Building glass production method, system and equipment

Abstract

The invention relates to the technical field of glass production, and provides a building glass production method, system and equipment, and the method comprises the steps: obtaining historical production information data sets of a plurality of different batches of building glass of the same type; constructing a multi-target dynamic loss function, and constructing a neural network model based on the loss function; training the neural network model according to the historical production information data set; and obtaining basic information of building glass to be produced, inputting the basic information into the neural network model to obtain optimal process control parameters, and producing the building glass according to the optimal process control parameters. According to the method, the robustness of the neural network model is improved, so that the optimal process control parameters matched with the current raw material information and the target performance information of the to-be-produced building glass are obtained, and the intelligent degree and the production benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of glass production, and more specifically, to a method, system, and equipment for producing architectural glass. Background Art

[0002] Architectural glass refers to transparent or translucent glass materials used for building components, which have characteristics such as daylighting, heat insulation, and sound insulation, and can be divided into various types, including ordinary flat glass, tempered glass, laminated glass, LOW-E glass, embossed glass, and frosted glass, etc.

[0003] After retrieval, some typical prior arts were found. For example, the application number CN2021110133635 discloses an intelligent production process method and production system for architectural engineering glass, which includes processes such as raw embryo cutting, edge grinding, cleaning, heat soaking, and production of composite glass. Another example is that the application number CN2023116763280 discloses a production control system and method for tempered glass, which is based on artificial intelligence technology of a deep neural network model to realize the analysis and classification of the preheating process of tempered glass, and determine whether to stop preheating to improve the efficiency and quality of the tempered glass production process. Still another example is that the application number CN2023102954723 discloses an intelligent production control method for tempered glass, which characterizes the lattice structure of nickel sulfide grains in tempered glass through the principle of photoluminescence, and based on the change of its crystal phase, judges the timing of self-explosion of tempered glass, so as to take out the tempered glass from the homogenizing furnace in advance to prevent pollution and damage to the homogenizing furnace.

[0004] The above three prior arts all relate to the architectural production methods of glass, but do not take into account multiple factors affecting the production efficiency of architectural glass, and it is difficult to obtain relatively ideal process control parameters, and there is still room for further improvement in its intelligence level and production efficiency. Summary of the Invention

[0005] Based on this, in order to solve the problems existing in the prior art, the present invention provides a method, system, and equipment for producing architectural glass, which constructs a neural network model according to multi-scale data information including defect information, energy consumption information, and environmental factor information, and obtains a historical production information dataset of multiple different batches of architectural glass of the same type to train the neural network model, taking into account multiple factors affecting the production efficiency of architectural glass, which can improve the robustness of the neural network model, so as to obtain the optimal process control parameters matching the current raw material information and target performance information of the architectural glass to be produced, improving the intelligence level and production efficiency. The specific technical solutions are as follows:

[0006] A method for producing architectural glass includes the following steps:

[0007] Obtain the type of building glass to be produced and obtain a historical production information dataset of multiple different batches of building glass of the same type;

[0008] Construct a multi-objective dynamic loss function based on defect information, energy consumption information, performance information, and environmental factor information, and construct a neural network model based on the loss function;

[0009] Train the neural network model according to the historical production information dataset;

[0010] Obtain the basic information of the building glass to be produced, input the basic information into the neural network model to obtain the optimal process control parameters, and produce the building glass according to the optimal process control parameters;

[0011] Among them, the historical production information dataset includes historical raw material information, historical process control parameters, historical defect information, historical performance information, historical energy consumption information, and historical environmental factor information, and the basic information includes current raw material information, target performance information, and environmental factor information during the production process.

[0012] By constructing a neural network model and training the neural network model with a historical production information dataset of multiple different batches of building glass of the same type, the building glass production method takes into account multiple factors affecting the production efficiency of building glass, can improve the robustness of the neural network model, thereby obtaining the optimal process control parameters that match the current raw material information and target performance information of the building glass to be produced, and improves the degree of intelligence and production efficiency.

[0013] Preferably, the specific method for training the neural network model according to the historical production information dataset includes the following steps:

[0014] Obtain the correlation degrees of multiple different batches of building glass of the same type with the building glass to be produced respectively;

[0015] Screen the historical production information dataset according to a preset correlation degree threshold and the correlation degrees;

[0016] Train the neural network model according to the screened historical production information dataset.

[0017] Preferably, the multi-objective dynamic loss function \(L = L 1 +L 2 +L 3 +β×||W defect W energy || 2 ;

[0018] Among them, \(L 1 =α defect ||wdefect || 2 +α energy ||w energy || 2 represents the multi - objective regularization term, α defect and α energy represent the defect loss weight coefficient and the energy consumption loss weight coefficient respectively, w defect and w energy represent the defect information feature vector and the energy consumption loss feature vector respectively, |||| represents the norm, L 2 = Relu(f(env)-threshold) represents the environment - sensitive penalty term, represents the comprehensive environment sensitivity evaluation function, Relu() represents the activation function, threshold represents the environmental factor safety threshold, M1 represents the number of types of environmental factors, λ i represents the evaluation value weight of the i - th type of environmental factor, e represents the natural constant, represents the intermediate variable, E i represents the actual value of the i - th environmental factor, E i ' represents the standard value of the i - th environmental factor, || represents the absolute value function, represents the performance loss term, α performance represents the performance loss weight coefficient, M2 represents the number of types of performance information, p i represents the actual value of the i - th type of performance information, p i ' represents the predicted value of the i - th type of performance information β×||W defect W energy || 2 represents the weight regularization term, β represents the regularization strength, W defect represents the defect loss weight matrix, W energy represents the energy consumption loss weight matrix.

[0019] Preferably, the correlation degree R = R'+R”+1 / R”', where R', R”, and R”' represent the raw material correlation coefficient, the performance parameter correlation coefficient, and the defect correlation coefficient respectively;

[0020] R' is obtained by calculating the weighted value of the Pearson correlation coefficients between the current raw materials of multiple types and the historical raw material feature information values;

[0021] R” is obtained by calculating the weighted value of the Pearson correlation coefficients between the target performance parameters of multiple types and the historical performance parameter feature information values;

[0022] R”' is obtained by calculating the weighted value of the mean square errors between the historical defect feature values of the historical defect information of multiple types and the corresponding defect feature standard values.

