Glass thermal cycle control method and related device

Abstract

The present application provides a glass thermal cycle control method and related devices. The present application obtains the equipment parameters of the target device; the equipment parameters are predicted in real time through a thermal cycle control model to obtain the glass yield prediction information of the target device under different parameter settings; the thermal cycle control model is a lightweight model installed in the target device for learning the real-time situation of the production environment; based on the glass yield prediction information, an optimization strategy for the running program parameters is generated; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that meets the preset glass finished product evaluation standard; the glass finished product evaluation standard is related to the equipment production demand information of the target device; the optimization strategy is used to adjust the running program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device. The present application can dynamically predict and control the running program parameters to control the thermal cycle speed of the target device and improve production efficiency and glass quality.

Description

Technical Field

[0001] The present application relates to the technical field of glass processing, and in particular to a glass thermal cycle control method and related devices. Background Art

[0002] In the glass processing industry, the production process of glass curtain walls is highly dependent on precise heating and cooling control. In particular, the reciprocating heating process in a single glass furnace.

[0003] In the related art, the reciprocating circulation speed of glass in a single furnace usually depends on a preset fixed schedule or manual adjustment based on experience. This method relies on a fixed time period or the intuitive judgment of the operator to set the travel speed of the glass. This method lacks flexibility and is difficult to adapt to the changing conditions in actual production. For example, furnaces of different lengths require different cycle times to ensure uniform heating. Glass of different thicknesses has different requirements for heating rates and cycle times due to their different thermal conductivity. This method is difficult to accurately adapt to the changes in different furnace lengths, travel program settings, and glass thickness in these situations, which can easily lead to problems such as low heating efficiency, increased energy consumption, and uneven glass quality in the glass processing process.

[0004] Therefore, it is urgent to design a new technical solution to solve at least one of the above technical problems. Summary of the invention

[0005] In order to solve or improve the above problems, the embodiments of the present application provide a glass thermal cycle control method and related devices.

[0006] In one embodiment of the present application, a glass thermal cycle control method is provided, comprising:

[0007] Obtaining equipment parameters of the target equipment; the equipment parameters at least include: furnace size, running program parameters; the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment;

[0008] The equipment parameters are predicted in real time through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings; the thermal cycle control model is a lightweight model installed in the target equipment for learning the real-time situation of the production environment;

[0009] Based on the glass yield prediction information, an optimization strategy for the walking program parameters is generated; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that reaches a preset glass finished product evaluation standard; the glass finished product evaluation standard is related to the equipment production demand information of the target equipment; the equipment production demand information at least includes: glass thickness and glass material composition;

[0010] The optimization strategy is adopted to adjust the walking program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0011] In another embodiment of the present application, a glass thermal cycle control device is provided, comprising at least the following units:

[0012] A collection unit, used to obtain equipment parameters of the target equipment; the equipment parameters at least include: furnace size, running program parameters; the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment;

[0013] A prediction unit, used to predict the equipment parameters in real time through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings; the thermal cycle control model is a lightweight model installed in the target equipment for learning the real-time situation of the production environment;

[0014] A generating unit, configured to generate an optimization strategy for the walking program parameters based on the glass yield prediction information; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that reaches a preset glass finished product evaluation standard; the glass finished product evaluation standard is related to equipment production demand information of the target equipment; the equipment production demand information at least includes: glass thickness and glass material composition;

[0015] The adjustment unit is used to adjust the walking program parameters of the target device by adopting the optimization strategy to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0016] In one embodiment of the present application, an electronic device includes:

[0017] at least one processor, memory, and input-output unit;

[0018] The memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the glass thermal cycle control method in the first embodiment.

[0019] In one embodiment of the present application, a computer-readable storage medium is provided, which includes instructions. When the instructions are executed on a computer, the computer executes the glass thermal cycle control method in the first embodiment.

[0020] In the technical solution provided by the embodiment of the present application, the equipment parameters of the target device are obtained; the equipment parameters at least include: furnace size, walking program parameters; the walking program parameters include a variety of process parameters associated with the production links and equipment structural components of the target device; the equipment parameters are predicted in real time by a thermal cycle control model to obtain the glass yield prediction information of the target device under different parameter settings; the thermal cycle control model is a lightweight model installed in the target device for learning the real-time situation of the production environment; based on the glass yield prediction information, an optimization strategy for the walking program parameters is generated; the optimization strategy includes the equipment parameters corresponding to the glass yield prediction value that reaches the preset glass finished product evaluation standard; the glass finished product evaluation standard is related to the equipment production demand information of the target device; the equipment production demand information at least includes: glass thickness, glass material composition; the walking program parameters of the target device are adjusted by the optimization strategy to control the thermal cycle speed of the target device and improve the production efficiency of the target device. The present application can dynamically predict and control the walking program parameters, so as to control the thermal cycle speed of the target device and improve the glass production efficiency and glass quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A schematic diagram of a flow chart of a glass thermal cycle control method provided in one embodiment of the present application;

[0023] Figure 2 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application provides the following embodiments to solve or improve at least some of the above problems. In order to enable people in the technical field to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0025] In some features described in the specification, claims and the above-mentioned drawings of this application, descriptions such as "first" and "second" are used to distinguish different components, parts, modules, devices, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to be different types. In addition, the embodiments described below are only some embodiments of the present application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0026] In order to solve the above technical problems, a glass thermal cycle control device provided in one embodiment of the present application includes the following units:

[0027] A collection unit, used to obtain equipment parameters of the target equipment; the equipment parameters at least include: furnace size, running program parameters; the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment;

[0028] A prediction unit, used to predict the equipment parameters in real time through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings; the thermal cycle control model is a lightweight model installed in the target equipment for learning the real-time situation of the production environment;

[0029] A generating unit, configured to generate an optimization strategy for the walking program parameters based on the glass yield prediction information; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that reaches a preset glass finished product evaluation standard; the glass finished product evaluation standard is related to equipment production demand information of the target equipment; the equipment production demand information at least includes: glass thickness and glass material composition;

[0030] The adjustment unit is used to adjust the walking program parameters of the target device by adopting the optimization strategy to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0031] The above-mentioned device has relevant functions or modules corresponding to the specific implementation methods involved in the method part of the embodiment of this application or other embodiments.

[0032] In the above device, the following is a detailed introduction to the functions and principles of each unit of the device:

[0033] The core function of the acquisition unit is to collect various parameter information related to the target equipment, and provide basic data support for subsequent prediction, optimization and adjustment work. Specifically, the furnace size parameters obtained can reflect the spatial range, heat distribution and other conditions during the thermal cycle. For example, a larger furnace size may mean a longer heat transfer path and more challenging thermal uniformity control. The walking program parameters cover many process parameters closely related to the production links and equipment structural components, such as the conveying speed of glass in different areas of the furnace, the residence time of each heating stage, the power setting of different heating elements, etc. These parameters directly affect the heating conditions of the glass during the production process, stress formation, and the final product quality and production efficiency.

[0034] The corresponding parameters are obtained by connecting and communicating with the control system, sensors, etc. of the target equipment. For the furnace size parameters, they may be read from the original design document of the equipment (if the equipment has the function of storing relevant design information), or obtained by actual measurement through the distance measuring sensor installed on the equipment (such as laser rangefinder, etc.). The walking program parameters are mainly extracted from the programmable logic controller (PLC) of the equipment. The PLC stores and controls the process parameter settings involved in each link of the entire production process. The acquisition unit reads these real-time or preset parameter values ​​from the PLC according to the established communication protocol (such as Modbus protocol, etc.) to ensure that accurate and complete equipment parameter information is obtained.

[0035] The prediction unit uses the thermal cycle control model to deeply analyze and process the equipment parameters obtained by the acquisition unit, and predicts in advance the yield rate of the target equipment when producing glass under different parameter settings. This enables production personnel to know the impact of different parameter combinations on the yield rate before actual production or when adjusting parameters during production, thereby providing a basis for formulating reasonable optimization strategies, avoiding production losses caused by blindly adjusting parameters, and helping to control the production process more scientifically and improve product quality and production efficiency.

[0036] The thermal cycle control model is a lightweight model that mines the complex internal relationship between equipment parameters and glass yield based on the learning of a large amount of past production data (including historical equipment parameters and corresponding actual glass yield data). In real-time prediction, the collected current equipment parameters are input into the model, and the model outputs the corresponding glass yield prediction information based on the learned rules and internal calculation logic (for example, for the model of the neural network structure, it is the forward propagation calculation process of each neuron layer). For example, the model may recognize that when the furnace size is a certain value, the transmission speed in the walking program parameters is within a certain range, and the heating power is at a certain level, the glass yield will be at a relatively high or low level, and then give the corresponding prediction value.

[0037] The generation unit generates a parameter optimization strategy for the running program that can improve product quality and yield rate based on the glass yield rate prediction information provided by the prediction unit and the established production requirements of the target equipment. The aim is to find the best parameter combination that meets the preset glass finished product evaluation standards (related to production requirements such as glass thickness and material composition), so that the produced glass can meet the quality requirements and ensure a high yield rate, optimize the production process as a whole, and maximize production benefits.

[0038] First, the evaluation standard of finished glass products is clarified. The standard is determined by the equipment production demand information. For example, for glass of a specific thickness and material composition, the finished product has corresponding qualified range requirements in terms of hardness, flatness, transparency and other quality indicators. Then, the glass yield prediction information is screened and analyzed to find out the equipment parameter combinations with high yield prediction values ​​and the corresponding glass finished products that can meet the qualified range of the above quality indicators. On this basis, the influence of each parameter on the yield and quality is comprehensively considered, and a specific walking program parameter optimization strategy is generated through certain mathematical methods (such as constructing an objective function and solving optimization problems, see the relevant content described above for details), and it is clarified how each process parameter should be adjusted, such as whether to appropriately increase or decrease the heating time, adjust the transmission speed, etc., in order to achieve the purpose of optimizing production.

[0039] The adjustment unit applies the optimization strategy formulated by the generation unit to the actual production of the target equipment. By making targeted adjustments to the running program parameters and changing key production links such as the thermal cycle speed of the equipment, the operating parameters of the equipment are optimized, ultimately improving the production efficiency of the target equipment, ensuring the production of more glass products that meet quality requirements, and improving the economic benefits of the entire production system.

