AI-based intelligent detection method and system for bubble defects in curved glass coating process

Abstract

An AI-based intelligent detection method and system for bubble defects in a curved glass film coating process, the curvature information of the curved glass surface is obtained and the region is divided; during the film coating process, images are collected at different process stages, suspected bubble areas are obtained after light suppression and abnormality extraction; the same area is rechecked for the suspected areas, the change characteristics between different stages are extracted, and the equipment operation parameters are synchronously obtained; the suspected areas are classified and determined according to the change characteristics, and the transient eliminable bubbles, stable residual bubbles, edge rebound type bubbles and pseudo defects are distinguished; according to the classification result, the corresponding processing action is controlled to be executed by the equipment, and rechecking is performed after processing; finally, the final state of all suspected areas on the workpiece is executed to release or determine unqualified.

Description

Technical Field

[0001] This invention relates to the field of curved glass coating inspection technology, and in particular to an AI-based intelligent detection method and system for bubble defects in the curved glass coating process. Background Technology

[0002] With the increasing demands for consistent appearance, structural strength, and surface protection in consumer electronics, automotive displays, and smart home panels, curved glass is being used more and more widely in cover plates, display windows, and decorative components. To improve the impact resistance, abrasion resistance, explosion-proof performance, optical properties, or decorative properties of curved glass surfaces, protective films, explosion-proof films, functional films, or decorative films are typically applied. Because curved glass surfaces exhibit continuous curvature changes and significant bending at the edges, the film application process is susceptible to uneven tension distribution, insufficient pressing, limited venting paths, and edge rebound, leading to under-film bubble defects.

[0003] Existing curved glass coating equipment generally includes a feeding mechanism, positioning fixture, vacuum adsorption platform, film unwinding mechanism, pre-lamination mechanism, main pressing mechanism, exhaust roller mechanism, and edge finishing mechanism. During operation, the curved glass is first fixed by the positioning fixture and vacuum adsorption platform. After unwinding, the film is pre-laminationd onto the curved glass. Then, the main pressing mechanism gradually completes the large-area lamination. Subsequently, the exhaust roller mechanism removes residual air from under the film layer, and the edge finishing mechanism completes edge lamination and shaping.

[0004] During the aforementioned coating process, various types of adhesion anomalies are prone to occur due to the large curvature of the curved glass surface, stress concentration at the edges, and the complex air venting paths between the film and the glass. Some of these anomalies are transient bubbles that disappear naturally during the coating process; others are stable residual bubbles that persist even after the main pressing and venting stages; still others are significantly reduced during the main pressing and venting stages, but reappear during or after the edge finishing stage due to edge rebound, forming edge-rebound bubbles. Furthermore, curved glass surfaces are prone to visual anomalies such as specular reflection, bright bands, adhesive lines, slight wrinkles, and edge shadows under different lighting and observation conditions. These anomalies are somewhat similar to real bubbles in images, easily leading to misjudgment.

[0005] In existing technologies, the detection of bubble defects after lamination of curved glass typically employs manual visual inspection or final inspection using machine vision. Manual visual inspection relies on operator experience and suffers from low efficiency, high subjectivity, poor consistency, and difficulty in adapting to high-speed automated production lines. Final inspection using machine vision usually involves photographing the entire curved glass sheet after lamination and then judging the presence of bubbles based on changes in brightness or abnormal contours in the image. While these methods improve detection efficiency to some extent, they still have the following shortcomings.

[0006] First, curved glass surfaces exhibit significant specular reflection characteristics, and the curvature varies in different areas, causing the same defect to appear differently under different viewing angles and lighting conditions. Single-shot imaging methods are prone to misidentifying false defects such as high-brightness reflections, light band distortion, adhesive lines, and slight wrinkles as real bubbles, resulting in false alarms; they are also prone to missing real bubbles due to localized reflection obstruction, resulting in false alarms.

