A method for detecting the quality of a glass surface spray coating

Abstract

The present application relates to the technical field of glass detection, in particular to a glass surface spraying quality detection method, comprising: generating a surface virtual topology model of a target glass substrate according to multispectral surface image data; determining a coating thickness uniformity index according to the surface virtual topology model; sending an environment parameter acquisition request message to an air flow sensor; receiving a first environment parameter response message from the air flow sensor; wherein the first environment parameter response message includes real-time air flow disturbance data; and calculating a defect formation risk probability according to the coating thickness uniformity index and the real-time air flow disturbance data. The present application can identify possible defect areas in advance by calculating the defect formation risk probability, providing a basis for subsequent processing and repair, and improving the stability and quality control of production.

Description

Technical Field

[0001] This invention relates to the field of glass testing technology, specifically a method for testing the quality of glass surface coating. Background Technology

[0002] Currently, traditional methods typically rely on visual observation or single-wavelength optical detection, which makes it difficult to fully capture the microscopic features of the coating layer, resulting in low accuracy in detecting coating thickness uniformity and defects. Furthermore, many traditional detection methods cannot monitor the spraying process in real time, making it impossible to make rapid adjustments when environmental conditions change, thereby increasing the risk of defects. In addition, traditional methods are usually based on static standards for evaluation and cannot be dynamically adjusted according to environmental factors such as real-time airflow disturbances, which may cause them to miss the opportunity to detect and correct potential problems in a timely manner.

[0003] Furthermore, due to the lack of effective data analysis tools, traditional methods often fail to accurately calculate the risk probability of defect formation, resulting in untimely identification of potential defect areas and affecting the effectiveness of subsequent processing and repair. Moreover, traditional detection methods can usually only provide macroscopic location after discovering defects, and it is difficult to perform sub-pixel-level precise positioning, which affects the efficiency of defect tracking and management. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting the quality of glass surface coating, wherein a patrol inspection mode is activated on the display interface of a quality inspection terminal, and a real-time monitoring screen is displayed in the patrol inspection mode; a selection operation of a target glass substrate in the real-time monitoring screen is detected; In response to the selection operation, an image acquisition request message is sent to the multispectral imaging array; and a first image acquisition response message is received from the multispectral imaging array; wherein the first image acquisition response message includes multispectral surface image data; A virtual topological model of the target glass substrate is generated based on multispectral surface image data; the coating thickness uniformity index is determined based on the virtual topological model. Sending an environmental parameter acquisition request message to an airflow sensor; and receiving a first environmental parameter response message from the airflow sensor; wherein the first environmental parameter response message includes real-time airflow disturbance data; The probability of defect formation risk is calculated based on coating thickness uniformity index and real-time airflow disturbance data; If the probability of defect formation exceeds the preset risk threshold, the coating thickness uniformity index is dynamically compensated and corrected based on the surface virtual topology model and real-time airflow disturbance data to obtain the target quality assessment index; a quality inspection report is then sent to the quality inspection terminal based on the target quality assessment index.

[0005] Preferably, the image acquisition request message includes identification information of the target glass substrate, and the image acquisition request message is used to instruct the multispectral imaging array to acquire surface images of the target glass substrate at multiple preset wavelengths.

[0006] Preferably, the target glass substrate is any glass substrate to be inspected as determined by the user; the surface virtual topology model is used to characterize the microscopic geometric features of the coating on the target glass substrate; the multispectral surface image data includes visible light band images and infrared band images; A virtual topological model of the target glass substrate is generated based on multispectral surface image data, including: Infrared band images are extracted from multispectral surface image data; among them, infrared feature images are used to characterize the differences in thermal conductivity inside the sprayed coating. Potential void regions inside the coating are identified based on infrared feature images, and void distribution data is obtained. The visible light band image is extracted from the multispectral surface image data to obtain the surface texture image; the surface texture image is used to characterize the reflective properties of the sprayed coating surface. An initial geometric mesh model is constructed based on the surface texture image; the initial geometric mesh model consists of multiple triangular facets. The void distribution data is mapped to the corresponding grid nodes in the initial geometric grid model to obtain an intermediate topology model with void properties; The intermediate topology model is smoothed by surface fitting to obtain the virtual topology model of the target glass substrate.

[0007] Preferably, the coating thickness uniformity index is used to indicate the degree of thickness dispersion of the sprayed coating at different locations; The coating thickness uniformity index is determined based on the surface virtual topology model, including: Multiple sampling paths are determined in the surface virtual topology model; wherein, the multiple sampling paths cover the central region and edge region of the target glass substrate; Calculate the rate of curvature change on each sampling path to obtain multiple sequences of rates of curvature change; The thickness estimate for each sampling path is calculated based on multiple curvature change rate sequences, resulting in multiple thickness estimates; Calculate the standard deviation of multiple thickness estimates to obtain the thickness discrete parameters; The coating thickness uniformity index is calculated based on the thickness dispersion parameter and the mean of multiple thickness estimates.

[0008] Preferably, real-time airflow disturbance data is used to indicate the motion vector and velocity variance of aerosol particles in the current environment; the defect formation risk probability is used to indicate the likelihood of orange peel texture or sagging defects occurring on the surface of the target glass substrate under the current airflow environment. The probability of defect formation risk is calculated based on coating thickness uniformity index and real-time airflow disturbance data, including: The airflow velocity variance is determined based on real-time airflow disturbance data; whereby the airflow velocity variance is used to indicate the degree of airflow instability. If the variance of the airflow velocity is greater than the preset variance threshold, the motion vector in the real-time airflow disturbance data is extracted to obtain the target motion vector field; The deposition trajectory of aerosol particles on the surface of the target glass substrate is determined based on the target motion vector field, and multiple predicted deposition trajectories are obtained. The number of matched trajectories is obtained by matching multiple predicted deposition trajectories with low curvature regions in the surface virtual topology model. The probability of defect formation is calculated based on the number of matching trajectories, coating thickness uniformity index, and preset risk weight coefficient.