[0040] The adjustment unit establishes a communication connection with the control system (such as PLC) of the target device, and transmits the specific parameter adjustment instructions in the optimization strategy to the control system according to the corresponding communication protocol. For example, if the optimization strategy requires increasing the heating power of a certain heating area, the adjustment unit will send the corresponding power increase value instruction to the PLC. After receiving the instruction, the PLC will control the corresponding heating element to change the power supply, thereby achieving parameter adjustment. Similarly, the adjustment of other walking program parameters such as transmission speed is also carried out in the same way. By adjusting the relevant parameters in a coordinated manner, the thermal cycle state of the equipment is changed, so that the production process can be developed in a more efficient and higher quality direction.

[0041] In general, the device forms a closed-loop system from parameter collection, yield prediction, optimization strategy generation to actual parameter adjustment through the coordinated cooperation between various units. It aims to continuously optimize the equipment parameters in the glass production process, improve the glass yield and production efficiency, and ensure that product quality meets production needs.

[0042] It should be added here that the control logic of the relevant device will be introduced in detail in the method embodiment below.

[0043] An embodiment of the present application provides a glass thermal cycle control method. The method of this embodiment can be performed by the glass thermal cycle control device in the above embodiment, and is implemented on the basis of the structure of the glass thermal cycle control device in the above embodiment. Figure 1 As shown, the method may include the following steps:

[0044] 101. Obtain device parameters of the target device; wherein the target object is colored glaze glass, and the acquisition method is high-resolution scanning;

[0045] 102. Real-time prediction of the equipment parameters is performed through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings;

[0046] 103. Generate an optimization strategy for the running program parameters based on the glass yield prediction information;

[0047] 104. Use the optimization strategy to adjust the walking program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0048] The equipment parameters at least include: furnace size and running program parameters. Furthermore, the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment.

[0049] Exemplarily, the walking program parameters include at least one of the following: heating time, rapid cooling time, cooling time, furnace entry interval, rapid cooling air pressure, cooling air pressure, furnace entry speed, furnace exit speed, furnace reciprocating speed, furnace adjustment speed, adjustment time, rapid cooling reciprocating speed, cooling reciprocating speed, and preheating time.

[0050] The following is a detailed introduction to the principles and effects of each step:

[0051] In step 101, device parameters of the target device are obtained.

[0052] As a collection method, high-resolution scanning can accurately obtain various detailed information of colored glaze glass, such as the thickness distribution of the colored glaze, the shape of the pattern, and some microscopic features of the glass substrate itself. This information is crucial for the subsequent thermal cycle regulation, because the presence of the colored glaze will affect the thermal conductivity of the glass. The furnace size is a basic physical parameter of the thermal cycle, which determines the distribution range and transfer path of heat in space. For colored glaze glass, the furnace size will affect the uniform heating degree of the colored glaze during the heating process.

[0053] The various process parameters in the walking program parameters are closely linked to specific production links. For example, the conveying speed of the glass in the furnace will affect its residence time in each heating area, and the power settings of different heating areas will determine the heating speed of different parts of the glass. These are closely related to the final quality indicators such as the color and glossiness of the colored glaze glass.

[0054] High-resolution scanning technology can be used to obtain fine features on the surface of the equipment, such as the distribution of temperature sensors inside the furnace, the precise position and status of heating elements, etc. This information can more accurately reflect the actual working conditions of the equipment, provide more detailed data input for subsequent models, and help improve the accuracy of prediction and optimization.

[0055] Therefore, it is possible to obtain parameter information including the detailed features of the colored glaze glass and the precise internal structure of the equipment, providing comprehensive and accurate data support for subsequent thermal cycle control, and avoiding optimization deviations caused by inaccurate or incomplete data. Due to the particularity of colored glaze glass, this collection method is helpful for optimizing the production of colored glaze glass, so that subsequent parameter prediction, optimization and adjustment are more in line with the production needs of colored glaze glass.

[0056] In step 102, the equipment parameters are predicted in real time through the thermal cycle control model to obtain the glass yield prediction information of the target equipment under different parameter settings.

[0057] The model is built based on a large amount of historical production data (including equipment parameters and actual yield of colored glaze glass). It learns the complex nonlinear relationship between equipment parameters and yield through machine learning algorithms (such as neural networks, regression analysis, etc.). For example, the model can learn the relationship between the size of the furnace and the uniformity of heating of various parts of the glass, as well as the influence of the transmission speed and heating power in the walking program parameters on the quality indicators such as color curing and glossiness of colored glaze glass, so as to predict the yield under different parameter settings.

[0058] When new equipment parameters are input into the model, the model performs real-time calculations based on the learned rules through internal calculation logic (such as forward propagation of neural networks). For example, for a given combination of furnace size, conveying speed, and heating power, the model can predict the yield rate of glazed glass under this parameter setting, including quality-related yield indicators such as whether the color is uniform and whether the glaze is firm.

[0059] Before actual production or parameter adjustment, a relatively accurate estimate of the glass yield under different parameter settings can be made, providing a reference for production decisions. For example, production personnel can judge in advance whether a certain parameter adjustment will increase or decrease the yield based on the forecast information, avoiding losses caused by blindly adjusting parameters. By comparing the yield forecast information under different parameter settings, it is possible to determine which parameters have a positive impact on the yield and which have a negative impact, thereby providing direction for subsequent optimization strategies.

[0060] In step 103, an optimization strategy for the running program parameters is generated based on the glass yield prediction information.

[0061] Colored glaze glass has its specific production requirements, such as the color brightness, firmness, and flatness of the glaze. According to the glass yield prediction information, those parameter combinations that can meet these quality standards and have a high yield are screened out. For example, if the prediction information shows that under a certain combination of conveying speed and heating power, the color uniformity and firmness of the colored glaze glass can meet the standards and have a high yield, then this combination will be given priority consideration. Considering the interrelationship between the parameters of the walking program, multiple process parameters are comprehensively optimized through mathematical models (such as multi-objective optimization models) or empirical rules. For example, appropriate adjustment of the conveying speed may require simultaneous adjustment of the heating power to ensure that the colored glaze glass can reach the best state in each link during the thermal cycle, thereby improving the overall yield. The generated optimization strategy is customized for the production characteristics and quality requirements of colored glaze glass, and can effectively improve the yield of colored glaze glass during the thermal cycle. For example, the optimization strategy may adjust the heating power and residence time for the curing link of the colored glaze, so that the colored glaze can better adhere to the glass surface and improve the quality of the finished product. While improving the yield rate, we take production efficiency into consideration and reasonably adjust the parameters of the walking program to avoid excessive pursuit of quality that may lead to long production cycles or high costs, thus achieving a balance between quality and efficiency.

[0062] In step 104, an optimization strategy is used to adjust the walking program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0063] The control system of the target device receives the parameter adjustment instructions in the optimization strategy and adjusts the parameters of the device's travel program according to these instructions. For example, when the optimization strategy requires reducing the conveying speed and increasing the power of a certain heating area, the control system will adjust the motor speed and the power output of the heating element accordingly, thereby changing the thermal cycle speed.

[0064] By adjusting the parameters of the travel program, such as the conveying speed and heating power, the thermal cycle speed of the glass in the furnace can be directly controlled. The appropriate thermal cycle speed is very important for the production of enameled glass, which can ensure that the enameled glass is cured under the appropriate temperature and time conditions while ensuring the quality of the glass substrate. For example, slowing down the conveying speed appropriately can allow the enameled glass to be heated in the heating area for enough time, so that the enameled glass can be fully cured and the quality of the finished product can be improved.

[0065] In this way, through the optimized parameter adjustment, the equipment can run at a more reasonable thermal cycle speed, reduce rework or waste caused by quality problems, and thus improve actual production efficiency. For example, in the production process of colored glaze glass, the optimized parameters can improve the first-time qualified rate of colored glaze, reduce the wasted time in the production cycle, and increase the output of qualified products per unit time. A stable thermal cycle speed helps to ensure the stability of the quality of colored glaze glass. Since the equipment parameters are adjusted according to the optimization strategy, the produced colored glaze glass is more stable in quality indicators such as color, glossiness, and glaze firmness, meeting the market's requirements for product quality consistency.

[0066] As an optional embodiment, before obtaining the equipment parameters of the target equipment in 101, multiple equipment sensors may be deployed; the multiple equipment sensors are respectively installed in the equipment structural components corresponding to the multiple production links of the target equipment. Furthermore, in 101, obtaining the equipment parameters of the target equipment may be implemented as follows: collecting multi-angle image information of the target equipment, and determining the furnace size of the target equipment based on the multi-angle image information; and collecting multi-dimensional walking program parameters from multiple equipment structural components as the equipment parameters.

[0067] Before obtaining the equipment parameters of the target equipment, multiple equipment sensors are deployed and installed in the equipment structural components corresponding to each production link in order to collect equipment-related data information more comprehensively, accurately and in real time.

[0068] In terms of collecting multi-angle image information of the target device to determine the size of the furnace, by arranging image acquisition sensors (such as cameras, etc.) at different angles, the complete appearance of the furnace can be captured, and the length, width, height, etc. of the furnace can be accurately calculated using image recognition and processing technology based on the geometric principles of object imaging and known reference information (if any). Images from different angles can complement and verify each other, overcoming the possible occlusion or viewing angle deviation problems of a single angle, making the acquired furnace size data more in line with the actual situation.

[0069] The reason why multiple equipment sensors are installed in different equipment structural components is that the target equipment has many production links and each component works together. The operating parameters at each component have a key impact on the overall production process and the quality of the final product. For example, installing a speed sensor at the glass conveying component can obtain the conveying speed of the glass in real time, installing a temperature sensor around the heating component can know the heating temperature of different areas, and installing current and voltage sensors on key power and control components can grasp the energy consumption and power of the equipment. These sensors collect parameters related to the walking program from multiple dimensions, such as time parameters in the process (such as the residence time of each link) and physical quantity parameters (such as power, speed, temperature, etc.), and then integrate them to form complete walking program parameters, which can accurately reflect the actual operating status of the equipment in each production link and the process execution, and provide rich and reliable basic data for the analysis and prediction of the subsequent thermal cycle control model. After all, thermal cycle control needs to comprehensively consider the relationship and interaction between various parameters. These parameters from different structural components are an important basis for analyzing this complex relationship.

[0070] Overall, this method can greatly improve the accuracy and comprehensiveness of equipment parameters. The determined furnace size is more in line with the actual situation, so that the subsequent size-based heat transfer analysis, thermal cycle range definition and other work have a reliable spatial parameter basis, avoiding the deviation of the thermal cycle process estimation due to dimensional data errors, which in turn affects the prediction of glass yield and the entire production optimization process.