[0007] Second, existing final inspection methods usually make judgments based only on static image information at a certain moment, which cannot reflect the evolution of abnormal areas in different process stages such as pre-bonding, main pressing, venting and shaping and edge finishing. Therefore, it is difficult to distinguish between transient eliminable bubbles and stable residual bubbles, and it is also difficult to identify edge rebound bubbles that reappear only in the edge finishing stage.

[0008] Third, existing detection methods typically leave the results at a static "defective / no defect" level, failing to establish an effective correlation with the process parameters of curved glass coating equipment. In other words, when an abnormality such as an air bubble is detected, existing solutions usually cannot accurately determine whether the abnormality is more likely caused by insufficient main pressing pressure, low venting roller pressure, abnormal film tension, or improper edge finishing parameters. Consequently, it is difficult to trigger targeted re-pressing, supplementary venting, or secondary edge finishing treatment based on the defect type.

[0009] Fourth, existing solutions typically lack post-processing re-inspection and workpiece-level final judgment mechanisms. Even if individual equipment has partial rework capabilities, it often simply performs a single repressurization or repressurization, lacking a process to confirm the processing effect, and even more so lacking a complete closed-loop control logic to comprehensively judge the status of all abnormal areas on the entire workpiece before deciding whether to release or reject it.

[0010] Therefore, there is an urgent need to provide an intelligent detection method for bubble defects in the coating process of curved glass, to achieve the following objectives: First, it can differentiate the detection area by combining the curvature information of the curved glass surface, thereby improving the detection accuracy of high curvature edge areas; second, it can re-inspect the same area of ​​the abnormal area in at least two different process stages, thereby distinguishing different types of bubble defects according to the evolution process; third, it can combine image features with equipment operating parameters to classify and judge the abnormal area in a way that is more in line with the actual process mechanism; fourth, it can trigger corresponding processing actions according to different defect categories, and perform re-inspection and confirmation after processing, and finally determine whether to release or reject the product based on the status of all abnormal areas on the workpiece, thereby forming a complete, clear and implementable closed loop of detection and processing. Summary of the Invention

[0011] The purpose of this invention is to overcome the aforementioned shortcomings in the prior art and provide an AI-based intelligent detection method, system, electronic device, and computer-readable storage medium for bubble defects in the curved glass coating process. The method, during the curved glass coating process, combines the curvature information of the curved glass surface, staged image acquisition results, re-inspection results of the same area, and equipment operating parameters to classify and determine suspected bubble areas. Based on the determination results, corresponding processing actions are executed, and the corresponding areas are re-inspected after processing. Finally, based on the final state of all suspected bubble areas on the workpiece, the curved glass is judged as either approved or rejected, thereby achieving a complete closed loop for bubble defect detection and processing in the curved glass coating process.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides an AI-based intelligent detection method for bubble defects in the coating process of curved glass, which is applied to a curved glass coating equipment. The curved glass coating equipment includes at least a positioning fixture, a vacuum adsorption platform, a film unwinding mechanism, a pre-lamination mechanism, a main pressing mechanism, an exhaust roller mechanism, an edge-gathering mechanism, an image acquisition unit, a light source unit, a contour acquisition unit, a parameter acquisition unit, and a control unit.

[0013] The method includes the following steps: S1. Position the curved glass to be coated on the positioning fixture, and use the vacuum adsorption platform to adsorb and fix the curved glass, obtain the surface curvature information of the curved glass, and establish the workpiece coordinate system and surface area mapping relationship. First, the curved glass to be coated is positioned on the positioning fixture, and the curved glass is adsorbed and fixed by the vacuum adsorption platform to obtain the surface curvature information of the curved glass and establish the workpiece coordinate system and surface area mapping relationship.

[0014] Furthermore, the surface curvature information can be obtained by online acquisition of the curved glass surface contour by the contour acquisition unit, or by calling pre-stored curved surface template data and performing registration correction in conjunction with the current positioning result. The control unit establishes a regional mapping relationship of the curved glass surface based on the curvature information, providing a positional basis for subsequent re-inspection of the same area.

[0015] S2. Divide the surface of the curved glass into detection areas based on the surface curvature information; The curved glass surface is divided into detection areas based on the surface curvature information. Preferably, the curved glass surface is divided into at least two regions: a high curvature edge region, a curvature transition region, and a gentle center region, and different detection rules are configured for different detection regions.