[0023] Preferably, the specific method for screening the historical production information dataset includes: screening and processing the historical production information of batches of architectural glass with a correlation less than a preset correlation threshold;

[0024] The specific method for training the neural network model according to the screened historical production information dataset includes: preprocessing the historical production information of multiple batches of architectural glass after screening and processing to generate a historical production information dataset, and training the neural network model according to the historical production information dataset to obtain the trained neural network model.

[0025] An architectural glass production system for implementing the architectural glass production method includes:

[0026] A production information acquisition module for acquiring the type of architectural glass to be produced and acquiring a historical production information dataset of multiple different batches of the same type of architectural glass;

[0027] A neural network construction module for constructing a multi-objective dynamic loss function based on defect information, energy consumption information, performance information, and environmental factor information, and constructing a neural network model based on the loss function;

[0028] A training module for training the neural network model according to the historical production information dataset;

[0029] A process parameter acquisition module for acquiring the basic information of the architectural glass to be produced, inputting the basic information into the neural network model to obtain the optimal process control parameters, and producing the architectural glass according to the optimal process control parameters;

[0030] Wherein, the historical production information dataset includes historical raw material information, historical process control parameters, historical defect information, historical performance information, historical energy consumption information, and historical environmental factor information, and the basic information includes current raw material information, target performance information, and environmental factor information during the production process.

[0031] Preferably, the training module includes:

[0032] A correlation acquisition unit for respectively acquiring the correlations between multiple different batches of the same type of architectural glass and the architectural glass to be produced;

[0033] A screening unit for screening the historical production information dataset according to a preset correlation threshold and the correlation;

[0034] A training unit for training the neural network model according to the screened historical production information dataset.

[0035] Preferably, the neural network construction module includes:

[0036] A loss function construction unit, configured to construct a multi-objective dynamic loss function L = L 1 +L 2 +β×||W defect W energy || 2 ;

[0037] A neural network model construction unit, configured to construct a neural network model based on the multi-objective dynamic loss function;

[0038] Wherein, L 1 =α defect ||w defect || 2 +α energy ||w energy || 2 represents a multi-objective regularization term, α defect , α energy respectively represent a defect loss weight coefficient and an energy consumption loss weight coefficient, w defect , w energy respectively represent a defect information feature vector and an energy consumption loss feature vector, |||| represents a norm, L 2 =Relu(f(env)-threshold) represents an environment-sensitive penalty term, represents a comprehensive environmental sensitivity evaluation function, Relu() represents an activation function, threshold represents an environmental factor safety threshold, M1 represents the number of types of environmental factors, λ i represents the evaluation value weight of the i-th type of environmental factor, e represents the natural constant, represents an intermediate variable, E i represents the actual value of the i-th environmental factor, E i ' represents the standard value of the i-th environmental factor, || represents an absolute value function, represents a performance loss term, α performance represents a performance loss weight coefficient, M2 represents the number of types of performance information, p i represents the actual value of the i-th type of performance information, p i ' represents the predicted value of the i-th type of performance information β×||W defect W energy || 2 represents a weight regularization term, β represents a regularization strength, W defect represents a defect loss weight matrix, W energy represents an energy consumption loss weight matrix.

[0039] Preferably, the correlation degree acquisition unit includes:

[0040] The first sub-unit is configured to obtain a raw material correlation coefficient R' by calculating a weighted value of Pearson correlation coefficients between current raw materials of multiple types and historical raw material characteristic information values;

[0041] The second sub-unit is configured to obtain a performance parameter correlation coefficient R'' by calculating a weighted value of Pearson correlation coefficients between target performance parameters of multiple types and historical performance parameter characteristic information values;

[0042] The third sub-unit is configured to obtain a defect correlation coefficient R''' by calculating a weighted value of mean square errors between historical defect characteristic values of historical defect information of multiple types and corresponding defect characteristic standard values;

[0043] Wherein, the correlation degree R = R' + R'' + 1 / R'''.

[0044] A building glass production device, comprising:

[0045] A controller;

[0046] A memory storing executable instructions;

[0047] Wherein, the executable instructions can run on the controller and implement the described building glass production method. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention can be further understood from the following description in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but the emphasis is placed on showing the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0049] Figure 1 is an overall flowchart of a building glass production method in an embodiment of the present invention;

[0050] Figure 2 is a flowchart of a specific method for training a neural network model in an embodiment of the present invention;

[0051] Figure 3 is a flowchart of a specific method for calculating a raw material correlation coefficient, a performance parameter correlation coefficient, and a defect correlation coefficient in an embodiment of the present invention;

[0052] Figure 4 is an overall structural schematic diagram of a building glass production system in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below in conjunction with its embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the protection scope of the present invention.

[0054] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly on the other element or there can also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are only for the purpose of illustration and do not represent the only implementation.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the description of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0056] The "first" and "second" mentioned in the present invention do not represent specific quantities and orders, but are only used for name distinction.

[0057] Before elaborating on the embodiments of the present invention, a brief introduction to the prior art will be given first.

[0058] Classified according to application scenarios and functions, architectural glass can be divided into safety protection glass (such as tempered glass, laminated glass, explosion-proof glass, etc.), energy-saving and environmental protection glass (such as insulating glass, low-emissivity glass, heat-absorbing glass, etc.), decorative art glass (such as painted / patterned glass, embossed glass, etc.) and functional characteristic glass (such as fireproof glass, soundproof glass, self-cleaning glass, etc.). For the production of architectural glass, it generally includes raw material preparation, cleaning and drying, melting and forming, heat treatment and strengthening (such as annealing and tempering treatment, etc.), deep processing and quality inspection and packaging. For the two steps of melting and forming and heat treatment and strengthening, there are many process control parameters involved, and these process control parameters often need to be adjusted according to production information such as raw materials, preset target performance parameters (such as light transmittance, impact resistance, flatness, hardness and toughness, etc.), quality requirements for surface and internal defects, etc., in order to meet production requirements and improve production efficiency.

[0059] The glass production methods disclosed in Application No. CN2021110133635, Application No. CN2023116763280, Application No. CN2023102954723, etc. do not take into account multiple factors affecting the production efficiency of architectural glass, that is, they do not comprehensively optimize process control parameters based on basic information such as raw material information, preset target performance information, and quality requirements for surface and internal defects during the glass production process. There is room for further optimization in terms of their intelligence level and production efficiency.