[0009] Preferably, the coating thickness uniformity index is dynamically compensated and corrected based on the surface virtual topology model and real-time airflow disturbance data to obtain the target quality evaluation index, including: The main direction of airflow is determined based on real-time airflow disturbance data; the main direction of airflow is used to indicate the main impact direction of airflow on the sprayed coating. In the surface virtual topology model, a target cross section perpendicular to the main airflow direction is determined, and the target cross section contour is obtained. The local sagging sensitivity coefficient is calculated based on the target cross-sectional profile; the local sagging sensitivity coefficient is used to indicate the tendency of the coating to accumulate liquid under airflow impact; The compensation gain coefficient is determined based on the local sag sensitivity coefficient and the probability of defect formation risk. The coating thickness uniformity index is weighted and adjusted based on the compensation gain coefficient to obtain the target quality evaluation index.

[0010] Preferably, before sending the image acquisition request message to the multispectral imaging array, the method further includes: The target material refractive index is obtained by querying the preset material mapping table based on the identification information of the target glass substrate. An initial digital twin is constructed based on the refractive index of the target material and the expected coating thickness of the target glass substrate; The detection accuracy parameters of the multispectral imaging array are dynamically reconstructed based on the defect confidence level in the initial digital twin and the preset detection mode gradient table. The detection accuracy parameters include scan line density and exposure gain coefficient; the dynamically reconstructed detection accuracy parameters are used to generate multispectral surface image data in the first image acquisition response message.

[0011] Preferably, the detection accuracy parameters of the multispectral imaging array are dynamically reconstructed based on the defect confidence level in the initial digital twin and a preset detection mode gradient table, including: The defect confidence level of each micro-region in the initial digital twin is calculated in real time, and the defect confidence levels calculated multiple times are compared with the preset confidence level threshold. When the defect confidence level is greater than or equal to the confidence level threshold, the line density gradient table and exposure gain gradient table of the corresponding detection area are queried according to the current transmission speed of the glass substrate and the previously generated digital twin. Based on the retrieved line density gradient table and exposure gain gradient table, calculate the signal-to-noise ratio of signal acquisition under different combinations of scan line density and exposure gain; Different combinations of scan line density and exposure gain are sorted from high to low according to signal-to-noise ratio; Select the scan line density and exposure gain combination that is one position higher than the currently configured scan line density and exposure gain combination, and update the currently configured scan line density and exposure gain of the multispectral imaging array.

[0012] Preferably, after sending the quality inspection report to the quality inspection terminal according to the target quality assessment indicators, the process also includes: If the target quality assessment index is lower than the preset qualified threshold, then the high confidence defect sites in the surface virtual topology model are located at the sub-pixel level according to the updated scan line density and exposure gain. The located subpixel coordinates are bound to the real-time position code of the target glass substrate to generate a defect physical marking instruction; By using a high-precision laser galvanometer to perform micro-etching marks at corresponding positions on the target glass substrate according to the physical marking instructions for defects, the physical source of defects can be traced.

[0013] Preferably, before sending an environmental parameter acquisition request message to the airflow sensor, the method further includes: When the glass production line is started, the material code and process parameters of the target glass substrate are received. The process parameters include the expected coating thickness, transfer speed, and second identification information, which includes the batch number and production line number. Based on the optical refractive index corresponding to each material code, a detection mode gradient table for different coating thicknesses and a mapping table between the corresponding material codes and the detection mode gradient table are constructed. The mapping table is used to indicate the mapping relationship between batch number, production line number, and line density gradient table and exposure gain gradient table.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention acquires images at different wavelengths using a multispectral imaging array, enabling a more comprehensive analysis of the microscopic features of the coating layer and improving the detection accuracy of coating thickness uniformity and defects. Furthermore, by combining real-time airflow disturbance data, it can dynamically adjust the coating thickness uniformity index according to environmental changes, promptly identify and correct potential problems, and reduce the risk of defects occurring. This invention calculates the probability of defect formation, enabling the pre-identification of potential defect areas and providing a basis for subsequent processing and repair, thereby improving production stability and quality control. In the event of non-conformity, it can perform sub-pixel-level defect localization, ensuring more accurate physical marking of defects, thus achieving effective defect tracking and management. Furthermore, by binding the physical defect markings with real-time location codes, it enables product traceability, facilitating effective quality control and accountability when problems arise. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall method steps in Embodiment 1 of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1, please refer to Figure 1 This invention provides a technical solution: a method for detecting the quality of glass surface coating, comprising: S1. Start the inspection mode on the display interface of the quality inspection terminal. The real-time monitoring screen is displayed in the inspection mode; the selection operation of the target glass substrate in the real-time monitoring screen is detected. S2. In response to the selection operation, send an image acquisition request message to the multispectral imaging array; and receive a first image acquisition response message from the multispectral imaging array; wherein the first image acquisition response message includes multispectral surface image data. S3. Generate a virtual topology model of the target glass substrate based on multispectral surface image data; determine the coating thickness uniformity index based on the virtual topology model. S4. Send an environmental parameter acquisition request message to the airflow sensor; and receive a first environmental parameter response message from the airflow sensor; wherein the first environmental parameter response message includes real-time airflow disturbance data; S5. Calculate the probability of defect formation risk based on the coating thickness uniformity index and real-time airflow disturbance data; S6. If the probability of defect formation is greater than the preset risk threshold, the coating thickness uniformity index is dynamically compensated and corrected based on the surface virtual topology model and real-time airflow disturbance data to obtain the target quality assessment index; a quality inspection report is sent to the quality inspection terminal based on the target quality assessment index.

