A precise control system for molten glass flow on a patterned glass production line

Abstract

This invention discloses a precise control system for molten glass flow on a patterned glass production line, relating to the technical field of patterned glass production equipment. The invention uses multiple sensors to collect real-time data on kiln temperature, flow channel pressure, and liquid level. After moving average filtering and outlier removal, a digital twin layer deployed on an industrial edge server synchronously simulates the molten glass flow state. The predictive control layer uses an improved LSTM network incorporating digital twin features to predict the flow rate trend over the next 300 seconds. When a flow rate deviation exceeding 0.5% is detected for five consecutive time intervals, a pre-adjustment mechanism is triggered. The pre-adjustment amount of the regulating valve is calculated based on the pure time lag, and control commands are issued in advance. Simultaneously, the system is linked with the patterned roller speed control system to adjust the speed in real time to maintain a constant molten glass accumulation. It also features virtual testing and online self-calibration functions, thereby achieving feedforward pre-adjustment control of the flow rate and effectively overcoming the problem of lag in traditional feedback control.

Description

Technical Field

[0001] This invention relates to the field of patterned glass production equipment technology, specifically to a precise control system for the flow rate of molten glass on a patterned glass production line. Background Technology

[0002] Patterned glass is a functional glass product widely used in architectural decoration, home design, and other fields. Its surface is pressed with various exquisite patterns by embossing rollers. In the production process of patterned glass, after the molten glass flows out of the kiln, it is transported through the flow channel to the embossing rollers for embossing. The stability of the molten glass flow rate directly determines the thickness uniformity and pattern depth consistency of the final product, and is a core process parameter affecting the quality of patterned glass products.

[0003] However, existing patterned glass production lines mostly use traditional PID feedback control for glass melt flow control. This method only adjusts after a flow deviation has occurred, resulting in significant response lag and an inability to predict and pre-adjust before flow fluctuations. Furthermore, traditional control systems do not fully consider the non-Newtonian fluid characteristics of glass melt and the complex heat conduction effects within the flow channel, leading to limited control accuracy and frequent quality problems such as uneven glass thickness and inconsistent pattern depth. In addition, optimizing traditional process parameters relies on repeated on-site adjustments, which is time-consuming and labor-intensive, and lacks a linkage mechanism with the embossing roller speed. When flow fluctuates, the embossing roller speed cannot be adjusted in time to maintain a constant glass melt accumulation, further exacerbating product quality instability.

[0004] In summary, existing glass melt flow control technologies on patterned glass production lines have significant shortcomings in terms of control accuracy, real-time response, product quality stability, and process debugging efficiency. There is an urgent need for a precise glass melt flow control system that can achieve feedforward pre-adjustment, integrate digital twin prediction, and has the functions of patterned roller linkage control and virtual testing, so as to comprehensively improve the production quality and process efficiency of patterned glass. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a precise control system for molten glass flow in a patterned glass production line. This system achieves precise control of molten glass flow by constructing a five-layer architecture comprising a physical entity layer, a data acquisition and transmission layer, a digital twin layer, a predictive control layer, and an execution layer. The system utilizes multiple sensors to collect real-time data on kiln temperature, flow channel pressure, and liquid level. After moving average filtering and outlier removal, the digital twin layer deployed on an industrial edge server synchronously simulates the molten glass flow state. The predictive control layer uses an improved LSTM network incorporating digital twin features to predict the flow rate trend over the next 300 seconds. When a flow rate deviation exceeding 0.5% is detected for five consecutive time intervals, a pre-adjustment mechanism is triggered. The pre-adjustment amount of the regulating valve is calculated based on pure time lag, and control commands are issued in advance. Simultaneously, the system is linked with the patterned roller speed control system to adjust the speed in real time to maintain a constant molten glass accumulation. It also features virtual testing and online self-calibration functions, thereby achieving feedforward pre-adjustment control of the flow rate and effectively overcoming the lag problem of traditional feedback control.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a precise control system for molten glass flow on a patterned glass production line, the system comprising the following components:

[0007] Physical entity layer: includes kiln, flow channel, flow control valve and embossing roller;

[0008] Data Acquisition and Transmission Layer: Used to acquire real-time production data from the physical entity layer;

[0009] Digital twin layer: Communicatively connected to the data acquisition and transmission layer, used to synchronously simulate the flow state of molten glass in the flow channel based on the real-time production data, and generate molten glass flow prediction data for future times;

[0010] Predictive control layer: Communicatively connected to the digital twin layer, used to predict the trend of glass melt flow rate change within a preset time window using an improved long short-term memory network, and generate pre-adjustment instructions;

[0011] Execution layer: Communicatively connected to the predictive control layer, used to adjust the opening of the flow regulating valve according to the pre-adjustment command.

[0012] Furthermore, the data acquisition and transmission layer includes four type B thermocouples evenly arranged circumferentially at the kiln outlet. Each thermocouple is inserted 50mm into the molten glass, with a measurement range of 800-1600℃ and an accuracy of ±0.5℃. A high-temperature resistant ceramic pressure sensor is installed at the inlet of the flow channel, with a measurement range of 0-1MPa and an accuracy of ±0.001MPa. A laser level sensor is installed 1.5m above the surface of the molten glass above the flow channel, with a measurement range of 0-500mm and an accuracy of ±0.1mm. A non-contact electromagnetic induction flow sensor is installed at the outlet of the flow channel, based on Faraday's law of electromagnetic induction, measuring the induced electric current generated when the molten glass flows through a magnetic field. The flow rate is calculated using a sensor that does not directly contact the molten glass. The measurement range is 0-100 m³ / h with an accuracy of ±0.2%. An ambient temperature sensor is installed near the flow channel, with a measurement range of -20-80℃ and an accuracy of ±0.1℃. All sensors communicate with the digital twin layer via Profinet industrial Ethernet. The data sampling frequency is 20Hz, and the data transmission delay is less than 1ms. The collected raw data is first processed by moving average filtering with a window size of 5. The filtered data is then subjected to outlier removal. The outlier judgment criterion is: the arithmetic mean of the three data points before and after the current data point is used as the benchmark. If the current value deviates from the benchmark by more than three times the standard deviation, the point is removed.