[0060] To optimize the intelligence level and production efficiency of glass production, an embodiment of the present invention provides an architectural glass production method. It constructs a neural network model through multi-scale data information including defect information, energy consumption information, and environmental factor information, and obtains a historical production information dataset of multiple different batches of architectural glass of the same type to train the neural network model. By taking into account multiple factors affecting the production efficiency of architectural glass, it can improve the robustness of the neural network model, thereby obtaining the optimal process control parameters that match the current raw material information and target performance information of the architectural glass to be produced, improving the intelligence level and production efficiency.

[0061] As Figure 1 shown, the architectural glass production method includes the following steps:

[0062] S1, obtain the type of the architectural glass to be produced and obtain a historical production information dataset of multiple different batches of architectural glass of the same type.

[0063] The historical production information dataset includes historical raw material information, historical process control parameters, historical defect information, historical performance information, historical energy consumption information, and historical environmental factor information. Preferably, a virtual digital model can be constructed based on digital twin technology, and simulation can be carried out based on the digital model to obtain a more abundant and reference-worthy historical production information dataset. Of course, the historical production information dataset of multiple different batches of architectural glass of the same type here can be the historical production information dataset of multiple different batches of architectural glass of the same type in the production service database of the same production manufacturer, or the historical production information dataset of multiple different batches of architectural glass of the same type in the production server databases of multiple production manufacturers.

[0064] The raw material information includes, but is not limited to, parameter information (including types and ratios) such as quartz sand, limestone, feldspar, and soda ash used in the production of architectural glass; the process control parameters include, but are not limited to, parameter information such as melting temperature and time, and temperature and time at each stage of annealing; the defect information includes, but is not limited to, parameter information on surface and internal defects of architectural glass (such as scratches, bubbles, unevenness, distortion, cracks, and defects); the performance information includes, but is not limited to, parameter information such as light transmittance, impact resistance, pressure resistance, hardness, and toughness; the energy consumption information includes, but is not limited to, parameter information such as gas consumption, cooling water consumption, and electric energy; the environmental factor information includes, but is not limited to, parameter information such as environmental temperature and humidity, and vibration value.

[0065] Obtaining the type of architectural glass to be produced and obtaining a historical production information dataset of multiple different batches of architectural glass of the same type has the function of improving the referenceability of the historical production information dataset, so as to better train the neural network model.

[0066] S2. Construct a multi-objective dynamic loss function based on the defect information, energy consumption information, performance information, and environmental factor information, and construct a neural network model based on the loss function.

[0067] The neural network model includes, but is not limited to, a CNN convolutional neural network model, a BP neural network model, a fusion neural network model of Transformer and CNN.

[0068] For the fusion neural network model of Transformer and CNN, it aims to combine the advantages of both to improve the performance of the model. The fusion methods include:

[0069] 1. Early layer fusion: At the early stage of the network, fuse the feature maps of CNN with certain layers of Transformer. This usually involves converting the output of a certain intermediate layer of CNN into a sequence format that Transfommer can process, and then inputting it into Transformer.

[0070] 2. Horizontal layer fusion: Perform fusion at multiple levels of the network. The outputs of CNN and Transformer at each level can be combined together to create a richer representation that contains both local and global information.

[0071] 3. Sequential fusion: First use CNN to process the input data to extract local features, and then pass these features as input to Transformer. This method allows Transformer to model long-range dependencies based on the already extracted local features.

[0072] 4. Parallel Fusion: CNN and Transformer run in parallel and process the input data separately. Then, the outputs of the two are combined in some form (such as concatenation, addition, or multiplication) to create a representation that contains both local and global information.

[0073] 5. Encoder-Decoder Architecture: In this architecture, CNN is used as the encoder to extract local features of the input data, while Transformer is used as the decoder to utilize these features and generate the output. The decoder can use the self-attention mechanism of Transformer to model long-range dependencies and produce task-related outputs.

[0074] 6. Cross-Teaching Method: In a semi-supervised learning scenario, labeled images and unlabeled images are used as inputs. For the labeled data, CNN and Transformer are supervised by the true labels respectively. For the unlabeled data, the predictions generated by CNN and Transformer are used to update the parameters of Transformer / CNN respectively. This method can encourage consistency between different networks and improve the performance of the model.

[0075] Of course, the neural network model can be a single type of neural network for deep learning, or a fused neural network after combining two different types of neural networks, such as a fused neural network model of Transformer and CNN, to improve accuracy, robustness, and / or shorten the training time and improve the training effect.

[0076] Among them, by combining two neural networks and using the information in the forward and backward transmissions, the context information in the input data can be more fully utilized, thereby improving the accuracy and robustness of the model. For example, a double-layer bidirectional neural network structure can consider the context information in both the front and the back at the same time, enhancing the robustness and stability of the model; or by combining two neural networks and using the context information in both the forward and backward transmissions, the problems of gradient disappearance and gradient explosion can be reduced, thereby shortening the training time and improving the training effect. This structure can also improve the generalization ability of the model.

[0077] When combining two neural networks, different feature fusion methods can be adopted, specifically including:

[0078] 1. Simple Addition or Concatenation: The features extracted by different neural networks are simply added or concatenated. For example, for the feature vectors extracted by two neural networks, they can be simply added, or they can be concatenated along the feature dimension to form a longer feature vector.

[0079] 2. Weighted addition or concatenation: Weighted addition or concatenation is based on simple addition or concatenation, where weights are applied to the features extracted by different neural networks to better fuse the features. For example, the features extracted by each neural network can be weighted according to its performance. This method can reduce the feature dimension and improve the generalization ability of the model, but accurate evaluation of the performance of different neural networks is required to determine the weighting coefficients.

[0080] 3. Feature crossing: Feature crossing is to cross the features extracted by different neural networks to generate new feature representations. For example, the features extracted by two neural networks can be multiplied element by element to generate a new feature vector. This method can increase the diversity of features and improve the expressive ability of the model, but it may increase the feature dimension and the complexity of the model.

