Intelligent control system and method in the production process of tempered glass
Through the intelligent control system, combined with the quality data and monitoring video of tempered glass, the parameters of the homogenized furnace are adjusted, and the self-destruction problem caused by differences in the homogenization treatment of tempered glass is solved, and the stability of the homogenized treatment and equipment life are improved.
Patent Information
- Application Number
- CN202410061650.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-16
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-01-16
AI Technical Summary
During the homogenization process of tempered glass, the differences between each piece of glass are ignored, resulting in self-destruction phenomenon interfering with the parameters of the homogenization furnace and reducing the reliability and life of the equipment.
The intelligent control system is adopted to collect the quality data and monitoring video of tempered glass, combine the homogenized furnace parameter data, perform multi-scale feature extraction and feature analysis, and adjust the homogenized furnace parameters to ensure the quality stability of the glass homogenization treatment.
Effectively reduce the impact of homogenized furnace on the quality of tempered glass, extend the service life of the equipment, and improve the stability and reliability of the homogenized treatment process.
Smart Images

Figure CN117872992B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent control of tempered glass production, and more specifically, to an intelligent control system and method in the tempered glass production process. Background Art
[0002] Tempered glass is a specially treated glass, also known as safety glass or strengthened glass. It has its physical properties enhanced through a heat treatment process, making it stronger, more impact-resistant and safer than ordinary glass. The process of tempered glass production mainly includes: cutting and processing, cleaning and testing, preheating and heating, rapid cooling or quenching, and homogenization. Among them, the homogenization of tempered glass is to eliminate the residual nickel sulfide inside the tempered glass to prevent it from self-explosion.
[0003] However, during the homogenization process of tempered glass, the differences between each piece of tempered glass are often ignored. Moreover, the self-explosion of tempered glass may interfere with the parameter performance of the homogenization furnace, which will shorten the service life of the homogenization furnace and affect the homogenization of the next piece of tempered glass. The self-explosion event will cause sudden temperature changes and stress release inside the homogenization furnace, which may cause fatigue damage to the furnace body and heating elements, reducing the reliability and life of the equipment.
[0004] Therefore, an intelligent control system and method for the tempered glass production process is desired, which adjusts the parameters of the homogenizing furnace based on the quality data and surface state of the tempered glass to be homogenized, thereby ensuring the quality of the tempered glass homogenization process. Summary of the invention
[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an intelligent control system and method in the production process of tempered glass, which first collects the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized are thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled include: temperature, time, heating power, and cooling rate; then, multi-scale feature extraction is performed on the quality data, the surface features of the tempered glass in the monitoring video are extracted, and feature analysis is performed on the parameter data, and then the classification results are obtained in combination with the extracted feature information. In this way, the influence of the homogenizing furnace on the quality of the tempered glass can be effectively reduced, thereby ensuring the stability of the quality during the homogenization process of the tempered glass.
[0006] According to a first aspect of the present application, an intelligent control system for a tempered glass production process is provided, comprising:
[0007] The tempered glass production data acquisition module is used to collect the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate;
[0008] A production data feature extraction module, used to extract feature information of quality data of the tempered glass to be homogenized, feature information of monitoring video of the tempered glass to be homogenized, and feature information of parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality feature vector, a tempered glass surface feature vector, and a homogenizing furnace parameter feature vector;
[0009] The homogenizing furnace classification result generating module is used to obtain the classification result based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector.
[0010] In the above-mentioned intelligent control system in the tempered glass production process, the intelligent control system in the tempered glass production process also includes a training module for training the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier;
[0011] Wherein, the training module includes:
[0012] A training data acquisition unit, used to acquire training data, wherein the training data includes quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes temperature, time, heating power, and cooling rate;
[0013] A tempered glass quality feature training unit, used for arranging the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension and then passing it through a multi-scale tempered glass quality feature extraction module to obtain a training tempered glass quality feature vector;
[0014] A video denoising unit, used for passing the surveillance video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a surveillance video after training and denoising;
[0015] A key frame acquisition unit, used to acquire a plurality of training tempered glass key frames from the training noise-reduced monitoring video;
[0016] A tempered glass surface feature training unit, used for arranging the plurality of training tempered glass key frames into a three-dimensional tempered glass input tensor according to the time dimension and then passing the tensor through a tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a training tempered glass surface feature map;
[0017] A feature map dimensionality reduction unit, used for performing global mean pooling processing along the channel dimension on each feature matrix in the training tempered glass surface feature map to obtain a training tempered glass surface feature vector;
[0018] A tempered glass training feature fusion unit, used for fusing the training tempered glass surface feature vector and the training tempered glass quality feature vector to obtain a tempered glass feature vector;
[0019] A homogenizing furnace parameter feature training unit, used for arranging the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector and then passing it through a parameter feature extraction module based on a sequence encoder to obtain a training homogenizing furnace parameter feature vector;
[0020] A homogenization process training feature fusion unit, used for performing feature fusion on the training homogenization furnace parameter feature vector and the training tempered glass feature vector to obtain a training homogenization furnace classification feature vector;
[0021] A metric factor calculation unit, used to calculate a topological dimension information metric factor of a matching node between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector;
[0022] A classification loss unit, used for passing the training homogenizing furnace classification feature vector through a classifier to obtain a classification loss function value;
[0023] A training unit is used to train the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier by taking the weighted sum of the topological dimension information measurement factor between the matching nodes and the classification loss function value as the loss function value.
[0024] According to a second aspect of the present application, an intelligent control method in a tempered glass production process is provided, comprising:
[0025] Collecting quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate;
[0026] Extracting characteristic information of quality data of the tempered glass to be homogenized, characteristic information of monitoring video of the tempered glass to be homogenized, and characteristic information of parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality characteristic vector, a tempered glass surface characteristic vector, and a homogenizing furnace parameter characteristic vector;
[0027] Based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector, a classification result is obtained.
[0028] In combination with the first aspect of the present application, in an intelligent control system in a tempered glass production process of the first aspect of the present application, the production data feature extraction module includes: a tempered glass quality feature acquisition unit, used to acquire multi-scale feature information of the quality data of the tempered glass to be homogenized to obtain the tempered glass quality feature vector; a tempered glass surface feature acquisition unit, used to perform feature extraction on the monitoring video of the tempered glass to be homogenized to obtain the tempered glass surface feature vector; a homogenizing furnace parameter feature acquisition unit, used to perform feature extraction on the parameter data of the homogenizing furnace to be controlled to obtain the homogenizing furnace parameter feature vector.