[0013] Furthermore, the digital twin layer is deployed on an industrial edge server equipped with an Intel Xeon Gold 6248R processor and an NVIDIA RTX A4000 graphics card. First, a 3D laser scanner is used to scan the actual flow channel, obtaining a geometric model with an accuracy of 0.05 mm. Then, a glass melt flow dynamics model is constructed based on computational fluid dynamics methods, and the Navier-Stokes equations are solved using the finite element method. A hexahedral structured mesh is used for mesh generation, and boundary conditions are set as velocity inlet and pressure outlet. The RNG k-ε model is selected as the turbulence model, and a power-law model is used to describe the flow. The non-Newtonian fluid characteristics of molten glass are described. The viscosity-temperature relationship is fitted using the VFT equation. The operating parameter database stores physical parameters such as viscosity, density, and specific heat capacity of 10 commonly used glass formulations at different production speeds. The digital twin layer receives data from various sensors in real time during operation. The furnace temperature, channel pressure, molten glass level, and ambient temperature are used as boundary conditions and input into the flow dynamics model. The three-dimensional velocity distribution, pressure distribution, and temperature distribution of molten glass in the channel are calculated through finite element analysis with a calculation cycle of 100ms, achieving synchronous operation with the physical entity. The absolute value of the relative deviation between the model calculation results and the actual measured values ​​is less than 1%.

[0014] Furthermore, the predictive control layer employs an improved long short-term memory network that integrates digital twin features to construct a traffic prediction model, wherein the input feature vector... The output layer calculation formula includes at least the current furnace temperature, flow channel pressure, glass liquid level, actual flow rate, and ambient temperature. ,in for Predicting the future Glass melt flow rate per second; The weighting coefficients for digital twin features are dynamically adjusted based on the average prediction error of the digital twin model over the past 100 time steps. A smaller average prediction error results in a higher weighting coefficient. The smaller the value; The output function of the basic LSTM network; for The input feature vector at time step; The future predicted by digital twin models The flow rate of molten glass per second.

[0015] Furthermore, the predictive control layer acquires the latest input data every second, updates the flow prediction curve, calculates the flow prediction value per second for the next 300 seconds, and then calculates the deviation between the flow prediction value and the set value at each prediction time. The deviation is defined as (predicted flow - set flow) / set flow × 100%. When the absolute value of the flow deviation at 5 consecutive prediction times exceeds 0.5%, the pre-adjustment mechanism is immediately triggered to calculate the pre-adjustment amount of the flow regulating valve. The calculation of the pre-adjustment amount takes into account the pure time lag of the system. Specifically, the execution time of the pre-adjustment command is advanced by at least one pure time lag, so that the action of the regulating valve is synchronized with the occurrence of flow fluctuations. The pure time lag is pre-determined through a step response experiment. The pure time lag at different production speeds is stored in the operating condition parameter database. The system automatically calls the corresponding pure time lag parameter according to the current production speed to ensure that the pre-adjustment command reaches the execution layer before the actual flow fluctuation occurs.

[0016] Furthermore, the calculation formula for the pre-adjustment amount of the flow regulating valve generated by the predictive control layer is as follows: ,in for The pre-adjustment amount of the flow control valve at any given time. This is the proportionality coefficient. For pre-adjustment time window, for Predicting the future glass melt flow rate per second For pure time delay, Set the flow rate of the molten glass.

[0017] Furthermore, the virtual experiment function of the digital twin layer allows process engineers to input different furnace temperatures, glass liquid levels, and regulating valve opening parameters. Based on the same glass liquid flow dynamics model in the digital twin layer, the system automatically generates corresponding flow rate change simulation curves. During the simulation, the viscosity changes of the glass liquid, the heat conduction effect of the inner wall of the flow channel, and the dynamic response characteristics of the regulating valve are considered. The simulation results include flow velocity, pressure, and temperature data at any position in the flow channel, as well as the flow rate change curve over time. Process engineers can compare the flow rate stability and control effect under different parameter combinations, select the optimal process parameter combination, and then directly write the optimal parameters into the non-volatile memory of the predictive control layer via the OPCUA protocol. The parameters are automatically loaded at the start of the next production cycle, keeping the current control parameters unchanged during the loading process to achieve seamless switching.

[0018] Furthermore, the execution layer employs a high-precision electric flow control valve with a nominal diameter of 150mm, an adjustment accuracy better than 0.1%, and a full stroke time of less than 1 second. The control valve has a built-in position feedback device that can provide real-time feedback on the actual opening degree with a position feedback accuracy of 0.02%. After receiving the pre-adjustment command from the predictive control layer, the control valve drives the valve core to move via a built-in servo motor, achieving precise adjustment of the opening degree. The control valve also has a fault self-diagnosis function, which can detect the motor current, valve core position, and communication status in real time. When a fault is detected, an alarm signal is immediately issued, and the control valve is switched to manual control mode while maintaining the last effective opening degree. A fault alarm is also sent to the host computer to ensure production safety.