[0081] 4. Feature selection: Feature selection is to select the most representative features from the features extracted by different neural networks to reduce the feature dimension and improve the generalization ability of the model. For example, methods such as correlation analysis and PCA can be used to screen the features and select the most representative features. This method can reduce the feature dimension and improve the generalization ability of the model, but some important feature information may be lost.

[0082] S3. Train the neural network model according to the historical production information dataset.

[0083] Preferably, as Figure 2 shown, in step S3, the specific method for training the neural network model according to the historical production information dataset includes the following steps:

[0084] S31. Obtain the correlation degrees of multiple different batches of building glass of the same type with the building glass to be produced respectively.

[0085] Specifically, for the correlation degree between each batch of building glass and the building glass to be produced, it can be calculated by the formula R = R' + R'' + 1 / R''', where R', R'', and R''' represent the raw material correlation coefficient, the performance parameter correlation coefficient, and the defect correlation coefficient respectively.

[0086] As Figure 3 shown, the specific methods for calculating the raw material correlation coefficient, the performance parameter correlation coefficient, and the defect correlation coefficient include the following steps:

[0087] S311. Obtain the raw material correlation coefficient by calculating the weighted value of the Pearson correlation coefficients between the current raw materials of multiple types and the historical raw material feature information values;

[0088] S312. Obtain the performance parameter correlation coefficient by calculating the weighted value of the Pearson correlation coefficients between the target performance parameters of multiple types and the historical performance parameter characteristic information values;

[0089] S313. Obtain the defect correlation coefficient by calculating the weighted value of the mean square errors between the historical defect characteristic values of multiple types of historical defect information and the corresponding defect characteristic standard values.

[0090] More specifically, the raw material correlation coefficient

[0091] where represents the correlation coefficient value of the i-th type of raw material, N1 represents the number of types of raw materials, n1 represents the number of dimensions of the characteristic information of the i-th type of raw material and n1>1, RMCx j represents the j-th characteristic information value of the i-th type of current raw material under a certain dimension, RMCy j represents the j-th characteristic information value of the i-th type of historical raw material under a certain dimension, λ 1i represents the adjustment weight factor of the correlation coefficient value of the i-th type of raw material, which can be set by technicians according to experience.

[0092] The number of dimensions can be understood as the number of types of characteristic information of a certain type of raw material; the characteristic information includes but is not limited to the chemical composition content of the raw material (such as the content of silicon dioxide, iron oxide and alumina in quartz sand), particle size distribution, purity and impurity content (such as the content of elements such as iron, aluminum, calcium, magnesium, potassium, sodium, etc.) and physical properties (such as specific gravity, moisture content, hardness, etc.).

[0093] The performance parameter correlation coefficient

[0094] where represents the correlation coefficient value of the i-th type of performance parameter, M2 represents the number of types of performance parameters, n2 represents the number of characteristic information of the i-th type of performance parameter and n2>1, PPx j represents the j-th characteristic information value of the i-th type of target performance parameter, PPy j represents the j-th characteristic information value of the i-th type of historical performance parameter, λ 2i represents the adjustment weight factor of the correlation coefficient value of the i-th type of performance parameter, which can be set by technicians according to experience.

[0095] The performance parameters refer to the performance information of corresponding types of architectural glass, including but not limited to light transmittance, impact resistance, pressure resistance, hardness, and toughness. The number of characteristic information of a certain type of historical performance parameters can be understood as the number of batches of architectural glass for which the performance parameters are detected and collected. For example, when collecting the performance parameters of architectural glass, it is preset to detect and collect the performance parameters of n2 pieces of architectural glass in each batch, and the number of types of performance parameters collected for each piece of architectural glass is M2. More specifically, assuming that the types of performance parameters include light transmittance, impact resistance, pressure resistance, hardness, and toughness, then M2 = 5, and 100 pieces of architectural glass are detected and collected for the characteristic information values of performance parameters in each batch, so n2 = 100.

[0096] The target performance information is a preset value and can be set by technicians according to production requirements or product order requirements. The n2 characteristic information values under the target performance parameters of the i-th type are all uniformly set values, that is, PPx 1 ,PPx 2 ,…,PPx n2 are all the same preset values.

[0097] Defect correlation coefficient

[0098] Among them, represents the mean square error value of the defect information of the i-th type, N3 represents the number of types of defect information, n3 represents the number of characteristic information of the historical defect information of the i-th type and n3 > 1, D j represents the j-th historical defect characteristic value under the historical defect information of the i-th type, D j ' represents the j-th defect characteristic standard value under the historical defect information of the i-th type, which can be set by technicians, λ 3i represents the adjustment weight factor of the mean square error value of the historical defect information of the i-th type, which can be set by technicians according to experience.

[0099] The defect information includes but not limited to defect parameter information (such as scratches, bubbles, unevenness, distortion, cracks, and defects) on the surface and inside of architectural glass. The j-th defect characteristic standard value under the historical defect information of the i-th type is a preset value, which can be set by technicians according to production requirements, product order requirements, or production experience.

[0100] The quantity of characteristic information for a certain type of historical defect information can be understood as the quantity of batch building glass for which defect information is detected and collected. For example, when collecting defect information of building glass, it is preset that for each batch of n3 pieces of building glass, defect information is detected and collected, and the quantity of types of defect information collected for each piece of building glass is N3. More specifically, assuming the types of defect information include scratches, bubbles, unevenness, distortion, cracks, and defects, then N3 = 6, and for each batch, 50 pieces of building glass are detected and collected for defect characteristic values, so n3 = 50.

[0101] S32. Screen the historical production information dataset according to the preset correlation threshold and the said correlation.

[0102] The specific method for screening the historical production information dataset includes: eliminating and screening the historical production information of batch building glass with a correlation less than the preset correlation threshold.

[0103] S33. Train the neural network model according to the screened historical production information dataset.

[0104] The specific method for training the neural network model according to the screened historical production information dataset includes: preprocessing the historical production information of multiple batches of building glass after the elimination and screening process to generate a historical production information dataset, and training the neural network model according to the historical production information dataset to obtain the trained neural network model.