[0018] It should be noted that on the display interface of the quality inspection terminal, the operator selects to enter the inspection mode. At this time, the system begins to monitor the actual status of the target glass substrate and displays the monitoring screen in real time. When the operator finds that a target glass substrate in the monitoring screen needs to be inspected in detail, he / she can use the mouse to select it. For example, if the quality inspector notices that there is abnormal light reflection at the edge of a glass substrate, he / she can select the substrate with the mouse. After receiving the selection operation, the system sends a request to the multispectral imaging array, requesting image acquisition of the selected glass substrate; for example, the system sends an instruction to the multispectral imaging device to request the capture of a multispectral image of the selected glass substrate; after the multispectral imaging array completes the image acquisition, it returns a message containing the captured multispectral surface image data; for example, the system receives data from the multispectral imaging array, which includes reflection information at different wavelengths. Based on the acquired multispectral image data, the system creates a virtual topological model of the glass substrate surface to analyze its surface features. For example, by processing the image data, the system constructs a three-dimensional model that shows the fine structure and coating distribution of the glass substrate surface. Using the virtual topological model, the system calculates the coating thickness uniformity index to assess whether it meets the standards. For example, after analyzing the model, the system finds that the coating thickness is uniform in most areas, but there is a significant deviation in a small area. The system sends requests to the airflow sensor to obtain current environmental parameters, such as airflow disturbance data, which is crucial for assessing the risk of coating defects. For example, the system queries the airflow sensor for the current airflow speed and direction to better understand factors that may affect coating quality. Combining coating thickness uniformity indicators and real-time airflow disturbance data, the system calculates the probability of defect formation. For example, the system's algorithm analysis shows that if the coating thickness is uneven and the airflow disturbance is large, the probability of defect formation is 15%. If the calculated probability of defect formation exceeds the preset risk threshold, the system will adjust the coating thickness uniformity index based on the virtual topology model and airflow data to obtain a more accurate quality assessment index. For example, if the risk probability exceeds 10%, the system will automatically adjust the coating thickness assessment, taking into account the influence of airflow to reduce the possibility of misjudgment. The system will send the adjusted target quality assessment index to the quality inspection terminal to generate a quality inspection report. For example, the quality inspector receives a report showing that although the coating quality of this glass substrate has some problems, the corrected assessment results indicate that it is still within the acceptable range.

[0019] In an optional embodiment, the image acquisition request message includes identification information of the target glass substrate, and the image acquisition request message is used to instruct the multispectral imaging array to acquire surface images of the target glass substrate at multiple preset wavelengths.

[0020] It should be noted that the image acquisition request message contains identification information of the target glass substrate, such as production batch number and serial number, to uniquely identify the glass substrate to be inspected. The main purpose of this message is to notify the multispectral imaging array to acquire images. For example, on a production line, multiple glass substrates are undergoing spray coating. Quality inspectors find that one substrate (numbered "GB12345") may have uneven coating, so they send an image acquisition request message, which includes "GB12345" as identification information to ensure that the system can accurately identify and focus on inspecting this substrate. The image acquisition request message not only contains identification information but also instructs the multispectral imaging array to acquire surface images of the target glass substrate at multiple preset wavelengths. This is because different wavelengths of light can reveal different characteristics of the material surface, such as coating thickness, color, and uniformity. For example, for a glass substrate identified as "GB12345", the system may instruct the multispectral imaging array to use multiple wavelengths such as 400nm, 500nm, 600nm, and 700nm for imaging. In this way, the imaging array can capture rich information about the coating, such as the reflectance changes at certain wavelengths, thereby helping to analyze the quality of the coating. Upon receiving an image acquisition request message, the multispectral imaging array will take pictures of the target glass substrate according to the indicated information, thereby acquiring surface image data covering different wavelengths. For example, once the request is received, the imaging array starts working; at a wavelength of 400nm, the imaging array may capture tiny imperfections on the glass surface, while at wavelengths of 500nm and 600nm, it may show differences in coating thickness. Finally, all of these image data will be stored and transmitted back to the quality inspection system.

[0021] In an optional embodiment, the target glass substrate is any glass substrate to be inspected as determined by the user; the surface virtual topology model is used to characterize the microscopic geometric features of the coating on the target glass substrate; the multispectral surface image data includes visible light band images and infrared band images; A virtual topological model of the target glass substrate is generated based on multispectral surface image data, including: Infrared band images are extracted from multispectral surface image data; among them, infrared feature images are used to characterize the differences in thermal conductivity inside the sprayed coating. Potential void regions inside the coating are identified based on infrared feature images, and void distribution data is obtained. The visible light band image is extracted from the multispectral surface image data to obtain the surface texture image; the surface texture image is used to characterize the reflective properties of the sprayed coating surface. An initial geometric mesh model is constructed based on the surface texture image; the initial geometric mesh model consists of multiple triangular facets. The void distribution data is mapped to the corresponding grid nodes in the initial geometric grid model to obtain an intermediate topology model with void properties; The intermediate topology model is smoothed by surface fitting to obtain the virtual topology model of the target glass substrate.

[0022] It should be noted that the target glass substrate can be any glass substrate specified by the user for inspection. During the production process, multiple glass substrates may need to undergo quality inspection. For example, a company is producing a batch of automotive windshields, and the quality inspector selects one of them as the target glass substrate for coating quality inspection. The surface virtual topology model refers to a three-dimensional digital model reconstructed based on multispectral surface image data (including visible and infrared bands) to characterize the microscopic geometry and internal structure (such as voids) of the coating surface of the target glass substrate. This model is composed of triangular meshes and includes derived attributes such as curvature and thickness estimation. The surface virtual topology model is used to characterize the microscopic geometric features of the coating on the target glass substrate. This model is built by analyzing image data to better understand the shape and structure of the coating. For example, by processing the collected image data, a virtual model is generated to show the thickness variation and surface undulation of the coating, helping quality inspectors to intuitively view the overall state of the coating. Multispectral surface image data includes images from the visible and infrared bands, providing information on the coating's performance under different spectra. For example, a quality inspection system acquires a set of images where the visible light images show the coating's color and reflectivity, while the infrared images reveal its thermal properties. Infrared images can be extracted from multispectral surface image data to analyze differences in thermal conductivity within the coating layer. For instance, quality inspectors extract infrared images from multispectral data and find that the thermal conductivity in certain areas is significantly lower than in other areas, which may indicate the presence of voids or defects in these areas. Gaussian filtering is applied to denoise infrared feature images; an adaptive threshold segmentation algorithm (such as the Otsu algorithm) is used to binarize the images, distinguishing between suspected void regions and normal regions; morphological opening operations are performed on the binary images to eliminate noise points and smooth region boundaries; connected components are extracted, and features such as area and roundness of each connected component are calculated; regions with areas smaller than a preset value (such as 10 pixels) or with insufficient roundness are removed, and the center coordinates and contours of the remaining connected components are used as void distribution data; for example, by analyzing infrared feature images, quality inspectors discovered several areas with abnormal temperatures, which were marked as potential void locations, forming a void distribution data record; visible light band images are extracted from multispectral surface image data to obtain surface texture images, thereby analyzing the reflective properties of the coating; for example, the quality inspection system extracts visible light images, showing the smoothness and reflectivity of the sprayed coating, helping to assess the visual quality of the coating; By employing structured light 3D reconstruction technology or photometric stereoscopic vision technology, multiple surface texture images from different angles or under different lighting conditions in the visible light band are used to calculate the height information of each pixel, thereby generating an initial geometric mesh composed of triangular facets to showcase the surface features of the sprayed coating. For example, quality inspectors can use the surface texture data to generate a 3D mesh model representing the subtle undulations and textures of the coating surface. The void distribution data is then mapped onto the corresponding mesh nodes in the initial geometric mesh model to form an intermediate topological model with void attributes. For example, combining the identified void area information with the mesh model generates a new model that indicates the void location information, helping to intuitively understand the integrity of the coating. The intermediate topology model is subjected to surface fitting and smoothing to obtain a virtual topology model of the target glass substrate surface. For example, after surface fitting, the final virtual topology model not only accurately reflects the geometric features of the coating, but also eliminates the noise caused by data acquisition, making the model smoother and more usable.

