Glass substrate surface defect detection method and system
By spraying volatile liquid combined with hot baking and laminar drying gas, and using infrared thermal imagers and AI models to identify surface defects of glass substrates, the problems of low efficiency and insufficient precision in existing technologies are solved, and efficient and accurate defect detection is achieved.
Patent Information
- Application Number
- CN202510917217.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-03
- Publication Date
- 2025-09-19
AI Technical Summary
Existing technologies have problems with low efficiency and insufficient precision in detecting surface defects on glass substrates. They are particularly difficult to identify tiny defects and shallow surface defects, and are easily affected by reflections from the glass surface, resulting in a high missed detection rate.
The system uses spraying of volatile liquid combined with thermal drying and laminar drying gas, monitors temperature changes with an infrared thermal imager to capture residual signals, combines differential enhancement processing with AI models for defect identification, and uses multi-dimensional feature data for classification and labeling.
It improves detection efficiency and accuracy, significantly reduces missed detection rate, and achieves efficient identification of tiny and shallow surface defects while taking into account the protection of the glass.
Smart Images
Figure FT_1 
Figure SMS_1 
Figure SMS_2
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of glass production quality detection, and in particular relates to a glass substrate surface defect detection method and system. Background Art
[0002] Glass substrates are a key material in precision manufacturing applications such as display panels and photovoltaic glass. Their surface quality directly impacts product yield and performance. Traditional defect detection methods, such as manual visual inspection or optical scanning, suffer from a trade-off between efficiency and accuracy. Manual inspection relies on subjective experience and is slow, making it difficult to meet the demands of large-scale production. Conventional optical inspection, while capable of automation, has a low recognition rate for minute defects (such as nanocracks and chemical residues) and is susceptible to interference from reflections on the glass surface, resulting in a high rate of missed detections. Existing technologies, in particular, need to further improve their detection efficiency for volatile contaminants or shallow surface defects. Summary of the Invention
[0003] In view of the problem that the efficiency of glass substrate surface quality inspection in the prior art needs to be improved, a glass substrate surface defect inspection method and system are proposed. The present invention provides the following technical solutions: A method for detecting surface defects of a glass substrate comprises the following steps: S1, spraying volatile liquid evenly on the glass surface; S2, uniformly heat-drying the glass surface and applying a uniform laminar flow of drying gas parallel to the glass surface; S3, determining the volatilization critical point by real-time monitoring of surface reflectivity or temperature changes, and triggering an imaging device to capture the residual signal image at the critical point; S4, performing differential enhancement processing on the residual image to generate a defect location distribution map; S5, scan only the marked area to obtain the three-dimensional morphology or spectral characteristics of the defect; S6, integrates multi-dimensional feature data, classifies defect types through AI models, and determines quality levels; S7, performs automated marking and sorting based on the grade results.
[0004] Preferably, the volatile liquid has a viscosity of ≤2 cP, a surface tension of ≤25 mN / m, and a boiling point of ≤85°C at 25°C.
[0005] Preferably, in step S1, the volatile liquid is a binary mixed solvent containing 0.001-0.1 wt% of a fluorescent dye, wherein the boiling point of the fast-volatile component is ≤60°C, and the boiling point of the slow-volatile component is ≥100°C.
[0006] Preferably, in step S2, during the heat drying, the surface temperature of the glass is controlled within the range of 40-70° C., and the flow rate of the uniform laminar drying gas is 0.5-3 m / s.
[0007] Preferably, the imaging device in step S3 is an infrared thermal imager, which realizes non-contact detection by capturing the temperature difference of the residual area.
[0008] Preferably, in step S5, the DLM is used to guide the high-precision detection equipment to scan the marked area.