[0016] S3. During the coating process of the curved glass coating equipment, images of the detection area are acquired at different process stages, and reflection suppression and abnormal area extraction are performed on the acquired images to obtain suspected bubble areas. Furthermore, the detection rules include at least one or more of the following: defect area release threshold, re-inspection priority, image acquisition frequency, and processing priority. Preferably, the defect area release threshold for high-curvature edge regions is lower than that for smooth center regions, and the re-inspection priority for high-curvature edge regions is higher than that for smooth center regions, thereby enabling the system to perform more rigorous and higher-frequency detection and processing on abnormal areas in high-curvature edge regions.

[0017] Subsequently, during the coating process of the curved glass coating equipment, images of the detection area are acquired at different stages of the process, and reflection suppression and abnormal area extraction are performed on the acquired images to obtain suspected bubble areas.

[0018] Preferably, the different process stages include a pre-bonding stage, a main pressing stage, an venting and shaping stage, and an edge-trimming stage. The pre-bonding stage is used to generate initial suspected bubble areas; the main pressing stage is used to observe whether the suspected bubble areas migrate or shrink as the pressing progresses; the venting and shaping stage is used to make a primary judgment on the suspected bubble areas; and the edge-trimming stage is used to identify edge-rebound bubbles.

[0019] The reflection suppression is preferably achieved through comparative processing between images at different illumination angles and / or between images at different polarization states, in order to reduce the interference of specular reflection, highlight bands, and reflection artifacts on the extraction of abnormal regions. The extraction of abnormal regions is preferably performed based on at least one of local brightness anomaly features, edge closure features, contour morphology features, and light transmission anomaly features, thereby obtaining suspected bubble regions.

[0020] S4. The suspected bubble area is re-inspected in the same area in subsequent process stages to obtain the change characteristics of the suspected bubble area between different process stages, and the equipment operating parameters of the corresponding process stages are obtained simultaneously. After identifying the suspected bubble area, the same area is re-inspected in subsequent process stages to obtain the change characteristics of the suspected bubble area between different process stages, and the equipment operating parameters of the corresponding process stages are obtained simultaneously.

[0021] The re-inspection of the same area is preferably achieved in the following way: based on at least one of the transformation relationship between the equipment coordinate system and the workpiece coordinate system, the reference position of the positioning fixture, the encoder displacement information and the local image registration result, the suspected bubble area identified in the previous process stage is mapped to the corresponding position in the next process stage, and the abnormal area extraction is re-executed in the corresponding re-inspection window to ensure that the objects compared in the previous and next stages are the same physical area.

[0022] The variation characteristics include at least one or more of the following: area change rate, center position drift, displacement along the pressing direction, change in contour roundness, change in brightness, and increase in edge springback. The equipment operating parameters include at least one or more of the following: membrane tension, preheating temperature, main pressing pressure, main pressing speed, exhaust roller pressure, vacuum adsorption state, edge-receiving pressure, edge-receiving speed, and edge temperature.

[0023] S5. Classify and determine the suspected bubble area based on the change characteristics to distinguish between transient eliminable bubbles, stable residual bubbles, edge rebound type bubbles and pseudo defects. Based on the changes and the equipment operating parameters, the suspected bubble areas are classified and determined to distinguish between transiently removable bubbles, stable residual bubbles, edge rebound bubbles, and pseudo-defects.

[0024] Among them, when the suspected bubble area shows a natural retreat trend during the main pressing stage or the exhaust shaping stage, and disappears during the exhaust shaping stage, it can be determined as a transient removable bubble.

[0025] If a suspected bubble area exists continuously in the main pressing stage and the degassing and shaping stage, it can be identified as a stable residual bubble. Preferably, it can be further identified as a stable residual bubble when the area change rate is within a preset second threshold (10%-20%) and the center position drift is less than a preset third threshold (1mm-2mm).