[0029] In combination with the first aspect of the present application, in an intelligent control system in a tempered glass production process of the first aspect of the present application, the metric factor calculation unit is used to: calculate the topological dimension information metric factor of the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector according to the following formula;
[0030]
[0031] in, represents the training homogenizing furnace parameter feature vector, represents the training tempered glass feature vector, represents the transposed vector of the training homogenizing furnace parameter feature vector, and is a column vector, and They are respectively the training homogenizing furnace parameter feature vectors And train the tempered glass feature vector No. The eigenvalues at the positions, and is the length of the eigenvector, Indicates the information metric factor of the topological dimension between the matching nodes.
[0032] In combination with the second aspect of the present application, in an intelligent control method in a tempered glass production process of the second aspect of the present application, based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector, a classification result is obtained, including: a tempered glass feature fusion unit, used to fuse the tempered glass quality feature vector and the tempered glass surface feature vector to obtain a tempered glass feature vector; a homogenizing process feature fusion unit, used to perform feature fusion on the tempered glass feature vector and the homogenizing furnace parameter feature vector to obtain a homogenizing furnace classification feature vector; a classification unit, used to pass the homogenizing furnace classification feature vector through a classifier to obtain a classification result. Wherein, the classification result is used to indicate whether the parameter data of the homogenizing furnace to be controlled needs to be adjusted.
[0033] Compared with the prior art, the present application provides an intelligent control system and method in the production process of tempered glass, which first collects the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized are thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled include: temperature, time, heating power, and cooling rate; then, multi-scale feature extraction is performed on the quality data, the surface features of the tempered glass in the monitoring video are extracted, and feature analysis is performed on the parameter data, and then the classification results are obtained by combining the extracted feature information. In this way, the influence of the homogenizing furnace on the quality of the tempered glass can be effectively reduced, thereby ensuring the stability of the quality during the homogenization process of the tempered glass. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.
[0035] Figure 1 The figure shows a schematic block diagram of an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0036] Figure 2 The figure illustrates a schematic block diagram of a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0037] Figure 3 The figure illustrates a schematic block diagram of a tempered glass quality feature acquisition unit in a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0038] Figure 4 The figure illustrates a schematic block diagram of a tempered glass surface feature acquisition unit in a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0039] Figure 5 The figure illustrates a schematic block diagram of a homogenizing furnace parameter feature acquisition unit in a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0040] Figure 6 The figure illustrates a schematic block diagram of a homogenizing furnace classification result generation module in an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0041] Figure 7 The diagram shows an architecture diagram of an intelligent control system in a tempered glass production process according to an embodiment of the present application.
[0042] Figure 8 The figure illustrates a flow chart of an intelligent control method in a tempered glass production process according to an embodiment of the present application. DETAILED DESCRIPTION
[0043] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.
[0044] Exemplary Systems
[0045] Figure 1 The figure shows a schematic block diagram of an intelligent control system in a tempered glass production process according to an embodiment of the present application. Figure 1As shown, according to the intelligent control system 100 in the tempered glass production process of the embodiment of the present application, it includes: a tempered glass production data acquisition module 110, which is used to collect quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized is thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate; a production data feature extraction module 120, which is used to extract feature information of the quality data of the tempered glass to be homogenized, feature information of the monitoring video of the tempered glass to be homogenized, and feature information of the parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality feature vector, a tempered glass surface feature vector, and a homogenizing furnace parameter feature vector; a homogenizing furnace classification result generation module 130, which is used to obtain a classification result based on the tempered glass quality feature vector, the tempered glass surface feature vector, and the homogenizing furnace parameter feature vector.
[0046] Tempered glass is a safety glass with high strength and impact resistance. It forms compressive stress on the surface of the glass, making it better able to resist cracking and shattering when subjected to stress. However, after tempering, the crystal structure of the nickel sulfide grains inside the tempered glass will change over time. During the transformation, the volume of its lattice will increase, and internal stress will be generated around it. When the internal stress accumulates to a certain level, it may cause the tempered glass to explode. In order to prevent the tempered glass from exploding during use, the tempered glass needs to be homogenized in a homogenizing furnace, and the detonation test and the elimination of residual nickel sulfide are carried out through the principle of hot dipping in the homogenizing furnace.
[0047] However, as described in the above background technology, in the homogenization process of tempered glass, the differences between each piece of tempered glass are often ignored, and the self-explosion phenomenon of tempered glass may interfere with the parameter performance of the homogenization furnace, which will shorten the service life of the homogenization furnace and affect the homogenization of the next piece of tempered glass. The self-explosion event will cause sudden temperature changes and stress release inside the homogenization furnace, which may cause fatigue damage to the furnace body and heating elements, and reduce the reliability and life of the equipment. Therefore, an intelligent control system and method in the tempered glass production process is desired, which adjusts the parameters of the homogenization furnace based on the quality data and surface state of the tempered glass to be homogenized, thereby ensuring the quality of the tempered glass homogenization.
[0048] In particular, computer vision is a discipline that studies how to enable computers to understand and interpret image and video data. It involves techniques and methods for extracting useful information from images or videos, analyzing and understanding image content. The goal of computer vision is to enable computers to simulate the functions of the human visual system, including image recognition, target detection, image segmentation, posture estimation, motion tracking, etc. Through computer vision, computers can perceive and understand objects, scenes, and actions in images, thereby achieving automated and intelligent visual tasks. Therefore, in the embodiment of the present application, the surface state of tempered glass during the homogenization process is monitored and analyzed by computer vision technology.
[0049] In the embodiment of the present application, the tempered glass production data acquisition module 110 is used to collect the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized is thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate. It should be understood that the production process of tempered glass mainly includes: cutting, processing, cleaning, preheating, homogenization, and cooling. Homogenization is a key step in the production process of tempered glass. By controlling the temperature and cooling rate, the glass surface is cooled rapidly, while the interior is still at a high temperature, thereby giving the tempered glass higher strength, impact resistance, and thermal stability. However, considering that the quality of tempered glass after different tempering treatments has certain differences, these tempered glasses may require different homogenizing furnace parameters during the homogenization process. Therefore, in order to better control the parameters of the homogenizing furnace, the quality data of the tempered glass to be homogenized is first collected. In addition, considering that the surface state of the tempered glass may change during the homogenization process, these surface states can reflect from the side whether the parameters of the homogenization furnace are suitable for the tempered glass in the homogenization process. Therefore, the homogenization state information of the tempered glass during the homogenization process can also be obtained through the monitoring video. Furthermore, real-time monitoring of the parameter information of the homogenization furnace can timely detect the abnormality of the homogenization parameters and make timely adjustments to ensure the quality of the homogenization of the tempered glass. Therefore, the parameters of the homogenization furnace are further collected in real time.