[0009] A glass substrate surface defect detection system is used to implement a glass substrate surface defect detection method, comprising: The initial screening and positioning module includes a precision atomizing nozzle for evenly spraying volatile liquids, a PID temperature-controlled heating plate for accelerating volatilization, a laminar air knife system for promoting volatilization, and an infrared thermal imager or high-speed camera for detecting volatile residues. Initial screening and positioning are performed using glass substrate surface defect detection methods. Precision inspection and review module, used to accurately detect defects; Execute the output module to mark the defects; The intelligent decision-making module is used to output the corresponding control information to the fine inspection and review module through the preliminary screening and positioning module, and to mark the defects and sort the corresponding glass substrates through the execution output module.
[0010] Preferably, the laminar air knife system includes a honeycomb guide plate and a wind speed sensor to ensure that the airflow uniformity deviation is ≤5%.
[0011] Preferably, the precision inspection and review module includes a six-axis robotic arm and a confocal probe or a multispectral sensor mounted thereon.
[0012] Preferably, the execution output module includes an ultraviolet laser marking machine and a pneumatic sorting robot.
[0013] Compared with the prior art, the present invention has the following beneficial effects: By spraying a low-viscosity, low-surface-tension volatile liquid, combined with capillary force, the system can efficiently penetrate cracks and pit defects, turning shallow surface defects that are difficult to capture with traditional optical inspection into detectable signals, thereby improving inspection efficiency. This strategy enables the use of a laser confocal microscope in high-confidence areas and switching to a hyperspectral probe in low-confidence areas. Combined with Hamiltonian path planning, this reduces the idle travel of the robotic arm, shortening the overall inspection cycle while ensuring the in-depth analysis requirements of key areas, and taking into account both the protection of the glass and the detection rate. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 It is a schematic diagram of the method structure of the present invention; DETAILED DESCRIPTION
[0015] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is clearly and completely described below in conjunction with the accompanying drawings of the present invention. The directional words mentioned in the following embodiments, such as "up", "down", "left" and "right", etc., are only referenced to the directions of the accompanying drawings. Therefore, the directional words used are used to illustrate rather than limit the invention.
[0016] The following is a supplementary explanation of the English abbreviations that appear in the plan: PID: Proportional-Integral-Derivative controller, used to precisely regulate the temperature of the heating plate and maintain a stable glass surface temperature through real-time feedback.
[0017] GNN: Graph Neural Network, a deep learning architecture that represents complex relationships between data through graph structures. It is used here to integrate multi-dimensional features of defects (such as thermal imaging, 3D morphology, and spectral information) and improve classification accuracy.
[0018] SNR: Signal-to-Noise Ratio, a key indicator for measuring signal quality. Here, it refers to an improvement of ≥8dB in the ratio of the defect signal to the background noise after differential enhancement processing.
[0019] NETD: Noise Equivalent Temperature Difference, a core performance parameter of infrared thermal imagers, indicates the minimum temperature difference that the device can resolve. Here, ≤50mK (milliKelvin) indicates high sensitivity.
[0020] Re: Reynolds Number, a dimensionless number used to determine the flow state in fluid mechanics. Here, the laminar gas Reynolds number Re<2000 indicates that the airflow is stable laminar flow.
[0021] EPDM: Ethylene Propylene Diene Monomer (EPDM), an aging-resistant and corrosion-resistant synthetic rubber, is used here as the suction cup material for pneumatic sorting robots.
[0022] DLM: Defect Location Map, an image generated through differential enhancement processing, contains defect coordinates, confidence scores, and preliminary morphological classification, providing precise guidance for subsequent precision inspection.
[0023] CLAHE: Contrast Limited Adaptive Histogram Equalization, an image enhancement algorithm that improves the visibility of defect features by local contrast stretching.
[0024] like Figure 1 As shown, a method for detecting surface defects of a glass substrate comprises the following steps: S1. Evenly spray a volatile liquid on the glass surface. The liquid should have a viscosity of ≤2cP, a surface tension of ≤25mN / m, and a boiling point of ≤85°C at 25°C, ensuring that it can quickly penetrate tiny defects on the glass surface and form a detectable signal during the volatilization process.
[0025] Specifically, an ultrasonic atomizing nozzle is used to control the droplet diameter to 10-50μm, forming a uniform liquid film with a thickness of ≤5μm; the viscosity parameter ensures that the capillary force efficiently penetrates into cracks ≥0.5μm, and the surface tension ensures complete lubrication of the burning pit defects.