[0026] When a suspected bubble area is located in a high curvature edge region, and after shrinking during the main pressing stage or the venting and shaping stage, it enlarges again during the edge closing stage, it can be identified as an edge rebound bubble.

[0027] When a suspected bubble area exhibits unstable contours, unclosed boundaries, or no reasonable correlation with the process direction under different lighting conditions, or shows significant attenuation after polarization switching, it can be identified as a false defect.

[0028] S6. Based on the classification judgment result, control the curved glass coating equipment to perform corresponding processing actions, and after the processing actions are performed, process and re-inspect the corresponding area; For transiently removable bubbles, it is preferable to perform a confirmation re-inspection; when the confirmation re-inspection result shows that the area has disappeared, the area is updated to the eliminated state; when the confirmation re-inspection result shows that the area still exists, it is updated to the pending state.

[0029] For stable residual bubbles, it is preferable to perform local repressurization, supplemental venting, or a combination of both. After treatment, a post-treatment re-inspection is performed. If the post-treatment re-inspection results show that the abnormal area has disappeared or the residual area is lower than the release threshold of the corresponding detection area, then it is updated to the post-treatment pass status. If the post-treatment re-inspection results show that the abnormal area has no significant change, then the treatment is upgraded.

[0030] For edge-rebound bubbles, it is preferable to perform at least one of the following treatment actions: secondary edge trimming, directional re-pressing of the edge, and auxiliary heating of the edge; after treatment, perform post-treatment re-inspection; if the post-treatment re-inspection results show that the abnormal area still exists and the preset maximum number of treatments has been reached, then update it to the final unqualified state.

[0031] For pseudo-defects, it is preferable not to trigger rework actions, but the recorded information can be retained for subsequent parameter optimization or quality traceability.

[0032] During processing and re-inspection, the control unit preferably establishes a re-inspection task record for each suspected bubble area. The re-inspection task record includes at least the area number, the workpiece number, the curvature zone, the initial appearance stage, the current status label, the number of re-inspections performed, the number of processing steps performed, and the current risk level. Furthermore, it is preferable to set a maximum number of processing steps and / or a maximum number of re-inspections for each suspected bubble area. If the release conditions are still not met after reaching the maximum number of processing steps or the maximum number of re-inspections, the area is determined to be in a final unqualified state.

[0033] S7. Update the status of the suspected bubble area based on the re-inspection results after processing, and perform a release or non-conformance judgment on the curved glass based on the final status of all suspected bubble areas on the workpiece.

[0034] The status of the suspected bubble area is updated based on the re-inspection results after processing, and the curved glass is judged to be released or rejected based on the final status of all suspected bubble areas on the workpiece.

[0035] Preferably, the control unit establishes a workpiece-level status table for each piece of curved glass. The workpiece-level status table includes at least the final category of all suspected bubble areas on the workpiece, whether they have been processed, whether they passed after processing, whether there are any pending anomalies, and whether there are any prohibited release anomalies. The curved glass is released only when the final status of all suspected bubble areas is "eliminated," "passed after processing," or "false defect." If any suspected bubble area's final status is a failed stable residual bubble or a failed edge-rebound bubble, the curved glass is rejected.

[0036] This invention also provides an intelligent detection system for bubble defects in curved glass coating for implementing the above method. The system includes: a curvature information acquisition module for acquiring the surface curvature information of the curved glass to be coated and establishing a workpiece coordinate system and surface region mapping relationship; a region division module for dividing the surface of the curved glass into detection regions based on the surface curvature information; an image acquisition module for acquiring images of the detection regions at different process stages during the coating process performed by the curved glass coating equipment; and a control module connected to the curvature information acquisition module, the region division module, and the image acquisition module, and for performing reflection suppression and abnormal region detection on the images acquired by the image acquisition module. Extraction is performed to identify suspected bubble areas; these areas are then re-inspected in subsequent processes to obtain their variation characteristics across different processes, and the corresponding equipment operating parameters are simultaneously acquired; based on these variation characteristics, the suspected bubble areas are classified to distinguish between transiently removable bubbles, stable residual bubbles, edge-rebound bubbles, and false defects; according to the classification results, the curved glass coating equipment is controlled to perform corresponding processing actions, and a post-processing re-inspection is conducted after each action; the status of the suspected bubble areas is updated based on the post-processing re-inspection results, and based on the final status of all suspected bubble areas on the workpiece, a release or rejection determination is made for the curved glass.