[0050] More specifically, when collecting the quality data of the tempered glass to be homogenized, sensors or measuring equipment can be used to measure the quality indicators such as thickness, flatness and impact resistance of the tempered glass after the tempering treatment. For example, devices such as thickness gauges, surface flatness gauges and impact testers can be used for measurement. These devices can convert the measurement results into digital signals or data for subsequent processing and analysis. Among them, these sensors can be directly installed at the appropriate position of the glass processing line, measure by physical or chemical principles, and convert the measurement results into digital signals or data. When collecting the monitoring video of the tempered glass to be homogenized, a camera or monitoring system can be used to shoot the monitoring video of the tempered glass to be homogenized. The camera can be installed near the homogenizing furnace or on the glass processing line to record the state of the glass surface in real time. When collecting the parameter data of the homogenizing furnace to be controlled, it can be collected by sensors or control systems. For example, devices such as temperature sensors, timers, power sensors and flow meters can be used to monitor and record the parameter values of the homogenizing furnace in real time.
[0051] In the embodiment of the present application, the production data feature extraction module 120 is used to extract feature information of quality data of the tempered glass to be homogenized, feature information of the monitoring video of the tempered glass to be homogenized, and feature information of parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality feature vector, a tempered glass surface feature vector, and a homogenizing furnace parameter feature vector. It should be understood that after collecting these data, feature extraction and analysis tasks can then be performed.
[0052] Specifically, Figure 2 The figure shows a schematic block diagram of a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application. Figure 2 As shown, the production data feature extraction module 120 includes: a tempered glass quality feature acquisition unit 121, used to acquire multi-scale feature information of the quality data of the tempered glass to be homogenized to obtain the tempered glass quality feature vector; a tempered glass surface feature acquisition unit 122, used to perform feature extraction on the monitoring video of the tempered glass to be homogenized to obtain the tempered glass surface feature vector; a homogenizing furnace parameter feature acquisition unit 123, used to perform feature extraction on the parameter data of the homogenizing furnace to be controlled to obtain the homogenizing furnace parameter feature vector.
[0053] First, the quality data of the tempered glass to be homogenized is subjected to feature extraction and analysis. Specifically, Figure 3 The figure shows a schematic block diagram of a tempered glass quality feature acquisition unit in a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application. Figure 3As shown, the tempered glass quality feature acquisition unit 121 includes: a quality data arrangement subunit 121-1, used to arrange the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension; a multi-scale feature acquisition subunit 121-2, used to pass the one-dimensional tempered glass quality input vector through a multi-scale tempered glass quality feature extraction module to obtain the tempered glass quality feature vector.
[0054] In particular, considering that different types of quality data are integrated into one vector, the organization and management of data are more convenient. In this way, quality indicators such as thickness, flatness and impact resistance of tempered glass can be represented in the same vector without storing and processing each data type separately. Therefore, the quality data of the tempered glass to be homogenized are arranged according to the data type dimension to obtain the one-dimensional tempered glass quality input vector.
[0055] Next, it should be understood that the quality characteristics of tempered glass may exist at different scales. For example, surface defects may be a problem at the microscopic scale, while overall uniformity may involve a problem at the macroscopic scale. Therefore, in order to capture the quality characteristics of tempered glass at different scales, the one-dimensional tempered glass quality input vector is feature extracted by a multi-scale tempered glass quality feature extraction module. That is, by using a multi-scale feature extraction module, the quality characteristics of tempered glass can be captured from different scales, so that the system can have a more comprehensive understanding of the quality status of tempered glass.
[0056] In a specific embodiment of the present application, the multi-scale feature acquisition subunit 121-2 is used to: input the one-dimensional tempered glass quality input vector into the first convolution layer of the multi-scale tempered glass quality feature extraction module to obtain a first-scale quality feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; input the one-dimensional tempered glass quality input vector into the second convolution layer of the multi-scale tempered glass quality feature extraction module to obtain a second-scale quality feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and use the cascade layer of the multi-scale tempered glass quality feature extraction module to cascade the first-scale quality feature vector and the second-scale quality feature vector to obtain the tempered glass quality feature vector.
[0057] Then, feature extraction and analysis are performed on the monitoring video of the tempered glass to be homogenized. Specifically, Figure 4 The figure shows a schematic block diagram of a tempered glass surface feature acquisition unit in a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application. Figure 4As shown, the tempered glass surface feature acquisition unit 122 includes: a video denoising subunit 122-1, which is used to pass the surveillance video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a denoised surveillance video; a key frame acquisition subunit 122-2, which is used to obtain multiple tempered glass key frames from the denoised surveillance video; a key frame arrangement subunit 122-3, which is used to arrange the multiple tempered glass key frames into a three-dimensional tempered glass input tensor according to the time dimension; a surface change feature acquisition subunit 122-4, which is used to pass the three-dimensional tempered glass input tensor through a tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a tempered glass surface feature map; a feature map dimension reduction subunit 122-5, which is used to perform global mean pooling processing along the channel dimension on each feature matrix in the tempered glass surface feature map to obtain the tempered glass surface feature vector. Wherein, the generative adversarial network includes a discriminator and a generator.
[0058] It should be understood that in actual surveillance videos, various noises may exist, such as image noise, motion blur, illumination changes, etc. These noises will interfere with the accurate analysis and judgment of the quality of tempered glass. Therefore, in order to improve the accuracy of feature extraction, a video denoiser based on an adversarial generative network is used to perform denoising on the surveillance video. In this way, by using a video denoiser based on an adversarial generative network, the noise level in the video can be reduced, the clarity and quality of the image can be improved, and the intelligent control system can more accurately analyze and judge the quality of the tempered glass. Among them, the adversarial generative network includes a discriminator and a generator, and the generator can generate a clearer video by learning the mapping relationship from the noisy video to the denoised video. The generator uses the input noisy video and the feedback signal of the discriminator to gradually optimize the generated video to make it closer to the real clear video. During the training process of the generator, the quality of the generated video is continuously improved by adversarial learning.
[0059] In a specific embodiment of the present application, the video denoising subunit 122-1, the video denoiser based on the adversarial generative network includes a discriminator and a generator, wherein the generator is used to generate a denoised surveillance video, the discriminator is used to calculate the difference between the denoised surveillance video and the real surveillance video, and the network parameters of the generator are updated by the directional propagation algorithm of gradient descent to obtain a generator with denoising function. Furthermore, the surveillance video of the tempered glass to be homogenized is input into the generator of the video denoiser based on the adversarial generative network to obtain the denoised surveillance video.