[0026] S2, uniformly heats the glass surface and applies a uniform laminar flow of drying gas parallel to the glass surface; controlling the glass surface temperature within the range of 40-70°C, combined with a uniform laminar flow of drying gas at a flow rate of 0.5-3m / s, can significantly amplify the difference in volatilization rate between defective areas and normal areas.
[0027] Specifically, the temperature field was optimized through finite element simulation, and the volatilization rate in the defective area was reduced to 1 / 4 of that in the normal area at 55°C; the laminar gas Reynolds number Re was <2000 (flow rate 1.2m / s + honeycomb guide plate aperture 1mm), and the airflow uniformity deviation was ≤3%.
[0028] S3 determines the volatilization critical point by real-time monitoring of surface reflectivity or temperature changes, and triggers the imaging device to capture the residual signal image at the critical point; an infrared thermal imager is used as the imaging device to achieve non-contact detection by capturing the temperature difference in the residual area, avoiding secondary damage that may be caused by mechanical contact.
[0029] Specifically, the thermal sensitivity (NETD) of the infrared thermal imager is ≤50mK, and shooting is triggered when the temperature difference ΔT is ≥0.5°C for three consecutive frames. The detection sensitivity of silicone oil contaminants is 5 times higher than that of visible light (due to the difference in thermal conductivity of 0.15W / m·K vs 1.4W / m·K).
[0030] S4 performs differential enhancement on the residual image to generate a defect location map (DLM). This process effectively suppresses background noise, highlights defect features, and provides precise guidance for subsequent precision inspection.
[0031] Specifically, the background subtraction + CLAHE contrast enhancement algorithm is used to improve the defect signal-to-noise ratio by ≥8dB; the DLM includes defect coordinates, confidence scores and preliminary morphological classification (point / line / surface).
[0032] S5 scans only the marked area to obtain the three-dimensional morphology or spectral characteristics of the defect; the DLM guides high-precision inspection equipment to scan the marked area. The selective scanning strategy greatly reduces the invalid inspection area and improves the overall inspection efficiency.
[0033] Specifically, high-precision equipment is allocated according to confidence levels: a laser confocal microscope (axial resolution 10nm) is used in the high-confidence area (>90%), and a hyperspectral probe (256 bands @ 5nm resolution) is used in the low-confidence area (<50%). The six-axis robotic arm generates the optimal trajectory of the Hamiltonian path based on the DLM coordinates, reducing the idle travel by 40%.
[0034] S6 integrates multi-dimensional feature data, classifies defect types and determines quality levels through AI models; the AI model adopts a deep learning architecture, which can automatically learn the complex mapping relationship between defect characteristics and quality levels to achieve intelligent decision-making.
[0035] Specifically, a dual-stream graph neural network (GNN) is used: the morphology stream processes point cloud data (50,000 points / defect), and the spectral stream analyzes the 400-2500nm reflectance curve; the attention mechanism weightedly fuses features and outputs the probability distribution of 12 types of defects, such as scratches, bubbles, and oil stains.
[0036] S7 performs automated marking and sorting based on grading results; the entire process implements closed-loop control, ensuring that defective products are handled promptly and improving the automation level of the production line.
[0037] Specifically, an ultraviolet laser marker (355nm) engraves a micro-pit QR code (depth ≤ 2μm) on the edge of the substrate, which contains the coordinates of the defect location; the vacuum suction cup of the pneumatic sorting robot is covered with a flexible layer of carbon nanotubes with an adsorption stress ≤ 0.1MPa.
[0038] Furthermore, the volatile liquid is a binary mixed solvent containing 0.001-0.1 wt% of a fluorescent dye, the boiling point of the fast-volatile component is ≤60°C, and the boiling point of the slow-volatile component is ≥100°C; Specifically, the fast-volatile component (acetone) produces the initial temperature difference signal, the slow-volatile component (butyl diglycol) carries the Rhodamine B dye retention defect, and the excitation light 488nm / emission light 520nm provides optical secondary verification.
