A method for on-line detection and evaluation of glass clarity

Abstract

The application discloses a kind of glass clarity on-line detection and evaluation method, it is related to glass manufacturing technical field, the method includes: using high-temperature industrial endoscope imaging system real-time acquisition glass liquid image;Through filtering denoising and adaptive threshold segmentation extraction bubble dynamic characteristics and establish database;Based on bubble number density, particle size distribution and rising rate, combine temperature and viscosity parameter to construct multidimensional evaluation model;Using time series analysis is carried out sliding average processing and deviation trend fitting, output final evaluation result;Evaluation result is fed back to control system, and automatically adjusts glass melting furnace heating power, liquid level and drawing speed.The application aims at solving the problem of glass clarity detection existing hysteresis and evaluation dimension single, by the above technical scheme, the real-time on-line monitoring and intelligent prediction of glass clarification state are realized, while greatly enhancing evaluation accuracy, product yield and energy utilization rate are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of glass manufacturing technology, and in particular to an online detection and evaluation method for glass clarity. Background Technology

[0002] With the rapid development of the optoelectronic information and precision optics industries, precise control of glass clarity in glass manufacturing processes has become a core factor determining the final product quality. Clarification, a crucial step in glass melting, hinges on eliminating air bubbles within the molten glass through optimized thermal regimes and controlled chemical kinetics, ensuring the glass substrate possesses extremely high optical uniformity and structural integrity. Especially in large-scale continuous production processes, the complex material circulation and chemical reactions within the glass melting furnace place stringent demands on the stability of the clarification effect.

[0003] Among them, the online detection and evaluation method for glass clarity, as a key technology for improving the closed-loop control capability of industrial production, focuses on achieving real-time monitoring of the clarification state of molten glass on the production line through advanced sensing technology and algorithm models. This technology acquires image or physical feature data of molten glass under high-temperature conditions, and combines mathematical modeling methods to quantitatively characterize the clarification efficiency, thereby providing an objective basis for adjusting core process parameters such as melting temperature, drawing speed, and pressure environment.

[0004] However, existing technologies mostly employ offline testing methods based on laboratory environments, statistically analyzing bubble parameters through static melting experiments. This fails to simulate the complex liquid flow field distribution in real production, resulting in significant lag in evaluation results. Furthermore, some technical solutions rely excessively on passive improvements to the physical structure of the refining tank, lacking active detection mechanisms for the internal state of the molten glass, and thus cannot provide real-time clarity feedback during production. In addition, traditional evaluation systems lack the ability to integrate and process multi-source heterogeneous data, making it difficult to achieve deep correlation between testing data and process control systems. This results in a lack of scientific data support for production scheduling, limiting further improvements in glass product yield.

[0005] Therefore, an online detection and evaluation method for glass clarity is desired. Summary of the Invention

[0006] The purpose of this invention is to provide an online detection and evaluation method for glass clarity, so as to solve the problems of outdated detection methods, lack of active online detection capabilities, and inability to deeply correlate multi-source data with process control in the prior art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] A method for online detection and evaluation of glass clarity includes:

[0009] Step 1, Acquire high-temperature image data of molten glass: Install a high-temperature industrial endoscope imaging system above the refining section or flow channel of the glass melting furnace. Use its built-in charge-coupled device sensor to acquire real-time images of the surface and depth of molten glass under high-temperature conditions. The sampling frequency is set to a preset sampling frequency, the single frame image resolution is set to a preset resolution, and the synchronous trigger controller ensures that the image acquisition is consistent with the light source pulse frequency.

[0010] Step 2, preprocess image data and extract bubble feature information: The median filtering algorithm is used to denoise the acquired original image, and the background subtraction algorithm is applied to eliminate the ripple interference caused by the flow of molten glass. The filter window size is set to a preset size. Then, the adaptive threshold segmentation algorithm is used to separate the bubble target in the image from the molten glass background. The equivalent diameter, circumference, roundness and coordinate position of each bubble are calculated to establish a real-time bubble dynamic feature database.

[0011] Step 3: Construct a multi-dimensional evaluation model for glass clarity: Based on the bubble dynamic characteristic database, calculate the bubble number density, bubble size distribution gradient, and average bubble rise rate per unit volume of molten glass. Combined with the current melting temperature and molten glass viscosity parameters, obtain the initial clarity index through a weighted summation method.

[0012] Step 4, Real-time correction of evaluation results and deviation analysis: Introduce time series analysis method, perform moving average processing on the initial clarity index within the predetermined time period, compare it with the preset standard clarity curve, calculate the deviation value, use the least squares method to fit the deviation evolution trend, and output the final online evaluation result of glass clarity.

[0013] Step 5, execute closed-loop feedback control of the process: The final online evaluation result of glass clarity is transmitted to the glass melting furnace automation control system through the industrial bus. Based on the difference between the evaluation result and the target value, the heating power, liquid level or drawing speed of the glass melting furnace clarification section is automatically adjusted to achieve closed-loop management of glass production quality.

[0014] In step 1, the high-temperature industrial endoscope imaging system is equipped with a high-temperature resistant water-cooling kit and a nitrogen purging device to ensure stable operation of the imaging system under a preset high-temperature environment. The field of view of its optical lens is a preset angle, the object distance range covers a preset distance range, and the light source uses a high-power narrow-band laser with the wavelength locked in a preset band to reduce the interference of the infrared radiation of the high-temperature glass melt itself on the imaging quality.