[0060] Considering that video data usually contains a large number of frames, feature extraction of the entire video may generate a large amount of data, increasing the complexity of data storage and processing. Acquiring key frames can reduce the amount of data and only select key moments in the video for extraction, thereby reducing the burden of data compression and storage. Therefore, multiple tempered glass key frames are obtained from the noise-reduced surveillance video.
[0061] Next, considering that the production process of tempered glass is a continuous process, its surface state changes over time. That is, the multiple tempered glass key frames are time-series. Therefore, in order to retain these time-series information, the multiple tempered glass key frames are arranged according to the time dimension to obtain the three-dimensional tempered glass input tensor. Then, the three-dimensional tempered glass input tensor is subjected to feature extraction through a three-dimensional convolutional neural network model. Among them, by using a three-dimensional convolutional neural network model, not only the time-series change characteristics of tempered glass can be extracted, but also the spatial characteristics of the tempered glass surface can be effectively extracted. The three-dimensional convolutional neural network has good feature extraction capabilities when processing image data, and can capture the texture, shape and other important features of the tempered glass surface.
[0062] In a specific embodiment of the present application, the surface change feature acquisition subunit 122-4 is used to: use each layer of the three-dimensional convolutional neural network model to perform the following on the input data in the forward transfer of the layer: perform convolution processing based on a three-dimensional convolution kernel on the input data to obtain a convolution feature map; perform pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; perform nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the input of the first layer of the three-dimensional convolutional neural network model is the three-dimensional tempered glass input tensor, and the output of the last layer of the three-dimensional convolutional neural network model is the tempered glass surface feature map.
[0063] Since each feature matrix in the tempered glass surface feature map represents the response of a specific surface feature at different positions, the information in each feature matrix can be integrated by performing global mean pooling to obtain a global surface feature representation. In this way, the overall characteristics of the tempered glass surface can be captured, rather than being limited to the characteristics of a specific position or local area. Therefore, each feature matrix in the tempered glass surface feature map is subjected to global mean pooling along the channel dimension to obtain the tempered glass surface feature vector.
[0064] Furthermore, feature extraction is performed on the parameter data of the homogenizing furnace to be controlled. Specifically, Figure 5 The figure shows a schematic block diagram of a homogenizing furnace parameter feature acquisition unit in a production data feature extraction module in an intelligent control system in a tempered glass production process according to an embodiment of the present application. Figure 5 As shown, the homogenizing furnace parameter feature acquisition unit 123 includes: a homogenizing furnace parameter arrangement subunit 123-1, which is used to arrange the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector; a homogenizing furnace parameter feature acquisition subunit 123-2, which is used to pass the one-dimensional homogenizing furnace parameter input vector through a parameter feature extraction module based on a sequence encoder to obtain the homogenizing furnace parameter feature vector. Wherein, the sequence encoder includes a one-dimensional convolution layer and a fully connected layer.
[0065] It should be understood that arranging the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector can uniformly represent the relationship between the parameters, simplify the input form, improve the model compatibility, facilitate parameter adjustment, and reduce the computational complexity. This helps the intelligent control system to adjust the parameters of the homogenizing furnace in real time to ensure the quality of the tempered glass homogenization process. Therefore, the parameter data of the homogenizing furnace to be controlled is first arranged into a one-dimensional homogenizing furnace parameter input vector.
[0066] Then, the one-dimensional homogenizing furnace parameter input vector is feature extracted by a sequence encoder including a one-dimensional convolution layer and a fully connected layer. Considering that the one-dimensional convolution layer can effectively extract local features in the input vector in the sequence encoder. The homogenizing furnace parameters in the production of tempered glass usually have a certain temporal relationship, and the local patterns and features in the parameter sequence can be captured by the one-dimensional convolution layer. In this way, important information about the homogenizing furnace parameters can be extracted, providing a more accurate feature representation for subsequent parameter adjustment. Moreover, the homogenizing furnace parameters in the production of tempered glass usually have a certain local correlation. The one-dimensional convolution layer can capture the local correlation features in the parameter sequence through the convolution operation. In this way, the interaction and influence between the parameters can be better understood, providing a more accurate basis for subsequent parameter adjustment and quality control. The fully connected layer can globally integrate the features output by the one-dimensional convolution layer. Through the operation of the fully connected layer, the local features can be integrated into a global feature vector. In this way, the overall features of the homogenizing furnace parameters can be captured, providing a more comprehensive feature representation for subsequent parameter adjustment and quality evaluation. That is, the parameter feature extraction module based on the sequence encoder, including the one-dimensional convolution layer and the fully connected layer, can extract the important features in the homogenizing furnace parameter input vector, perform dimensional transformation, capture local correlation, and realize the integration of global features. This helps the intelligent control system to extract and represent the features of the homogenizing furnace parameters, and provide more accurate and comprehensive feature representation for the quality control of tempered glass homogenization.
[0067] In a specific embodiment of the present application, the homogenizing furnace parameter feature acquisition subunit 123-2 is used to: use the fully connected layer of the sequence encoder to perform fully connected encoding on the one-dimensional homogenizing furnace parameter input vector using the following formula to extract the high-dimensional implicit features of the eigenvalues at each position in the one-dimensional homogenizing furnace parameter input vector, wherein the formula is: Y=W⊗X+B, wherein X is the one-dimensional homogenizing furnace parameter input vector, Y is the output vector, W is the weight matrix, B is the bias vector, and ⊗ represents matrix multiplication; and use the one-dimensional convolution layer of the sequence encoder to perform one-dimensional convolution encoding on the one-dimensional homogenizing furnace parameter input vector using the following formula to extract the high-dimensional implicit correlation features of the correlation between the eigenvalues at each position in the one-dimensional homogenizing furnace parameter input vector, wherein the formula is:
[0068]
[0069] in, a is the first convolution kernel in x Width in direction, is the first convolution kernel parameter vector, is the local vector matrix that operates with the convolution kernel function, w is the size of the first convolution kernel, X represents the one-dimensional homogenizing furnace parameter input vector, Indicates that one-dimensional convolution encoding is performed on the input vector.
[0070] In the embodiment of the present application, the homogenizing furnace classification result generating module 130 is used to obtain a classification result based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector.
[0071] Specifically, Figure 6 The figure shows a schematic block diagram of a homogenizing furnace classification result generation module in an intelligent control system in a tempered glass production process according to an embodiment of the present application. Figure 6 As shown, the homogenizing furnace classification result generating module 130 includes: a tempered glass feature fusion unit 131, used to fuse the tempered glass quality feature vector and the tempered glass surface feature vector to obtain a tempered glass feature vector; a homogenizing process feature fusion unit 132, used to perform feature fusion on the tempered glass feature vector and the homogenizing furnace parameter feature vector to obtain a homogenizing furnace classification feature vector; and a classification unit 133, used to pass the homogenizing furnace classification feature vector through a classifier to obtain a classification result. The classification result is used to indicate whether the parameter data of the homogenizing furnace to be controlled needs to be adjusted.