[0071] Collecting multi-dimensional walking program parameters can enable production personnel to have an in-depth understanding of the specific conditions of each production link, to know whether the actual working status of each component is consistent with the preset process parameters, to promptly discover potential abnormalities, such as abnormal temperature fluctuations in a heating area, unstable glass transmission speed, and other problems, and to take measures to make adjustments in advance. For the subsequent thermal cycle control model, multi-dimensional and accurate walking program parameter input can help the model to more accurately learn the intrinsic connection between different parameter combinations and glass yield, and improve the accuracy of the model's prediction information on glass yield under different parameter settings. The walking program parameter optimization strategy generated based on such accurate prediction information is more targeted and effective. It will eventually be applied to actual production, which can better control the thermal cycle speed of the target equipment, significantly improve the production efficiency of the target equipment and the quality of the finished glass products, reduce the generation of defective and waste products, and improve the overall production efficiency.

[0072] As an optional embodiment, in 101, determining the furnace size of the target device based on the multi-angle image information includes:

[0073] Perform image preprocessing on the multi-angle image information; identify key features about the furnace body size in the multi-angle image information; the key features at least include furnace body contour features; the furnace body contour features are used to indicate the positions of furnace body edges and / or corners; perform image registration on the key features, and construct a corresponding furnace body three-dimensional model based on the registered key feature map; extract the furnace body size of the target device based on the furnace body three-dimensional model; the furnace body size is calculated based on a preset furnace body morphology mapping relationship.

[0074] When processing multi-angle image information, convolutional neural networks (CNNs) in deep learning can be used to complete image preprocessing and key feature recognition. For example, the relevant architecture of the generative adversarial network (GAN) is used for image denoising, enhancement and other preprocessing operations to make the image clearer and the details more prominent, laying a good foundation for subsequent feature extraction. For key feature recognition, CNN-based target detection algorithms such as Faster R-CNN and YOLO series are used. After being trained with a large amount of annotated furnace image data, these algorithms can accurately identify the contour features of the furnace body in the image, such as edges and corners. Compared with traditional manual feature extraction methods, they are more robust and accurate, and can adapt to images under complex shooting conditions such as different lighting and angles.

[0075] In the image registration process, we introduce registration methods based on feature descriptors, such as SIFT (Scale Invariant Feature Transform), SURF (Speeded Up Robust Features) and other algorithms combined with deep learning feature matching methods. We first use traditional algorithms to extract feature points in the image and generate feature descriptors, and then use deep learning models to further match and screen these feature descriptors to improve the accuracy and efficiency of registration, so as to better align and integrate key features in images at different angles.

[0076] When constructing the 3D model of the furnace, we use the multi-view stereo vision (MVS) algorithm in the 3D reconstruction technology combined with the depth estimation method of deep learning. The traditional MVS algorithm calculates depth information through the geometric relationship between images. Now we can use the deep learning model to pre-train the depth estimation of multi-angle images, provide more accurate initial depth values ​​for the MVS algorithm, speed up the construction of the 3D model and improve the model accuracy, and finally obtain a high-quality 3D model of the furnace.

[0077] When extracting the furnace body dimensions based on the furnace body three-dimensional model, the furnace body shape mapping relationship is set and the spatial measurement algorithm and model analysis algorithm in computer graphics are used to automatically and accurately obtain the furnace body's length, width, height, internal cavity dimensions and other parameters. At the same time, virtual reality (VR) or augmented reality (AR) technology can be combined to visualize the furnace body three-dimensional model, making it easier for operators to intuitively confirm the size information and perform virtual parameter adjustments, analysis and other operations, which is more in line with the needs of modern and intelligent production management.

[0078] The above implementation method can improve the accuracy of image information processing. Whether it is image preprocessing or key feature recognition, it can effectively reduce the errors caused by factors such as image quality problems and shooting angles, making the extracted key features such as furnace body contour more accurate and reliable, and providing a high-quality data basis for subsequent model construction. The optimization of the image registration link ensures the accurate alignment of images at different angles, improves the accuracy of building a three-dimensional model of the furnace body, and enables the three-dimensional model to more realistically reflect the actual shape and structure of the furnace body, avoiding model distortion caused by inaccurate registration. With the help of the improved three-dimensional reconstruction and size extraction method, the furnace body size data obtained is more accurate and comprehensive, and can better serve subsequent production links such as thermal cycle control. Moreover, the visual display method facilitates the understanding and application of furnace body size by production personnel, facilitates timely discovery of the difference between size data and actual production needs, and makes production adjustments and optimizations in advance, improves the scientificity and accuracy of the entire glass production process, thereby improving production efficiency and product quality, reducing production failures and quality problems caused by factors such as inaccurate size, and enhancing the competitiveness of enterprises in the field of glass production.

[0079] Furthermore, use high-resolution and high-precision cameras, such as industrial-grade 3D cameras or digital SLR cameras with high-pixel and low-distortion lenses. These cameras can capture clearer and more detailed images, providing rich details for subsequent size determination. Ensure the stability and consistency of the shooting environment. During the shooting process, to avoid camera shake, you can use auxiliary equipment such as tripods and gimbals to fix the camera. At the same time, control the lighting conditions of the shooting environment to ensure that the light is uniform and sufficient, and reduce the impact of shadows and reflections on image quality. For example, set up multiple evenly distributed light sources indoors to avoid strong light directly on the surface of the furnace to obtain high-quality multi-angle images.

[0080] It is also important to note that all parts of the furnace should be fully covered, and photographed from different directions (such as the front, the back, the left and right sides, the top and the bottom), and that images from adjacent angles should have a certain overlap area. This can avoid the loss of some furnace structure information due to missing shooting angles, and the overlapping area will help with subsequent image stitching and feature matching. Determine the appropriate shooting distance and angle interval based on the shape and size of the furnace. For large furnaces, a longer shooting distance and smaller angle interval may be required to ensure that each image contains enough furnace information; for furnaces with complex shapes, special attention should be paid to the shooting of special structural parts (such as corners, protruding parts, etc.) to ensure that these parts can be clearly presented in images from multiple angles.

[0081] Use appropriate filtering algorithms to remove noise from the image. For example, Gaussian filtering can be used for Gaussian noise, and median filtering can be used for salt and pepper noise. These filtering algorithms can make the image smoother without destroying the main features of the image, which is conducive to subsequent feature extraction and analysis.

[0082] Optionally, image enhancement technology is used to highlight the contour and key features of the furnace body. For example, the contrast of the image is adjusted by histogram equalization to make the edges and details of the furnace body more obvious; or a sharpening filter is used to enhance the edge information of the image and improve the recognition of the features.

[0083] Check and correct possible geometric distortion in the image. For images taken with a wide-angle lens, barrel or pincushion distortion may occur. This distortion can be corrected using camera calibration techniques and distortion correction algorithms (such as checkerboard-based calibration methods) to ensure that the geometric shape in the image truly reflects the actual shape of the furnace.

[0084] Before image registration, the image is normalized, for example, the pixel size, brightness and color space of the image are unified within a standard range. This helps to improve the accuracy and efficiency of the image registration algorithm and reduce the registration error caused by the difference between images. For the extraction of furnace contour features, edge detection algorithms such as the Canny edge detection algorithm can be used. The Canny algorithm can accurately detect the edges of the furnace according to the gradient information of the image, and by setting a suitable threshold, the accuracy of edge detection can be controlled to reduce the appearance of false edges. Combined with corner detection algorithms, such as the Harris corner detection algorithm or the Shi-Tomasi corner detection algorithm, the position of the corners of the furnace is determined. Corner points are important feature points of the furnace contour and play a key role in building a three-dimensional model and calculating the size of the furnace. These algorithms can find pixels with significant changes in the image, that is, corner points, and accurately locate the position of the corner points. A small convolutional neural network (CNN) can be trained to extract the features of the furnace. The CNN is trained with a large amount of furnace image data so that it can learn various feature patterns of the furnace. For example, the network can automatically extract the texture features and shape features of the furnace body. These features can be combined with traditional edge and corner features to provide more comprehensive feature information and improve the accuracy of feature extraction.

[0085] Feature-based image registration methods are used, such as SIFT (Scale Invariant Feature Transform) or SURF (Speeded Up Robust Features) algorithms. These algorithms can extract feature points with scale and rotation invariance in images at different angles, and then determine the transformation relationship between images through feature point matching, thereby achieving accurate image registration. After feature point matching, optimization algorithms such as least squares are used to calculate the optimal transformation parameters between images, such as translation, rotation, and scaling parameters. This can further improve the accuracy of image registration, ensure that images at different angles can be accurately spliced ​​together, and provide a basis for building an accurate three-dimensional model of the furnace.

[0086] The multi-view stereo vision (MVS) algorithm is used to construct the three-dimensional model of the furnace. The MVS algorithm can calculate the three-dimensional spatial position of each pixel in the image through the principle of triangulation based on the registered multi-angle images, thereby generating a three-dimensional model of the furnace. When using the MVS algorithm, it is important to select appropriate parameters, such as parallax range, matching window size, etc., to ensure that the generated three-dimensional model can accurately reflect the actual shape and size of the furnace. Combined with three-dimensional reconstruction software or tools, such as MeshLab, Open3D, etc., the constructed three-dimensional model is optimized and repaired. These software can smooth the model, remove redundant data, fill holes, etc., to improve the quality of the three-dimensional model, which is more conducive to the accurate extraction of the furnace size.

[0087] Therefore, in the three-dimensional model, the coordinate system and geometric relationship of the model are used to extract the size of the furnace body. For example, the length, width or height of the furnace body is determined by measuring the distance between two opposite edges in the three-dimensional model of the furnace body. You can use the measurement tools provided by the three-dimensional modeling software, or write a custom measurement script to ensure the accuracy of the size measurement. For the measurement of irregular-shaped furnace bodies or internal structure dimensions, the method of segmenting the model and measuring separately can be used. The three-dimensional model of the furnace body is segmented according to different structural parts, and then the dimensions of each part are measured separately, and finally these dimensional information are integrated to obtain complete furnace body dimension data. In addition, if there are design drawings of the furnace body or known standard dimensions as a reference, the extracted dimensions can be compared and verified with the reference dimensions. If there is a deviation, analyze the cause of the deviation, such as image acquisition error, model construction error, etc., and calibrate the dimension extraction method or model to improve the accuracy of dimension determination.

[0088] As an optional embodiment, in 101, after determining the furnace body size of the target device based on the multi-angle image information, the position offset of the position before and after the door is opened can be measured by a displacement sensor deployed at the furnace door and / or the furnace body frame; a corresponding motion trajectory is generated based on the position offset, and a first furnace body shape variable matching the motion trajectory is calculated according to the geometric shape of the structural component corresponding to the position; the pressure change in the furnace body is measured by a pressure sensor, and the second furnace body shape variable of the target device is predicted based on the pressure change and the material composition in the furnace body; and deformation compensation parameters are generated based on the first furnace body shape variable and the second furnace body shape variable.