[0039] In step S2, the glass surface temperature is controlled within the range of 40-70°C, and the flow rate of the uniform laminar drying gas is 0.5-3 m / s; Specifically, the temperature-flow rate combination was optimized using the response surface methodology: the critical point time window reached 200±10ms at 55℃+1.2m / s, and the wind speed sensor used real-time feedback to adjust the PID parameters to maintain a fluctuation of ≤3%.
[0040] The imaging device in step S3 is an infrared thermal imager, which realizes non-contact detection by capturing the temperature difference of the residual area; Specifically, the medium-wave infrared lens (3-5μm band) penetrates the ambient water vapor interference and captures transient temperature changes at a 100Hz frame rate.
[0041] In step S5, the high-precision detection equipment is guided by the DLM to scan the marked area; Specifically, the multispectral probe and the confocal microscope are switched through a magnetic interface (time < 2s), and the piezoelectric ceramic active damper built into the robotic arm joint suppresses the amplitude to ±0.5μm.
[0042] A glass substrate surface defect detection system is used to implement a glass substrate surface defect detection method, comprising: The initial screening and positioning module includes a precision atomizing nozzle for evenly spraying volatile liquids, a PID-controlled heating plate for accelerated volatilization, a laminar air knife system for enhanced volatilization, and an infrared thermal imager or high-speed camera for detecting volatile residues. Initial screening and positioning are performed using glass substrate surface defect detection methods. The module integrates a precision control unit to ensure precise adjustment of spraying, heating, and airflow parameters. Specifically, the laminar air knife system includes a honeycomb deflector (1mm aperture / 10:1 thickness ratio) and a hot-wire anemometer. A PID controller adjusts the fan speed in real time to maintain a constant flow rate.
[0043] The precision inspection and review module is used for accurate defect detection. It includes a six-axis robotic arm equipped with a confocal probe or multispectral sensor. The six-axis robotic arm features high degrees of freedom of movement, enabling comprehensive inspection of complex glass substrates. Specifically, the six-axis robotic arm (with a repeatability of ±1μm) is equipped with a weight reduction mechanism and has a maximum load capacity of 15kg. The multispectral sensor covers the visible and near-infrared range (400-2500nm) with a spectral resolution of 5nm.
[0044] The execution output module is used to mark defects; it includes a UV laser marker and a pneumatic sorting robot. The UV laser marker can clearly mark the glass surface, while the pneumatic sorting robot enables rapid sorting. Specifically, a UV laser (3W / 355nm) etches the QR code through a galvanometer system. The vacuum sorting robot is equipped with a pressure sensor with a contact force of ≤0.5N to prevent ultra-thin glass from breaking.
[0045] The intelligent decision-making module, which outputs control information from the initial screening and positioning module to the refined inspection and review module, then executes the output module to mark defects and sort the corresponding glass substrates. This module uses a graph neural network (GNN) to process the spatial distribution characteristics of defects, improving defect classification accuracy. Specifically, the GNN input layer integrates: 1) thermal image residual intensity, 2) confocal 3D point cloud, and 3) spectral absorption peak position. Edge weights in the graph convolution layer are dynamically calculated based on defect spacing, achieving an output classification accuracy exceeding 99%.
[0046] Furthermore, the laminar air knife system includes honeycomb deflectors and wind speed sensors to ensure airflow uniformity deviation of ≤5%. This design effectively eliminates airflow turbulence and improves the stability of the volatilization process. Specifically, the deflectors have an aspect ratio of 5:1, and the honeycomb cell inner wall roughness Ra is ≤0.8μm. The wind speed sensor has a sampling rate of 1kHz, and data fusion Kalman filtering eliminates noise.
[0047] Furthermore, the precision inspection and verification module includes a six-axis robotic arm equipped with a confocal probe or multispectral sensor. The multispectral sensor can simultaneously acquire spectral information in multiple bands, providing rich data for defect composition analysis. Specifically, it uses push-broom hyperspectral imaging with a spatial resolution of 20μm / pixel and an SNR greater than 200:1. The built-in halogen tungsten lamp has a color temperature of 2856K ± 5%.