[0072] It should be understood that both the quality data and surface state of tempered glass have an important influence on the parameter adjustment of the homogenizing furnace. The quality data includes indicators such as physical properties, dimensional accuracy, and surface flatness of the tempered glass, while the surface state includes information such as defects, pollution, and temperature distribution on the glass surface. Fusion of these two aspects of information can more comprehensively describe the characteristics of tempered glass and provide a more accurate basis for parameter adjustment. Therefore, in order to more comprehensively obtain the characteristic information of tempered glass, the tempered glass quality feature vector and the tempered glass surface feature vector are feature fused.
[0073] In a specific embodiment of the present application, the tempered glass feature fusion unit 131 is used to fuse the tempered glass quality feature vector and the tempered glass surface feature vector using the following cascade formula to obtain a tempered glass feature vector; wherein the cascade formula is: ,in, represents the quality feature vector of the tempered glass, represents the surface feature vector of the tempered glass, represents the tempered glass feature vector, Represents a cascade function.
[0074] Next, considering that the quality and surface state of tempered glass are affected by the parameters of the homogenizing furnace, and the parameter adjustment of the homogenizing furnace is affected by the characteristics and requirements of the tempered glass, by fusing the tempered glass feature vector and the homogenizing furnace parameter feature vector, we can capture the correlation between them, better understand the impact of parameters on the quality of tempered glass, and thus achieve more accurate parameter adjustment.
[0075] In a specific embodiment of the present application, the homogenization process feature fusion unit 132 is used to: fuse the tempered glass feature vector and the homogenization furnace parameter feature vector using the following cascade formula to obtain a homogenization furnace classification feature vector; wherein the cascade formula is: ,in, represents the tempered glass feature vector, represents the homogenizing furnace parameter feature vector, represents the homogenizing furnace classification feature vector, Represents a cascade function.
[0076] Then, consider that there is a complex relationship between the quality data and surface state of tempered glass and the parameters of the homogenizing furnace. By using a classifier, this complex relationship can be modeled and learned using a machine learning algorithm. The classifier can learn the correlation between features from a large amount of training data, so that it can better understand and predict whether the parameters of the homogenizing furnace to be controlled need to be adjusted. Therefore, the homogenizing furnace classification feature vector is feature classified by the classifier. Among them, the classifier can classify the homogenizing furnace classification feature vector into different categories, such as need to be adjusted and not need to be adjusted. In this way, corresponding control strategies can be adopted according to different classification results to meet the quality requirements of tempered glass.
[0077] In a specific embodiment of the present application, the classification result generating unit 133 is used to: use the classifier to process the homogenizing furnace classification feature vector according to the following formula to obtain the classification result; wherein the formula is: ,in, arrive is the weight matrix, arrive is the bias vector, is the homogenizing furnace classification feature vector, represents the normalized exponential function, Indicates the classification result.
[0078] It is worth mentioning that in addition to the above-mentioned use of classifiers, a decision-making system based on a rule engine can also be used. This method uses a rule engine to make decisions and adjust the parameters of the homogenizing furnace. A rule engine is a rule-based system that can make decisions based on a predefined set of rules. In the tempered glass production process, a set of rules can be formulated to determine whether the parameters of the homogenizing furnace need to be adjusted based on the quality data and surface state of the tempered glass to be homogenized.
[0079] The rule engine can make decisions based on the conditions and actions in the rule set. The condition part can include information such as the quality indicators, surface conditions, and environmental conditions of the tempered glass, which is used to describe the current state of the homogenizing furnace to be controlled. The action part can include adjusting the parameters of the homogenizing furnace, such as temperature, speed, etc. By formulating a set of rules, the rule engine can determine whether the parameters of the homogenizing furnace need to be adjusted based on the current state information.
[0080] The advantage of the rule engine is that it can make flexible decisions based on actual conditions and empirical knowledge. Rules can be defined based on the knowledge and experience of experts and can be quickly adjusted and updated. In addition, the rule engine can also support human-computer interaction, allowing operators to add, modify or delete rules as needed.
[0081] Further, it should be understood that before monitoring the tempered glass to be homogenized, it is also necessary to train the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier. That is, in the intelligent control system in the tempered glass production process of the embodiment of the present application, a training module is also included, and the training module is used to train the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier.
[0082] Specifically, in a specific embodiment of the present application, the training module 200 includes: a training data acquisition unit 201, which is used to collect training data, wherein the training data includes quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized is thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate; a tempered glass quality feature training unit 202, which is used to arrange the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension, and then extract the quality data through a multi-scale tempered glass quality feature extraction unit. A module is used to obtain a training tempered glass quality feature vector; a video denoising unit 203 is used to pass the monitoring video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a training denoised monitoring video; a key frame acquisition unit 204 is used to obtain a plurality of training tempered glass key frames from the training denoised monitoring video; a tempered glass surface feature training unit 205 is used to arrange the plurality of training tempered glass key frames into a three-dimensional tempered glass input tensor according to the time dimension and then pass the tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a training tempered glass surface feature map; a feature map dimension reduction unit 206 is used to convert each feature matrix in the training tempered glass surface feature map into a three-dimensional tempered glass input tensor; The array is used for performing global mean pooling processing along the channel dimension to obtain a training tempered glass surface feature vector; a tempered glass training feature fusion unit 207 is used to fuse the training tempered glass surface feature vector and the training tempered glass quality feature vector to obtain a tempered glass feature vector; a homogenizing furnace parameter feature training unit 208 is used to arrange the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector and then pass it through a parameter feature extraction module based on a sequence encoder to obtain a training homogenizing furnace parameter feature vector; a homogenization process training feature fusion unit 209 is used to perform feature fusion on the training homogenizing furnace parameter feature vector and the training tempered glass feature vector to obtain a training homogenizing furnace classification feature vector; a measurement The factor calculation unit 210 is used to calculate the information-based measurement factor of the topological dimension between the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector; the classification loss unit 211 is used to pass the training homogenizing furnace classification feature vector through a classifier to obtain a classification loss function value; the training unit 212 is used to train the multi-scale tempered glass quality feature extraction module, the video denoiser based on the generative adversarial network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier using the weighted sum of the information-based measurement factor of the topological dimension between the matching nodes and the classification loss function value as the loss function value.