[0023] In an alternative embodiment, the coating thickness uniformity index is used to indicate the degree of thickness dispersion of the sprayed coating at different locations; The coating thickness uniformity index is determined based on the surface virtual topology model, including: Multiple sampling paths are determined in the surface virtual topology model; wherein, the multiple sampling paths cover the central region and edge region of the target glass substrate; Calculate the rate of curvature change on each sampling path to obtain multiple sequences of rates of curvature change; The thickness estimate for each sampling path is calculated based on multiple curvature change rate sequences, resulting in multiple thickness estimates; Calculate the standard deviation of multiple thickness estimates to obtain the thickness discrete parameters; The coating thickness uniformity index is calculated based on the thickness dispersion parameter and the mean of multiple thickness estimates.

[0024] It should be noted that the coating thickness uniformity index is a comprehensive parameter used to quantify the dispersion of the coating thickness distribution. It is usually derived from the standard deviation and mean of the thickness estimates calculated from multiple sampling paths in a virtual topology model. The smaller the value, the more uniform the thickness. Uniform coating thickness helps improve the performance and appearance of products. For example, in automobile manufacturing, if the UV protection coating on the windshield is uneven in thickness, it will lead to poor protection performance in some areas, thus affecting its service life. In the surface virtual topology model, multiple sampling paths are selected; these paths need to cover the central and edge areas of the target glass substrate to comprehensively evaluate the coating thickness; for example, quality inspectors design several sampling paths on the virtual model, some of which are located in the center of the glass and others along the edge, so as to ensure data obtained from different locations. For each sampling path, the rate of curvature change is calculated; the change in curvature can reflect the change in coating thickness; for example, using the acquired virtual model, the quality inspection system analyzed the curvature data of each path. For example, the curvature change is small at the center of the path, while there may be obvious curvature fluctuations at the edge due to the different coating thicknesses. Based on multiple curvature change rate sequences, the thickness estimate for each sampling path is calculated. The conversion relationship is based on a pre-established curvature-thickness calibration model, which is obtained experimentally: standard samples of known thicknesses are scanned to establish a mapping relationship between their surface curvature change rate and the true thickness. Linear or polynomial regression fitting can be used. In actual inspection, the measured curvature change rate is substituted into the model to estimate the coating thickness at the corresponding location. These estimates provide coating thickness information for each location. For example, quality inspectors deduce the thickness estimate for each sampling path from the curvature change; for example, the thickness of the center path is 1.2 mm, while the thickness of the edge path may only be 0.9 mm. By statistically analyzing multiple thickness estimates, their standard deviations are calculated to obtain the thickness dispersion parameter. The smaller the standard deviation, the more uniform the coating thickness. For example, if the thickness estimates obtained by the quality inspector are 1.2 mm, 1.1 mm, 0.9 mm, 1.0 mm, and 1.1 mm, then the calculated standard deviation of these values ​​is 0.1 mm, indicating relatively good thickness consistency. Based on the mean of the thickness dispersion parameter and the thickness estimate, the coating thickness uniformity index is calculated to form the final quality assessment. For example, the quality inspector combines the calculated mean thickness (e.g., 1.1 mm) with the standard deviation (0.1 mm) to obtain a uniformity index of 0.09. This value is used to determine whether the coating meets the company's quality standards.

[0025] In an optional embodiment, real-time airflow disturbance data is used to indicate the motion vector and velocity variance of aerosol particles in the current environment; the defect formation risk probability is used to indicate the likelihood of orange peel texture or flow defects occurring on the surface of the target glass substrate under the current airflow environment. The probability of defect formation risk is calculated based on coating thickness uniformity index and real-time airflow disturbance data, including: The airflow velocity variance is determined based on real-time airflow disturbance data; whereby the airflow velocity variance is used to indicate the degree of airflow instability. If the variance of the airflow velocity is greater than the preset variance threshold, the motion vector in the real-time airflow disturbance data is extracted to obtain the target motion vector field; The deposition trajectory of aerosol particles on the surface of the target glass substrate is determined based on the target motion vector field, and multiple predicted deposition trajectories are obtained. The number of matched trajectories is obtained by matching multiple predicted deposition trajectories with low curvature regions in the surface virtual topology model. The probability of defect formation risk is calculated based on the number of matching trajectories, coating thickness uniformity index, and preset risk weight coefficient. The formula for calculating the probability of defect formation is as follows: ; in, Indicates the probability of a defect occurring. Indicates the number of matched trajectories. This indicates the total number of predicted sedimentary trajectories. Indicators representing coating thickness uniformity This represents the normalized variance of airflow velocity. All of these represent preset risk weight coefficients.

