection signal are feature extracted to obtain the corresponding bubble features; the bubble features include at least one of bubble density features, bubble area features and bubble depth features, and then based on a first preset model, a bubble grade prediction model is obtained according to the bubble features and the grade data, and based on a second preset model, a bubble feature prediction model is obtained according to the melting process data and the bubble features, and then the target bubble grade and restriction conditions are obtained, at least one of the bubble density, bubble area and bubble depth is traversed based on a random network search algorithm, and the bubble grade prediction model is called for prediction to obtain The target bubble level is a bubble level, and the bubble parameters corresponding to the bubble level consistent with the target bubble level are used as target bubble characteristic parameters, the bubble parameters include at least one of bubble density, bubble area and bubble depth, and the target bubble characteristic parameters include at least one of target bubble density, target bubble area and target bubble depth. Finally, a target vector is generated according to the target bubble characteristic parameters, and the bubble characteristic prediction model is called to set the process preconditions and ranges, and a group of process parameters are screened out according to the Monte Carlo algorithm and the target vector, and the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters. Compared with the method of optimizing process parameters by trial and error, the method of the present invention improves the optimization efficiency of the optical glass melting process, reduces the optimization cost of the optical glass melting process, and realizes the digital tuning of the optical glass melting process.

[0071] As a further implementation of the method, the bubble features include bubble density features, bubble area features and bubble depth features, and the steps of extracting features from the image data and the ultrasonic reflection signal to obtain the corresponding bubble features include:

[0072] Step S21, extracting features from the image data according to a pre-generated convolutional neural network CNN to obtain bubble density features and bubble area features.

[0073] It should be noted that this application adopts a deep learning algorithm, and inputs the convolutional neural network CNN for feature extraction of the upper and lower images of the bubbles. It uses three convolutional layers, a ReLU activation layer and a fully connected layer, back-propagates the update error, and classifies the images according to the bubble density and bubble area.

[0074] Step S22, extracting features of the ultrasonic reflection signal according to the pre-generated recurrent neural network RNN ​​to obtain bubble depth features.

[0075] It should be noted that, for the upper and lower surface ultrasonic signal reflection wave signals, this application uses a recurrent neural network RNN ​​to extract features, mark the audio of different bubble depths separately, and perform audio recognition and classification.

[0076] In the above embodiment, in order to obtain the corresponding bubble features, feature extraction is performed on the image data according to the pre-generated convolutional neural network CNN to obtain the bubble density feature and the bubble area feature, and feature extraction is performed on the ultrasonic reflection signal according to the pre-generated recurrent neural network RNN ​​to obtain the bubble depth feature.

[0077] As a further implementation of the method, the second preset model is a random forest model, and the step of obtaining a bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics includes:

[0078] Step S31, preprocessing the smelting process data according to the smelting process data to obtain preprocessed data.

[0079] It can be understood that the preprocessing operations include missing value checking and KNN nearest neighbor filling, 3σ principle to eliminate outliers, data standardization, etc.

[0080] Step S32, dividing the preprocessed data and bubble features into a training set and a test set according to a preset ratio, and setting and optimizing the hyperparameters of the random forest model.

[0081] Among them, hyperparameters include the number of random forest decision trees, the maximum number of features, the maximum depth, and the minimum number of sample splits.

[0082] It should be noted that the preset ratio can be 8:2 or 7:3; the hyperparameter optimization method uses the grid search algorithm Grid Search.

[0083] Step S33, training the random forest model according to the training set to obtain a trained random forest model.

[0084] It should be noted that in step S33, the evaluation index is the mean square error MSE.

[0085] Step S34, testing the trained random forest model according to the test set to obtain a random forest model that passes the test, and using the random forest model that passes the test as a bubble feature prediction model.

[0086] It should be noted that in step S34, the evaluation indicators selected are R2 and RMSE.

[0087] In the above embodiment, in order to obtain the bubble feature prediction model, the smelting process data is first preprocessed according to the smelting process data to obtain preprocessed data, and then the preprocessed data and the bubble features are divided into a training set and a test set according to a preset ratio, and the hyperparameters of the random forest model are set and optimized, and the hyperparameters include the number of random forest decision trees, the maximum number of features, the maximum depth and the minimum number of sample splits, and then the random forest model is trained according to the training set to obtain a trained random forest model, and finally the trained random forest model is tested according to the test set to obtain a tested random forest model, and the tested random forest model is used as the bubble feature prediction model.

[0088] As a further implementation method of the method, the process preconditions and ranges are set, and a set of process parameters are screened out according to the Monte Carlo algorithm and the target vector, and the steps of the bubble characteristic parameters corresponding to the set of process parameters being closest to the target bubble characteristic parameters include:

[0089] Step S41, in the initial state, the parameters in the process parameter set Φ corresponding to the smelting process data are set to a random state, and annealing is performed for 500 iterations.