[0083] In particular, in the technical solution of the present application, considering that the training homogenizing furnace parameter feature vector and the training tempered glass feature vector come from different data sources and encoding processes, they have differences in high-dimensional implicit feature distribution and feature scale. Specifically, the training homogenizing furnace parameter feature vector is extracted from the parameter data of the homogenizing furnace to be controlled, and these parameters are usually measured by sensors. Therefore, the high-dimensional implicit feature distribution of the training homogenizing furnace parameter feature vector may vary within this smaller range, forming a relatively simple feature distribution manifold. On the other hand, the training tempered glass feature vector is obtained by processing the quality data and monitoring video of the tempered glass to be homogenized. The quality data of the tempered glass includes indicators such as thickness, flatness and impact resistance, while the monitoring video provides the surface features of the tempered glass. These data sources usually have higher dimensions and contain rich information such as visual features and surface texture. Since the training homogenizing furnace parameter feature vector and the training tempered glass feature vector come from different data sources and encoding processes, their corresponding feature distribution manifolds in the high-dimensional feature space are different. Therefore, when fusing the training homogenizing furnace parameter feature vector and the training tempered glass feature vector, the correlation between their internal element sub-dimensions in the high-dimensional space is also considered. Based on this, in the technical solution of the present application, the topological dimension informationization metric factor between the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector is calculated as the loss function value, so that when fusing the training homogenizing furnace parameter feature vector and the training tempered glass feature vector, the correlation between the internal element sub-dimensions of the training homogenizing furnace parameter feature vector and the training tempered glass feature vector in the high-dimensional space is also considered.
[0084] In a specific embodiment of the present application, the metric factor calculation unit 210 is used to: calculate the topological dimension information metric factor of the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector according to the following formula;
[0085]
[0086] in, represents the training homogenizing furnace parameter feature vector, represents the training tempered glass feature vector, represents the transposed vector of the training homogenizing furnace parameter feature vector, and is a column vector, and They are respectively the training homogenizing furnace parameter feature vectors And train the tempered glass feature vector No. The eigenvalues at the positions, and is the length of the eigenvector, Indicates the information metric factor of the topological dimension between the matching nodes.
[0087] That is, when the training homogenizing furnace parameter feature vector and the training tempered glass feature vector are fused to obtain the training homogenizing furnace classification feature vector, due to the different feature scales between the high-dimensional implicit feature distribution of the training homogenizing furnace parameter feature vector and the high-dimensional implicit correlation features of the training tempered glass feature vector, the corresponding feature distribution manifolds in the high-dimensional feature space are also not the same, that is, the feature distribution and feature shapes of the two feature vectors in the high-dimensional space are inconsistent, which makes it difficult to match and align the information between the feature vectors, and it is impossible to effectively perform feature fusion and classification.
[0088] In order to solve this problem, the present application proposes a method based on the information-based measurement factor of the topological dimension between matching nodes. The method can calculate the information-based measurement factor of the topological dimension between matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector, and use the information-based measurement factor of the topological dimension between matching nodes as the loss function value to measure the information between the internal element sub-dimensions of the feature vector in the high-dimensional space, so that each element in the feature vector can express its topological structure or topological relationship in the high-dimensional space, thereby performing topological dimension informationization of the feature vector, thereby realizing the correlation between sub-regions and correlation migration within sub-regions of the encoded feature vector in the feature space, that is, the feature vectors between different sub-regions can influence and adjust each other, thereby promoting two-way approximation between feature vectors, and that the correlation expression between feature vectors is closer and consistent, thereby satisfying the density of the feature expression of the training homogenizing furnace classification feature vector under the probability density, that is, each element in the training homogenizing furnace classification feature vector can reflect the similarity or distance between the two feature vectors, thereby improving the accuracy of classification. In this way, the manifold differences between the feature vectors extracted by the encoder can be adapted to the convex monotonicity on the projections of each sub-dimension, thereby improving the feature association ability of the feature vector for training the homogenous furnace classification feature vector.
[0089] Further, Figure 7 The figure shows the architecture of the intelligent control system in the tempered glass production process according to the embodiment of the present application. Figure 7As shown, first, the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled are collected, wherein the quality data of the tempered glass to be homogenized are thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled include: temperature, time, heating power, and cooling rate; secondly, the quality data of the tempered glass to be homogenized is arranged into a one-dimensional tempered glass quality input vector according to the data type dimension, and then passed through a multi-scale tempered glass quality feature extraction module to obtain a tempered glass quality feature vector; then, the monitoring video of the tempered glass to be homogenized is passed through a video denoiser based on a generative adversarial network to obtain a denoised monitoring video, and a plurality of tempered glass key frames are obtained from the denoised monitoring video; then, the plurality of tempered glass key frames are arranged into a three-dimensional tempered glass input tensor according to the time dimension, and then passed through A tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model is used to obtain a tempered glass surface feature map; further, each feature matrix in the tempered glass surface feature map is subjected to global mean pooling processing along the channel dimension to obtain a tempered glass surface feature vector; then, the tempered glass quality feature vector and the tempered glass surface feature vector are fused to obtain a tempered glass feature vector; at the same time, the parameter data of the homogenizing furnace to be controlled is arranged as a one-dimensional homogenizing furnace parameter input vector and then passed through a parameter feature extraction module based on a sequence encoder to obtain a homogenizing furnace parameter feature vector; then, the tempered glass feature vector and the homogenizing furnace parameter feature vector are feature fused to obtain a homogenizing furnace classification feature vector; finally, the homogenizing furnace classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the parameter data of the homogenizing furnace to be controlled needs to be adjusted.
[0090] In summary, according to the intelligent control system 100 in the tempered glass production process of the embodiment of the present application, it is explained, which first collects the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized are thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled include: temperature, time, heating power, and cooling rate; then, multi-scale feature extraction is performed on the quality data, the surface features of the tempered glass in the monitoring video are extracted, and feature analysis is performed on the parameter data, and then the classification result is obtained by combining the extracted feature information. In this way, the influence of the homogenizing furnace on the quality of the tempered glass can be effectively reduced, thereby ensuring the stability of the quality during the homogenization process of the tempered glass.
[0091] As described above, the intelligent control system 100 in the tempered glass production process according to the embodiment of the present application can be implemented in various terminal systems, such as a server deployed with an intelligent control algorithm in the tempered glass production process. In one example, the intelligent control system 100 in the tempered glass production process can be integrated into the terminal system as a software module and / or a hardware module. For example, the intelligent control system 100 in the tempered glass production process can be a software module in the operating system of the terminal system, or can be an application developed for the terminal system; of course, the intelligent control system 100 in the tempered glass production process can also be one of the many hardware modules of the terminal system.