[0015] The image data preprocessing process in step 2 also involves a background subtraction algorithm. By extracting the mean background value of a predetermined number of frames and subtracting it from the current frame, the interference of ripples caused by the flow of molten glass is eliminated, improving the contrast of bubble extraction. The adaptive threshold segmentation algorithm dynamically sets the segmentation threshold by calculating the mean and variance of grayscale values ​​in local image regions, ensuring that even under uneven lighting conditions, it can accurately extract tiny bubbles with diameters smaller than a preset size threshold.

[0016] The bubble dynamic feature extraction process in step 2 also includes trajectory tracking. The Hungarian matching algorithm is used to associate bubbles in two adjacent frames of images, calculate the displacement vector of the bubble in the two-dimensional plane, and then deduce the three-dimensional motion trend of the bubble inside the glass melt. The equivalent diameter of the bubble is calculated by dividing the bubble pixel area by pi, taking the square root, and then multiplying by a predetermined scaling factor.

[0017] The mathematical expression of the multi-dimensional evaluation model for glass clarity in step 3 includes a bubble removal rate function. The clarity index is defined as equal to a first preset constant minus the ratio of the current bubble density to the initial bubble density, multiplied by a temperature correction coefficient. This temperature correction coefficient is determined by the Arrhenius equation and reflects the influence of the molten glass temperature on gas diffusion and bubble rising resistance.

[0018] The first formula in step 3 is the clarification efficiency calculation formula, which is equal to the quotient obtained by multiplying a predetermined ratio by a second preset constant and subtracting the total number of bubbles in the observation area from the total number of bubbles in the reference area. The second formula is the average bubble diameter calculation formula, which is equal to the sum of the diameters of all identified bubbles during the observation time divided by the total number of bubbles. The third formula is the viscosity compensation model, which is equal to the original clarity multiplied by a preset power of the ratio of the glass melt viscosity to the standard viscosity, thereby eliminating the evaluation error caused by temperature fluctuations. The fourth formula is the bubble removal rate function, which can reflect the actual effect of bubble removal in real time.

[0019] In step 4, the time series analysis method employs a long short-term memory neural network model. It uses historical clarification data within a first predetermined historical period as input features to predict the clarification trend within a second predetermined prediction period, achieving a prediction accuracy greater than a preset accuracy threshold. The window length for the moving average processing is dynamically adjusted according to the production line's pulling speed. If the pulling speed changes by a preset proportion, the window length is adjusted proportionally to ensure the real-time nature of the evaluation results.

[0020] In step 4, the standard clarity curve is determined by the clarity trajectory of a high-quality glass sample measured in a laboratory environment, including the bubble elimination rate constant and the final residual bubble density index. If the deviation value generated during the comparison process exceeds the preset deviation threshold, the system will automatically issue an early warning signal and record the current process parameter status.

[0021] In step 5, the industrial bus uses the industrial Ethernet protocol with real-time characteristics, and the data transmission delay is less than a preset delay threshold. The heating power adjustment step in the automated control system is set to a preset adjustment step, the liquid level adjustment accuracy is a preset accuracy range, and the pulling speed adjustment range is set within a preset floating range based on the current value.

[0022] It also includes a self-diagnostic process for sensor data acquisition, which monitors the return water temperature of the water-cooling kit and the nitrogen purging flow rate to determine whether the imaging system is within a safe operating range. If the return water temperature exceeds a preset safe temperature threshold, the system will automatically retract the endoscope and cut off the light source.

[0023] It also includes establishing a big data platform for glass production containing a predetermined number of records, deeply correlating the online detected clarity data with the transmittance, stress, and bubble defect rate of the finished glass, and continuously optimizing the weight coefficients in the evaluation model using a decision tree algorithm. This method is applied to an ultra-thin float glass production line. For display glass substrates within a preset thickness range, multiple sets of high-temperature industrial endoscope imaging systems are arranged laterally in the clarification section of the glass melting furnace to form a detection array, achieving complete coverage of the clarification state of the glass melt across the entire width area. A coordinate space transformation is performed to map the image coordinates of all bubbles to a unified physical coordinate system of the glass melting furnace, and the weight coefficients of the evaluation model in step 3 are continuously optimized using a decision tree algorithm.

[0024] The method is applied to an ultra-thin float glass production line. For display glass substrates within a preset thickness range, it achieves online bubble detection accuracy at a preset accuracy level, with a single evaluation cycle less than a preset time threshold, and completely covers the full width area of ​​the glass melting furnace refining section.

[0025] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0026] This invention enables real-time monitoring of the bubble state inside molten glass by deploying a high-temperature industrial endoscope imaging system in the glass melting furnace and flow channel. Compared with the traditional offline method of sampling and cooling in the laboratory before statistical analysis, the feedback cycle is significantly shortened, making it possible to intervene in the production process in real time. It also enables continuous monitoring in complex dynamic environments: through narrow-band laser illumination and adaptive image algorithms, the influence of high-temperature radiation on image quality is effectively overcome, and stable detection performance can be maintained in the dynamic environment of high-speed flow of molten glass, ensuring the continuity of the production process.