[0019] Furthermore, the digital twin layer has an online self-calibration function. The system automatically collects the actual flow data under stable operating conditions and the predicted flow data of the digital twin model every hour, calculates the deviation between the two, and automatically adjusts the viscosity coefficient and thermal conductivity coefficient in the glass melt flow dynamics model when the deviation exceeds 1%. The adjustment range is 0.5 times and 0.3 times the deviation amount, respectively, when the deviation exceeds 1%. The adjustment direction is to reduce the model prediction deviation. The larger the deviation, the larger the adjustment range. The adjusted model parameters are automatically saved to the operating condition parameter database for subsequent calculations. The self-calibration process does not require manual intervention and does not affect normal production.

[0020] Furthermore, the system also has a linkage function with the embossing roller speed control system. The predictive control layer sends the flow prediction value to the embossing roller speed control system in real time. The embossing roller speed control system automatically adjusts the speed of the embossing roller according to the flow prediction value. The adjustment ratio is determined according to the glass strip thickness setting value. When the flow increases, the embossing roller speed is increased accordingly, and when the flow decreases, the embossing roller speed is decreased accordingly, so that the amount of glass liquid accumulated on the embossing roller remains constant. The response time of the linkage control is less than 1 second, which effectively avoids the problems of uneven glass thickness and inconsistent pattern depth caused by flow fluctuations.

[0021] Compared with existing technologies, this precise control system for molten glass flow on a patterned glass production line has the following advantages:

[0022] I. This invention achieves feedforward pre-regulation control of molten glass flow rate through deep integration of a digital twin layer and a predictive control layer. The predictive control layer updates the flow prediction curve every second, calculating the predicted flow rate value per second for the next 300 seconds. When the absolute value of the flow deviation at five consecutive prediction times exceeds 0.5%, the pre-regulation mechanism is immediately triggered. Furthermore, the calculation of the pre-regulation amount fully considers the pure system time lag, ensuring that the pre-regulation command reaches the execution layer before the actual flow fluctuation occurs. This effectively overcomes the problem of lag response in traditional feedback control and significantly improves the accuracy and real-time performance of flow control.

[0023] Second, this invention significantly improves product quality and reduces process debugging costs through the linkage function between the system and the embossing roller speed control system, as well as the virtual testing function of the digital twin layer. The predictive control layer sends the predicted flow rate value to the embossing roller speed control system in real time. When the flow rate changes, the embossing roller speed is automatically adjusted to keep the amount of molten glass accumulated on the embossing roller constant. The linkage response time is less than 1 second, effectively avoiding problems such as uneven glass thickness and inconsistent pattern depth caused by flow fluctuations. At the same time, process personnel can compare the control effects under different parameter combinations within 5 minutes through virtual testing, and directly send the optimal parameters to the actual control system without on-site debugging.

[0024] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0026] Figure 1 This is an overall flowchart of a precise control system for molten glass flow on a patterned glass production line.

[0027] Figure 2 A flowchart of production data acquisition and preprocessing for a glass melt flow precision control system on a patterned glass production line;

[0028] Figure 3 This is a flow prediction and pre-adjustment control flowchart for a precise control system for molten glass flow on a patterned glass production line. Detailed Implementation

[0029] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0030] Example 1

[0031] First, a physical layer is constructed, including a furnace for fully melting soda-lime-silica glass raw materials into a uniform, bubble-free molten glass; a flow channel for smoothly conveying the high-temperature molten glass from the furnace outlet to the embossing station; a flow regulating valve for precisely controlling the outflow of molten glass per unit time; and an embossing roller for pressing the glass into a specified thickness of 3mm and a preset pattern. According to the production process requirements of conventional 3mm thick soda-lime-silica embossed glass, the initial production speed is set to 12 meters per minute, and the molten glass flow rate is set to 18 cubic meters per hour. This matches the production capacity requirements of this product specification and the embossing process window, laying the foundation for subsequent stable production.

[0032] Then, a data acquisition and transmission layer is deployed. Four Type B thermocouples are evenly arranged along the circumference at the kiln outlet to comprehensively collect the temperature of the molten glass from different locations at the outlet, avoiding errors caused by uneven temperature fields in single-point measurements. A high-temperature resistant ceramic pressure sensor is installed at the inlet of the flow channel to monitor the static pressure of the molten glass in real time, indirectly reflecting the liquid level and flow potential energy. A laser level sensor is installed 1.5m above the surface of the molten glass above the flow channel. This installation height avoids damage to the sensor from the strong radiation of the high-temperature molten glass and ensures that the level measurement accuracy reaches the millimeter level. A non-contact electromagnetic flow sensor is installed at the outlet of the flow channel, which can accurately measure the actual outflow without direct contact with the high-temperature molten glass, completely solving the problems of corrosion and scaling of contact sensors. An ambient temperature sensor is installed near the flow channel. Used to collect ambient temperature data around the flow channel, eliminating the influence of ambient temperature fluctuations on the viscosity and flow characteristics of molten glass. All sensors establish communication connections with the digital twin layer via Profinet industrial Ethernet, ensuring that production data is transmitted in real time with millisecond-level latency, meeting the real-time requirements of industrial sites. The raw production data collected by the sensors is first processed by a moving average filter with a window size of 5 to effectively filter out high-frequency noise caused by factors such as electromagnetic interference and equipment vibration in the industrial site, making the data curve smoother. After filtering, outlier removal is performed. The arithmetic mean of the three data points before and after the current data point is used as a benchmark. If the current value deviates from the benchmark by more than three times the standard deviation, it is removed. This eliminates erroneous data caused by sudden sensor failures and instantaneous interference, ensuring that the data input to the digital twin layer is true and reliable. The processed data is transmitted to the digital twin layer in real time.