[0092] Alternatively, in another example, the intelligent control system 100 in the tempered glass production process and the terminal system may also be separate systems, and the intelligent control system 100 in the tempered glass production process may be connected to the terminal system via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.
[0093] Exemplary Methods
[0094] Figure 8 The figure shows a flow chart of the intelligent control method in the tempered glass production process according to the embodiment of the present application. Figure 8 As shown, according to the intelligent control method in the tempered glass production process of the embodiment of the present application, the method includes: S110, collecting quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized are thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled include: temperature, time, heating power, and cooling rate; S120, extracting feature information of the quality data of the tempered glass to be homogenized, feature information of the monitoring video of the tempered glass to be homogenized, and feature information of the parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality feature vector, a tempered glass surface feature vector, and a homogenizing furnace parameter feature vector; S130, based on the tempered glass quality feature vector, the tempered glass surface feature vector, and the homogenizing furnace parameter feature vector, to obtain a classification result.
[0095] Here, those skilled in the art can understand that the specific functions and operations of each step in the intelligent control method in the above-mentioned tempered glass production process have been described in the above reference. Figure 1 The description of the intelligent control system in the tempered glass production process has been introduced in detail, and therefore, its repeated description will be omitted.
[0096] Those skilled in the art will appreciate that various aspects of the present application may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of the present application may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.
[0097] The present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that the mention of "one embodiment" or "an embodiment" or "an alternative embodiment" twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.
[0098] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present disclosure belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology and should not be interpreted in an idealized or highly formal sense, unless explicitly defined as such herein.
[0099] The above is an explanation of the present disclosure and should not be considered as a limitation thereof. Although several exemplary embodiments of the present disclosure are described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present disclosure. Therefore, all such modifications are intended to be included within the scope of the present disclosure as defined in the claims. It should be understood that the above is an explanation of the present disclosure and should not be considered to be limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present disclosure is defined by the claims and their equivalents.
Claims
1. An intelligent control system for tempered glass production process, characterized in that: include: The tempered glass production data acquisition module is used to collect the quality data of the tempered glass to be homogenized, the monitoring video of the tempered glass to be homogenized, and the parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate; A production data feature extraction module, used to extract feature information of quality data of the tempered glass to be homogenized, feature information of monitoring video of the tempered glass to be homogenized, and feature information of parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality feature vector, a tempered glass surface feature vector, and a homogenizing furnace parameter feature vector; A homogenizing furnace classification result generating module, used to obtain a classification result based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector; Wherein, the production data feature extraction module includes: A tempered glass quality feature acquisition unit, used to acquire multi-scale feature information of quality data of the tempered glass to be homogenized to obtain the tempered glass quality feature vector; A tempered glass surface feature acquisition unit, used for extracting features from the monitoring video of the tempered glass to be homogenized to obtain a surface feature vector of the tempered glass; A homogenizing furnace parameter feature acquisition unit, used for extracting features from the parameter data of the homogenizing furnace to be controlled to obtain a homogenizing furnace parameter feature vector; Wherein, the tempered glass quality characteristic acquisition unit comprises: A quality data arrangement subunit, used for arranging the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension; A multi-scale feature acquisition subunit, used for passing the one-dimensional tempered glass quality input vector through a multi-scale tempered glass quality feature extraction module to obtain the tempered glass quality feature vector; Wherein, the tempered glass surface feature acquisition unit comprises: A video denoising subunit, used for passing the surveillance video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a denoised surveillance video; A key frame acquisition subunit, used to acquire a plurality of tempered glass key frames from the noise-reduced surveillance video; A key frame arrangement subunit, used for arranging the plurality of tempered glass key frames into a three-dimensional tempered glass input tensor according to a time dimension; A surface change feature acquisition subunit, used for passing the three-dimensional tempered glass input tensor through a tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a tempered glass surface feature map; A feature map dimensionality reduction subunit, used for performing global mean pooling processing along the channel dimension on each feature matrix in the surface feature map of the tempered glass to obtain a surface feature vector of the tempered glass; Wherein, the homogenizing furnace parameter feature acquisition unit includes: A homogenizing furnace parameter arrangement subunit, used for arranging the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector; A homogenizing furnace parameter feature acquisition subunit, used for passing the one-dimensional homogenizing furnace parameter input vector through a parameter feature extraction module based on a sequence encoder to obtain the homogenizing furnace parameter feature vector; Wherein, the homogenizing furnace classification result generation module includes: A tempered glass feature fusion unit, used for fusing the tempered glass quality feature vector and the tempered glass surface feature vector to obtain a tempered glass feature vector; A homogenization process feature fusion unit, used for performing feature fusion on the tempered glass feature vector and the homogenization furnace parameter feature vector to obtain a homogenization furnace classification feature vector; A classification unit, used for passing the homogenizing furnace classification feature vector through a classifier to obtain a classification result; Among them, the intelligent control system in the tempered glass production process also includes a training module for training the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier; Wherein, the training module includes: A training data acquisition unit, used to acquire training data, wherein the training data includes quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes temperature, time, heating power, and cooling rate; A tempered glass quality feature training unit, used for arranging the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension and then passing it through a multi-scale tempered glass quality feature extraction module to obtain a training tempered glass quality feature vector; A video denoising unit, used for passing the surveillance video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a surveillance video after training and denoising; A key frame acquisition unit, used to acquire a plurality of training tempered glass key frames from the training noise-reduced surveillance video; A tempered glass surface feature training unit, used for arranging the plurality of training tempered glass key frames into a three-dimensional tempered glass input tensor according to the time dimension and then passing the tensor through a tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a training tempered glass surface feature map; A feature map dimensionality reduction unit, used for performing global mean pooling processing along the channel dimension on each feature matrix in the training tempered glass surface feature map to obtain a training tempered glass surface feature vector; A tempered glass training feature fusion unit, used for fusing the training tempered glass surface feature vector and the training tempered glass quality feature vector to obtain a tempered glass feature vector; A homogenizing furnace parameter feature training unit, used for arranging the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector and then passing it through a parameter feature extraction module based on a sequence encoder to obtain a training homogenizing furnace parameter feature vector; A homogenization process training feature fusion unit, used for performing feature fusion on the training homogenization furnace parameter feature vector and the training tempered glass feature vector to obtain a training homogenization furnace classification feature vector; A metric factor calculation unit, used to calculate a topological dimension information metric factor of a matching node between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector; A classification loss unit, used for passing the training homogenizing furnace classification feature vector through a classifier to obtain a classification loss function value; A training unit, used to train the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier by taking the weighted sum of the topological dimension information metric factor between the matching nodes and the classification loss function value as the loss function value; Wherein, the metric factor calculation unit is used to: calculate the topological dimension information metric factor of the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector according to the following formula; ; in, represents the training homogenizing furnace parameter feature vector, represents the training tempered glass feature vector, represents the transposed vector of the training homogenizing furnace parameter feature vector, and is a column vector, and They are respectively the training homogenizing furnace parameter feature vectors And training tempered glass feature vector No. The eigenvalues at the positions, and is the length of the eigenvector, A measure factor representing the informationization of the topological dimension between the matching nodes; The classification result is used to indicate whether the parameter data of the homogenizing furnace to be controlled needs to be adjusted.