[0027] This invention not only counts the number of bubbles, but also comprehensively considers the bubble size gradient, motion vector, and physical properties of the molten glass to construct a multi-dimensional evaluation model. This comprehensive evaluation method is more in line with the physical nature of glass clarification than simply counting bubbles, and the accuracy of the evaluation results is significantly improved. By introducing a long short-term memory neural network and time series analysis, it can not only reflect the current clarification state, but also predict future trends. Through dynamic comparison with the standard curve, the system can identify minute quality fluctuations, achieving a technological leap from passive detection to active prediction.

[0028] This invention, through deep integration with the automated control system of glass melting furnaces, directly transforms evaluation results into control commands. This closed-loop control based on real-time data reduces the randomness and lag caused by manual experience-based adjustments, making the adjustment of core parameters such as glass melt level and heating power more scientific. It effectively reduces the bubble defect rate in finished glass and improves product yield. At the same time, by optimizing the matching of heating power and drawing speed, it reduces unnecessary energy consumption and realizes a green, efficient, and intelligent production mode.

[0029] This invention constructs a comprehensive production big data system. By establishing a big data platform of a predetermined scale, it enables in-depth mining of testing data and finished product quality indicators. This data correlation mechanism provides objective measured data support for the optimization of process formulas and the improvement of glass melting furnace structures, and promotes the continuous evolution of glass manufacturing technology towards digitalization and intelligence. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall technical solution architecture of an online detection and evaluation method for glass clarity according to an embodiment of the present invention;

[0031] Figure 2 This is a core principle framework diagram of a multi-dimensional evaluation model and closed-loop feedback control based on bubble dynamic characteristics and glass liquid physical parameters in an online detection and evaluation method for glass clarity according to an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1

[0034] To address the complex physical environment of high temperature, high radiation, and high dynamic flow within a glass melting furnace, this embodiment provides an online detection and evaluation method for glass clarity, the implementation of which strictly follows these steps:

[0035] Please refer to Figure 1 In step 1, high-temperature image data of the molten glass is acquired. A high-temperature industrial endoscopic imaging system is installed at specific locations above the refining section and flow channel of the glass melting furnace. The core components of the high-temperature industrial endoscopic imaging system include a high-temperature resistant probe, an optical imaging lens, a charge-coupled device (CCD) sensor, a high-temperature resistant water-cooling kit, a nitrogen purging device, and a high-power narrow-band laser. The high-temperature resistant probe adopts a double-layer cooling sleeve structure made of 316L stainless steel, with circulating cooling water in the inner layer and compressed nitrogen in the outer layer. The water-cooling kit is maintained by a high-pressure water pump to ensure that the sensor operating temperature remains below 45°C in a high-temperature environment of 1600°C. The nitrogen purging device not only plays an auxiliary cooling role, but more importantly, it forms a positive pressure layer at the foreground of the optical lens to prevent condensation of volatiles from the molten glass on the lens surface.

[0036] The optical imaging lens has a field of view of 60 degrees and an object distance range of 150 mm to 2500 mm, enabling clear imaging of deep bubbles in molten glass. To eliminate strong infrared radiation interference generated by the high-temperature molten glass itself, the system uses a high-power narrow-band laser with a wavelength locked at 532 nm as the active illumination source, coupled with a 532 nm narrow-band filter in the optical path with a bandwidth of 10 nm, thereby blocking most of the infrared background noise.

[0037] The image acquisition process is precisely regulated by a synchronous trigger controller. The sampling frequency is set to 30Hz, and the single-frame image resolution is set to 2048×2048 pixels. The synchronous trigger controller ensures precise synchronization between the shutter opening time of the charge-coupled device (CCD) sensor and the pulse emission time of the laser by outputting a TTL level signal; the pulse width is set to 5ms. The acquired raw image data is transmitted to the image processing workstation in real time via the Camera Link protocol.

[0038] The system also includes a self-diagnostic process for the sensor data acquisition status in step 1. A high-precision thermocouple is installed at the return water inlet of the water-cooling kit to monitor the return water temperature, and a mass flow meter is installed on the nitrogen pipeline to monitor the purging flow rate. The system reads the return water temperature and flow rate data in real time. If the temperature exceeds the preset safe temperature threshold of 50°C, or the flow rate falls below the preset flow rate threshold of 20 L / min, the system control unit immediately triggers the emergency protection logic, automatically driving the electric displacement stage to retract the endoscope probe from the kiln observation hole, simultaneously cutting off the laser power supply, and issuing a hardware fault alarm.

[0039] In step 2, the image data is preprocessed and bubble feature information is extracted. Since the original image is inevitably affected by electromagnetic interference and non-uniform background lighting during acquisition, a median filtering algorithm is first used for noise reduction. The filtering window size is set to 5×5 pixels. The median filtering process specifically involves: extracting the grayscale values ​​of 25 pixels within the window, arranging them in ascending order, and taking the 13th value as the output value of the center pixel.

[0040] Subsequently, a background subtraction algorithm is applied to eliminate the ripple interference caused by the molten glass flow. This algorithm constructs a static background model by extracting the pixel mean of 100 consecutive frames. The current frame image is then subtracted from the static background model at the pixel level, i.e.:

[0041]

[0042] in, For difference images, For the current frame, The background mean. The column coordinates of the pixels. Here are the row coordinates of the pixels. This operation effectively highlights dynamically changing bubble targets while suppressing static furnace wall reflections.