[0033] Next, a digital twin layer was built and deployed on an industrial edge server equipped with an Intel Xeon Gold 6248R processor and an NVIDIA RTX A4000 graphics card. This hardware configuration possesses powerful parallel computing and graphics rendering capabilities, enabling real-time solving and 3D visualization of the glass melt flow dynamics model. First, a 3D laser scanner was used to perform a comprehensive scan of the actual flow channel, obtaining a flow channel geometry model with an accuracy of 0.05mm. This accurately reproduced all details of the flow channel, such as internal curvature, inner wall roughness, and corner structures, avoiding distortion of simulation results due to model geometric deviations. Then, a glass melt flow dynamics model was constructed based on computational fluid dynamics methods, employing a power-law model to describe the non-Newtonian fluid characteristics of the glass melt, accurately reflecting the shear-thinning flow of the glass melt. The relationship between viscosity and temperature is fitted using the VFT equation to accurately characterize the nonlinear characteristics of glass melt viscosity as a function of temperature. Physical parameters such as viscosity, density, and specific heat capacity of ordinary soda-lime silica glass formulations at a production speed of 12 meters per minute are retrieved from the operating condition parameter database to ensure that the model parameters are perfectly matched with the current actual production conditions. During the operation of the digital twin layer, real-time data on furnace temperature, channel pressure, glass melt level, and ambient temperature are received and processed by the data acquisition and transmission layer. These data are used as boundary conditions and input into the glass melt flow dynamics model. Finite element analysis is used to calculate the three-dimensional velocity distribution, pressure distribution, and temperature distribution of the glass melt in the channel, which intuitively displays the glass melt flow state at any position inside the channel and promptly detects potential problems such as uneven flow velocity, local dead zones, and temperature stratification within the channel.

[0034] Next, the predictive control layer is activated. This layer employs an improved Long Short-Term Memory network that incorporates digital twin features to construct a traffic prediction model; its output layer calculation formula is as follows: ,in for Predicting the future Glass melt flow rate per second; Weighting coefficients for digital twin features; The output function of the basic LSTM network; for The input feature vector at time step; The future predicted by digital twin models This method, which calculates glass melt flow rate per second, combines the real-time nature of pure data-driven prediction with the physical accuracy of a digital twin model, effectively improving the accuracy and robustness of flow rate prediction. In this embodiment, the improved Long Short-Term Memory (LSTM) network contains two hidden layers, each with 64 LSTM units. The input features are an 8-dimensional time series of the past 60 seconds, including furnace temperature, flow channel pressure, glass melt level, actual flow rate, and ambient temperature. The output is the predicted flow rate for the next 1-300 seconds. The network uses the Adam optimizer with mean squared error as the loss function, trained on 30 consecutive days of historical production data until the relative error on the validation set is less than 0.3%. (Digital twin feature weighting coefficients are also mentioned.) Based on the average prediction error of the digital twin model over the past 100 time steps, a dynamic lookup table is used to determine: when the average prediction error is <0.5%, When it is 0.5% to 1% When >1% The improved Long Short-Term Memory (LSTM) network output layer calculation formula is used to perform the fusion calculation of the flow prediction value. The weights of the pure data-driven prediction results and the digital twin model prediction results are reasonably allocated to make the final flow prediction value more closely match the actual production situation. The prediction control layer obtains the latest input data every second, updates the flow prediction curve, calculates the glass melt flow prediction value per second for the next 300 seconds, and anticipates the flow change trend over a longer period, reserving sufficient response time for flow pre-adjustment. Then, the deviation between the flow prediction value and the set value at each prediction moment is calculated, and the flow fluctuation is monitored in real time. When the absolute value of the flow deviation exceeds 0.5% for five consecutive prediction moments, the pre-adjustment mechanism is immediately triggered. This triggering condition effectively avoids erroneous adjustments caused by instantaneous flow fluctuations, ensuring the stability of the production process. The system's pure lag time parameter, pre-determined through a step response experiment at a production speed of 12 meters per minute, is retrieved from the operating condition parameter database. The system's lag characteristics are fully considered when calculating the adjustment amount to prevent overshoot or under-adjustment. The pre-adjustment amount of the flow control valve is calculated using the flow control valve pre-adjustment amount calculation formula; the formula is: ,in for The pre-adjustment amount of the flow control valve at any given time. This is the proportionality coefficient. For pre-adjustment time window, for Predicting the future glass melt flow rate per second For pure time delay, A set value for the glass melt flow rate is set, and a corresponding pre-adjustment command is generated and sent to the execution layer. In this embodiment, the execution time of the pre-adjustment command is one pure time delay before the predicted fluctuation occurs. That is, if a deviation is predicted to occur at t+8 seconds, the command is issued immediately at t seconds, so that the regulating valve completes its action exactly 8 seconds later. The adjustment amount is calculated using the above formula, where the predicted flow rate value is the average deviation from t+8 to t+38 seconds.