[0089] When the furnace door is opened and closed, the displacement sensors deployed at the furnace door and / or the furnace frame can accurately record the change in position, that is, the position offset. These displacement sensors sense the change in the position of an object based on physical principles (such as capacitive, inductive or photoelectric displacement measurement principles). Based on these position offsets, the motion trajectory of the furnace door or related structural components can be depicted. Since the structural components of the furnace body have a specific geometric shape, for example, the furnace door may be rectangular, the deformation amount of the furnace body during the opening and closing of the furnace door can be calculated by geometrically analyzing the motion trajectory and combining the geometric shape and size information of the components, that is, the first furnace body deformation amount. This is based on the principle of geometric kinematics, linking the displacement change with the actual deformation of the furnace structure.

[0090] The pressure sensor is installed in the furnace to measure the pressure change in the furnace. According to the ideal gas state equation PV = nRT, when the material composition of the furnace is known (the amount of material n and the gas constant R can be determined) and the temperature (assuming that the temperature can be measured or known within a certain range), the change in pressure Delta P can reflect the change in the volume of the furnace. Because the volume of the furnace is closely related to the shape and size of the furnace, through this physical relationship, the change in the shape of the furnace caused by the pressure change can be predicted, that is, the second furnace shape variable. This is to use the principles of thermal and mechanical science to infer the deformation of the furnace from the perspective of internal pressure changes.

[0091] The deformation compensation parameters are generated by combining the first furnace body deformation variables and the second furnace body deformation variables. This is because the furnace body will be deformed due to the combined influence of multiple factors (such as mechanical movement and internal pressure changes) during actual operation. The purpose of the deformation compensation parameters is to quantify the impact of these deformations on the production process, so that these deformation factors can be considered in subsequent equipment parameter adjustment, thermal cycle control and other links, and the equipment operation parameters can be compensated and adjusted, thereby improving the accuracy of the production process and product quality.

[0092] In this way, by considering the deformation of the furnace body, the equipment parameters can be compensated more accurately. For example, in thermal cycle control, accurate furnace size is very important for the calculation of heat distribution and heat transfer. The equipment parameters after considering the deformation can more realistically reflect the actual state of the furnace body, thereby improving the accuracy of the thermal cycle control model in predicting the equipment operation state. For the glass production process, accurate equipment parameters help to ensure uniform heating and quality stability of the glass. Deformation compensation can avoid problems such as uneven temperature and unstable pressure caused by furnace body deformation, thereby improving the glass yield, optimizing the quality of glass products, and reducing product defects caused by furnace body deformation. Real-time monitoring of the furnace body deformation and compensation can timely discover potential problems of the equipment, such as damage to the furnace body structure that may be caused by excessive deformation. Through reasonable parameter adjustment, the damage to the furnace body caused by unreasonable stress distribution can be reduced, thereby extending the service life of the equipment and reducing equipment maintenance costs.

[0093] For example, displacement sensors and pressure sensors are connected to the IoT platform to achieve real-time data transmission and remote monitoring. On the IoT platform, a large amount of furnace displacement and pressure data can be collected and combined with other equipment parameters (such as temperature, production time, etc.) for big data analysis. Through machine learning algorithms (such as the long short-term memory network LSTM in deep learning for analyzing time series data), the complex relationship between furnace deformation and various factors can be mined. For example, an association model between furnace deformation variables and production batches, different types of glass products, etc. is established to more accurately predict furnace deformation and generate more optimized deformation compensation parameters.

[0094] The finite element simulation software is used to establish a three-dimensional finite element model of the furnace body. The data obtained by the displacement sensor and the pressure sensor are input into the finite element model as boundary conditions or loading conditions. Through simulation calculation, the stress distribution and deformation of the furnace body can be analyzed more accurately. At the same time, the finite element simulation results are fused with the actual sensor data, and the two data are weighted by using a data fusion algorithm (such as Kalman filtering) to obtain more accurate information on the furnace body deformation. This method combines the advantages of theoretical simulation and actual measurement, and can improve the reliability of deformation compensation parameters.

[0095] Smart materials (such as fiber Bragg grating sensors, shape memory alloys, etc.) are embedded in the furnace structure materials. These smart materials can sense physical quantities such as their own strain and temperature changes. The perception data of these smart materials are combined with the data of displacement sensors and pressure sensors to form a perception system with multi-source data fusion. For example, fiber Bragg grating sensors can directly measure the strain inside the furnace material, providing more direct internal information for the calculation of the furnace shape variable. Through the self-sensing characteristics of smart materials and multi-sensor data fusion, more comprehensive and accurate monitoring of furnace deformation can be achieved, thereby generating more effective deformation compensation parameters.

[0096] Further optionally, the thermal cycle control model is a lightweight model installed in the target device for learning the real-time situation of the production environment.

[0097] For example, the model can also be deployed in cloud computing nodes, or deployed in edge extreme nodes.

[0098] It is understandable that cloud computing provides powerful computing resources and storage capabilities. The thermal cycle control model is deployed on the cloud computing node, and the model can take advantage of the high-performance processors, large-capacity memory, and high-speed storage devices of the cloud computing center. When the target device collects the device parameters, these data are transmitted to the cloud computing node through the network. The model on the cloud computing node processes and analyzes a large amount of real-time data based on its powerful computing power. In this process, the model can access the massive historical data stored in the cloud, which includes information such as glass yield under different equipment parameter combinations, for model training and optimization. At the same time, the distributed computing architecture of cloud computing allows the model to operate in parallel on multiple computing nodes, improving the speed and efficiency of prediction.

[0099] From the perspective of learning the real-time situation of the production environment, cloud computing nodes can receive data from multiple target devices, which enables the model to learn a wider range of changes in the production environment. For example, the differences in equipment parameters in different regions and production batches and the corresponding changes in glass yield can be used as material for model learning, so that the model can better adapt to the complex and changing production environment.

[0100] Edge computing is a computing mode that stores computing and data close to the data source or user end. When the thermal cycle control model is deployed on the edge computing node, the edge computing node is directly connected to the target device, and the data is processed locally without transmitting all data to the cloud or remote data center. The edge computing node performs rapid prediction and analysis locally based on the real-time device parameters of the target device. This is based on the certain computing power of the edge computing node itself, which can run a lightweight thermal cycle control model.

[0101] For learning the real-time situation of the production environment, edge computing nodes can quickly perceive and process local environmental changes around the target device. For example, at the production site, the immediate state changes of the equipment (such as sudden temperature fluctuations, equipment failures, etc.) can be captured by the edge computing node in a timely manner, and based on the rapid calculation of the local model, the corresponding prediction and adjustment strategy can be quickly given. Because data processing is performed locally, the impact of network transmission delay on real-time performance is reduced, allowing the model to respond to changes in the production environment more promptly. In this way, multiple target devices can share the resources of the cloud computing center, and enterprises do not need to configure high-end computing hardware for each device to run complex models. This reduces the hardware cost and maintenance cost of the equipment. At the same time, the pay-as-you-go model of cloud computing allows enterprises to pay according to actual usage, further optimizing the cost structure.

[0102] Since cloud computing nodes can process large amounts of data, the model can be continuously trained and optimized with new data from different devices, thereby improving the accuracy and generalization of the model. For example, the model can better learn the complex relationship between equipment parameters and glass yield under different production conditions, providing support for generating more accurate optimization strategies.

[0103] Through the cloud computing platform, production managers can remotely monitor and manage target equipment from anywhere with an Internet connection. They can view equipment parameters, glass yield prediction information, and model operation status in real time, making it easier to make decisions and adjustments in a timely manner.

[0104] At the production site, edge computing nodes can achieve rapid data processing and decision-making, reducing delays caused by data transmission and cloud processing. For example, when urgent adjustments need to be made to the parameters of the equipment's running program to cope with sudden changes in the production environment, edge computing nodes can quickly generate and execute optimization strategies, effectively avoiding production accidents or product quality degradation.

[0105] From the perspective of data security, some production data may contain sensitive information of the enterprise, such as specific process parameters of the equipment, production formulas, etc. Edge computing limits data processing to the local area, reduces the risk of privacy leakage during data transmission, and better protects the company's business secrets. Even in the case of unstable network connection or failure of the cloud computing center, the edge computing node can still operate independently to ensure the basic control function of the target device. This improves the autonomy and reliability of the production system and reduces dependence on external networks and cloud computing centers.

[0106] In 102, the equipment parameters are predicted in real time through the thermal cycle control model to obtain the glass yield prediction information of the target equipment under different parameter settings. As an optional embodiment, in 102, the equipment parameters are sampled by uniform design method; the spatial correlation characteristics between the various production links in the target equipment are extracted from the sampled furnace body size; the process correlation characteristics between different process parameters are extracted from the sampled walking program parameters; different process parameters are mutually correlated with different production links; the temporal dependency between the spatial correlation characteristics and the process correlation characteristics is obtained; according to the spatial correlation characteristics, the process correlation characteristics and the corresponding temporal dependency, the glass yield prediction information of the target equipment under different parameter settings is predicted.

[0107] In this application, the principle of the thermal cycle control model is mainly based on the following aspects: heat transfer mainly includes three modes: heat conduction, heat convection and heat radiation. In the thermal cycle control model, it is necessary to consider the heat transfer process between different media and components. For example, in the heating stage, heat is transferred from the heating source to the target object or medium by conduction; in the cooling stage, heat is dissipated to the surrounding environment by convection or radiation. Understanding the law and rate of heat transfer helps to accurately control the temperature change during the thermal cycle, thereby realizing the thermal management of the entire system. Different substances have different thermophysical properties, such as specific heat capacity, thermal conductivity, thermal expansion coefficient, etc. These properties will affect the temperature response and thermal deformation of the substance during the thermal cycle. For example, a substance with a large specific heat capacity needs to absorb or release more heat during heating or cooling, and the temperature change is relatively slow; while a substance with high thermal conductivity can transfer heat faster and make the temperature distribution more uniform. In the thermal cycle control model, it is necessary to consider the thermophysical properties of the target device and related media in order to more accurately predict and control the temperature change and thermal deformation during the thermal cycle. The heat balance equation is the basic equation that describes the balance of heat income and expenditure during the thermal cycle. According to the law of conservation of energy, in a closed or open system, the input heat equals the output heat plus the heat stored or released inside the system. By establishing a heat balance equation, the heat transfer and conversion of each link in the thermal cycle process can be quantitatively analyzed, thus providing a theoretical basis for the thermal cycle control model. For example, in a simple heating-cooling cycle, the heat input in the heating stage is equal to the sum of the heat absorbed by the object and the heat lost to the surrounding environment during the heating process; the heat released by the object in the cooling stage is equal to the sum of the heat absorbed by the cooling medium and the heat lost to the surrounding environment during the cooling process. The thermal cycle control model involves multiple parameters, such as heating power, cooling rate, heat exchange coefficient, time constant, etc. The accurate estimation of these parameters is crucial to the accuracy and reliability of the model. These parameters can be estimated and optimized through experimental data or actual operation data. For example, the least squares method, maximum likelihood estimation and other methods are used to fit the parameters in the model according to the actual measured temperature data, so that the model can better reflect the actual thermal cycle process.