[0043] Next, an adaptive thresholding segmentation algorithm is used to separate the bubble target from the molten glass background. This algorithm calculates the average grayscale value of a local image region (32×32 pixels). and variance Dynamically set the segmentation threshold The specific calculation formula is as follows:

[0044]

[0045] in, The correction factor is set to 0.2. This local dynamic thresholding ensures that even under uneven lighting conditions, tiny bubbles with a diameter of less than 0.1 mm can still be accurately extracted.

[0046] After binarization segmentation, the system labels and calculates geometric features of the connected components. For each identified bubble, its equivalent diameter, perimeter, roundness, and coordinate position are calculated. The equivalent diameter of the bubble is calculated by extracting the pixel area occupied by the bubble. ,Will Divide by pi Take the square root and then multiply by the predetermined proportionality factor. The proportionality coefficient Obtained through pre-calibration imaging of a standard-sized object, with units of mm / pixel. Bubble roundness is calculated as follows: in Let be the pixel perimeter of the bubble.

[0047] The bubble dynamic feature extraction process also includes trajectory tracking. The system uses the Hungarian matching algorithm to associate bubbles in two adjacent frames, specifically frames t and t+1. A cost matrix is ​​established, where each element is defined as the combined Euclidean distance of the bubble in terms of spatial location, area, and roundness. By solving for the minimum weighted matching of this cost matrix, the unique identifier of the bubble is determined. The displacement vector of the bubble in the two-dimensional image plane is calculated. By combining the object-side magnification of the imaging system, the three-dimensional motion trend of the bubble within the molten glass was derived, including the rising velocity component and the flow velocity component. All extracted feature information was stored in real-time in a bubble dynamic feature database. This database uses a relational structure, with each record containing a timestamp, bubble ID, and coordinates. equivalent diameter Roundness Displacement vector .

[0048] Please refer to Figure 2 In step 3, a multi-dimensional evaluation model for glass clarity is constructed. This model is based on a bubble dynamic characteristic database and comprehensively considers the physical distribution law of bubbles and the rheological properties of molten glass. First, the number density of bubbles per unit volume of molten glass is calculated. The calculation method is as follows: a virtual three-dimensional spatial window is set within the observation area. Count the total number of bubbles that exist instantaneously within the window. ,but .

[0049] Next, the bubble size distribution gradient is calculated. The equivalent diameter of the bubble is divided into 10 intervals, such as 0.01-0.1 mm, 0.1-0.2 mm, etc., and the proportion of bubbles in each interval is counted to construct a particle size distribution function. The quantitative index of the bubble size distribution gradient reflects the dynamic balance between the elimination of large bubbles and the growth of small bubbles.

[0050] The system also obtains the current melting temperature. And the viscosity parameters of the molten glass calculated based on temperature. The relationship between the viscosity of molten glass and temperature follows the VFT equation:

[0051]

[0052] in , , An empirical constant for a specific glass composition. The asymptotic intercept of the logarithmic viscosity at infinite temperature is limited to this VFT equation. This is the sensitivity coefficient of viscosity to temperature changes. The temperature singularity that causes viscosity to diverge.

[0053] Based on the above parameters, the initial clarification index is obtained through a weighted summation method. The specific implementation of step 3 involves the following core mathematical models:

[0054] Formula 1 is the clarification efficiency calculation formula, which is used to measure the effectiveness of bubble removal in the current area. The specific description is as follows:

[0055]

[0056] in, Indicates clarification efficiency. The predetermined proportionality constant is set to 0.85. The second preset constant is set to 1.0. This indicates the total number of bubbles in the current observation area. This indicates the total number of bubbles in the initial feeding area or a specific reference area.

[0057] Formula 2 is the formula for calculating the average diameter of bubbles, which is used to characterize the macroscopic size characteristics of bubble clusters. The specific description is as follows:

[0058]

[0059] in, Indicates the average diameter. Indicates the first The equivalent diameter of each identified bubble. This represents the total number of bubbles identified during the observation period.

[0060] Furthermore, the model incorporates a viscosity compensation mechanism, as shown in Formula 3. Since temperature fluctuations significantly alter the resistance to molten glass movement, the compensated clarity... The calculation is as follows:

[0061]

[0062] in, Viscosity at standard process temperature The preset power coefficient is set to 0.33. This refers to the viscosity parameter of molten glass. This is the initial clarification level. This is used to eliminate evaluation errors caused by fluctuations in the thermal regime.

[0063] Formula 4 is the bubble removal rate function. The multi-dimensional evaluation model of glass clarity also includes the bubble removal rate function, which defines the clarity index as equal to a first preset constant minus the ratio of the current bubble density to the initial bubble density, multiplied by a temperature correction coefficient. The temperature correction coefficient is determined by the Arrhenius equation, specifically in the form:

[0064]

[0065] in, This is a temperature correction factor. It is the apparent activation energy for gas diffusion. It is a universal gas constant. The melting temperature is denoted as . This coefficient reflects the physical contribution of increased temperature to improving the gas diffusion rate and reducing the resistance to bubble rise. Its bubble removal rate function formula is:

[0066]

[0067] in, Clarity index, This is the first preset constant. Given the current bubble density, The initial bubble density, Let R be the apparent activation energy for gas diffusion, and R be the universal gas constant. The clarity index output by this function, which is the melting temperature, reflects the actual effect of bubble removal in real time.