[0037] The present invention also provides an electronic device, the electronic device including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method.

[0038] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Beneficial effects

[0039] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention does not rely solely on a single static final inspection after the curved glass is coated, but instead captures images of the inspection area at different stages of the coating process and re-inspects the same area for suspected bubble areas. This allows for a more accurate determination based on the evolution of abnormal areas at different stages of the process, avoiding the misjudgment and missed judgment problems caused by single-moment detection in the prior art.

[0040] 2. This invention introduces curvature information of the curved glass surface and divides the curved glass surface into regions based on this curvature information. Different detection rules, judgment thresholds, and processing priorities are applied to high-curvature edge areas, curvature transition areas, and gentle central areas. Since high-curvature edge areas are high-incidence areas for bubble defects, especially edge-rebound bubbles, this invention improves the detection sensitivity and processing targeting of these areas through a regional differentiation strategy, thus making it more suitable for the actual needs of curved glass coating processes.

[0041] 3. This invention not only extracts image features of suspected bubble areas, but also further extracts the variation features of these areas across different process stages. Simultaneously, it collects equipment operating parameters such as main pressing pressure, main pressing speed, venting roller pressure, membrane tension, and edge-receiving pressure. These variation features, along with the equipment operating parameters, are used for classification and determination. This approach makes the classification results more consistent with the actual process mechanism, effectively distinguishing between transiently removable bubbles, stable residual bubbles, edge-rebound bubbles, and false defects.

[0042] 4. This invention employs differentiated processing actions for different types of abnormal areas. For transiently removable bubbles, a confirmation re-inspection is performed; for stable residual bubbles, local repressurization, supplementary venting, or a combination of these processes are performed; for edge-rebound bubbles, at least one of the following processes is performed: secondary edge tightening, directional edge repressurization, and auxiliary edge heating; for pseudo-defects, no rework is triggered. Compared to the existing technology's approach of "uniformly alarming or uniformly rejecting anomalies," this invention's processing strategy is more targeted, more conducive to improving yield, and reducing unnecessary rework.

[0043] 5. This invention does not determine the pass / fail status of an entire workpiece solely based on whether a single abnormal area has disappeared. Instead, it performs a final workpiece-level judgment based on the final state of all suspected bubble areas on the workpiece. Only when all suspected bubble areas have been eliminated, passed after processing, or are identified as pseudo-defects can the workpiece be released; otherwise, it is deemed unqualified. This workpiece-level final judgment mechanism makes the final release conclusion more comprehensive and reliable, which is beneficial for improving the consistency of finished product quality.

[0044] 6. This invention forms a complete closed-loop logic from curvature information acquisition, region division, phased image acquisition, re-inspection of the same region, classification and judgment, differential processing, post-processing re-inspection to final judgment at the workpiece level. This closed-loop logic not only improves the accuracy of bubble defect detection in the curved glass coating process, but also enables the detection results to directly serve the optimization of equipment process actions and defect diversion control, thus possessing good engineering feasibility and industrial application value. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the overall logic processing of the present invention; Figure 3 This is a schematic diagram of the structure of a system provided in an embodiment of the present invention. Detailed Implementation

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Any equivalent substitutions, improvements, combinations, or simplifications made by those skilled in the art based on this specification within the spirit and principles of the present invention should fall within the scope of protection of the present invention.

[0047] In this specification, unless otherwise expressly defined, "curved glass" includes, but is not limited to, double-curved glass, four-curved glass, 2.5D glass, 3D glass, and irregularly shaped cover glass with locally high curvature edges; "coating" includes the process of applying protective film, explosion-proof film, functional film, decorative film, and other film materials to the surface of curved glass; "bubble defects" include closed or semi-closed cavities formed by residual air between the film layer and the curved glass, local cavities regenerated after edge rebound, and strip-shaped or cluster-shaped abnormal areas formed by insufficient air venting; "pseudo-defects" include reflections, bright bands, adhesive lines, slight wrinkles, foreign object obstruction, edge shadows, etc., which are not real bubbles but appear as abnormal areas in the image.