[0105] By observing the formula R = R' + R'' + 1 / R''', it can be seen that the correlation combines the raw material correlation coefficient, the performance parameter correlation coefficient, and the defect correlation coefficient, considering the correlation coefficient between historical performance information and target performance information, the correlation coefficient between current raw material information and historical raw material information, and the mean square error between historical defect characteristic values and preset defect characteristic standard values. Screening the historical production information dataset according to the preset correlation threshold and the said correlation can eliminate the historical production information of batch building glass with a large difference between historical raw material information and current raw material information, historical performance information and target performance information, and large overall defects, that is, eliminating and screening the historical production information of batch building glass that does not meet the requirements, so as to ensure the value of the historical production information dataset, improve its quality and reliability, and thus improve the training effect and accuracy of the neural network model through the historical production information dataset.

[0106] S4. Obtain the basic information of the building glass to be produced, input the basic information into the neural network model to obtain the optimal process control parameters, and produce the building glass according to the optimal process control parameters.

[0107] Here, the basic information includes current raw material information, target performance information, and environmental factor information of the production process. For the neural network model, process control parameters are used as the output, and raw material information and environmental factor information are used as the input. A loss function is constructed based on defect information, performance information, environmental factor information, and energy consumption information. Through iterative training, the loss function is minimized to obtain the optimal process control parameters.

[0108] In the production of architectural glass in the prior art, when a neural network based on deep learning is used to obtain optimal process parameters, the loss function often only considers one of the defects, performance, or energy consumption losses in glass production. This method optimizes for a single production target and cannot comprehensively improve the comprehensive production efficiency of architectural glass production.

[0109] Compared with the prior art, the architectural glass production method of the present invention constructs a multi-objective dynamic loss function based on defect information, energy consumption information, performance information, and environmental factor information. It not only overcomes the problem that the prior art architectural glass production method cannot comprehensively improve the comprehensive production efficiency of architectural glass production due to optimizing for a single production target, but also is beneficial to improving the comprehensive production efficiency of architectural glass according to actual production needs. By incorporating environmental factor information into the loss function, it can also take into account the possible impacts of environmental factor information in the production process on the quality and efficiency of architectural glass such as defects, energy consumption, and performance, ensuring the robustness and accuracy of the neural network model.

[0110] In summary, the architectural glass production method trains the neural network model by constructing a neural network model and obtaining a historical production information data set of multiple different batches of architectural glass of the same type, taking into account multiple factors that affect the production efficiency of architectural glass, which can improve the robustness of the neural network model, thereby obtaining the optimal process control parameters that match the current raw material information and target performance information of the architectural glass to be produced, and improving the degree of intelligence and production efficiency.

[0111] In addition, by screening the historical production information data set according to the preset correlation threshold and the correlation degree, the historical production information of batches of architectural glass with a large difference between historical raw material information and current raw material information, historical performance information and target performance information, and a large overall defect can be excluded, that is, the historical production information of batches of architectural glass that do not meet the requirements is excluded and screened to ensure the value of the historical production information data set, improve its quality and reliability, and thus improve the training effect and accuracy of the neural network model through the historical production information data set.

[0112] As a preferred technical solution, in step S2, the multi-objective dynamic loss function L = L 1 +L 2 +L3 +β×||W defect W energy || 2 ;

[0113] Wherein, L 1 =α defect ||w defect || 2 +α energy ||w energy || 2 represents a multi-objective regularization term, α defect , α energy respectively represent the defect loss weight coefficient and the energy consumption loss weight coefficient, w defect , w energy respectively represent the defect information feature vector and the energy consumption loss feature vector, |||| represents the norm, L 2 =Relu(f(env)-threshold) represents an environment-sensitive penalty term, represents the comprehensive environmental sensitivity evaluation function, Relu() represents the activation function, threshold represents the environmental factor safety threshold, M1 represents the number of types of environmental factors, λ i represents the evaluation value weight of the i-th type of environmental factor, which can be set by technicians, e represents the natural constant, represents an intermediate variable, E i represents the actual value of the i-th environmental factor, E i ' represents the standard value of the i-th environmental factor, || represents the absolute value function, represents the performance loss term, α performance represents the performance loss weight coefficient, M2 represents the number of types of performance information, p i represents the actual value of the i-th type of performance information, p i ' represents the predicted value of the i-th type of performance information, β×||W defect W energy || 2 represents the weight regularization term, β represents the regularization strength, which can be set by technicians, and the initial value is between 0.01 and 0.1, W defect represents the defect loss weight matrix, W energy represents the energy consumption loss weight matrix.

[0114] The weight regularization term β×||W defect W energy || 2 is mainly used to impose the Frobenius norm constraint on the defect loss weight matrix and the energy consumption loss weight matrix to prevent overfitting during the training process of the neural network model.

[0115] The environment-sensitive penalty term L2 = Relu(f(env) - threshold) integrates the eigenvalue features of environmental factors in multiple dimensions, creates a non-linear threshold response mechanism by setting the activation function, and triggers gradient penalty when the comprehensive environment deteriorates and exceeds the safety threshold of environmental factors, increasing the constraint strength. One of the main functions of setting the environment-sensitive penalty term is to construct a dynamically updated loss function according to the changes in the actual values of environmental factors and increase the model constraint strength, enabling the neural network model to adapt to the diversity of input sample data. Specifically,

[0116] For the energy consumption loss weight coefficient α energy and the defect loss weight coefficient α defect as well as the performance loss weight coefficient can all be set by technicians according to experience.

[0117] In data analysis and machine learning, a feature vector is usually a vector composed of multiple features. Each feature may have different measurement units or numerical ranges, which can lead to uneven weight distribution among features during the model training process, affecting the performance and accuracy of the model. Feature vector normalization eliminates the dimensionality impact between features by mapping the numerical range of each feature to the same interval, making it easier for the model to capture the relationships and patterns between features. Therefore, before calculating the loss function according to the formula L = L 1 + L 2 + β × ||W defect W energy || 2 preferably, the defect information feature vector and the energy consumption loss feature vector can also be normalized respectively by the method of normalization processing to unify the numerical ranges between different features, avoiding problems such as unstable model training or slow convergence speed caused by excessive differences between features.

[0118] Similarly, when training the neural network model, the sample data feature vectors in the historical production information dataset can also be normalized to improve the training speed and stability of the model, and improve the generalization ability and accuracy of the model. Through feature vector normalization, the impact of dimensional differences between features on the model can be effectively avoided, making it easier for the model to capture the laws and patterns in the data.