[0026] It should be noted that real-time airflow disturbance data is used to monitor the motion vector and velocity variance of aerosol particles in the current environment. By analyzing this information, the impact of airflow on the spraying operation can be understood. For example, in the spraying workshop, sensors are installed to monitor the airflow in real time. The data shows the changes in the movement direction and speed of particles in the airflow. This data helps operators determine whether the spraying environment is stable. The variance of airflow velocity is an important indicator for measuring the degree of airflow instability; a large variance indicates that the airflow fluctuates greatly, which may lead to coating defects; for example, if the recorded airflow velocity variance is 0.5 m / s^2, which exceeds the preset threshold (such as 0.3 m / s^2), it indicates that the current coating environment is unstable and there is a risk of forming defects. When the airflow velocity variance exceeds a preset threshold, motion vectors are extracted from real-time airflow disturbance data to obtain the target motion vector field. For example, in this case, the motion vectors extracted by the system show that the airflow exhibits obvious vortex motion in a certain area, indicating that the airflow in this area is complex and may affect the uniformity of the coating. Based on the target motion vector field, the deposition trajectory of aerosol particles on the target glass substrate surface is calculated, generating multiple predicted deposition trajectories. For example, the system simulates multiple particle deposition trajectories, showing small particle aggregation areas that may form on the glass surface, which may become breeding grounds for defects. The predicted deposition trajectory is matched with low-curvature regions in a virtual surface topology model to assess potential defect locations. Low-curvature regions are defined as those in the virtual topology model where the absolute value of the local average curvature is less than a preset threshold (e.g., ...). For continuous regions of ), the matching algorithm is as follows: calculate the grid patch that each predicted deposition trajectory passes through. If the patch belongs to a low curvature region, it is counted as one match. For example, assuming that low curvature regions are identified as more susceptible to defects, the system finds that three predicted deposition trajectories overlap with these regions, which means that there is a higher risk of defect formation in these places. Based on the number of matching trajectories, the coating thickness uniformity index, and the preset risk weight coefficient, the probability of defect formation is calculated. The probability of defect formation is a quantified probability value (usually between 0 and 1) used to represent the likelihood that specific types of defects such as "orange peel texture" or "sag" will occur on the surface of the target glass substrate under the current coating thickness uniformity state and real-time airflow disturbance environment. For example, if three matching trajectories are identified and the coating thickness uniformity index is relatively low, combined with the risk weight coefficient (for example, set to 0.7), the final calculated probability of defect formation is 80%. This means that under the current environmental conditions, there is a high probability that orange peel texture or sagging defects will form on the surface of the target glass.

[0027] In an optional embodiment, the coating thickness uniformity index is dynamically compensated and corrected based on a surface virtual topology model and real-time airflow disturbance data to obtain a target quality evaluation index, including: The main direction of airflow is determined based on real-time airflow disturbance data; the main direction of airflow is used to indicate the main impact direction of airflow on the sprayed coating. In the surface virtual topology model, a target cross section perpendicular to the main airflow direction is determined, and the target cross section contour is obtained. The local sagging sensitivity coefficient is calculated based on the target cross-sectional profile; the local sagging sensitivity coefficient is used to indicate the tendency of the coating to accumulate liquid under airflow impact; The compensation gain coefficient is determined based on the local sag sensitivity coefficient and the probability of defect formation risk. The coating thickness uniformity index is weighted and adjusted based on the compensation gain coefficient to obtain the target quality evaluation index.

[0028] It should be noted that by analyzing real-time airflow disturbance data, the main direction of airflow can be determined; this direction affects the formation and quality of the coating during the spraying process; for example, suppose that in a spraying workshop, the data collected by the sensor shows that the airflow is mainly directed to the upper right, which means that during the spraying process, aerosol particles will mainly move in this direction and exert an impact on the coating formation. In the surface virtual topology model, a target cross section perpendicular to the main airflow direction is determined. This cross section can help analyze the influence of airflow on the coating. For example, if the main airflow direction is the upper right, then the target cross section perpendicular to it may be a line from the lower left to the upper right. This cross section will be used to further analyze the coating's response to the airflow. Based on the target cross-sectional profile, the local sagging sensitivity coefficient is calculated. The local sagging sensitivity coefficient is a dimensionless parameter calculated based on the geometric features (such as depression depth and curvature change) of a specific cross-sectional profile in the surface virtual topology model. It is used to assess the tendency of the coating in this local area to accumulate liquid (coating) and form sagging defects under the impact of airflow. For example, if the analysis of the target cross-section shows that the local sagging sensitivity coefficient of a certain part is high, it indicates that under the action of airflow, liquid is more likely to accumulate in this area, resulting in uneven coating or defects. By combining the local sagging sensitivity coefficient and the probability of defect formation risk, a compensation gain coefficient is calculated. This coefficient is used to adjust the parameters in the spraying process to reduce the risk of defects. For example, if the local sagging sensitivity coefficient is high and the probability of defect formation risk is also high, the system may obtain a higher compensation gain coefficient, which means that more aggressive measures need to be taken to improve the coating quality. The coating thickness uniformity index is weighted and adjusted according to the compensation gain coefficient to obtain the final target quality evaluation index. For example, if the original coating thickness uniformity index is 0.8, after adjusting the compensation gain coefficient, the new evaluation index is improved to 0.9. This indicates that the coating quality has been improved and is closer to the ideal state.

[0029] In an optional embodiment, before sending an image acquisition request message to the multispectral imaging array, the method further includes: The target material refractive index is obtained by querying the preset material mapping table based on the identification information of the target glass substrate. An initial digital twin is constructed based on the refractive index of the target material and the expected coating thickness of the target glass substrate; The detection accuracy parameters of the multispectral imaging array are dynamically reconstructed based on the defect confidence level in the initial digital twin and the preset detection mode gradient table. The detection accuracy parameters include scan line density and exposure gain coefficient; the dynamically reconstructed detection accuracy parameters are used to generate multispectral surface image data in the first image acquisition response message.