2. An intelligent control method in a tempered glass production process, used for the intelligent control system in a tempered glass production process according to claim 1, characterized in that: include: Collecting quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes: temperature, time, heating power, and cooling rate; Extracting characteristic information of quality data of the tempered glass to be homogenized, characteristic information of monitoring video of the tempered glass to be homogenized, and characteristic information of parameter data of the homogenizing furnace to be controlled to obtain a tempered glass quality characteristic vector, a tempered glass surface characteristic vector, and a homogenizing furnace parameter characteristic vector; Based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector, a classification result is obtained; The feature information of the quality data of the tempered glass to be homogenized, the feature information of the monitoring video of the tempered glass to be homogenized, and the feature information of the parameter data of the homogenizing furnace to be controlled are extracted to obtain the tempered glass quality feature vector, the tempered glass surface feature vector, and the homogenizing furnace parameter feature vector, including: Acquiring multi-scale feature information of quality data of the tempered glass to be homogenized to obtain a quality feature vector of the tempered glass; Performing feature extraction on the monitoring video of the tempered glass to be homogenized to obtain a surface feature vector of the tempered glass; Performing feature extraction on the parameter data of the homogenizing furnace to be controlled to obtain a feature vector of the homogenizing furnace parameters; Wherein, obtaining multi-scale feature information of the quality data of the tempered glass to be homogenized to obtain the quality feature vector of the tempered glass includes: Arranging the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension; Passing the one-dimensional tempered glass quality input vector through a multi-scale tempered glass quality feature extraction module to obtain the tempered glass quality feature vector; The process of extracting features from the monitoring video of the tempered glass to be homogenized to obtain the surface feature vector of the tempered glass includes: Passing the surveillance video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a denoised surveillance video; Acquire multiple tempered glass key frames from the noise-reduced surveillance video; Arranging the plurality of tempered glass key frames into a three-dimensional tempered glass input tensor according to a time dimension; Passing the three-dimensional tempered glass input tensor through a tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a tempered glass surface feature map; Performing global mean pooling processing on each feature matrix in the tempered glass surface feature map along the channel dimension to obtain the tempered glass surface feature vector; Among them, feature extraction is performed on the parameter data of the homogenizing furnace to be controlled to obtain the parameter feature vector of the homogenizing furnace, including: Arranging the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector; The one-dimensional homogenizing furnace parameter input vector is passed through a parameter feature extraction module based on a sequence encoder to obtain the homogenizing furnace parameter feature vector; Wherein, based on the tempered glass quality feature vector, the tempered glass surface feature vector and the homogenizing furnace parameter feature vector, a classification result is obtained, including: Merging the tempered glass quality feature vector and the tempered glass surface feature vector to obtain a tempered glass feature vector; Performing feature fusion on the tempered glass feature vector and the homogenizing furnace parameter feature vector to obtain a homogenizing furnace classification feature vector; Passing the homogenizing furnace classification feature vector through a classifier to obtain a classification result; The intelligent control method in the tempered glass production process further includes a training step: training the multi-scale tempered glass quality feature extraction module, the video denoiser based on the adversarial generative network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier; Wherein, the training step includes: Collecting training data, wherein the training data includes quality data of the tempered glass to be homogenized, monitoring video of the tempered glass to be homogenized, and parameter data of the homogenizing furnace to be controlled, wherein the quality data of the tempered glass to be homogenized includes thickness, flatness, and impact resistance, and the parameter data of the homogenizing furnace to be controlled includes temperature, time, heating power, and cooling rate; Arranging the quality data of the tempered glass to be homogenized into a one-dimensional tempered glass quality input vector according to the data type dimension and then passing it through a multi-scale tempered glass quality feature extraction module to obtain a training tempered glass quality feature vector; Passing the monitoring video of the tempered glass to be homogenized through a video denoiser based on a generative adversarial network to obtain a trained denoised monitoring video; Acquire multiple training tempered glass key frames from the training denoised surveillance video; Arranging the plurality of training tempered glass key frames into a three-dimensional tempered glass input tensor according to the time dimension and then passing the tensor through a tempered glass surface feature extraction module based on a three-dimensional convolutional neural network model to obtain a training tempered glass surface feature map; Performing global mean pooling processing on each feature matrix in the training tempered glass surface feature map along the channel dimension to obtain a training tempered glass surface feature vector; Merging the training tempered glass surface feature vector and the training tempered glass quality feature vector to obtain a tempered glass feature vector; Arranging the parameter data of the homogenizing furnace to be controlled into a one-dimensional homogenizing furnace parameter input vector and then passing it through a parameter feature extraction module based on a sequence encoder to obtain a training homogenizing furnace parameter feature vector; Performing feature fusion on the training homogenizing furnace parameter feature vector and the training tempered glass feature vector to obtain a training homogenizing furnace classification feature vector; Calculating an information-based metric factor of a topological dimension of matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector; Passing the training homogenizing furnace classification feature vector through a classifier to obtain a classification loss function value; The multi-scale tempered glass quality feature extraction module, the video denoiser based on the generative adversarial network, the tempered glass surface feature extraction module based on the three-dimensional convolutional neural network model, the parameter feature extraction module based on the sequence encoder, and the classifier are trained by taking the weighted sum of the topological dimension information metric factor between the matching nodes and the classification loss function value as the loss function value; The step of calculating the topological dimension information-based metric factor between the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector comprises: calculating the topological dimension information-based metric factor between the matching nodes between the training homogenizing furnace parameter feature vector and the training tempered glass feature vector using the following formula; ; in, represents the training homogenizing furnace parameter feature vector, represents the training tempered glass feature vector, represents the transposed vector of the training homogenizing furnace parameter feature vector, and is a column vector, and They are respectively the training homogenizing furnace parameter feature vectors And training tempered glass feature vector No. The eigenvalues at the positions, and is the length of the eigenvector, A measure factor representing the informationization of the topological dimension between the matching nodes; The classification result is used to indicate whether the parameter data of the homogenizing furnace to be controlled needs to be adjusted.