[0119] When performing feature vector normalization, the most commonly used methods include min-max normalization and Z-score standardization. Min-max normalization maps the numerical range of each feature to between [0, 1], retaining the distribution information of the original data; while Z-score standardization maps the data to a normal distribution with a mean of 0 and a standard deviation of 1, which is applicable to the case where the data conforms to a normal distribution.

[0120] When inputting the basic information into the neural network model to obtain the optimal process control parameters, for the performance loss term in the loss function Preferably, p i ' represents the predicted value of the performance information of the i-th type, and p i represents the actual value of the performance information of the i-th type, which can be replaced by the characteristic information value under the target performance parameters of the i-th type.

[0121] By constructing the multi-objective dynamic loss function and incorporating environmental factor information into the loss function, the possible impacts of environmental factor information in the production process on the quality benefits of building glass such as defects, energy consumption, and performance can be taken into account, ensuring the robustness and accuracy of the neural network model. At the same time, it also overcomes the problem that the existing production methods of building glass cannot comprehensively improve the comprehensive production benefits of building glass production due to optimizing for a single production target, and is conducive to improving the comprehensive production benefits of building glass according to actual production requirements. Based on the constructed multi-objective dynamic loss function and neural network model, by adjusting the defect loss weight coefficient, energy consumption loss weight coefficient, and performance loss weight coefficient, the optimal process parameters that meet the actual comprehensive production benefits required by building glass manufacturers can be obtained based on the multi-objective dynamic loss function, and the overall intelligent level of the system is also improved.

[0122] An embodiment of the present invention further provides a building glass production system for implementing the above-mentioned building glass production method, as Figure 4 shown, including a production information acquisition module, a neural network construction module, a training module, and a process parameter acquisition module.

[0123] The production information acquisition module is used to acquire the type of building glass to be produced and obtain a historical production information data set of multiple different batches of building glass of the same type; the neural network construction module is used to construct a multi-objective dynamic loss function according to defect information, energy consumption information, performance information, and environmental factor information, and construct a neural network model based on the loss function;

[0124] The training module is used to train the neural network model according to the historical production information data set; the process parameter acquisition module is used to acquire the basic information of the building glass to be produced, input the basic information into the neural network model to obtain the optimal process control parameters, and produce the building glass according to the optimal process control parameters;

[0125] Among them, the historical production information data set includes historical raw material information, historical process control parameters, historical defect information, historical performance information, historical energy consumption information, and historical environmental factor information, and the basic information includes current raw material information, target performance information, and environmental factor information during the production process.

[0126] For the process control parameters, specifically, they can be the relevant parameters of several annealing stages in the glass production process, such as the temperature ranges, heating rates, and heating times of primary annealing and secondary annealing in the heating stage, the temperature range and holding time in the holding stage, the cooling rate and temperature range in the slow cooling stage, and the cooling rate and cooling method in the rapid cooling stage. Correspondingly, the performance information includes but is not limited to residual stress parameters, structural relaxation degree, grain size, grain boundary clarity, elastic modulus, coefficient of thermal expansion, light transmittance, and refractive index.

[0127] Preferably, the training module includes a correlation degree acquisition unit, a screening unit, and a training unit.

[0128] The correlation degree acquisition unit is used to respectively acquire the correlation degrees between multiple different batches of architectural glass of the same type and the architectural glass to be produced; the screening unit is used to screen the historical production information data set according to a preset correlation degree threshold and the correlation degrees; the training unit is used to train the neural network model according to the screened historical production information data set.

[0129] Specifically, the correlation degree acquisition unit includes a first sub-unit, a second sub-unit, and a third sub-unit.

[0130] The first sub-unit is used to obtain a raw material correlation coefficient R' by calculating the weighted value of the Pearson correlation coefficients between the current raw materials of multiple types and the historical raw material characteristic information values; the second sub-unit is used to obtain a performance parameter correlation coefficient R'' by calculating the weighted value of the Pearson correlation coefficients between the target performance parameters of multiple types and the historical performance parameter characteristic information values; the third sub-unit is used to obtain a defect correlation coefficient R''' by calculating the weighted value of the mean square errors between the historical defect characteristic values of multiple types of historical defect information and the corresponding defect characteristic standard values.

[0131] After calculating the correlation degree R = R' + R'' + 1 / R''', the historical production information of the batch of architectural glass with a correlation degree less than the preset correlation degree threshold is removed and screened, and the historical production information of multiple batches of architectural glass after the removal and screening process is preprocessed to generate a historical production information data set.

[0132] The preprocessing method includes but is not limited to data cleaning, partitioning, normalization, and enhancement. The partitioning of data refers to dividing the data into a training set, a validation set, and a test set. Since data cleaning, partitioning, normalization, and enhancement belong to conventional technical means in the art, they will not be elaborated here.

[0133] Screening the historical production information dataset according to the preset correlation threshold and the correlation degree can eliminate the historical production information of batch building glass with a large difference between historical raw material information and current raw material information, historical performance information and target performance information, and a large overall defect, that is, eliminating and screening the historical production information of batch building glass that does not meet the requirements, so as to ensure the value of the historical production information dataset, improve its quality and reliability, and thus improve the training effect and accuracy of the neural network model through the historical production information dataset.

[0134] Regarding the environmental factor information in the production process, preferably, it can be obtained by prediction. The specific steps include: first, extracting the historical environmental factor information from the historical production information dataset generated after preprocessing the historical production information of multiple batches of building glass after elimination and screening, and performing fitting processing according to the historical environmental information of each type to obtain the fitting curve function of the historical environmental factor information. This fitting curve function is the curve function of the characteristic value of the historical environmental factor information with respect to time. Then, according to the fitting curve function corresponding to the historical environmental factor information of each type, obtaining the predicted value of the historical environmental factor information and using it as the environmental factor information in the production process.

[0135] Of course, in order to simplify the calculation, the current environmental factor information can also be directly used as the environmental factor information in the production process of the building glass to be produced. More specifically, for example, if the production process of the building glass specifically targets the melting or forming and annealing stage, the environmental factor information at a certain moment before the start of melting or forming and annealing can be used as the environmental factor information in the production process of the building glass to be produced.