[0090] It should be noted that the process parameter set Φ includes the smelting charging time (t), charging weight (G), the temperature of each part of the smelting furnace (T1~TN , stirring rate (v And other parameters.

[0091] Step S42, after annealing is completed, the corresponding first vector product is calculated based on the corresponding bubble characteristic vector and the target vector in the current state, and it is determined whether the difference between the modulus length of the first vector product and the modulus length of the target vector is within a preset range, and whether the modulus length of the first vector product meets the preset conditions. If so, the corresponding process parameter values ​​are used as a set of process parameters.

[0092] It should be noted that in this application, the preset range is The default condition is Specifically, after annealing is completed, the first vector product with the target vector is calculated based on the bubble density, area, and depth in this state. The modulus length of the first vector product is Representing the angle difference between vectors at small angles can characterize the prediction accuracy of the vectors. If the modulus length of the new vector (i.e., the first vector product) is | The difference between | and the target vector is Within, and the first vector product modulus length is less than or equal to If the value is within the range of 0, the parameters are saved, otherwise, step S43 is executed.

[0093] Step S43: Otherwise, according to the original data level Q iIn [c*Q i ,d*Q i ] randomly select 500 parameter points, c is less than d, and c and d are both constants greater than zero. For each parameter point, determine whether the difference between the modulus of the corresponding bubble feature vector and the modulus of the target vector is within a preset range. If so, calculate the corresponding second vector product according to the corresponding bubble feature vector and the target vector, and determine whether the modulus of the second vector product is less than the modulus of the second vector product calculated last time.

[0094] Step S44, if the modulus of the second vector product is smaller than the modulus of the second vector product calculated last time, update the process parameters; otherwise, determine whether to accept the current parameter point based on the difference between the modulus of the second vector product and the modulus of the second vector product calculated last time.

[0095] It should be noted that in this application, c is 0.8 and d is 1.2. Specifically, according to the magnitude of the original data , in [0.8 , 1.2 ] Randomly select 500 points within the range, and these 500 points are called random points, where At each random point, the density, area, and depth of the bubble are calculated to correspond to the vector difference between the vector and the target vector. If the vector difference is If the vector product modulus is less than the last calculated vector product modulus, then ,in, If the second vector product modulus length calculated this time and the second vector product modulus length calculated last time are respectively, the process parameters are updated, otherwise the probability Choose to accept this parameter point, where The number of iterations or other parameters preset by the user.

[0096] Step S45, iterate 1000 times or meet the iteration end point to obtain the final process parameters, and use the final process parameters as a set of process parameters.

[0097] It can be understood that the iteration endpoint is the preset range and preset condition mentioned in step S42.

[0098] In the above embodiment, in order to screen the process parameters, in the initial state, the parameters in the process parameter set Φ corresponding to the smelting process data are set to a random state, and annealing is performed for 500 iterations; after the annealing is completed, the corresponding first vector product is calculated according to the corresponding bubble feature vector and the target vector in the current state, and it is determined whether the difference between the modulus length of the first vector product and the modulus length of the target vector is within a preset range, and whether the modulus length of the first vector product meets the preset conditions. If both are true, the corresponding process parameter values ​​are taken as a set of process parameters; otherwise, according to the original data magnitude Qi In [c*Q i ,d*Q i ] randomly select 500 parameter points, c is less than d, and c and d are both constants greater than zero. For each parameter point, determine whether the difference between the modulus length of the corresponding bubble feature vector and the modulus length of the target vector is within the preset range. If so, calculate the corresponding second vector product according to the corresponding bubble feature vector and the target vector, and determine whether the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time. If the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time, update the process parameters. Otherwise, update the process parameters according to the formula Determine whether to accept the current parameter point; after 1000 iterations or when the iteration end point is met, obtain the final process parameters, and use the final process parameters as a set of process parameters.

[0099] As a further implementation of the method, after the step of obtaining the bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics, the method further includes:

[0100] Step S51, obtaining target smelting process data.

[0101] The target smelting process data include target smelting charging time, target charging weight, target temperature T1~TN at each location of the smelting furnace, and target stirring rate.

[0102] Step S52, inputting the target smelting process data into the bubble characteristic prediction model to obtain the predicted characteristic data of the bubbles.

[0103] Among them, the predicted feature data includes predicted density data, predicted area data and predicted depth data.

[0104] Step S53, inputting the predicted feature data into the bubble level prediction model to obtain the corresponding predicted bubble level.