[0030] It should be noted that, based on the identification information of the target glass substrate, a preset material mapping table is consulted to obtain the refractive index of the material. This refractive index is very important for subsequent optical inspection because it affects the propagation and reflection characteristics of light. For example, if the target glass substrate is identified as "AG-123", the system looks up the refractive index of the glass in the material mapping table and finds that it is 1.5. This means that when performing optical imaging, the system can perform accurate calculations and compensation based on this refractive index. Based on the obtained material refractive index and the expected coating thickness of the target glass substrate, an initial digital twin is constructed. This digital twin is a virtual model that reflects the optical properties and surface state of the actual material. For example, assuming the expected coating thickness is 250 micrometers and the refractive index is 1.5, the system generates a digital twin containing these parameters, which will be used for subsequent defect analysis and detection. In the initial digital twin, the defect confidence level is obtained by analyzing the current coating state. This confidence level value is used to refer to a preset detection mode gradient table to evaluate the required detection accuracy. For example, if the digital twin analyzes a defect confidence level of 0.7 (i.e., 70% probability of a defect), the recommended detection accuracy parameters for this confidence level in the detection mode gradient table include higher scan line density and exposure gain coefficient to ensure more accurate imaging. Based on the defect confidence level and the detection mode gradient table, detection accuracy parameters such as scan line density and exposure gain coefficient are dynamically adjusted. These adjustments improve the accuracy and reliability of multispectral imaging. The detection mode gradient table is a pre-generated data table that defines the recommended combination of scan line density and exposure gain coefficient that the multispectral imaging array should use to achieve the optimal detection signal-to-noise ratio under different coating thicknesses, glass refractive indices, and defect confidence levels. For example, through dynamic reconstruction, the originally set scan line density is increased from 100 lines to 200 lines, while the exposure gain coefficient is increased from 1.0 to 1.5 to enhance the image detail capture capability and ensure clear imaging even under high defect confidence levels. Multispectral imaging is performed using adjusted detection accuracy parameters to generate multispectral surface image data in the first image acquisition response message; this image data will be used for subsequent quality assessment and defect detection; for example, with the adjusted parameters, the system completes the acquisition of multispectral images, generating a set of high-resolution image data; these images can clearly show the uniformity of the coating and any potential defects, such as bubbles, runs, or uneven coatings.

[0031] In an optional embodiment, the detection accuracy parameters of the multispectral imaging array are dynamically reconstructed based on the defect confidence level in the initial digital twin and a preset detection mode gradient table, including: The defect confidence level of each micro-region in the initial digital twin is calculated in real time, and the defect confidence levels calculated multiple times are compared with the preset confidence level threshold. When the defect confidence level is greater than or equal to the confidence level threshold, the line density gradient table and exposure gain gradient table of the corresponding detection area are queried according to the current transmission speed of the glass substrate and the previously generated digital twin. Based on the retrieved line density gradient table and exposure gain gradient table, calculate the signal-to-noise ratio of signal acquisition under different combinations of scan line density and exposure gain; Different combinations of scan line density and exposure gain are sorted from high to low according to signal-to-noise ratio; Select the scan line density and exposure gain combination that is one position higher than the currently configured scan line density and exposure gain combination, and update the currently configured scan line density and exposure gain of the multispectral imaging array.

[0032] It should be noted that the defect confidence level of each micro-region in the initial digital twin needs to be calculated in real time. This calculation is based on the optical properties of the digital twin model and the current coating state, and aims to assess the probability of defects in different regions. For example, suppose that on a coated glass substrate, the digital twin analysis finds defect confidence levels of 0.6, 0.4 and 0.8 for several regions. This means that the first region has a 60% probability of having a defect, the second region has a lower probability, and the third region has an 80% probability of having a defect. The calculated defect confidence level is compared with a preset confidence threshold to determine which areas need further detection and processing. For example, if the preset confidence threshold is 0.7, then according to the above results, only the third area (0.8) needs to be focused on because its confidence level exceeds the threshold. When a defect confidence level in a certain area is found to be greater than or equal to the confidence threshold, the system queries the line density gradient table and exposure gain gradient table for the corresponding detection area based on the current transmission speed of the glass substrate and the previously generated digital twin, in order to formulate an appropriate detection strategy. For example, assuming the transmission speed of the glass substrate is 1 meter / minute, the system queries the recommended line density gradient table for an area with a confidence level of 0.8, which shows that 150 lines should be used at this speed, while the exposure gain gradient table suggests using a gain of 2.0. Based on the obtained line density and exposure gain combinations, the signal-to-noise ratio (SNR) of signal acquisition under different combinations is calculated; the higher the SNR, the better the image quality; for example, assuming the system tries multiple combinations of scan line density and exposure gain, such as 100 lines with 1.5 gain, 150 lines with 2.0 gain, and 200 lines with 2.5 gain; after calculation, the SNR of each combination is obtained, among which the combination of 150 lines and 2.0 gain has the highest SNR; The system sorts all combinations by signal-to-noise ratio from highest to lowest and selects the combination that is better than the current configuration to update the settings of the multispectral imaging array. For example, if the current configuration is 100 lines and 1.5 gain, after sorting, the combination of 150 lines and 2.0 gain has a higher signal-to-noise ratio and is therefore selected as the new configuration. The system will automatically update the parameters of the imaging device to improve the quality of subsequent imaging.

[0033] In an optional embodiment, after sending the quality inspection report to the quality inspection terminal according to the target quality assessment indicators, the method further includes: If the target quality assessment index is lower than the preset qualified threshold, then the high confidence defect sites in the surface virtual topology model are located at the sub-pixel level according to the updated scan line density and exposure gain. The located subpixel coordinates are bound to the real-time position code of the target glass substrate to generate a defect physical marking instruction; By using a high-precision laser galvanometer to perform micro-etching marks at corresponding positions on the target glass substrate according to the physical marking instructions for defects, the physical source of defects can be traced.