[0108] With the development of big data technology and machine learning algorithms, data-driven modeling methods have also been widely used in thermal cycle control. By collecting a large amount of temperature, pressure, flow and other data during the thermal cycle, a thermal cycle control model is established using machine learning algorithms (such as neural networks, support vector machines, decision trees, etc.). This method does not require an in-depth understanding of the physical mechanism of the thermal cycle process, but directly mines the relationship between input and output from the data, and has strong adaptability and predictive ability. For example, by learning and analyzing historical data, a mapping relationship between equipment parameters and glass yield can be established, thereby achieving optimal control of thermal cycles in the glass production process.

[0109] Feedback control is one of the commonly used control strategies in thermal cycle regulation models. By installing temperature sensors, pressure sensors and other detection elements in the thermal cycle system, the system's temperature, pressure and other state variables are monitored in real time, and these measured values ​​are fed back to the controller. The controller calculates the corresponding control quantity, such as heating power, cooling water volume, etc., based on the deviation between the set value and the feedback value to adjust the operating state of the system so that the system output can track the set value. For example, in a constant temperature control system, when the temperature sensor detects that the actual temperature is lower than the set temperature, the controller will increase the heating power to increase the temperature; when the actual temperature is higher than the set temperature, the controller will reduce the heating power or increase the cooling water volume to lower the temperature, thereby maintaining temperature stability.

[0110] Feedforward control is a prediction-based control strategy that calculates the corresponding control quantity in advance according to the changes in the system input or interference signal to compensate for the impact of these changes on the system output. In thermal cycle control, feedforward control can be used to predict and compensate for some measurable interference factors, such as changes in ambient temperature and load. For example, when the ambient temperature rises, the feedforward controller can reduce the heating power or increase the cooling water volume in advance to offset the impact of ambient temperature changes on the system temperature and improve the control accuracy and stability of the system.

[0111] The goal of optimal control theory is to find a control strategy that optimizes system performance indicators under certain constraints. In thermal cycle control, glass yield, energy consumption, equipment life, etc. can be used as performance indicators, and the optimal control parameters and control strategies can be solved by establishing corresponding optimization models. For example, by using optimal control algorithms such as dynamic programming and variational methods, thermal cycle parameters can be dynamically adjusted under different production stages and working conditions to achieve efficient, energy-saving and high-quality operation of the glass production process.

[0112] Specifically, in 102, the uniform design method is a sampling method that can obtain sufficient information with a smaller number of tests in a multi-factor, multi-level experimental design. In the thermal cycle control model, this method is used to sample equipment parameters (furnace size and walking program parameters) to ensure that the samples are evenly distributed in the parameter space. The purpose of this is to more comprehensively cover various possible parameter combinations, avoid local oversampling and ignore other areas, and provide a rich and representative data foundation for the subsequent accurate mining of the relationship between parameters.

[0113] For example, when the uniform design method is used for sample sampling, in order to ensure that the sampling method can cover the combination range of all key parameters, the following strategies and algorithms can be used to avoid sampling bias, thereby improving the training effect of the model. First, parameter definition and discretization. First, clarify the value range of all key parameters (such as furnace length, walking program settings, glass thickness, etc.). Reasonably discretize each parameter, that is, divide the range of each parameter into multiple intervals according to the actual situation, so that these intervals can be evenly covered in subsequent sampling. Alternatively, Latin hypercube sampling (LHS) is an effective uniform sampling method that can ensure that the value of each parameter is evenly distributed within its range. The specific approach is to divide the value range of each parameter into the same number of sub-intervals, and then randomly select a value so that each sub-interval is selected at least once. Alternatively, space-filling design methods, such as Sobol sequence, Halton sequence, etc., these low-difference sequences can help achieve uniform distribution in multidimensional space, which helps to fully explore the parameter combination space. Alternatively, response surface method (RSM) is a common strategy for optimization and modeling. By building a preliminary response surface model, the areas where key parameters affect the results can be identified, thus providing a basis for further uniform design. Alternatively, the samples generated by cross-validation and model evaluation need to be cross-validated to evaluate the robustness and predictive ability of the model to ensure the rationality of the sampling method. By training and validating different samples, the distribution and selection of samples can be adjusted to make them more uniform. Alternatively, in the initial iteration of the model, the Monte Carlo method can be used for random sampling to generate a large number of samples and evaluate their impact on the model, and then the best performing samples can be selected for uniform design. Through these strategies, the uniform distribution of samples can be effectively achieved, reducing the impact of sampling bias on the model training effect.

[0114] Furthermore, in the extraction of spatial correlation features (in terms of furnace size), the furnace size determines the spatial layout of each production link in the target equipment. By analyzing the furnace size, we can understand the spatial relationship between different production links. For example, spatial factors such as the relative position and distance between the heating area and the cooling area, and the length of the glass transmission path in the furnace will affect the thermal cycle process. Extracting these spatial correlation features can provide the thermal cycle control model with information about how heat is distributed and transferred in space, because different spatial layouts will lead to different heat transfer paths and efficiencies.

[0115] Next, in the process correlation feature extraction (walking program parameters), the different process parameters in the walking program parameters represent the various links in the production process. The process correlation features between these process parameters are extracted to clarify the sequence and mutual influence between different process links. For example, the time and temperature parameters of the glass in the heating link are interrelated with the parameters of the cooling link, and the parameter changes in one link will have a chain reaction on the subsequent links. These process correlation features help to build a complete production process model, so that the thermal cycle control model can understand the dynamic changes of the production process.

[0116] Since glass production is a time-ordered process, different production links and process parameters influence each other on the time axis. To obtain the temporal dependency between spatial correlation features and process correlation features is to clarify how the spatial layout and process flow interact with each other at different time points. For example, in the early stage of heating, the spatial layout inside the furnace will affect the initial distribution of heat, and over time, changes in process parameters (such as heating power and transmission speed) will change the distribution of heat in space. This temporal dependency is the core part of the thermal cycle control model, which can simulate the dynamic evolution of the entire production process from start to finish.

[0117] Further optionally, parameter association rules are formulated in combination with the professional knowledge of glass production and the production rules accumulated over a long period of time. For example, according to the principle of heat conduction, during the heating process, if the glass thickness increases, in order to ensure the uniform temperature inside the glass, it is necessary to appropriately reduce the transmission speed and increase the heating time. These rules clarify the mutual correlation between different parameters under different production conditions, and are an important basis for extracting process correlation features. For example, an expert system is established to organize the experience and knowledge of glass production experts into a knowledge base. The knowledge base stores knowledge such as parameter adjustment rules, troubleshooting methods, and correlations between parameters in various production scenarios. Through the reasoning engine of the expert system, the process correlation features between parameters are automatically extracted according to the current production parameters and status. For example, when there is a problem of uneven temperature on the glass surface, the expert system can infer from the knowledge in the knowledge base that it may be caused by the improper matching between the transmission speed and the heating power, and extract the correlation features between the two parameters.

[0118] Finally, based on the extracted spatial correlation features, process correlation features and the temporal dependency between them, the thermal cycle control model can simulate the entire glass production process under different parameter settings. By simulating these processes, we can analyze whether the thermal cycle in the glass production process is reasonable under various parameter combinations, whether there is local overheating or overcooling, whether the production process is smooth, and other factors. These factors directly affect the quality of the glass, and then predict the glass yield of the target equipment under different parameter settings. For example, if the simulation process shows that under a certain parameter combination, the temperature of the glass changes evenly during heating and cooling, and the various links of the production process are closely connected, then the predicted glass yield may be high; conversely, if the temperature changes sharply and the production link is interrupted during the simulation, the yield may be low.

[0119] In this way, by considering the spatial correlation, process correlation and time-series dependency between equipment parameters, the thermal cycle control model can simulate the actual process of glass production more accurately. This comprehensive analysis makes the predicted glass yield closer to the actual situation, avoids the prediction error caused by considering only a single parameter or a simple parameter combination, and provides a more reliable basis for production decisions. It can clearly show the relationship between different parameters and their impact on the glass yield, which helps production personnel find the key parameter combinations and parameter adjustment directions that can improve the yield. For example, through analysis, it can be determined which combination of walking program parameters can achieve the best thermal cycle effect under a specific furnace space layout, thereby providing specific guidance for optimizing equipment parameter settings and improving production efficiency and product quality. This prediction method based on correlation characteristics and time-series dependency makes the thermal cycle control model more versatile and adaptable. No matter how the specific structure and parameters of the equipment change, as long as the corresponding correlation characteristics and time-series relationships can be extracted, the model can analyze and predict them. This has important application value for glass production equipment of different types and sizes, and can quickly adapt to new production environments and equipment updates.

[0120] Further optionally, after predicting the glass yield prediction information of the target device under different parameter settings, the method further includes:

[0121] The dynamic configuration information of the thermal cycle control model is obtained by minimizing the error function; the minimizing error function is expressed as the following formula: ;

[0122] in, is the sample size, is the standard glass yield of the i-th sample, The predicted glass yield of the i-th sample, represents the change in the adjacent sample error of the i-th sample, Yes all The average value of and is a hyperparameter that weighs the importance of these two parts. It is used to control the weight of the sum of squares of the error between the predicted value and the true value. The weight used to control the error change trend, C represents the model complexity, is the weight hyperparameter of the model complexity penalty term;

[0123] The thermal cycle control model is controlled in real time based on the dynamic configuration information to improve the prediction accuracy of the thermal cycle control model.