[0048] The output module further includes a UV laser marker and a pneumatic sorting robot. The pneumatic sorting robot uses vacuum suction to grasp glass substrates, avoiding damage caused by mechanical clamping. Specifically, a multi-stage vacuum generator (ultimate vacuum -90kPa) combined with EPDM rubber suction cups allows for non-destructive grasping of curved glass (radius of curvature ≥300mm).
[0049] After the above specific settings, perform specific tests: The test data of the conventional detection scheme is: It can be seen that the present invention can greatly improve the detection efficiency while ensuring that the detection rate of 1 μm microcracks is almost unchanged, and also has high stability and reliability.
Claims
1. A method for detecting surface defects of a glass substrate, characterized in that: The following steps are involved: S1, spraying volatile liquid evenly on the glass surface; S2, uniformly heat-drying the glass surface and applying a uniform laminar flow of drying gas parallel to the glass surface; S3, determining the volatilization critical point by real-time monitoring of surface reflectivity or temperature changes, and triggering an imaging device to capture the residual signal image at the critical point; S4, performing differential enhancement processing on the residual image to generate a defect location distribution map; S5, scan only the marked area to obtain the three-dimensional morphology or spectral characteristics of the defect; S6, integrates multi-dimensional feature data, classifies defect types through AI models, and determines quality levels; S7, performs automated marking and sorting based on the grade results.
2. The method for detecting surface defects of a glass substrate according to claim 1, wherein: The volatile liquid has a viscosity of ≤2 cP at 25°C, a surface tension of ≤25 mN / m, and a boiling point of ≤85°C.
3. The method for detecting surface defects of a glass substrate according to claim 1 or 2, wherein: In step S1, the volatile liquid is a binary mixed solvent containing 0.001-0.1 wt% of a fluorescent dye, wherein the boiling point of the fast-volatile component is ≤60°C, and the boiling point of the slow-volatile component is ≥100°C.
4. The method for detecting surface defects of a glass substrate according to claim 1, wherein: In step S2, during the heat drying, the surface temperature of the glass is controlled within the range of 40-70°C, and the flow rate of the uniform laminar drying gas is 0.5-3 m / s.
5. The method for detecting surface defects of a glass substrate according to claim 1, wherein: The imaging device in step S3 is an infrared thermal imager, which realizes non-contact detection by capturing the temperature difference of the residual area.
6. The method for detecting surface defects of a glass substrate according to claim 1, wherein: In step S5, the DLM guides the high-precision detection device to scan the marked area.
7. A glass substrate surface defect detection system, used to implement the glass substrate surface defect detection method according to any one of claims 1 to 6, characterized in that: include: A primary screening and positioning module, comprising a precision atomizing nozzle for uniformly spraying a volatile liquid, a PID temperature-controlled heating plate for accelerating volatilization, a laminar air knife system for promoting volatilization, and an infrared thermal imager or a high-speed camera for detecting volatile residues, and performing primary screening and positioning using the glass substrate surface defect detection method according to any one of claims 1 to 6; Precision inspection and review module, used to accurately detect defects; Execute the output module to mark the defects; The intelligent decision-making module is used to output the corresponding control information to the fine inspection and review module through the preliminary screening and positioning module, and to mark the defects and sort the corresponding glass substrates through the execution output module.
8. The glass substrate surface defect detection system according to claim 7, wherein: The laminar air knife system includes a honeycomb guide plate and an air speed sensor to ensure that the air flow uniformity deviation is ≤5%.
9. The glass substrate surface defect detection system according to claim 7, wherein: The precision inspection and review module includes a six-axis robotic arm and a confocal probe or a multispectral sensor mounted thereon.
10. The glass substrate surface defect detection system according to claim 7, wherein: The execution output module includes a UV laser marking machine and a pneumatic sorting robot.