[0068] In step 4, the evaluation results and deviation analysis are corrected in real time. The system introduces time series analysis methods to process the initial clarification index within a predetermined time period, such as 300 consecutive seconds. Specifically, a Long Short-Term Memory (LSTM) neural network model is used, with the first predetermined historical time period, such as historical clarification data, temperature fluctuation data, and pull speed data from the past hour, as the input feature vector. The LTM neural network contains forget gates, input gates, and output gates, which can effectively capture the nonlinear evolution characteristics of the clarification index over time.

[0069] The system uses a long short-term memory neural network to predict the trend of clarity changes within a second predetermined prediction period of 15 minutes. The prediction accuracy is evaluated by mean square error, and the prediction accuracy is required to be greater than a preset accuracy threshold of 95%.

[0070] Simultaneously, a moving average is applied to the real-time calculated indicators. The window length for the moving average is dynamically adjusted based on the production line's pulling speed. The logic is as follows: if the pulling speed changes by a preset proportion, such as increasing by 10%, the window length is reduced by the same proportion, such as decreasing by 10%, to ensure the real-time nature of the evaluation results under different production loads.

[0071] The system compares the processed clarity index with a preset standard clarity curve. The standard clarity curve is an ideal clarification trajectory measured from high-quality glass samples under simulated laboratory conditions, including the bubble elimination rate constant and the final residual bubble density at different temperature points. The comparison process generates deviation values. The system uses the least squares method to linearly fit the deviation values ​​of 20 consecutive sampling points and calculates the slope of the deviation evolution trend. If the deviation value exceeds a preset deviation threshold of 0.05, or if the deviation trend slope shows a continuously increasing trend, the system automatically sends an early warning signal to the central control room and fully records the current liquid level, heating current, feeding rate, and other process parameters, forming a quality anomaly traceability file.

[0072] In step 5, closed-loop feedback control of the process is executed. The final online evaluation result of glass clarity is transmitted to the glass melting furnace automation control system via an industrial Ethernet protocol such as Profinet or EtherNet / IP. This industrial Ethernet protocol has extremely high real-time characteristics, with a data transmission delay of less than 20ms.

[0073] The automated control system automatically adjusts process parameters according to the difference between the evaluation result and the target value, using PID control logic. The heating power adjustment step is set to 10kW, and the liquid level adjustment accuracy is controlled within ±0.1mm. The pulling speed adjustment range is set within ±5% of the current setting.

[0074] If the evaluation results show a decrease in clarity, the system executes the following adjustment sequence: First, it increases the electric flux heating power in the pre-refining zone to raise the local temperature and reduce viscosity; if the temperature rise reaches its upper limit, it appropriately reduces the drawing speed to extend the residence time of the molten glass in the high-temperature zone; simultaneously, it fine-tunes the operating frequency of the feeder to maintain a stable liquid level. This closed-loop management based on real-time online detection data eliminates the lag of manual intervention and ensures the stability of glass quality.

[0075] This embodiment also includes establishing a big data platform for glass production. This platform aggregates historical data containing over one million records, deeply correlating online-detected clarity data with the transmittance, surface stress index, and bubble defect rate of the finished glass detected at the cold end. Decision tree algorithms such as Random Forest are used to continuously train and optimize the weight coefficients of the evaluation model in step 3, thereby increasing the correlation coefficient between the evaluation indicators and the final product quality. Greater than 0.92.

[0076] This method is applied to ultra-thin float glass production lines, achieving an online bubble detection accuracy of 0.05mm for display glass substrates with a thickness ranging from 0.1mm to 1.1mm, with a single evaluation cycle of less than 10 seconds. By horizontally arranging four sets of high-temperature industrial endoscopes in the refining section of the glass melting furnace, complete coverage of the glass molten state across the entire width is achieved, effectively avoiding localized refining defects caused by liquid flow deviation.

[0077] Example 2

[0078] This embodiment provides an online detection and evaluation method for glass clarity based on dual-spectral imaging. As a supplement to Embodiment 1, it has been specifically optimized in terms of hardware configuration and signal processing logic to meet the production needs of high-volatility specialty glasses such as borosilicate glass.

[0079] In step 1, high-temperature image data of the molten glass is acquired. This embodiment employs a dual-spectrum high-temperature industrial endoscope system, which incorporates two independent charge-coupled device (CCD) sensors, paired with a short-wave infrared filter in the 900-1700nm band and a 532nm visible light narrowband filter, respectively. Because borosilicate glass generates a large amount of boron volatile fumes during the melting process, visible light imaging is easily affected by scattering interference. The short-wave infrared band has strong fog-penetrating ability, enabling the observation of deep bubbles in the molten glass through the fumes.

[0080] The sampling frequency was set to 60Hz to capture the dynamics of high-speed flowing bubbles. A synchronous trigger controller outputs two synchronization pulses to control the shutters of the visible light sensor and the short-wave infrared sensor, respectively. The image resolution was set to 1280×1024 pixels.

[0081] In step 2, the image data is preprocessed and bubble feature information is extracted. The system employs an enhancement algorithm based on pixel-level fusion. First, bilinear interpolation is used to spatially align two images with different resolutions or fields of view. Then, Laplacian pyramid transform is used to decompose the visible light image and the infrared image into sub-bands of different frequencies. In the low-frequency sub-band, a weighted average fusion rule is used to preserve the background contour; in the high-frequency sub-band, the principle of taking the largest absolute value is used to extract the edge details of the bubbles.