[0124] Exemplarily, the dynamic adjustment of the above model can start with a smaller learning rate, such as 0.001 or 0.01. A learning rate that is too large may lead to unstable training, while a learning rate that is too small may lead to slow convergence. Gradually reduce the learning rate during training, and use a learning rate decay strategy (such as halving the learning rate every certain epoch) to improve the convergence of the model. Adaptive learning rate optimization algorithms (such as Adam, RMSprop, etc.) can automatically adjust the learning rate according to the change of the gradient, thereby improving training efficiency. You can try different numbers of hidden layer nodes (such as 10, 20, 50, 100, etc.) through the grid search method, and use cross-validation to evaluate the performance of each configuration. In general, the number of hidden layer nodes can be set to 1 to 2 times the number of input features, but the specific number needs to be adjusted according to the complexity of the data set. Start training with fewer hidden layer nodes, gradually increase the number of nodes, and observe the performance of the validation set until the model performance no longer improves.

[0125] As for the activation function, ReLU (Rectified Linear Unit) can be used because it performs well when training deep networks and is computationally efficient. If there is a "dead neuron" problem (that is, some neurons always output 0 during training), you can consider using Leaky ReLU. Sigmoid / Tanh are suitable for smaller networks or when the output needs to be within a specific range, but they are prone to gradient vanishing problems in deep networks.

[0126] Set a range for each hyperparameter and try all possible combinations. This method can find the global optimal solution, but the computational cost is high. Randomly select hyperparameter combinations for training. Compared with grid search, random search is more effective in high-dimensional space and has a lower computational cost. Use Bayesian model to select hyperparameters, and gradually select the optimal hyperparameter combination by building a probabilistic model between hyperparameters and model performance. Compared with grid and random search, Bayesian optimization can explore the hyperparameter space more efficiently. At each hyperparameter selection, use K-fold cross validation to evaluate the performance of the model to ensure that the selected hyperparameters perform well under different data partitions. Plot the learning curves of training and validation loss to observe whether the model is overfitting or underfitting, and adjust the hyperparameters accordingly. Monitor the model performance on the validation set and use early stopping to avoid overfitting. When faced with a large number of input features, feature selection methods (such as LASSO regression, tree model importance evaluation, etc.) can be used to reduce the number of features and retain the features that have the greatest impact on prediction. Transform or combine features based on domain knowledge to extract meaningful features to reduce model complexity. The experiment records the hyperparameter settings and results of each experiment for subsequent analysis and comparison. The hyperparameters are continuously adjusted according to the experimental results to form an iterative optimization process. Through the above strategy, appropriate hyperparameters can be effectively selected during the training process of the BP neural network to improve the training efficiency and prediction accuracy of the model, ensuring that the model can stably predict the glass yield when dealing with complex nonlinear relationships and a large number of input features.

[0127] In practice, when building and training the above models, it is crucial to ensure the generalization ability of the model to avoid overfitting, especially when facing a variety of different combinations of furnace lengths and glass thicknesses. The following methods can be used to improve the generalization ability of the model and ensure its stable prediction ability under different parameters. By increasing the diversity of training data, the model can learn a wider range of features. Adding L1 (Lasso) or L2 (Ridge) regularization terms to the loss function can effectively limit the complexity of the model and avoid overfitting. Randomly "discarding" a part of the neurons during training reduces the model's dependence on specific neurons, thereby improving the robustness of the model. Divide the dataset into K subsets and perform K training and validation, using a different subset as the validation set each time. In this way, the performance of the model can be evaluated more comprehensively and a more stable generalization ability can be obtained. Monitor the loss of the validation set during training, and stop training early when the validation loss no longer decreases. This can prevent the model from overfitting on the training set. Choose an appropriate network structure to avoid overly complex models. Model complexity can be reduced by reducing the number of hidden layers or the number of neurons in each layer. Choosing an appropriate activation function (such as ReLU, Leaky ReLU, etc.) can improve the nonlinear expression ability of the model, but avoid using overly complex activation functions. You can consider using an integration of multiple models (such as Bagging, Boosting, etc.) to improve the overall prediction stability and generalization ability by combining the prediction results of multiple models. Improve the generalization ability of the model by systematically searching the hyperparameter space, such as GridSearch / Random Search, to find the best learning rate, batch size, regularization parameters, etc. During the training process, regularly use an independent validation set to evaluate the model performance to ensure that the model performs well on unseen data. Plot the learning curves of training and validation loss to observe whether the model shows signs of overfitting (such as training loss continues to decrease while validation loss increases). Use the test set to evaluate the prediction ability of the model under different combinations of furnace length and glass thickness to ensure that the model can maintain stable prediction performance under different conditions. By analyzing the impact of each feature on the prediction results, ensure that the model can reasonably capture the impact of different parameters on the yield rate. Through the above method, the generalization ability of the BP neural network model can be effectively improved, the risk of overfitting can be reduced, and it can be ensured that the model can stably predict the yield rate when facing different combinations of furnace lengths and glass thicknesses.

[0128] In 103, the optimization strategy is used to adjust the walking program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0129] The optimization strategy includes equipment parameters corresponding to the predicted glass yield rate that meets the preset glass finished product evaluation standard; the glass finished product evaluation standard is related to the equipment production demand information of the target equipment. Further optionally, the equipment production demand information at least includes: glass thickness and glass material composition.

[0130] In the glass production process, the equipment's program parameters directly affect the thermal cycle speed, which in turn plays a key role in the quality and production efficiency of the finished glass. The optimization strategy is generated based on an in-depth analysis of the relationship between equipment parameters and glass yield.

[0131] In principle, the equipment production demand information (such as glass thickness and glass material composition) determines the evaluation criteria for finished glass products. The heat conduction time and method of glass with different thicknesses will vary. Thicker glass may require slower thermal cycle speeds and longer heating times to ensure uniform internal and external temperatures. The glass material composition affects the thermophysical properties of the glass, such as specific heat capacity, thermal expansion coefficient, etc. Different material compositions have different characteristics of heat absorption, transfer and release during heating and cooling.

[0132] After the glass yield rate under different parameter settings is predicted by the thermal cycle control model, the equipment parameter combinations that can meet the preset glass product evaluation standards will be screened out as optimization strategies. The equipment parameter adjustments corresponding to these parameter combinations can achieve fine control of the thermal cycle speed. For example, if the optimization strategy requires increasing the heating power or reducing the glass transmission speed, the purpose is to produce the glass at a suitable thermal cycle speed so that the glass can achieve an ideal temperature change curve during the heating and cooling process, thereby meeting the glass product evaluation standards.

[0133] Specifically, the control of the thermal cycle speed is achieved by adjusting the various process parameters in the walking program parameters. These process parameters are interrelated and jointly affect the speed and effect of the thermal cycle. For example, adjusting the power of the heating area and the residence time of the glass in the heating area can control the thermal cycle speed in the heating stage; changing the flow rate of the cooling medium and the conveying speed of the glass in the cooling area can control the thermal cycle speed in the cooling stage. By adjusting these parameters in a coordinated manner through optimization strategies, the thermal cycle speed can be matched with the production requirements of the glass.

[0134] This method of controlling the thermal cycle speed by adjusting the walking program parameters based on the optimization strategy will bring many positive effects. In terms of product quality, since the thermal cycle speed is precisely controlled, it can better meet the evaluation standards of glass products. The thickness uniformity and material properties of the glass can meet the expected requirements. For example, the quality indicators of the glass such as hardness, transparency, and flatness will be more stable and meet the standards. By ensuring that the glass is produced under appropriate thermal cycle conditions, glass defects such as cracking, deformation, and internal stress concentration caused by uneven thermal stress and rapid temperature changes can be reduced, thereby improving the quality of the glass products. In terms of production efficiency, a reasonable thermal cycle speed can shorten the production cycle. When the thermal cycle speed is precisely matched with the glass production requirements, it can avoid the situation where the production time is too long due to too slow thermal cycles, or a large number of defective products need to be reworked due to too fast thermal cycles. The optimized thermal cycle speed enables the equipment to produce more qualified glass products per unit time, thereby improving the production efficiency of the target equipment.

[0135] In addition, from the perspective of long-term operation and energy consumption of the equipment, adjusting the walking program parameters through optimization strategies can make the equipment operate under more reasonable working conditions. This helps to extend the service life of the equipment and reduce the wear and damage of the equipment caused by unreasonable thermal cycle processes. At the same time, precise thermal cycle control can avoid energy waste, such as avoiding overheating or overcooling, thereby reducing energy consumption, improving energy utilization efficiency, and further improving the economic benefits of production.

[0136] For example, when determining the equipment parameters in the optimization strategy, first, the evaluation criteria for finished glass products are analyzed in detail. Taking glass thickness as an example, if the thickness of the glass to be produced is required to be a specific value, such as 5mm, this means that the heating and cooling rates during the thermal cycle need to be adjusted to a degree that allows the glass to be evenly heated and cooled in the thickness direction. For different glass material components, their thermal conductivity, specific heat capacity and other thermophysical properties are different, and appropriate thermal cycle parameters need to be determined based on these properties. For example, glass with a high silicon content may require a higher heating temperature and a longer heating time to reach the softening point because its thermal conductivity is relatively low. Through the analysis of these production requirements, the approximate range of equipment parameters is preliminarily determined, such as the range of values ​​of parameters such as heating power, heating time, cooling rate, and glass conveying speed. Using physical models and empirical formulas, an association model between glass product quality indicators and equipment parameters is established. For example, according to the heat conduction equation, the temperature distribution inside the glass is related to the heating power, heating time, and the thermophysical properties of the glass itself. Mathematical relationships can be established between quality indicators of glass, such as hardness and transparency, and key parameters in the thermal cycle process. For example, hardness may be related to cooling rate and final cooling temperature, and transparency may be related to temperature uniformity during heating. These association models can provide a theoretical basis for determining equipment parameters. The glass yield information under different parameter settings predicted by the thermal cycle control model is used to screen out parameter combinations with high yields and that meet the glass finished product evaluation standards. For example, if the model predicts that under the parameter combination of heating power P1, heating time T1, and transmission speed V1, the glass yield reaches more than 90%, and the thickness and material properties of the produced glass meet the evaluation standards, then this parameter combination is a potential optimized parameter combination.

[0137] In addition to yield and quality standards, the stability and robustness of parameter combinations must also be considered. In the actual production process, equipment parameters may be affected by some external factors (such as ambient temperature changes, power supply fluctuations, etc.). Therefore, it is necessary to select parameter combinations that can still maintain a high yield and stable quality when fluctuating within a certain range. For example, by simulating parameter changes under different interference factors, observing the fluctuations in glass yield and quality, select parameter combinations with smaller fluctuations as equipment parameters in the optimization strategy.