[0048] This invention is primarily applied to curved glass coating equipment. The curved glass coating equipment may include a feeding mechanism, positioning fixtures, a vacuum adsorption platform, a film unwinding mechanism, a release layer peeling mechanism, a pre-alignment mechanism, a pre-lamination mechanism, a main pressing mechanism, an exhaust roller mechanism, an edge-gathering mechanism, an image acquisition unit, a light source unit, a contour acquisition unit, a parameter acquisition unit, and a control unit. Depending on the actual equipment structure, it may also include a local re-pressing mechanism, an edge heating mechanism, a sorting mechanism, a buffer mechanism, an industrial control computer, and an edge computing terminal.

[0049] The image acquisition unit can be one or more industrial cameras, preferably area array industrial cameras, but line array industrial cameras can also be used depending on the production line cycle time and installation space. The light source unit can include one or more of the following: bar light source, ring light source, coaxial light source, dome light source, side-slanted light source, and polarizer assembly. The contour acquisition unit can include one or more of the following: line laser contour sensor, structured light scanning assembly, depth camera, or contact contour measurement device. The parameter acquisition unit can include temperature sensor, pressure sensor, displacement sensor, tension sensor, vacuum pressure sensor, angle sensor, encoder, photoelectric switch, etc.

[0050] The control unit preferably includes an industrial controller, an industrial computer, or a combination of a PLC and a host computer, used to acquire curvature information, image information, and process parameters, perform area division, same area re-inspection, change feature extraction, fusion judgment, and processing action output, and to perform linkage control with the main pressing mechanism, exhaust roller mechanism, edge gathering mechanism, local re-pressing mechanism, and sorting mechanism.

[0051] For ease of explanation, this invention divides the curved glass coating process into a pre-lamination stage, a main pressing stage, an venting and shaping stage, and an edge finishing stage. It should be noted that different equipment may have different names for these process stages; for example, they may be designated as a preheating stage, a secondary pressing stage, or a final inspection stage. However, as long as the technical essence still involves re-inspecting the same area in at least two different process stages, and combining image evolution information with equipment operating parameters for fusion judgment, they all fall within the protection scope of this invention. Example

[0052] This embodiment provides an AI-based intelligent detection method for bubble defects in the coating process of curved glass. To ensure a thorough explanation, this embodiment focuses on describing the main technical contents of the invention.

[0053] In this embodiment, the product to be processed is a piece of curved glass to be coated, and the film material is a roll-shaped functional film. After the curved glass is conveyed to the positioning station by the feeding mechanism, it first enters the workpiece fixing process. Specifically, the positioning fixture limits at least two sides and one end of the curved glass to form a stable planar positioning reference; the vacuum adsorption platform adsorbs the bottom of the curved glass to suppress workpiece slippage, warping, or posture changes during the coating process. The control unit reads the vacuum pressure sensor signal, and when the vacuum degree reaches the preset lower limit value, it determines that the workpiece fixing is complete; if the vacuum degree does not reach the preset value, the control unit outputs an adsorption abnormality signal and suspends the subsequent coating operation.

[0054] Appendix Figure 1 The overall process is shown. (See attached document.) Figure 1As shown, after the workpiece is fixed, the control unit establishes a workpiece coordinate reference. Preferably, an equipment coordinate system is established using at least three reference points on the positioning fixture, and a workpiece coordinate system is established using the geometric center of the curved glass, edge feature points, or a preset reference edge. The equipment coordinate system and the workpiece coordinate system are mapped to each other through coordinate transformation relationships. This coordinate relationship is used for area mapping and re-inspection alignment between different subsequent process stages.