[0105] In the above embodiment, in order to predict the bubble level according to the smelting process data, the target smelting process data is first obtained, and the target smelting process data includes the target smelting feeding time, the target feeding weight, the target temperature T1~TN at each part of the smelting furnace, and the target stirring rate. Then, the target smelting process data is input into the bubble characteristic prediction model to obtain the predicted characteristic data of the bubbles, and the predicted characteristic data includes the predicted density data, the predicted area data and the predicted depth data. Then, the predicted characteristic data is input into the bubble level prediction model to obtain the corresponding predicted bubble level.

[0106] The embodiment of the present application also discloses an optical glass melting process prediction system.

[0107] refer to Figure 2, optical glass melting process prediction system, including:

[0108] A data acquisition module, used to acquire melting process data and bubble evaluation data of the optical glass, wherein the bubble evaluation data includes image data, ultrasonic reflection signal and grade data of the optical glass;

[0109] A feature extraction module is used to extract features from image data and ultrasonic reflection signals to obtain corresponding bubble features; the bubble features include bubble density features, bubble area features and bubble depth features;

[0110] A bubble level prediction model generation module, used to obtain a bubble level prediction model based on the first preset model and according to bubble characteristics and level data;

[0111] A bubble characteristic prediction model generation module, used to obtain a bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics;

[0112] A target bubble characteristic parameter generation module is used to obtain a target bubble level and restriction conditions, traverse at least one of bubble density, bubble area and bubble depth based on a random network search algorithm, and call the bubble level prediction model to predict, obtain the bubble level, and use the bubble parameter corresponding to the bubble level consistent with the target bubble level as the target bubble characteristic parameter; the target bubble characteristic parameter includes at least one of the target bubble density, the target bubble area and the target bubble depth;

[0113] The process parameter screening module is used to generate a target vector based on the target bubble density, target bubble area and target bubble depth, and call the bubble characteristic prediction model to set the process preconditions and ranges, and screen out a set of process parameters based on the Monte Carlo algorithm and the target vector. The bubble characteristic parameters corresponding to the set of process parameters are closest to the target bubble characteristic parameters.

[0114] The optical glass melting process prediction system of the present invention can implement any one of the optical glass melting process prediction methods, and the specific working process of the optical glass melting process prediction system of the present invention can refer to the corresponding process in the above-mentioned optical glass melting process prediction method.

[0115] The embodiment of the present application also discloses a computer device.

[0116] 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 melting process prediction methods is implemented.

[0117] The embodiment of the present application also discloses a computer-readable storage medium.

[0118] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any one of the optical glass melting process prediction methods.

[0119] 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.

[0120] The above are all preferred embodiments of the present application, and are not intended to limit the scope of protection of the present application. This specification (including abstract and drawings) Any feature disclosed in the invention, unless otherwise stated, can be replaced by other equivalent or alternative features having similar purposes. That is, unless otherwise stated, each feature is only an example in a series of equivalent or similar features.

Claims

1. A method for predicting optical glass melting process, characterized in that: include: Acquire melting process data and bubble evaluation data of the optical glass, wherein the bubble evaluation data includes image data, ultrasonic reflection signal and grade data of the optical glass; the image data includes image data of at least one side of the optical glass, the ultrasonic reflection signal includes ultrasonic reflection signal of at least one side of the optical glass, the melting process data includes at least one of melting feeding time t, feeding weight G, temperature T1-TN at each location of the melting furnace, and stirring rate v, and the grade data includes the grade corresponding to the bubbles in the optical glass; Performing feature extraction on the image data and the ultrasonic reflection signal to obtain corresponding bubble features; the bubble features include at least one of a bubble density feature, a bubble area feature, and a bubble depth feature; Based on the first preset model, and according to the bubble characteristics and the grade data, a bubble grade prediction model is obtained; Based on the second preset model, and according to the smelting process data and the bubble characteristics, a bubble characteristic prediction model is obtained; Obtaining a target bubble level and restriction conditions, traversing at least one of bubble density, bubble area and bubble depth based on a random network search algorithm, and calling the bubble level prediction model to predict, obtaining the bubble level, and using the bubble parameters corresponding to the bubble level consistent with the target bubble level as the target bubble characteristic parameters; the bubble parameters include at least one of bubble density, bubble area and bubble depth, and the target bubble characteristic parameters include at least one of target bubble density, target bubble area and target bubble depth; Generate a target vector according to the target bubble characteristic parameters, call the bubble characteristic prediction model, set process preconditions and ranges, and select a group of process parameters according to the Monte Carlo algorithm and the target vector, wherein the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters; The step of setting the process preconditions and ranges, and selecting a set of process parameters according to the Monte Carlo algorithm and the target vector, wherein the bubble characteristic parameters corresponding to the set of process parameters are closest to the target bubble characteristic parameters, comprises: In the initial state, the parameters in the process parameter set Φ corresponding to the smelting process data are set to a random state, and annealing is performed for 500 iterations; After the annealing is completed, the first vector product is calculated according to the corresponding bubble characteristic vector and the target vector in the current state, and it is determined whether the difference between the modulus length of the first vector product and the modulus length of the target vector is within a preset range, and whether the modulus length of the first vector product meets a preset condition. If yes, the corresponding process parameter value is used as a set of process parameters; Otherwise, randomly select 500 parameter points in [c*Qi,d*Qi] according to the original data magnitude Qi, where c is less than d, and both c and d are constants greater than zero. For each parameter point, determine whether the difference between the modulus length of the corresponding bubble feature vector and the modulus length of the target vector is within the preset range. If so, calculate the corresponding second vector product according to the corresponding bubble feature vector and the target vector, and determine whether the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time; If the modulus length of the second vector product is less than the modulus length of the second vector product calculated last time, the process parameters are updated; otherwise, whether to accept the current parameter point is determined according to the difference between the modulus length of the second vector product and the modulus length of the second vector product calculated last time; Iterate 1000 times or meet the iteration endpoint to obtain the final process parameters, and use the final process parameters as a set of process parameters.