[0138] Conduct small-scale experiments on actual production equipment, adjust the equipment's running program parameters according to the selected parameter combination, and observe the glass production process and finished product quality. For example, produce a batch of glass first, and record the actual glass yield, thickness deviation, material performance changes, and other data under the new parameter combination. Verify the effectiveness of the parameter combination by comparing it with the predicted results.

[0139] If the experimental results do not meet expectations, for example, the yield rate does not reach the predicted value or there is a problem with the glass quality, it is necessary to analyze the reasons and optimize the parameter combination. It may be that there is an error in the model prediction, or there are some unconsidered factors in actual production that affect the production process. According to the analysis results, fine-tune the equipment parameters, such as appropriately changing the heating power, adjusting the transmission speed, etc., and then conduct experimental verification again until the most suitable equipment parameter combination is found, so that it can stably meet the glass finished product evaluation standards and achieve a high yield rate.

[0140] Further optionally, in 103, before generating the optimization strategy of the walking program parameters based on the glass yield prediction information, a corresponding multi-branch deviation function can be constructed according to the glass yield prediction information and each glass finished product evaluation standard. Specifically, the multi-branch deviation function is at least expressed as the following formula: ;in, is the learning rate, n is the number of iterations, is the target optimization function, is the heating time, For the optimized heating time control parameters, is the heating time control parameter obtained by iterating n times, is the heating power, is the heating power control parameter obtained by iterating n times, is the optimized heating power control parameter, is the transmission speed, is the transmission speed control parameter obtained by iterating n times, is the transmission speed control parameter obtained by iterating n times. Further, in 103, the walking program parameter corresponding to the optimal glass yield rate when the glass finished product evaluation standard is qualified is selected through the multi-branch deviation function. The multi-branch deviation function is used to optimize the balance between the glass yield rate and each glass finished product evaluation standard.

[0141] Specifically, the purpose of constructing a multi-branch deviation function is to comprehensively consider the glass yield prediction information and various glass finished product evaluation standards. The glass yield prediction information reflects the probability of a qualified product under different equipment parameter settings, while the glass finished product evaluation standards cover key factors that determine product quality, such as glass thickness and material composition. The construction of this function is based on the control of multiple key parameters in the glass production process (such as heating time, heating power, and transmission speed). By introducing these parameters and their iterative update relationship in the target optimization function, we can find a balance point that can simultaneously meet the yield and quality standards.

[0142] From the formula structure, the learning rate controls the step size of the parameter iteration update, which determines the amplitude of the parameter adjustment in each iteration. The number of iterations n is used to record the progress of the optimization process. As the iterations proceed, the equipment parameters (heating time t, heating power p, transmission speed v) are continuously updated to try to find the optimal solution. The target optimization function J is a function that comprehensively measures the glass yield and quality deviation. It minimizes the deviation between the yield and the quality standard by adjusting the equipment parameters.

[0143] In the multi-branch deviation function, for each key equipment parameter, the difference between its current iteration value and the optimized value is considered. For example, for the heating time control parameter, t n + 1 (Optimized heating time control parameters) and t n The relationship between} (heating time control parameters obtained by iterating n times) reflects the adjustment strategy for heating time. Through this iterative update, the function can dynamically adjust the equipment parameters to find the optimal parameter combination that can achieve the optimal glass yield rate while meeting the glass product evaluation standards. The various parameters in the function are interrelated and jointly affect the glass production process. There are complex interactions between the three parameters of heating time, heating power and conveying speed. For example, increasing the heating power may shorten the heating time required to reach the target temperature, but if the conveying speed remains unchanged, it may cause uneven heating of the glass and affect the quality. The multi-branch deviation function optimizes the balance between the glass yield rate and each glass product evaluation standard by simultaneously considering the adjustment of these parameters and their combined impact on the yield rate and quality standards.

[0144] Through the multi-branch deviation function, the optimization of the walking program parameters can be decided more scientifically. It provides a quantitative method to find an optimal solution among many possible parameter combinations, so that the glass yield is as high as possible, while ensuring that the quality of the glass products meets the evaluation standards. For example, in the face of different production needs and equipment status, the function can automatically calculate the optimal adjustment direction and amplitude of parameters such as heating time, heating power and transmission speed based on the current glass yield prediction information and quality requirements. It effectively balances the relationship between the glass yield and the evaluation standards of each glass product. In the past, we may only focus on improving the yield and ignore the quality standards, or follow the quality standards too strictly, resulting in a decrease in the yield. The multi-branch deviation function can find an optimal balance between the two, ensuring that while improving the glass yield, the quality indicators such as glass thickness and material composition can still stably meet the requirements. This helps to produce high-quality, high-yield glass products and improve production efficiency. This function-based optimization strategy has strong flexibility and can adapt to different production scenarios and equipment conditions. Whether it is the debugging of new equipment, the upgrading of old equipment, the change of raw materials, the adjustment of product specifications, etc., the optimal walking program parameter combination suitable for the current scenario can be quickly found by adjusting the parameters in the multi-branch deviation function (such as learning rate, initial equipment parameters, etc.), thereby achieving effective control and optimization of the glass production process.

[0145] 104. Use the optimization strategy to adjust the walking program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0146] In the embodiment of the present application, the optimization strategy includes a combination of equipment parameters that have been screened and verified, and these parameters are mainly walking program parameters, such as heating time, heating power, transmission speed, etc. When these parameters are adjusted according to the optimization strategy, the heating and cooling of the glass during the thermal cycle will be directly changed. For example, increasing the heating power will speed up the heating rate of the glass, extending the heating time can keep the glass at a high temperature for a longer time, and adjusting the transmission speed will change the residence time of the glass in each heating and cooling area. The comprehensive adjustment of these parameters can accurately control the thermal cycle rate to match it with the optimal production conditions of the glass. From the perspective of heat transfer and thermal balance, the control of the thermal cycle rate is based on the regulation of heat input, output and heat accumulation process inside the glass. By adjusting the walking program parameters, the conduction, convection and radiation processes of heat in the glass can be changed. For example, an appropriate transmission speed can ensure that the glass fully absorbs heat in the heating area and dissipates heat in the cooling area in time, thereby maintaining a stable and efficient thermal cycle process. The thermal cycle rate is closely related to production efficiency. In glass production, a reasonable thermal cycle rate can make the production process more compact and efficient. When the thermal cycle speed is accelerated and the quality of glass can be guaranteed to be unaffected during this process, more glass products can be produced per unit time. For example, if the total time of glass in the entire thermal cycle process (including heating, forming, cooling, etc.) is shortened by optimizing parameter adjustment while maintaining the glass yield and quality, then within the same production time, the equipment can produce more qualified glass, thereby improving production efficiency.

[0147] In addition, the optimized thermal cycle speed can also reduce rework and scrap caused by unstable production process or quality problems. If the thermal cycle speed is not appropriate, it may cause quality problems in the glass, such as excessive internal stress, uneven surface, etc. These problems will make the product scrap or require rework. By controlling the thermal cycle speed, these problems can be effectively avoided, the one-time pass rate of the product can be increased, and production efficiency can be indirectly improved.

[0148] After adjusting the walking program parameters through the optimization strategy to control the thermal cycle speed, the most direct effect is a significant increase in output. Because the thermal cycle process is more reasonable and efficient, the equipment can complete more batches of glass production in a unit time, thereby increasing the total output of glass. For example, originally 1,000 pieces of glass could be produced a day, but after optimization, the output may be increased to 1,200 pieces or even more. The optimized thermal cycle speed can shorten the production cycle of a single glass product. This means that the time from raw material input to finished product output is reduced, and the turnover speed of the equipment is accelerated. For example, it originally took 10 hours to produce a piece of glass, but after optimization, it may only take 8 hours, which can greatly improve the utilization rate of equipment and overall production efficiency for large-scale production.

[0149] Reasonable control of the thermal cycle speed can make the glass more evenly heated during the production process and the cooling process more stable, thereby improving the quality stability of glass products. For example, the fluctuation range of quality indicators such as glass thickness uniformity, hardness, and transparency is reduced, and the product quality is more consistent, meeting the market's requirements for high-quality glass products. The optimized thermal cycle process can effectively reduce the quality defects of glass. For example, it can avoid problems such as glass breakage caused by thermal stress concentration and surface unevenness caused by uneven temperature, reduce the scrap rate and rework rate, and improve the product qualification rate.

[0150] When the thermal cycle rate is properly controlled, the equipment operates under more stable conditions, and the wear and fatigue of various components will be reduced. For example, the heating elements and cooling systems will not be damaged due to frequent overuse or rapid cooling and heating, thereby extending the service life of the equipment and reducing the cost of equipment maintenance and replacement. Accurate thermal cycle rate control can avoid energy waste. For example, by optimizing the heating time and power, the heat can be fully absorbed by the glass instead of being lost due to overheating. At the same time, the cooling resources can be reasonably used during the cooling process, reducing energy consumption and improving energy efficiency.

[0151] As an optional embodiment, during the production process, the current furnace length, travel program settings and glass thickness information are input into the trained thermal cycle control model in real time, and the thermal cycle control model outputs the predicted optimal cycle speed. The control system automatically adjusts the speed of the travel motor according to the prediction results to achieve dynamic control. In the real-time prediction and control process, the following steps and strategies can be taken to ensure that the prediction results of the neural network model accurately reflect the current production status and reliably adjust the speed of the travel motor accordingly:

[0152] Use incremental learning or online learning methods to continuously update the model as new data arrives. This ensures that the model can adapt to changes and new situations in the production process. Regularly retrain the model with new production data to maintain the accuracy and adaptability of the predictions. Use a high-frequency data acquisition system to ensure real-time monitoring of production status (such as sensor data, environmental parameters, etc.) and use this data as input to the neural network. Clean, normalize and feature extract the collected data to improve the input quality of the model. Use ensemble learning methods to combine the prediction results of multiple models to improve the stability and accuracy of the prediction. Introduce uncertainty assessment in the model to help judge the reliability of the prediction results by estimating the confidence interval of the prediction.

[0153] Adjust the speed of the travel motor in real time based on the prediction results and monitor the actual output. Incorporate actual feedback into the loop to form a closed-loop control system that can correct deviations in a timely manner. Optimize the control algorithm to ensure that the control system has a fast feedback loop and can adjust according to the new prediction results in the shortest time. Use lightweight models or quantized models to reduce inference time and ensure that the neural network can generate prediction results in a short time. Embed the neural network on the edge computing device to reduce the latency between the cloud computing or central processing unit.