[0055] Contour sampling is performed on the curved glass surface. Preferably, a line laser contour sensor scans along the length of the curved glass and performs several equally spaced cross-sectional samplings in the width direction to obtain surface height distribution data of the curved glass. The control unit reconstructs the surface contour model of the curved glass based on the height distribution data and further calculates the curvature characteristics at various locations on the surface. For standardized products, standard curved surface templates in the database can also be directly called, and rigid body registration or local correction can be performed in combination with the current workpiece positioning information.

[0056] The control unit divides the curved glass surface into regions based on curvature characteristics. Preferably, the curved glass surface is divided into a high-curvature edge region, a curvature transition region, and a gentle central region. The high-curvature edge region refers to the area near the curved edges, rounded corners, and areas with significant local bends around the glass; the curvature transition region refers to the intermediate strip-shaped area that transitions from the gentle central region to the edge curved surface region; the gentle central region refers to the area with minimal curvature change and an overall tendency towards flatness. The extent of each region can be automatically generated based on a curvature threshold or predefined based on a standard template.

[0057] In this embodiment, differentiated detection rules are applied to different regions. For high-curvature edge areas, the control unit sets a lower defect area judgment threshold, a higher number of re-inspections, and a higher processing priority. For curvature transition areas, in addition to considering area, the weight of contour stability and brightness changes is increased. For smooth center areas, the allowable residual threshold is appropriately increased to avoid over-processing minor optical fluctuations that do not affect appearance and reliability. This differentiated regional setting better meets the actual quality control requirements of strict edge control and stable center control in curved glass coating processes.

[0058] After the area division is completed, the film unwinding mechanism outputs the film, the release layer peeling mechanism separates the release layer from the functional film, and the pre-alignment mechanism completes the pre-alignment of the film with the curved glass based on the edge of the film, the edge of the workpiece, or visual markings. After the pre-alignment is completed, the pre-lamination mechanism makes one side of the film's starting edge or starting point contact the curved glass first, entering the pre-lamination stage.

[0059] During the pre-bonding stage, the image acquisition unit takes an initial image of the curved glass surface. Preferably, the image acquisition unit acquires at least two images of the same area under different illumination angles; more preferably, images can also be acquired under different polarization states to reduce artifact interference caused by curved surface reflection. For high-curvature edge areas, it is preferable to use oblique strip light combined with a polarizer to improve the prominence of local cavities at the edges; for gently sloping central areas, it is preferable to use a more uniform surface light source or a dome light source to obtain stable background brightness.

[0060] After the initial image acquisition, the control unit performs image preprocessing. This preprocessing includes at least brightness normalization, local contrast enhancement, noise suppression, and reflection suppression. These processes reduce the interference of curved glass reflections on subsequent suspected area extraction.

[0061] After image preprocessing, the control unit extracts suspected bubble regions within each detection area. Preferably, this can be achieved by extracting local brightness anomaly features, edge closure features, contour morphology features, and light transmission anomaly features, and then obtaining candidate connected regions based on adaptive threshold segmentation. The area, perimeter, roundness, aspect ratio, average gray level, edge sharpness, and spatial relationship with the region boundary are further calculated for each candidate connected region. A differential threshold is applied to filter these regions based on their respective curvature zones, retaining those that meet the criteria to form a set of suspected bubble regions for the pre-fitting stage. For each retained suspected bubble region, the control unit generates a unique region identifier number, recording its workpiece number, curvature zone, first appearance stage, center position, area, and morphological features, forming a first-stage suspected region mask. This mask is used for target localization during subsequent re-inspection of the same region, and is not immediately used as the final defect conclusion.

[0062] It should be noted that suspected areas formed during the pre-lamination stage are not directly identified as final bubble defects. This is because some localized cavities formed during the initial contact of the membrane material may be naturally expelled during subsequent main lamination and venting processes. If these cavities are rejected immediately after the initial image acquisition, false alarms are likely. Therefore, in this embodiment, suspected areas during the pre-lamination stage are used as the starting point for subsequent cross-stage tracking.