2. The optical glass melting process prediction method according to claim 1, characterized in that: The bubble features include bubble density features, bubble area features and bubble depth features. The step of extracting features from the image data and the ultrasonic reflection signal to obtain corresponding bubble features includes: Performing feature extraction on the image data according to a pre-generated convolutional neural network (CNN) to obtain the bubble density feature and the bubble area feature; The ultrasonic reflection signal is subjected to feature extraction according to a pre-generated recurrent neural network (RNN) to obtain the bubble depth feature.

3. The optical glass melting process prediction method according to claim 1, characterized in that: The first preset model is a SVM model.

4. The optical glass melting process prediction method according to claim 1, characterized in that: The second preset model is a random forest model, and the step of obtaining a bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics includes: preprocessing the smelting process data according to the smelting process data to obtain preprocessed data; Dividing the preprocessed data and the bubble features into a training set and a test set according to a preset ratio, and setting and optimizing the hyperparameters of the random forest model; the hyperparameters include the number of random forest decision trees, the maximum number of features, the maximum depth, and the minimum number of sample splits; Training the random forest model according to the training set to obtain a trained random forest model; The trained random forest model is tested according to the test set to obtain a random forest model that passes the test, and the random forest model that passes the test is used as a bubble feature prediction model.

5. The optical glass melting process prediction method according to claim 1, characterized in that: The image data is the image of the upper and lower surfaces of the optical glass, the ultrasonic reflection signal is the ultrasonic reflection signal of the upper and lower surfaces of the optical glass, and the smelting process data includes at least one of the smelting feeding time t, the feeding weight G, the temperature T1~TN at each location of the smelting furnace, and the stirring rate v.

6. The optical glass melting process prediction method according to claim 1, characterized in that: After the step of obtaining the bubble characteristic prediction model based on the second preset model and according to the smelting process data and the bubble characteristics, the method further includes: Obtaining target smelting process data; the target smelting process data includes target smelting charging time, target charging weight, target temperature T1~TN at each part of the smelting furnace, and target stirring rate; Inputting the target smelting process data into the bubble characteristic prediction model to obtain predicted characteristic data of the bubble; the predicted characteristic data includes predicted density data, predicted area data and predicted depth data; The predicted characteristic data is input into the bubble level prediction model to obtain the corresponding predicted bubble level.

7. Optical glass melting process prediction system, characterized in that: include: A data acquisition module, used to acquire melting process data and bubble evaluation data of the optical glass, wherein the bubble evaluation data includes image data, ultrasonic reflection signal and grade data of the optical glass; A feature extraction module, used to extract features from the image data and the ultrasonic reflection signal to obtain corresponding bubble features; the bubble features include at least one of a bubble density feature, a bubble area feature, and a bubble depth feature; A bubble level prediction model generation module, used to obtain a bubble level prediction model based on a first preset model and according to the bubble characteristics and the level data; A bubble characteristic prediction model generation module, used for obtaining a bubble characteristic prediction model based on a second preset model and according to the smelting process data and the bubble characteristics; A target bubble characteristic parameter generation module is used to obtain a target bubble level and restriction conditions, traverse at least one of bubble density, bubble area and bubble depth based on a random network search algorithm, and call the bubble level prediction model to predict, obtain the bubble level, and use the bubble parameter corresponding to the bubble level consistent with the target bubble level as the target bubble characteristic parameter; the target bubble characteristic parameter includes at least one of the target bubble density, the target bubble area and the target bubble depth; A process parameter screening module is used to generate a target vector according to the target bubble density, the target bubble area and the target bubble depth, and call the bubble characteristic prediction model, set process preconditions and ranges, and screen out a group of process parameters according to the Monte Carlo algorithm and the target vector, wherein the bubble characteristic parameters corresponding to the group of process parameters are closest to the target bubble characteristic parameters.

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