[0136] Preferably, the neural network construction module includes a loss function construction unit and a neural network model construction unit.

[0137] The loss function construction unit is used to construct a multi-objective dynamic loss function \(L = L\) 1 + \(L\) 2 + \(\beta\times||W\) defect \(W\) energy || 2 ; The neural network model construction unit is used to construct a neural network model based on the multi-objective dynamic loss function;

[0138] Among them, \(L\) 1 = \(\alpha\) defect ||w defect || 2 + \(\alpha\) energy ||w energy || 2 represents the multi-objective regularization term, \(\alpha\) defect 、\(\alpha\) energyrespectively represent the defect loss weight coefficient and the energy consumption loss weight coefficient, w defect , w energy respectively represent the defect information feature vector and the energy consumption loss feature vector, |||| represents the norm, L 2 = Relu(f(env)-threshold) represents the environmental sensitivity penalty term, represents the comprehensive environmental sensitivity evaluation function, Relu() represents the activation function, threshold represents the environmental factor safety threshold, M1 represents the number of types of environmental factors, λ i represents the evaluation value weight of the i-th type of environmental factor, e represents the natural constant, represents the intermediate variable, E i represents the actual value of the i-th environmental factor, E i ' represents the standard value of the i-th environmental factor, || represents the absolute value function, represents the performance loss term, α performance represents the performance loss weight coefficient, M2 represents the number of types of performance information, p i represents the actual value of the i-th type of performance information, p i ' represents the predicted value of the i-th type of performance information β×||W defect W energy || 2 represents the weight regularization term, β represents the regularization strength, the initial value is between 0.01 - 0.1, W defect represents the defect loss weight matrix, W energy represents the energy consumption loss weight matrix.

[0139] The weight regularization term β×||W defect W energy || 2 is mainly used to impose the Frobenius norm constraint on the defect loss weight matrix and the energy consumption loss weight matrix, preventing overfitting in the training process of the neural network model.

[0140] The environmental sensitivity penalty term L 2 = Relu(f(env)-threshold) integrates the eigenvalue of environmental factors in multiple dimensions, creates a non-linear threshold response mechanism by setting the activation function, and triggers gradient penalty when the comprehensive environment deteriorates and exceeds the environmental factor safety threshold, increasing the constraint strength. One of the main functions of setting the environmental sensitivity penalty term is that it can construct a dynamically updated loss function according to the change of the actual value of environmental factors and increase the model constraint strength, enabling the neural network model to adapt to the diversity of input sample data. Specifically,

[0141] For the energy consumption loss weight coefficient α energy, defect loss weight coefficient α defect and performance loss weight coefficient β performance can both be set by technicians according to experience. Preferably, α energy : α defect : α performance = 0.1: 0.4: 0.5 or α energy : α defect : α performance = 0.1: 0.6: 0.3. The initial value of the regularization strength is between 0.01 - 0.1 and the regularization strength gradually decays with the number of training epochs. The regularization strength where μ 0 represents the initial value of the regularization strength, and Epoch represents the number of training epochs.

[0142] Preferably, where μ 0 represents the initial value of the regularization strength, Epoch represents the number of training epochs, and α' represents the regularization strength adjustment coefficient. Both α' and μ 0 can be set by technicians according to experience. Through the formula the regularization strength can be made to gradually decay with the number of training epochs, and when it is fused with the comprehensive environmental sensitivity evaluation function, the regularization strength can be dynamically adjusted according to environmental changes; when the comprehensive environment is harsh, by weakening the regularization strength, the underfitting of model training caused by the influence of the harsh environment can be avoided, and the accuracy and generalization ability of the model can be improved. Generally speaking, by using the formula to adjust the regularization strength, which fuses the number of training epochs and the comprehensive environmental sensitivity, the robustness and generalization ability of the neural network model can be improved, and the probabilities of underfitting and overfitting occurring during the training process can be reduced. On this basis, through the trained neural network model, the optimal process parameters of the comprehensive production benefit that meet the actual requirements of the building glass manufacturer can be obtained based on the multi-objective dynamic loss function, and the overall intelligent level of the system and the overall production benefit of building glass production are also improved.

[0143] In summary, by constructing the multi-objective dynamic loss function and incorporating environmental factor information into the loss function, the building glass production system can take into account the possible impacts of environmental factor information in the production process on the quality and benefits of building glass, such as defects, energy consumption, and performance, ensuring the robustness and accuracy of the neural network model. At the same time, it also overcomes the problem that the existing building glass production methods cannot comprehensively improve the comprehensive production benefits of building glass production due to optimizing for a single production target, which is beneficial to improving the comprehensive production benefits of building glass according to actual production requirements. Based on the constructed multi-objective dynamic loss function and neural network model, by adjusting the defect loss weight coefficient, energy consumption loss weight coefficient, and performance loss weight coefficient, the optimal process parameters that meet the actual requirements of building glass manufacturers for comprehensive production benefits can be obtained based on the multi-objective dynamic loss function, and the overall intelligence level of the system is also improved.

[0144] An embodiment of the present invention further provides a building glass production device, which includes: a controller; a memory storing executable instructions; wherein, the executable instructions can run on the controller and implement the described building glass production method.

[0145] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0146] The above-described embodiments merely represent several implementation manners of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent should be subject to the appended claims.

Claims

1. A method for producing architectural glass, characterized in that: The architectural glass production method comprises the following steps: Obtain the type of architectural glass to be produced and obtain a historical production information dataset of multiple different batches of architectural glass of the same type; Construct a multi-objective dynamic loss function based on defect information, energy consumption information, performance information and environmental factor information, and build a neural network model based on the loss function; Training the neural network model according to the historical production information data set; Obtaining basic information of the architectural glass to be produced, inputting the basic information into the neural network model to obtain optimal process control parameters, and producing the architectural glass according to the optimal process control parameters; Among them, the historical production information data set includes historical raw material information, historical process control parameters, historical defect information, historical performance information, historical energy consumption information and historical environmental factor information, and the basic information includes current raw material information, target performance information and environmental factor information of the production process.

2. A method for producing architectural glass according to claim 1, characterized in that: The specific method for training the neural network model according to the historical production information data set comprises the following steps: Respectively obtaining correlations between a plurality of different batches of architectural glass of the same type and the architectural glass to be produced; Screening the historical production information data set according to a preset correlation threshold and the correlation; The neural network model is trained based on the screened historical production information data set.