[0082] For feature extraction, a bubble recognition model based on a deep convolutional neural network was employed to target the small, dense bubbles commonly found in molten borosilicate glass. This model, running on a local GPU accelerator and trained on 30,000 pre-annotated high-temperature bubble images, can automatically distinguish between real bubbles, ripples on the molten glass surface, and reflected light spots from the furnace wall. A refractive index correction parameter was introduced into the calculation of the equivalent bubble diameter. Considering the refraction of light by the curved surface of the molten glass, the formula is modified as follows:

[0083]

[0084] in, This is the corrected true equivalent diameter of the bubble. The diameter of the bubble pixel measured in the image. This is the proportionality coefficient.

[0085] In step 3, a multi-dimensional evaluation model for glass clarity is constructed. This embodiment focuses on the kinetic analysis of bubble removal. A bubble removal rate constant is introduced. The calculation formula is:

[0086]

[0087] in, The initial bubble density upon entering the clarification section. The current bubble density at the flow channel inlet. This refers to the nominal residence time of the molten glass.

[0088] The multi-dimensional evaluation model also includes a bubble size evolution index. The decomposition efficiency of the clarifying agent is assessed by calculating the rate of change of the average bubble diameter with respect to process position. If small bubbles rapidly merge, grow, and float to the surface, it indicates that the clarifying agent is working effectively.

[0089] In step 4, the evaluation results and deviation analysis are corrected in real time. Considering the frequent process changes in special glass production, the system uses an adaptive Kalman filter algorithm to smooth the clarity index. The state transition matrix of the Kalman filter is updated in real time according to the current production output.

[0090] The deviation analysis module incorporates an expert system. When the deviation value exceeds the threshold, the system not only issues an alarm but also calls upon a preset knowledge base to provide specific process adjustment suggestions based on the sign of the deviation and its rate of evolution, such as "suggest increasing the hydrogen-oxygen ratio of No. 3 septic tank" or "suggest increasing the back pressure of the clarifier tank."

[0091] In step 5, closed-loop feedback control of the process is executed. In this embodiment, communication with the host computer system is achieved through the OPC UA protocol. For ultra-thin glass production, the closed-loop control algorithm adds compensation logic for fluctuations in the drawing speed. If the drawing speed is slightly adjusted due to the needs of the downstream forming process, the control system will automatically compensate the heating power of the refining section in reverse to maintain a constant unit heat load on the molten glass, thereby maintaining stable clarity.

[0092] Example 3

[0093] This embodiment provides an online detection and evaluation method for glass clarity based on multi-point array distribution, mainly targeting the wide-range detection needs of ultra-large float glass melting furnaces with a daily output of over 600 tons.

[0094] In step 1, high-temperature image data of the molten glass is acquired. Six sets of high-temperature industrial endoscope imaging systems are evenly arranged along the width of the glass melting furnace refining section to form a detection array. The installation spacing of each endoscope is set to 1500mm. The video streams from all endoscopes are connected to the central server through a 10 Gigabit fiber optic switch.

[0095] The sampling frequency is uniformly set to 25Hz. A network time synchronization protocol ensures that the timestamp synchronization accuracy of each device is within 1ms. The resolution of a single frame image is 1920×1080 pixels.

[0096] In step 2, image data is preprocessed and bubble feature information is extracted. The system employs a distributed parallel computing architecture. The video stream corresponding to each endoscope is subjected to median filtering and adaptive threshold segmentation by an independent computing node. To handle bubbles in overlapping areas of multi-camera field of view, the system performs coordinate space transformation, mapping the image coordinates of all bubbles to a unified physical coordinate system of the glass melting furnace.

[0097] For bubbles moving across different fields of view, a distributed Hungarian algorithm is used for global tracking. When a bubble leaves the field of view of camera 1 and enters the field of view of camera 2, the system determines it as the same target by the consistency of the predicted trajectory, thus avoiding duplicate counting when calculating density.

[0098] In step 3, a multi-dimensional evaluation model for glass clarity is constructed. This embodiment introduces a clarity uniformity index. The calculation method is as follows: calculate the standard deviation of the local clarity measured at 6 detection points. .

[0099]

[0100] in, This represents the average clarity across the entire kiln width. This indicator can sensitively reflect quality fluctuations caused by uneven lateral temperature distribution in the kiln.

[0101] In step 4, the evaluation results and deviation analysis are corrected in real time. The system utilizes multi-source data fusion technology to incorporate real-time resistivity data measured by the glass melt electrode into the correction model. Changes in resistivity can indirectly reflect the distribution of alkali metal ions in the glass melt and the blocking effect of bubbles on the conductive path.

[0102] Deviation analysis employs an anomaly diagnosis model based on cluster analysis. Real-time multi-point detection vectors are projected onto a low-dimensional feature space, and the Mahalanobis distance between the current state point and the center of the normal operating cluster is calculated to determine whether the production state deviates from the process window.

[0103] In step 5, closed-loop feedback control of the process is implemented. The automated control system achieves precise zonal control of the clarification state by independently adjusting six groups of laterally arranged burners or electric heating electrodes. If the local clarity on a certain side is too low, the system automatically increases the fuel flow rate or current intensity on that side. This lateral differential adjustment capability is key to maintaining high yield in large-scale glass production.

[0104] Example 4

[0105] This embodiment describes an online detection and evaluation method for glass clarity applied to an all-electric melting glass furnace, focusing on solving the problem of monitoring deep, high-density bubbles.