[0035] After receiving the pre-adjustment command, the execution layer drives the valve core to move smoothly through the built-in high-precision servo motor, achieving micron-level precise adjustment of the flow control valve opening and accurately controlling the outflow of molten glass. The valve's built-in position feedback device provides real-time feedback on the actual opening, forming a closed-loop control circuit to promptly correct the deviation between the valve core position and the commanded opening, ensuring that the adjustment accuracy meets production requirements. The control valve also operates a fault self-diagnosis function, which monitors the motor current, valve core position, and communication status in real time, comprehensively monitoring the valve's operating status. If faults such as motor overload, valve core jamming, or communication interruption are detected, an audible and visual alarm signal is immediately issued, and the control valve is automatically switched to manual control mode to prevent production interruptions or product quality accidents caused by control valve failure, thus ensuring production safety.

[0036] During system operation, the digital twin layer synchronously performs online self-calibration, automatically collecting actual flow data under stable operating conditions and predicted flow data from the digital twin model every hour, calculating the deviation between the two to evaluate the prediction accuracy of the digital twin model. When the deviation exceeds 1%, the viscosity coefficient and thermal conductivity coefficient in the glass melt flow dynamics model are automatically adjusted to correct model parameter drift caused by factors such as fluctuations in glass raw material composition, scaling on the inner wall of the flow channel, and slow changes in kiln conditions. The adjustment range is determined according to the magnitude of the deviation; the larger the deviation, the larger the adjustment range. Under the premise of ensuring model stability, the deviation is quickly corrected. The adjusted model parameters are automatically saved to the operating condition parameter database for subsequent flow simulation and flow prediction calculations, continuously improving the accuracy of the digital twin model.

[0037] Process engineers can optimize process parameters through the virtual testing function of the digital twin layer. Multiple sets of process parameters can be compared without interrupting actual production, avoiding raw material waste and capacity loss associated with actual trial production. By inputting different kiln temperatures, glass liquid levels, and regulating valve openings, the system automatically generates corresponding flow rate change simulation curves. The simulation comprehensively considers the viscosity changes of the glass liquid, the thermal conduction effect of the flow channel wall, and the dynamic response characteristics of the regulating valve, ensuring the simulation results closely resemble actual production conditions. It outputs flow velocity, pressure, and temperature data at any location within the flow channel, as well as flow rate change curves over time, providing comprehensive analytical support for process engineers. By comparing the flow rate stability and control effects under different parameter combinations and selecting the optimal combination, process engineers can directly send the optimal parameters to the actual control system for execution, quickly optimizing and adjusting process parameters to improve production efficiency and product quality.

[0038] Simultaneously, the system activates the linkage function with the embossing roller speed control system to achieve synchronous matching between the glass melt flow rate and the embossing roller speed. This solves the problem of speed adjustment lagging behind flow rate changes in traditional control methods. The predictive control layer sends the predicted flow rate value to the embossing roller speed control system in real time. The embossing roller speed control system automatically adjusts the speed of the embossing roller according to the adjustment ratio determined by the 3mm glass strip thickness setting value. When the glass melt flow rate increases, the embossing roller speed is increased accordingly, and when the glass melt flow rate decreases, the embossing roller speed is decreased accordingly. This keeps the amount of glass melt accumulated on the embossing roller constant, which not only ensures the uniformity of the embossed glass thickness but also effectively avoids problems such as overflow and pattern deformation caused by excessive glass melt accumulation, as well as broken plates and material shortages caused by insufficient accumulation. This significantly improves the quality of embossing and the continuity of production.

[0039] Example 2

[0040] First, a physical layer adapted to the properties of high borosilicate glass is constructed. This includes a dedicated high borosilicate glass furnace capable of reaching melting temperatures above 1650℃, used to fully melt the high borosilicate glass raw materials into a uniform, bubble-free, and stone-free high-quality molten glass. A high-temperature and corrosion-resistant alumina flow channel is used to smoothly transport the high-temperature, high-viscosity molten glass from the furnace outlet to the embossing station, avoiding problems such as corrosion, deformation, and leakage that occur with ordinary flow channels under long-term high temperatures. A large-diameter, high-precision flow control valve is configured to precisely control the large flow rate output of the high borosilicate glass, meeting the requirements of high-speed operation. The production flow requirements, along with the embossing rollers whose surfaces have undergone special hardening treatment and are adapted to the pressing and molding characteristics of high borosilicate glass, are used to hydraulically press the glass into glass strips of a specified thickness of 5mm and with clear, pre-set patterns. In accordance with the high-speed production process requirements for 5mm thick high borosilicate patterned glass, the initial production speed is set to 8 meters per minute, and the glass liquid flow rate is set to 22 cubic meters per hour. This combination of parameters can fully leverage the capacity advantage of high-speed production while matching the high viscosity and slow molding speed of high borosilicate glass liquid, providing a foundation for subsequent stable production and high-quality molding.