3. A method for producing architectural glass according to claim 2, characterized in that: The multi-objective dynamic loss function L = L1 + L2 + L3 + β × || W defect W energy || 2 ; Where L1 = α defect ||w defect || 2 +α energy ||w energy || 2 represents the multi-objective regularization term, α defect , α energy They represent the defect loss weight coefficient and energy loss weight coefficient respectively, w defect、 w energy represent the defect information feature vector and energy loss feature vector respectively, || || represents the norm, L2=Relu(f(env)-threshold) represents the environmental sensitivity penalty term, represents the comprehensive environmental sensitivity evaluation function, Relu() represents the activation function, threshold represents the environmental factor safety threshold, M1 represents the number of environmental factors, and λ i represents the evaluation value weight of the i-th type of environmental factor, e represents the natural constant, represents the intermediate variable, E i represents the actual value of the i-th environmental factor, E i ' represents the standard value of the i-th environmental factor, || represents the absolute value function, represents the performance loss term, α performance represents the performance loss weight coefficient, M2 represents the number of types of performance information, and p i represents the actual value of the performance information of the ith category, p i ' represents the predicted value of the performance information of the i-th category β×||W defect W energy || 2 represents the weight regularization term, β represents the regularization strength, W defect represents the defect loss weight matrix, W energy Represents the energy loss weight matrix.

4. A method for producing architectural glass according to claim 3, characterized in that: Correlation R = R' + R" + 1 / R"', R', R", R"' respectively represent the raw material correlation coefficient, performance parameter correlation coefficient and defect correlation coefficient; R' is obtained by calculating the weighted value of the Pearson correlation coefficient between the current raw materials and the historical raw material characteristic information values ​​of multiple types; R” is obtained by calculating the weighted value of the Pearson correlation coefficient between multiple types of target performance parameters and historical performance parameter characteristic information values; R'' is obtained by calculating the weighted value of the mean square error between the historical defect feature values ​​of multiple types of historical defect information and the corresponding defect feature standard values.

5. A method for producing architectural glass according to claim 4, characterized in that: The specific method for screening the historical production information data set includes: removing and screening the historical production information of batches of building glass with a correlation less than a preset correlation threshold; The specific method for training the neural network model according to the screened historical production information data set includes: preprocessing the historical production information of multiple batches of architectural glass after screening to generate a historical production information data set, and training the neural network model according to the historical production information data set to obtain the trained neural network model.

6. A building glass production system, used to implement the building glass production method according to any one of claims 1 to 5, characterized in that: The architectural glass production system comprises: A production information acquisition module, used to acquire the type of architectural glass to be produced and to acquire historical production information data sets of multiple different batches of architectural glass of the same type; A neural network building module is used to build a multi-objective dynamic loss function based on defect information, energy consumption information, performance information and environmental factor information, and to build a neural network model based on the loss function; A training module, used for training the neural network model according to the historical production information data set; A process parameter acquisition module, used to acquire basic information of the architectural glass to be produced, input the basic information into the neural network model to obtain optimal process control parameters, and produce the architectural glass according to the optimal process control parameters; Among them, the historical production information data set includes historical raw material information, historical process control parameters, historical defect information, historical performance information, historical energy consumption information and historical environmental factor information, and the basic information includes current raw material information, target performance information and environmental factor information of the production process.

7. A building glass production system as claimed in claim 6, characterized in that: The training module includes: A correlation acquisition unit, used to respectively acquire correlations between a plurality of different batches of architectural glass of the same type and the architectural glass to be produced; A screening unit, used to screen the historical production information data set according to a preset correlation threshold and the correlation; A training unit is used to train the neural network model according to the screened historical production information data set.

8. A building glass production system as claimed in claim 7, characterized in that: Neural network building blocks include: The loss function construction unit is used to construct a multi-objective dynamic loss function L = L1 + L2 + L3 + β × || W based on defect information, energy consumption information and environmental factor information defect W energy || 2 ; A neural network model building unit, used to build a neural network model based on the multi-objective dynamic loss function; Where L1 = α defect ||w defect || 2 +α energy ||w energy || 2 represents the multi-objective regularization term, α defect , α energy They represent the defect loss weight coefficient and energy loss weight coefficient respectively, w defect、 w energy represent the defect information feature vector and energy loss feature vector respectively, || || represents the norm, L2=Relu(f(env)-threshold) represents the environmental sensitivity penalty term, represents the comprehensive environmental sensitivity evaluation function, Relu() represents the activation function, threshold represents the environmental factor safety threshold, M1 represents the number of environmental factors, and λ i represents the evaluation value weight of the i-th type of environmental factor, e represents the natural constant, represents the intermediate variable, E i represents the actual value of the i-th environmental factor, E i ' represents the standard value of the i-th environmental factor, || represents the absolute value function, represents the performance loss term, α performance represents the performance loss weight coefficient, M2 represents the number of types of performance information, and p i represents the actual value of the performance information of the ith category, p i ' represents the predicted value of the performance information of the i-th category β×||W defect W energy || 2 represents the weight regularization term, β represents the regularization strength, W defect represents the defect loss weight matrix, W energy Represents the energy loss weight matrix.

9. A building glass production system as claimed in claim 8, characterized in that: The relevance acquisition unit includes: A first subunit is used to obtain a raw material correlation coefficient R' by calculating a weighted value of a Pearson correlation coefficient between multiple types of current raw materials and historical raw material characteristic information values; The second subunit is used to obtain a performance parameter correlation coefficient R" by calculating a weighted value of a Pearson correlation coefficient between multiple types of target performance parameters and historical performance parameter characteristic information values; The third subunit is used to obtain a defect correlation coefficient R'' by calculating a weighted value of a mean square error between historical defect feature values ​​of multiple types of historical defect information and corresponding defect feature standard values; Among them, the correlation degree R=R'+R"+1 / R"'.

10. A building glass production equipment, characterized in that: The architectural glass production equipment comprises: Controller; A memory storing executable instructions; The executable instructions can be run on the controller and implement the method for producing architectural glass as described in any one of claims 1 to 5.

Images