[0106] In step 1, high-temperature image data of the molten glass is acquired. Due to the deep liquid level and large vertical gradient inside the all-electric melting furnace, the imaging system uses a high-temperature resistant probe with zoom function, whose focal length adjustment range is 10mm to 50mm. By driving the focusing lens group with a stepper motor, the system can achieve alternating scanning imaging of the 0-100mm surface layer of the molten glass and the 100-500mm deep layer.

[0107] The light source employs a high-power near-infrared laser with a wavelength of 1064nm, which has stronger penetrating power. Since there is no flame interference inside the all-electric melting furnace, the background is relatively simple, but the electromagnetic environment near the electrodes is extremely harsh. Therefore, all control and signal cables of the imaging system utilize multi-layer shielding structures and are equipped with industrial-grade surge suppressors.

[0108] In step 2, image data is preprocessed and bubble feature information is extracted. Considering the influence of electromagnetic convection on the bubble's upward path in the all-electric melting furnace, a fluid dynamics constraint model is incorporated into the bubble trajectory tracking. Using the flow field vector pre-calculated by computational fluid dynamics as prior information, the Hungarian matching algorithm is assisted in target association, significantly improving the bubble recognition accuracy in turbulent regions.

[0109] In step 3, a multi-dimensional evaluation model for glass clarity is constructed. A unique indicator in this embodiment is the bubble residence time distribution. By statistically analyzing the time it takes for bubbles of different sizes to move from the deep layer to the surface, the kinetic parameters of the clarification process are calculated. The proportionality constant in the clarification efficiency formula... Dynamic adjustments are made based on the electrical power load of the all-electric melting furnace.

[0110] In step 4, the evaluation results and deviation analysis are corrected in real time. Derivative analysis of the deviation evolution trend is introduced. If the rate of decrease in clarity exceeds the preset change rate threshold, the system determines it as a sudden process disturbance, such as fluctuations in raw material composition, and immediately triggers emergency intervention logic.

[0111] In step 5, closed-loop feedback control of the process is implemented. The system directly controls the electrode regulator of the all-electric melting furnace, changing the current distribution ratio of each electrode layer, and controlling the vertical upward floating force of the bubbles by adjusting the vertical temperature gradient. Data transmission adopts a redundant ring network structure with high anti-interference capability.

[0112] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

[0113] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for online detection and evaluation of glass clarity, characterized in that, Includes the following steps: Step 1: Install a high-temperature industrial endoscope imaging system above the refining section or flow channel of the glass melting furnace. The built-in charge-coupled device sensor of the high-temperature industrial endoscope imaging system acquires the original images of the surface and deep layers of the glass melt in real time under high temperature environment and transmits them to the image processing workstation. Step 2, preprocess image data and extract bubble feature information: use median filtering algorithm to denoise the acquired original image, apply background subtraction algorithm to eliminate the ripple interference caused by the flow of glass liquid, use adaptive threshold segmentation algorithm to separate the bubble target in the image from the glass liquid background, and establish a real-time bubble dynamic feature database. Step 3: Based on the real-time bubble dynamic feature database, calculate the bubble number density, bubble size distribution gradient and average bubble rise rate per unit volume of molten glass, and obtain the current melting temperature and molten glass viscosity parameters in real time. The melting temperature and molten glass viscosity parameters are fused by weighted accumulation. A clarification efficiency calculation model involving the bubble removal rate function and a viscosity compensation model to eliminate temperature fluctuation errors are introduced to obtain the initial clarification index reflecting the quality state of molten glass. Step 4: Introduce time series analysis method to process the initial clarity index within the predetermined time period by moving average to eliminate short-term random fluctuation interference. Compare the processed index with the standard clarity curve determined by high-quality glass samples, calculate the deviation value reflecting the difference between the actual production state and the ideal trajectory, and use the least squares method to perform linear fitting on the deviation value of continuous sampling points to analyze the deviation evolution trend, and output the final online evaluation result of glass clarity. Step 5: Implement closed-loop feedback control of the process. Based on the final online evaluation results of glass clarity, implement closed-loop feedback control of the process parameters of the glass melting furnace.

2. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: In step 1, the high-temperature industrial endoscope imaging system is equipped with a high-temperature water-cooling kit and a nitrogen purging device. The high-temperature industrial endoscope imaging system acquires images through a built-in charge-coupled device sensor, uses a synchronous trigger controller to ensure that the image acquisition frequency is consistent with the light source pulse frequency, uses a high-power narrow-band laser for active illumination, and uses a narrow-band filter set in the optical path to block the infrared background noise generated by the high-temperature glass melt itself. The high-temperature water-cooling kit adopts a double-layer cooling sleeve structure. The inner layer is circulated with cooling water to maintain the sensor's operating temperature within a preset safe range, while the outer layer is circulated with compressed nitrogen and forms a positive pressure layer at the very front of the optical imaging lens to prevent glass melt volatiles from condensing on the lens surface. The wavelength of a high-power narrow-band laser is locked in a specific visible light band, and the bandwidth of the narrow-band filter is set to a predetermined bandwidth, thereby blocking most of the infrared background radiation from interfering with the contrast of bubble imaging. The synchronous trigger controller ensures that the shutter opening time of the charge-coupled device sensor is precisely synchronized with the pulse emission time of the high-power narrow-band laser by outputting a level signal.

3. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: In step 2, when preprocessing image data, the median filtering algorithm specifically extracts the gray values ​​of all pixels within the filtering window and sorts them in ascending order, selecting the median value as the output value of the center pixel; the background subtraction algorithm constructs a static background model by extracting the pixel mean of a predetermined number of consecutive frames of images, performs pixel-level subtraction between the current frame image and the static background model, highlights the dynamically changing bubble target and suppresses static furnace wall reflection; The adaptive threshold segmentation algorithm introduces a correction coefficient and dynamically calculates the segmentation threshold of each region by combining the gray-scale distribution characteristics of the local region, thereby enabling the extraction of microbubble targets under uneven lighting conditions.

4. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: When extracting bubble feature information in step 2, a trajectory tracking algorithm is used to associate bubbles in two adjacent frames of images. A cost matrix containing a comprehensive weight of spatial location, area, and roundness is established, and the minimum weight matching of the cost matrix is ​​solved to determine the unique number of the bubble. The equivalent diameter of the bubble is obtained by extracting the pixel area occupied by the bubble, dividing the pixel area by pi, taking the square root, and then multiplying by a pre-calibrated scaling factor. The three-dimensional motion trend includes the velocity component of the bubble in the upward direction and the velocity component in the direction of glass melt flow. The real-time bubble dynamic feature database adopts a relational structure. Each record contains a timestamp, a unique bubble number, coordinates, equivalent diameter, roundness, and displacement vector.

5. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: In step 3, when calculating the number density of bubbles per unit volume of molten glass, a virtual three-dimensional spatial window is set within the observation area, and the total number of bubbles instantaneously present within the three-dimensional spatial window is counted. The bubble particle size distribution gradient is constructed by dividing the equivalent diameter of bubbles into multiple intervals and counting the proportion of bubbles in each interval to construct a particle size distribution function, which reflects the dynamic equilibrium state between bubble elimination and bubble growth. The viscosity parameter of molten glass is calculated based on the melting temperature through a nonlinear mapping equation between viscosity and temperature, reflecting the rheological characteristics of a specific glass composition under the current thermal regime. The viscosity compensation mechanism formula is as follows: ， in, For the clarity after compensation, Viscosity at standard process temperature The preset power coefficient is set to 0.

33. This refers to the viscosity parameter of molten glass. This represents the initial clarity.

6. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: The clarification efficiency calculation model involved in step 3 is used to measure the effectiveness of bubble removal in the current area. It is expressed as follows: clarification efficiency equals a predetermined proportional constant multiplied by the difference between a second preset constant and the ratio, where the ratio is the total number of bubbles in the current observation area divided by the baseline total number of bubbles in the reference area. The bubble removal rate function defines the clarification index as equal to the first preset constant minus the ratio of the current bubble density to the initial bubble density, multiplied by a temperature correction coefficient determined by the Arrhenius equation. The temperature correction coefficient is determined by the Arrhenius equation, specifically in the following form: ， in, This is a temperature correction factor. It is the apparent activation energy for gas diffusion. It is a universal gas constant. The melting temperature is the coefficient that reflects the physical contribution of temperature increase to improving gas diffusion rate and reducing bubble rising resistance. Its bubble removal rate function formula is as follows: ， in, Clarity index, This is the first preset constant. Given the current bubble density, The initial bubble density, Let R be the apparent activation energy for gas diffusion, and R be the universal gas constant. This refers to the melting temperature.

7. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: The time series analysis method in step 4 uses a long short-term memory neural network model. First, historical clarification data, temperature fluctuation data, and pull speed data within a predetermined historical period are used as input feature vectors. The nonlinear evolution characteristics of clarification indicators over time are captured through the forget gate, input gate, and output gate structure inside the long short-term memory neural network, and the trend of clarification change within the predetermined prediction period is predicted. Then, the initial clarity index is processed by moving average to eliminate short-term random fluctuations. The window length of the moving average is dynamically adjusted according to the pulling speed of the production line. If the pulling speed changes by a preset proportion, the window length is adjusted by the same proportion to ensure the real-time nature of the evaluation results under different production loads.

8. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: In step 4, if the deviation value exceeds the preset deviation threshold or the deviation trend shows that the deviation is likely to continue to expand, the system will automatically send an early warning signal to the central control room and record the current liquid level, heating current and feeding speed to form a quality anomaly traceability file.

9. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: When performing process closed-loop feedback control in step 5, if the final online evaluation result of glass clarity shows a decrease in clarity, the automated control system executes an adjustment sequence, prioritizing the increase of electric flux heating power in the front zone of the clarifying section. If the temperature rise reaches the upper limit, the drawing speed is reduced to prolong the residence time of the molten glass in the high-temperature zone, and the operating frequency of the feeder is adjusted simultaneously to maintain a stable liquid level.

10. The method for online detection and evaluation of glass clarity according to claim 1, characterized in that: It also includes establishing a big data platform for glass production, which deeply correlates the online detected clarity data with the transmittance, stress, and bubble defect rate of the finished glass, and uses a decision tree algorithm to continuously optimize the weight coefficients in the evaluation model. This method is applied to ultra-thin float glass production lines. For display glass substrates within a preset thickness range, multiple sets of high-temperature industrial endoscope imaging systems are arranged laterally in the clarification section of the glass melting furnace to form a detection array, achieving complete coverage of the clarification state of the glass melt across the entire width area. Coordinate space transformation is then performed to map the image coordinates of all bubbles to a unified physical coordinate system of the glass melting furnace.

Images