[0041] Then, a data acquisition and transmission layer adapted to the high-borosilicate high-temperature production environment is deployed. Four type B thermocouples are evenly arranged along the circumference at the outlet of the high-borosilicate glass furnace. Due to the high melting temperature of high-borosilicate glass and the extremely uneven temperature field distribution at the outlet, multi-point arrangement can comprehensively collect glass melt temperature data from different directions, effectively avoiding the large errors caused by single-point measurement. A high-temperature resistant ceramic pressure sensor capable of withstanding 1600℃ is installed at the inlet of the high-temperature resistant flow channel to monitor the static pressure of the glass melt at the inlet of the flow channel in real time, indirectly reflecting the liquid level height and flow potential energy changes of the glass melt. A laser level sensor is installed 1.5m above the surface of the molten glass, above the flow channel. This installation height effectively avoids damage to the sensor's optical components from the strong heat radiation generated by the borosilicate glass, while ensuring millimeter-level accuracy in level measurement. A non-contact electromagnetic flow sensor adapted to the high-temperature characteristics of borosilicate glass is installed at the flow channel outlet. This sensor accurately measures the actual outflow without direct contact with the high-temperature, highly corrosive glass, completely solving the problem of contact sensors being easily corroded and scaled, leading to measurement failure. An ambient temperature sensor is installed near the flow channel to collect data. The ambient temperature data around the flow channel is crucial because the viscosity of high borosilicate glass melt is far more sensitive to temperature changes than that of ordinary soda-lime silica glass. Even small fluctuations in ambient temperature can significantly affect the flow characteristics of the glass melt. Collecting this data provides a basis for correction in subsequent model calculations. All sensors establish communication connections with the digital twin layer via Profinet Industrial Ethernet, ensuring that production data is transmitted in real time with millisecond-level latency, meeting the real-time control requirements under high-speed production conditions. The raw production data collected by the sensors is first processed by a moving average filter with a window size of 5 to effectively filter out high-frequency noise caused by electromagnetic interference from high-power equipment operation, kiln combustion fluctuations, and equipment vibrations in the industrial field, making the data curve smoother and more stable. After filtering, outlier removal is performed. The arithmetic mean of the three data points before and after the current data point is used as a benchmark. If the current value deviates from the benchmark by more than three times the standard deviation, it is removed. This eliminates erroneous data caused by sudden sensor failures and instantaneous strong interference, ensuring that the data input to the digital twin layer is true and reliable, providing an accurate data foundation for subsequent flow simulation and flow prediction. The processed data is transmitted to the digital twin layer in real time.

[0042] Next, a digital twin layer adapted for borosilicate glass was built and run. This digital twin layer was deployed on an industrial edge server equipped with an Intel Xeon Gold 6248R processor and an NVIDIA RTX A4000 graphics card. This hardware configuration possesses powerful parallel computing and graphics rendering capabilities, enabling real-time solving and 3D visualization of complex flow dynamics models of borosilicate glass liquid, meeting the real-time requirements of high-speed production. First, a 3D laser scanner was used to perform a comprehensive scan of the high-temperature resistant flow channel, obtaining a flow channel geometric model with an accuracy of 0.05mm. This accurately reproduced all details of the flow channel, including internal curvature, inner wall roughness, corner structure, and minor wear and deformation caused by long-term use. Due to the high viscosity of borosilicate glass liquid, even small geometric deviations in the flow channel can significantly affect the flow state of the glass liquid. A high-precision geometric model can effectively avoid distortion of simulation results. Then, a flow dynamics model of borosilicate glass liquid was built based on ANSYS Fluent software, using a power-law model to describe the flow dynamics of borosilicate glass. The non-Newtonian fluid characteristics of molten glass accurately reflect the more significant shear-thinning flow law of high borosilicate glass. The relationship between viscosity and temperature is fitted using the VFT equation, precisely characterizing the drastic nonlinear characteristics of the viscosity of high borosilicate glass changing with temperature. Physical parameters such as viscosity, density, and specific heat capacity of high borosilicate glass formulations at a production speed of 8 meters per minute are retrieved from the operating condition parameter database to ensure that the model parameters are perfectly matched with the current actual production conditions. During the operation of the digital twin layer, real-time data on furnace temperature, channel pressure, glass liquid level, and ambient temperature are received after processing by the data acquisition and transmission layer. These data are used as boundary conditions and input into the high borosilicate glass flow dynamics model. Finite element analysis is used to calculate the three-dimensional velocity distribution, pressure distribution, and temperature distribution of high borosilicate glass liquid in the channel, intuitively displaying the glass liquid flow state at any position inside the channel. This allows for the timely detection of problems such as uneven flow velocity, local dead zones, and temperature stratification that may exist in the channel. These problems can easily lead to defects such as streaks and bubbles in the glass ribbon during the production of high borosilicate glass.

[0043] Next, a predictive control layer adapted for high-speed borosilicate glass production is activated. This layer employs an improved long short-term memory network (LSTM) incorporating digital twin features to construct a flow prediction model for borosilicate glass melt. This combines the real-time nature of pure data-driven prediction with the physical accuracy of the digital twin model, effectively addressing the issues of large fluctuations in high borosilicate glass production conditions and large errors in pure data-driven prediction. This significantly improves the accuracy and robustness of flow prediction. The improved LTM output layer calculation formula is used to fuse the pure data-driven prediction and the digital twin model prediction, rationally allocating the weights of the two prediction results to ensure the final flow prediction more closely reflects the actual production conditions of high borosilicate glass. The predictive control layer acquires the latest input data every second, updates the flow prediction curve, and calculates the flow prediction value per second for the next 300 seconds. Due to the high viscosity of high borosilicate glass melt, the system exhibits pure time lag. A longer prediction window allows sufficient response time for flow pre-regulation. The deviation between the predicted flow value and the set value at each prediction moment is then calculated, and flow fluctuations are monitored in real time. When the absolute value of the flow deviation exceeds 0.5% for five consecutive prediction moments, the pre-regulation mechanism is immediately triggered. This triggering condition effectively avoids erroneous regulation caused by instantaneous flow fluctuations, ensuring the stability of the high-speed production process. The pure lag time parameter of the high borosilicate glass production system, pre-determined through step response experiments at a production speed of 8 meters per minute, is retrieved from the operating parameter database. The system's lag characteristics are fully considered when calculating the regulation amount to prevent overshoot or undershoot. The pre-regulation amount of the large-diameter flow control valve is calculated using the flow control valve pre-regulation calculation formula, and a corresponding pre-regulation command is generated and sent to the execution layer.