[0034] It should be noted that the coating quality needs to be evaluated. If the evaluation index is lower than the set pass threshold, the system will initiate subsequent processing procedures. This evaluation may be based on multiple factors, such as the uniformity, thickness, and adhesion of the coating. For example, suppose the preset pass threshold is 0.75, but after testing, the coating quality evaluation index is only 0.65. This means that there is a problem with the coating and further processing is required. After confirming the quality failure, the system uses updated scan line density and exposure gain to perform more precise sub-pixel-level localization of the high-confidence defect sites identified in the virtual topology model. This step aims to improve the accuracy of defect localization to ensure the effectiveness of subsequent processing. For example, in the virtual topology model, a region is marked as a high-confidence defect with original coordinates (100, 200). After sub-pixel-level analysis, the system may determine the actual defect location to be (100.3, 200.7), which is a more precise localization. The located subpixel coordinates are bound to the real-time position code of the target glass substrate. The real-time position code is usually provided by a sensor or encoder to ensure that the exact location of the defect in the physical world matches the location in the virtual model. For example, assuming that the glass substrate moves during transportation due to vibration or other factors, the real-time position code shows that the current position of the substrate is (100.5, 201.2). Through calculation and binding, the system combines the subpixel coordinates (100.3, 200.7) with the real-time position (100.5, 201.2) to determine the final processing position. Based on the bound coordinate information, a defect physical marking instruction is generated. This instruction will guide the subsequent laser processing equipment to mark the defect at the correct location. For example, based on the previous calculations, the defect physical marking instruction generated by the system may include "perform micro-abrasion marking at location (100.8, 201.9)" to ensure that the laser equipment can accurately reach the location of the defect. Using a high-precision laser galvanometer, micro-etching marks are applied to the corresponding positions on the target glass substrate according to the generated defect physical marking instructions. These marks can leave physical traces on the glass surface, providing a basis for subsequent quality traceability. For example, after receiving the instruction, the laser device accurately etches at the position (100.8, 201.9) to form a tiny mark. This mark can not only serve as a visual identification of defects, but also provide important information in subsequent quality control and product traceability.

[0035] In an optional embodiment, before sending an environmental parameter acquisition request message to the airflow sensor, the method further includes: When the glass production line is started, the material code and process parameters of the target glass substrate are received. The process parameters include the expected coating thickness, transfer speed, and second identification information, which includes the batch number and production line number. Based on the optical refractive index corresponding to each material code, a detection mode gradient table for different coating thicknesses and a mapping table between the corresponding material codes and the detection mode gradient table are constructed. The mapping table is used to indicate the mapping relationship between batch number, production line number, and line density gradient table and exposure gain gradient table.

[0036] It should be noted that when the production line starts up, the system first receives the material code of the target glass substrate and related process parameters; this information is crucial for the subsequent spraying process; for example, assuming the material code of the target glass substrate is "GB-123", the process parameters include an expected coating thickness of 200 micrometers, a transmission speed of 1 meter / minute, and secondary identification information such as the batch number "BATCH001" and the production line number "LINE01"; Based on the optical refractive index corresponding to different material codes, the system will construct a detection mode gradient table for each material for different coating thicknesses. These gradient tables contain the required detection standards and parameters for different coating thicknesses, such as line density and exposure gain. For example, for material code "GB-123", its optical refractive index is 1.5. The system may construct a detection mode gradient table showing that the recommended line density is 150 lines and the exposure gain is 2.0 for a thickness of 200 micrometers. These parameters will be used for quality inspection after coating. The mapping table indicates the relationship between different batch numbers and production line numbers and their corresponding line density gradient tables and exposure gain gradient tables. This mapping provides a quick lookup function for different configurations during the production process, ensuring that different batches and production lines can achieve consistent quality inspection standards. For example, the mapping table may record that "BATCH001" and "LINE01" correspond to a line density gradient table with 150 lines and an exposure gain of 2.0, suitable for 200-micron coatings with material code "GB-123". This allows the production line to quickly look up relevant inspection parameters when it receives new batch or production line information. During the actual spraying process, the production line uses information from the mapping table to guide the quality inspection after spraying, ensuring that the coating meets the expected quality standards. For example, after spraying is completed, the inspection system will look up the line density and exposure gain corresponding to "BATCH001" and "LINE01" according to the mapping table, perform quality inspection, and ensure that the thickness and uniformity of the sprayed coating meet the requirements.

[0037] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for inspecting the quality of glass surface coating, characterized in that, include: Start the inspection mode on the display interface of the quality inspection terminal. The inspection mode displays the real-time monitoring screen. A selection operation was detected on the target glass substrate in the real-time monitoring screen; In response to the selection operation, an image acquisition request message is sent to the multispectral imaging array; In addition, receiving a first image acquisition response message from a multispectral imaging array; wherein the first image acquisition response message includes multispectral surface image data; A virtual topological model of the target glass substrate is generated based on multispectral surface image data; the coating thickness uniformity index is determined based on the virtual topological model. Sending an environmental parameter acquisition request message to an airflow sensor; and receiving a first environmental parameter response message from the airflow sensor; wherein the first environmental parameter response message includes real-time airflow disturbance data; The probability of defect formation risk is calculated based on coating thickness uniformity index and real-time airflow disturbance data; If the probability of defect formation exceeds the preset risk threshold, the coating thickness uniformity index is dynamically compensated and corrected based on the surface virtual topology model and real-time airflow disturbance data to obtain the target quality assessment index; a quality inspection report is then sent to the quality inspection terminal based on the target quality assessment index.

2. The method for detecting the quality of glass surface coating according to claim 1, characterized in that, The image acquisition request message includes the identification information of the target glass substrate. The image acquisition request message is used to instruct the multispectral imaging array to acquire surface images of the target glass substrate at multiple preset wavelengths.

3. The method for detecting the quality of glass surface coating according to claim 2, characterized in that, The target glass substrate is any glass substrate to be inspected as determined by the user; the surface virtual topology model is used to characterize the micro-geometric features of the coating on the target glass substrate. Multispectral surface image data includes images in the visible light band and images in the infrared band; A virtual topological model of the target glass substrate is generated based on multispectral surface image data, including: Infrared band images are extracted from multispectral surface image data; among them, infrared feature images are used to characterize the differences in thermal conductivity inside the sprayed coating. Potential void regions inside the coating are identified based on infrared feature images, and void distribution data is obtained. The visible light band image is extracted from the multispectral surface image data to obtain the surface texture image; the surface texture image is used to characterize the reflective properties of the sprayed coating surface. An initial geometric mesh model is constructed based on the surface texture image; the initial geometric mesh model consists of multiple triangular facets. The void distribution data is mapped to the corresponding grid nodes in the initial geometric grid model to obtain an intermediate topology model with void properties; The intermediate topology model is smoothed by surface fitting to obtain the virtual topology model of the target glass substrate.