[0154] Combining neural networks with traditional control algorithms (such as PID control), real-time adjustments can be made based on the prediction results to further improve the accuracy and stability of control. According to changes in the production environment, the control strategy is adaptively adjusted to respond to dynamic changes. Therefore, a monitoring system is established to continuously track the production status and the predicted output of the model, and anomalies can be quickly identified and adjusted. Anomaly detection and processing mechanisms are designed to ensure that they can automatically switch to safe mode or adopt backup control strategies when predictions are inaccurate or the system fails. Through the above methods, it can be ensured that the neural network can make accurate predictions in the real-time prediction and control process, and can efficiently and reliably adjust the speed of the travel motor to adapt to high-frequency production adjustments and feedback processes. Regularly evaluate the accuracy of the model's prediction results, and adjust the model parameters according to actual production feedback, such as increasing training data, optimizing the network structure, etc., to ensure that the model continues to be effective.

[0155] The optimal parameters can be selected through the above exemplary steps. For specific parameter values, please refer to Tables 1 to 3.

[0156] Table 1

[0157]

[0158] Table 2

[0159]

[0160] Table 3

[0161]

[0162] The parameters in the above table are similar to the walking program parameters introduced above, so they will not be elaborated here.

[0163] In the embodiment of the present application, the equipment parameters of the target device are obtained; the equipment parameters at least include: furnace size, walking program parameters; the walking program parameters include a variety of process parameters associated with the production links and equipment structural components of the target device; the equipment parameters are predicted in real time by a thermal cycle control model to obtain the glass yield prediction information of the target device under different parameter settings; the thermal cycle control model is a lightweight model installed in the target device for learning the real-time situation of the production environment; based on the glass yield prediction information, an optimization strategy for the walking program parameters is generated; the optimization strategy includes the equipment parameters corresponding to the glass yield prediction value that reaches the preset glass finished product evaluation standard; the glass finished product evaluation standard is related to the equipment production demand information of the target device; the equipment production demand information at least includes: glass thickness, glass material composition; the walking program parameters of the target device are adjusted by the optimization strategy to control the thermal cycle speed of the target device and improve the production efficiency of the target device. The present application can dynamically predict and control the walking program parameters, so as to control the thermal cycle speed of the target device and improve the glass production efficiency and glass quality.

[0164] Based on the same implementation principle, the embodiment of the present application provides a glass thermal cycle control device, which at least includes the following units:

[0165] A collection unit, used to obtain equipment parameters of the target equipment; the equipment parameters at least include: furnace size, running program parameters; the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment;

[0166] A prediction unit, used to predict the equipment parameters in real time through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings; the thermal cycle control model is a lightweight model installed in the target equipment for learning the real-time situation of the production environment;

[0167] A generating unit, configured to generate an optimization strategy for the walking program parameters based on the glass yield prediction information; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that reaches a preset glass finished product evaluation standard; the glass finished product evaluation standard is related to equipment production demand information of the target equipment; the equipment production demand information at least includes: glass thickness and glass material composition;

[0168] The adjustment unit is used to adjust the walking program parameters of the target device by adopting the optimization strategy to control the thermal cycle speed of the target device and improve the production efficiency of the target device.

[0169] The specific implementation of each unit in the above device is similar to the previous embodiment and will not be expanded here.

[0170] Based on the same implementation principle, the embodiment of the present application also provides an electronic device, Figure 2 is a structural block diagram of the electronic device 300, such as Figure 2 As shown, the electronic device 300 includes: a processor 301, a memory 302, a communication interface 303, a communication bus 304 and a controller 305; wherein the processor 301, the memory 302 and the communication interface 303 communicate with each other via the communication bus 304; the memory 302 is used to store computer programs; the processor 301 is used to execute the programs stored in the memory 302 to realize the corresponding processing functions; the communication bus 304 is used for communication between the electronic device 300 and other devices; the controller 305 is used to realize the glass thermal cycle control method described in the above method embodiment.

[0171] In the embodiment of the present application, the communication bus 304 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 304 may be divided into an address bus, a data bus, a control bus, etc., and the specific form is not limited. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0172] The memory 302 may include a random access memory (RAM) or a non-volatile memory (non-volatile memory), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far away from the processor 301 .

[0173] The processor 301 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other editable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the specific form can be determined according to actual needs.

[0174] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement each step that can be executed by an electronic device in the above method embodiment.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A glass thermal cycle control method, characterized in that: The method comprises: Obtaining equipment parameters of the target equipment; the equipment parameters at least include: furnace size, running program parameters; the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment; The equipment parameters are predicted in real time through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings; the thermal cycle control model is a lightweight model installed in the target equipment for learning the real-time situation of the production environment; Based on the glass yield prediction information, an optimization strategy for the walking program parameters is generated; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that reaches a preset glass finished product evaluation standard; the glass finished product evaluation standard is related to the equipment production demand information of the target equipment; the equipment production demand information at least includes: glass thickness and glass material composition; The optimization strategy is used to adjust the walking program parameters of the target device to control the thermal cycle speed of the target device and improve the production efficiency of the target device; Before obtaining the device parameters of the target device, the method further includes: Deploy multiple equipment sensors; the multiple equipment sensors are respectively installed in equipment structural components corresponding to multiple production links of the target equipment; The obtaining of device parameters of the target device includes: Collecting multi-angle image information of the target device, and determining the furnace size of the target device based on the multi-angle image information; Collecting walking program parameters of multiple dimensions from multiple equipment structural components as the equipment parameters; After determining the furnace size of the target device based on the multi-angle image information, the method further includes: By deploying a displacement sensor at the furnace door and / or the furnace frame, the position offset before and after the door is opened is measured; Generate a corresponding motion trajectory based on the position offset, and calculate a first furnace body shape variable matching the motion trajectory according to the geometric shape of the structural component corresponding to the position; Measuring the pressure change in the furnace body by a pressure sensor, and predicting the second furnace body shape amount of the target device based on the pressure change and the material composition in the furnace body; A deformation compensation parameter is generated based on the first furnace body deformation amount and the second furnace body deformation amount.

2. The method according to claim 1, characterized in that The step of determining the furnace size of the target device based on the multi-angle image information includes: Performing image preprocessing on the multi-angle image information; Identify key features about the size of the furnace body in the multi-angle image information; the key features at least include furnace body contour features; the furnace body contour features are used to indicate the positions of the edges and / or corners of the furnace body; Performing image registration on the key features, and constructing a corresponding three-dimensional model of the furnace body based on the registered key feature images; The furnace body size of the target device is extracted based on the furnace body three-dimensional model; the furnace body size is calculated based on a preset furnace body shape mapping relationship.

3. The method according to claim 1, characterized in that The walking program parameters include at least one of the following: heating time, rapid cooling time, cooling time, furnace entry interval, rapid cooling wind pressure, cooling wind pressure, furnace entry speed, furnace exit speed, furnace reciprocating speed, furnace adjustment speed, adjustment time, rapid cooling reciprocating speed, cooling reciprocating speed, and preheating time.

4. The method according to claim 1, characterized in that: The real-time prediction of the equipment parameters by the thermal cycle control model to obtain the glass yield prediction information of the target equipment under different parameter settings includes: The equipment parameters are sampled using a uniform design method; Extract the spatial correlation characteristics between various production links in the target equipment from the sampled furnace size; Extract the process correlation characteristics between different process parameters from the sampled walking program parameters; different process parameters are correlated with different production links; Acquire the temporal dependency relationship between the spatial correlation feature and the process correlation feature; According to the spatial correlation characteristics, the process correlation characteristics and the corresponding temporal dependency, the glass yield prediction information of the target device under different parameter settings is predicted.

5. The method according to claim 4, characterized in that After predicting the glass yield prediction information of the target equipment under different parameter settings, it also includes: The dynamic configuration information of the thermal cycle control model is obtained by minimizing the error function; the minimizing error function is expressed as the following formula: ; in, is the sample size, is the standard glass yield of the i-th sample, The predicted glass yield of the i-th sample, represents the change in the adjacent sample error of the i-th sample, Yes all The average value of and is a hyperparameter that weighs the importance of these two parts. It is used to control the weight of the sum of squares of the error between the predicted value and the true value. The weight used to control the error change trend, C represents the model complexity, is the weight hyperparameter of the model complexity penalty term; The thermal cycle control model is controlled in real time based on the dynamic configuration information to improve the prediction accuracy of the thermal cycle control model.

6. A glass thermal cycle control device, characterized in that: The device at least comprises the following units, wherein A collection unit, used to obtain equipment parameters of the target equipment; the equipment parameters at least include: furnace size, running program parameters; the running program parameters include a variety of process parameters associated with the production links and equipment structural components of the target equipment; A prediction unit, used to predict the equipment parameters in real time through a thermal cycle control model to obtain glass yield prediction information of the target equipment under different parameter settings; the thermal cycle control model is a lightweight model installed in the target equipment for learning the real-time situation of the production environment; A generating unit, configured to generate an optimization strategy for the walking program parameters based on the glass yield prediction information; the optimization strategy includes equipment parameters corresponding to the glass yield prediction value that reaches a preset glass finished product evaluation standard; the glass finished product evaluation standard is related to equipment production demand information of the target equipment; the equipment production demand information at least includes: glass thickness and glass material composition; An adjustment unit, used to adjust the walking program parameters of the target device by adopting the optimization strategy to control the thermal cycle speed of the target device and improve the production efficiency of the target device; The acquisition unit is further used to: deploy a plurality of equipment sensors before acquiring equipment parameters of the target equipment; the plurality of equipment sensors are respectively installed in equipment structural components corresponding to a plurality of production links of the target equipment; When acquiring the device parameters of the target device, the acquisition unit is specifically used to: acquire multi-angle image information of the target device, and determine the furnace body size of the target device based on the multi-angle image information; acquire multi-dimensional walking program parameters from multiple device structural components as the device parameters; wherein, after determining the furnace body size of the target device based on the multi-angle image information, it also includes: measuring the position offset of the position before and after opening the door by means of a displacement sensor deployed at the furnace door and / or the furnace body frame; generating a corresponding motion trajectory based on the position offset, and calculating a first furnace body shape variable matching the motion trajectory according to the geometric shape of the structural component corresponding to the position; measuring the pressure change in the furnace body by means of a pressure sensor, and predicting the second furnace body shape variable of the target device based on the pressure change and the material composition in the furnace body; generating deformation compensation parameters based on the first furnace body shape variable and the second furnace body shape variable.

7. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing the glass thermal cycle control method according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer software program, and when the computer software program is executed by a processor, the glass thermal cycle control method according to any one of claims 1 to 5 is implemented.

Images