[0044] After receiving the pre-adjustment command, the large-diameter, high-precision electric flow control valve at the execution layer drives the valve core to move smoothly through the built-in high-power servo motor, achieving micron-level precise adjustment of the valve opening. This accurately controls the outflow of high borosilicate glass liquid, avoiding water hammer effects caused by excessively fast valve core movement that could lead to fluctuations in the glass liquid flow. The valve's built-in position feedback device provides real-time feedback on the actual opening, forming a complete closed-loop control circuit. This promptly corrects deviations between the valve core position and the commanded opening, ensuring that the adjustment accuracy meets the requirements of high-speed production. The valve also features a continuous self-diagnostic function, monitoring motor current, valve core position, and communication status in real time. This comprehensive monitoring of the valve's operating status is crucial because interruptions in high borosilicate glass production can cause significant economic losses and equipment damage risks. If faults such as motor overload, valve core jamming, or communication interruption are detected, an audible and visual alarm signal is immediately issued, and the system automatically switches to manual control mode. This allows for timely intervention by operators, ensuring production safety and continuity.

[0045] During system operation, the digital twin layer synchronously performs online self-calibration, automatically collecting actual high borosilicate glass melt flow data and predicted flow data from the digital twin model under stable operating conditions every hour, calculating the deviation between the two, and thus evaluating the prediction accuracy of the digital twin model. Due to the large fluctuations in the composition of high borosilicate glass raw materials and the faster scaling rate on the inner wall of the flow channel, model parameters are prone to drift. When the deviation exceeds 1%, the viscosity coefficient and thermal conductivity coefficient in the high borosilicate glass melt flow dynamics model are automatically adjusted to correct model errors caused by factors such as fluctuations in raw material composition, scaling on the inner wall of the flow channel, and slow changes in kiln operating conditions. The adjustment range is determined according to the magnitude of the deviation; the larger the deviation, the larger the adjustment range. Under the premise of ensuring model stability, the deviation is quickly corrected. The adjusted model parameters are automatically saved to the operating condition parameter database for subsequent flow simulation and flow prediction calculations, continuously improving the accuracy and reliability of the digital twin model.

[0046] Process engineers can optimize high-speed production process parameters through the virtual testing function of the digital twin layer. Multiple sets of process parameters can be compared without interrupting actual production, avoiding the significant raw material waste, equipment wear, and capacity loss associated with actual trial production of borosilicate glass. By inputting different combinations of kiln discharge temperature, glass liquid level in the flow channel, and regulating valve opening, the system automatically generates corresponding simulation curves of borosilicate glass liquid flow rate changes. The simulation process fully considers the drastic viscosity changes of borosilicate glass liquid with temperature, the high-temperature heat conduction effect of the flow channel wall, and the dynamic response characteristics of large-diameter regulating valves, ensuring that the simulation results closely resemble actual production conditions. It outputs flow velocity, pressure, and temperature data and flow fluctuation curves at key locations in the flow channel, providing process engineers with comprehensive and detailed analytical data. By comparing the flow stability and embossing effect under different parameter combinations, and determining the optimal process parameters, engineers can directly send them to the actual control system for execution, quickly optimizing and adjusting process parameters, shortening the process debugging cycle, and improving the stability and product quality of high-speed production.

[0047] Simultaneously, the system activates the linkage function with the embossing roller speed control system to achieve synchronous matching between the glass melt flow rate and the embossing roller speed. This solves the problem of speed adjustment lagging behind flow rate changes in traditional control methods. This problem can lead to severe uneven glass strip thickness and embossing texture defects in high-speed production of high borosilicate glass. The predictive control layer sends the predicted flow rate of high borosilicate glass melt to the embossing roller speed control system in real time. The embossing roller speed control system automatically and synchronously adjusts the speed of the embossing roller according to the adjustment ratio determined by the 5mm glass strip thickness setting value. When the glass melt flow rate increases, the embossing roller speed is increased proportionally; when the flow rate decreases, the embossing roller speed is decreased proportionally. This keeps the accumulation of high borosilicate glass melt on the embossing roller constant, effectively avoiding quality problems and production accidents such as uneven glass strip thickness, unclear embossing texture, glass melt overflow, and broken plates under high-speed production, significantly improving the yield and production efficiency of high borosilicate patterned glass.

[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A precise control system for molten glass flow rate on a patterned glass production line, characterized in that, The system includes the following components: Physical entity layer: includes kiln, flow channel, flow control valve and embossing roller; Data Acquisition and Transmission Layer: Used to acquire real-time production data from the physical entity layer; Digital twin layer: Communicatively connected to the data acquisition and transmission layer, used to synchronously simulate the flow state of molten glass in the flow channel based on the real-time production data, and generate molten glass flow prediction data for future times; Predictive control layer: Communicatively connected to the digital twin layer, used to predict the trend of glass melt flow rate change within a preset time window using an improved long short-term memory network, and generate pre-adjustment instructions; Execution layer: Communicatively connected to the predictive control layer, used to adjust the opening of the flow regulating valve according to the pre-adjustment command.

2. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The data acquisition and transmission layer has four Type B thermocouples evenly arranged along the circumference at the kiln outlet, a high-temperature resistant ceramic pressure sensor installed at the inlet of the flow channel, a laser level sensor installed 1.5m above the glass melt surface above the flow channel, a non-contact electromagnetic induction flow sensor installed at the outlet of the flow channel, and an ambient temperature sensor installed near the flow channel. All sensors communicate with the digital twin layer via Profinet industrial Ethernet. The raw data collected is first processed by moving average filtering with a window size of 5. The filtered data is then subjected to outlier removal, with the outlier judgment criterion being that it exceeds three times the standard deviation of the average of the three preceding and following data.

3. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The digital twin layer is deployed on an industrial edge server equipped with an Intel Xeon Gold 6248R processor and an NVIDIA RTX A4000 graphics card. First, a 3D laser scanner is used to scan the actual flow channel, obtaining a geometric model with an accuracy of 0.05 mm. Then, a glass melt flow dynamics model is constructed based on computational fluid dynamics. The Navier-Stokes equations are solved using the finite element method, with a hexahedral structured mesh and boundary conditions set as velocity inlet and pressure outlet. The RNG k-ε turbulence model is used, and a power-law model is employed to describe the non-Newtonian fluid characteristics of the glass melt. The viscosity-temperature relationship is fitted using the VFT equation. A database of operating parameters stores physical parameters such as viscosity, density, and specific heat capacity for 10 commonly used glass formulations at different production speeds. During operation, the digital twin layer receives data from various sensors in real time, inputting furnace temperature, flow channel pressure, glass melt level, and ambient temperature as boundary conditions into the flow dynamics model. Finite element analysis is then used to calculate the three-dimensional velocity, pressure, and temperature distributions of the glass melt within the flow channel.

4. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The predictive control layer employs an improved long short-term memory network that incorporates digital twin features to construct a traffic prediction model. Its output layer calculation formula is as follows: ,in for Predicting the future Glass melt flow rate per second; Weighting coefficients for digital twin features; The output function of the basic LSTM network; for The input feature vector at time step; The future predicted by digital twin models The flow rate of molten glass per second.

5. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The predictive control layer acquires the latest input data every second, updates the flow prediction curve, calculates the flow prediction value per second for the next 300 seconds, and then calculates the deviation between the flow prediction value and the set value at each prediction time. When the absolute value of the flow deviation at five consecutive prediction times exceeds 0.5%, the pre-adjustment mechanism is immediately triggered to calculate the pre-adjustment amount of the flow regulating valve. The calculation of the pre-adjustment amount takes into account the pure time lag of the system. The pure time lag is pre-determined through a step response experiment. The pure time lag at different production speeds is stored in the operating condition parameter database. The system automatically calls the corresponding pure time lag parameter according to the current production speed.

6. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The formula for calculating the pre-adjustment amount of the flow regulating valve generated by the predictive control layer is as follows: ,in for The pre-adjustment amount of the flow control valve at any given time. This is the proportionality coefficient. For pre-adjustment time window, for Predicting the future glass melt flow rate per second For pure time delay, Set the flow rate of the molten glass.

7. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The virtual experiment function of the digital twin layer allows process engineers to input different furnace temperatures, glass liquid levels, and regulating valve opening parameters. Based on the same glass liquid flow dynamics model in the digital twin layer, the system automatically generates corresponding flow rate change simulation curves. During the simulation, the viscosity changes of the glass liquid, the heat conduction effect of the inner wall of the flow channel, and the dynamic response characteristics of the regulating valve are considered. The simulation results include flow velocity, pressure, and temperature data at any position in the flow channel, as well as the flow rate change curve over time. Process engineers can compare the flow rate stability and control effect under different parameter combinations, select the optimal process parameter combination, and then directly write the optimal parameters into the non-volatile memory of the predictive control layer via the OPCUA protocol. The parameters are automatically loaded at the start of the next production cycle, and the current control parameters remain unchanged during the loading process.

8. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The execution layer adopts a high-precision electric flow regulating valve. The regulating valve has a built-in position feedback device, which can provide real-time feedback on the actual opening degree. After receiving the pre-adjustment command from the predictive control layer, the regulating valve drives the valve core to move through the built-in servo motor to achieve precise adjustment of the opening degree. The regulating valve also has a fault self-diagnosis function, which can detect the motor current, valve core position and communication status in real time. When a fault is detected, an alarm signal is immediately issued and the regulating valve is switched to manual control mode.

9. The glass melt flow precision control system for a patterned glass production line according to claim 1, characterized in that, The digital twin layer has an online self-calibration function. The system automatically collects the actual flow data and the predicted flow data of the digital twin model under stable operating conditions every hour, calculates the deviation between the two, and automatically adjusts the viscosity coefficient and thermal conductivity coefficient in the glass melt flow dynamics model when the deviation exceeds 1%. The adjustment range is determined according to the preset adjustment function. The larger the deviation, the larger the adjustment range. The adjusted model parameters are automatically saved to the operating condition parameter database for subsequent calculations.

10. A precise control system for molten glass flow rate on a patterned glass production line according to claim 1, characterized in that, The system also has a linkage function with the embossing roller speed control system. The predictive control layer sends the flow prediction value to the embossing roller speed control system in real time. The embossing roller speed control system automatically adjusts the speed of the embossing roller according to the flow prediction value. The adjustment ratio is determined according to the glass strip thickness setting value. When the flow increases, the speed of the embossing roller is increased accordingly, and when the flow decreases, the speed of the embossing roller is decreased accordingly, so that the amount of glass liquid accumulated on the embossing roller remains constant.

Images