4. The method for detecting the quality of glass surface coating according to claim 3, characterized in that, Coating thickness uniformity index is used to indicate the degree of thickness dispersion of the sprayed coating at different locations; The coating thickness uniformity index is determined based on the surface virtual topology model, including: Multiple sampling paths are determined in the surface virtual topology model; wherein, the multiple sampling paths cover the central region and edge region of the target glass substrate; Calculate the rate of curvature change on each sampling path to obtain multiple curvature change rate sequences; The thickness estimate for each sampling path is calculated based on multiple curvature change rate sequences, resulting in multiple thickness estimates; Calculate the standard deviation of multiple thickness estimates to obtain the thickness discrete parameters; The coating thickness uniformity index is calculated based on the thickness dispersion parameter and the mean of multiple thickness estimates.

5. The method for detecting the quality of glass surface coating according to claim 4, characterized in that, Real-time airflow disturbance data is used to indicate the motion vector and velocity variance of aerosol particles in the current environment; the defect formation risk probability is used to indicate the likelihood of defects forming on the surface of the target glass substrate under the current airflow environment; The probability of defect formation risk is calculated based on coating thickness uniformity index and real-time airflow disturbance data, including: The airflow velocity variance is determined based on real-time airflow disturbance data; whereby the airflow velocity variance is used to indicate the degree of airflow instability. If the variance of the airflow velocity is greater than the preset variance threshold, the motion vector in the real-time airflow disturbance data is extracted to obtain the target motion vector field; The deposition trajectory of aerosol particles on the surface of the target glass substrate is determined based on the target motion vector field, and multiple predicted deposition trajectories are obtained. The number of matched trajectories is obtained by matching multiple predicted deposition trajectories with low curvature regions in the surface virtual topology model. The probability of defect formation is calculated based on the number of matching trajectories, coating thickness uniformity index, and preset risk weight coefficient.

6. The method for detecting the quality of glass surface coating according to claim 5, characterized in that, Dynamic compensation and correction of coating thickness uniformity indicators are performed based on the surface virtual topology model and real-time airflow disturbance data to obtain target quality evaluation indicators, including: The main direction of airflow is determined based on real-time airflow disturbance data; the main direction of airflow is used to indicate the main impact direction of airflow on the sprayed coating. In the surface virtual topology model, the target cross section perpendicular to the main airflow direction is determined, and the target cross section contour is obtained; The local sagging sensitivity coefficient is calculated based on the target cross-sectional profile; the local sagging sensitivity coefficient is used to indicate the tendency of the coating to accumulate liquid under airflow impact; The compensation gain coefficient is determined based on the local sag sensitivity coefficient and the probability of defect formation risk. The coating thickness uniformity index is weighted and adjusted based on the compensation gain coefficient to obtain the target quality evaluation index.

7. The method for detecting the quality of glass surface coating according to claim 6, characterized in that, Before sending the image acquisition request message to the multispectral imaging array, the following is also included: The target material refractive index is obtained by querying the preset material mapping table based on the identification information of the target glass substrate. An initial digital twin is constructed based on the refractive index of the target material and the expected coating thickness of the target glass substrate; The detection accuracy parameters of the multispectral imaging array are dynamically reconstructed based on the defect confidence level in the initial digital twin and the preset detection mode gradient table. The detection accuracy parameters include scan line density and exposure gain coefficient; the dynamically reconstructed detection accuracy parameters are used to generate multispectral surface image data in the first image acquisition response message.

8. The method for detecting the quality of glass surface coating according to claim 7, characterized in that, Based on the defect confidence level in the initial digital twin and the preset detection mode gradient table, the detection accuracy parameters of the multispectral imaging array are dynamically reconstructed, including: The defect confidence level of each micro-region in the initial digital twin is calculated in real time, and the defect confidence levels calculated multiple times are compared with the preset confidence level threshold. When the defect confidence level is greater than or equal to the confidence level threshold, the line density gradient table and exposure gain gradient table of the corresponding detection area are queried according to the current transmission speed of the glass substrate and the previously generated digital twin. Based on the retrieved line density gradient table and exposure gain gradient table, calculate the signal-to-noise ratio of signal acquisition under different combinations of scan line density and exposure gain; Different combinations of scan line density and exposure gain are sorted from high to low according to signal-to-noise ratio; Select the scan line density and exposure gain combination that is one position higher than the currently configured scan line density and exposure gain combination, and update the currently configured scan line density and exposure gain of the multispectral imaging array.

9. A method for detecting the quality of glass surface coating according to claim 8, characterized in that, After sending the quality inspection report to the quality inspection terminal based on the target quality assessment indicators, it also includes: If the target quality assessment index is lower than the preset qualified threshold, then the high confidence defect sites in the surface virtual topology model are located at the sub-pixel level according to the updated scan line density and exposure gain. The located subpixel coordinates are bound to the real-time position code of the target glass substrate to generate a defect physical marking instruction; By using a high-precision laser galvanometer to perform micro-etching marks at corresponding positions on the target glass substrate according to the physical marking instructions for defects, the physical source of defects can be traced.

10. A method for detecting the quality of glass surface coating according to claim 9, characterized in that, Before sending an environmental parameter acquisition request message to the airflow sensor, the process also includes: When the glass production line is started, the material code and process parameters of the target glass substrate are received. The process parameters include the expected coating thickness, transfer speed, and second identification information, which includes the batch number and production line number. Based on the optical refractive index corresponding to each material code, a detection mode gradient table for different coating thicknesses and a mapping table between the corresponding material codes and the detection mode gradient table are constructed. The mapping table is used to indicate the mapping relationship between batch number, production line number, and line density gradient table and exposure gain gradient table.

Images