[0063] The main pressing stage then begins. The main pressing mechanism presses and advances the membrane material along a preset path. The main pressing mechanism can be a roller type, a plate type, a flexible pressure head type, or a combination thereof. During the pressing process, the control unit simultaneously collects process parameters such as main pressing pressure, main pressing speed, current position, pressing angle, membrane tension, and vacuum adsorption status.

[0064] During the main pressing stage, the image acquisition unit performs a re-inspection of the suspected areas identified in the pre-bonding stage. Specifically, the control unit first maps the suspected areas from the pre-bonding stage from the first-stage image coordinates to the predicted positions in the current main pressing stage based on the fixture reference, workpiece coordinate mapping relationship, and the current position of the pressing mechanism; then, a re-inspection window is generated near the predicted position; and then, combined with the glass edge contour, local grayscale structure, visible fixture marks, or other auxiliary features, the re-inspection window is locally registered and corrected to determine the same area corresponding to the previous stage in the current stage.

[0065] Within the re-inspection window, the control unit re-executes image preprocessing and abnormal region extraction to obtain candidate regions corresponding to the main pressing stage. Subsequently, the control unit compares the same region in the pre-bonding stage and the main pressing stage, extracting the first set of change features. These change features may include the area change rate, center position drift, displacement along the pressing direction, contour roundness change, and brightness change. Specifically, the area change rate reflects whether the suspected area shrinks, enlarges, or remains essentially unchanged under the pressing effect; the center position drift and displacement along the pressing direction reflect whether local air is moved by the main pressing; the contour roundness change is used to distinguish between stable cavities and non-closed wrinkles; and the brightness change is used to assist in determining the degree of film adhesion.

[0066] As attached Figure 2 As shown, in this embodiment, if a suspected area significantly shrinks during the main pressing stage relative to the pre-bonding stage, and the entire area shifts along the pressing direction, then this area is preferentially marked as a transiently removable candidate area. If a suspected area has a very small area change and its position remains essentially unchanged, it is marked as a stable residual bubble candidate area. Preferably, when the area change rate is within a preset second threshold (10%-20%) and the center position drift is less than a preset third threshold (1mm-2mm), it is determined to be a stable residual bubble. If a suspected area is located in a high-curvature edge region, although it shrinks to some extent, it remains close to the edge, and the long axis of its contour is relatively consistent with the tangent direction of the edge, then it is marked as an edge rebound risk candidate area. The above markings are not final conclusions, but are further verified during the air supply and exhaust shaping stage and the edge finishing stage.

[0067] After the main pressing stage is completed, the workpiece enters the venting and shaping stage. The venting roller mechanism rolls and vents the film layer along the attached path or according to a preset zone path. The control unit synchronously collects information on the venting roller pressure, roller speed, number of roller passes, local contact duration, and roller path. For high-curvature edge areas, a higher number of local passes or a lower cycle time can be preferably set to improve the edge venting effect.

[0068] During the exhaust molding stage, the image acquisition unit performs a re-inspection of suspected and high-risk areas that still exist from the previous stage. The control unit uses the same area mapping method described above to map the suspected areas corresponding to the main pressing stage onto the exhaust molding stage image, and re-extracts the abnormal areas within the corresponding re-inspection window, thereby obtaining the re-inspection results of the exhaust molding stage.

[0069] The control unit extracts a second set of change features during the exhaust shaping stage, which may include the area reduction ratio of the exhaust shaping stage relative to the main pressing stage, the cumulative area reduction ratio relative to the pre-bonding stage, local contrast changes, changes in boundary closure degree, and whether the area has completely disappeared. For areas that have completely disappeared, they can be preferentially identified as eliminated anomalies, and these bubbles are identified as transiently removable bubbles. For areas that still exist and whose area change is less than a preset threshold and whose position drift is less than a preset threshold, they can be preliminarily identified as candidate areas for stable residual bubbles. For areas that have significantly decreased in size but still retain residual cores near high curvature edges, or when the suspected bubble area is located in a high curvature edge area, they are further included in the key review list for the edge finishing stage and listed as areas to be reviewed for edge finishing.

[0070] In this embodiment, the control unit establishes a temporal feature sequence for each suspected area. The temporal feature sequence includes at least the area A1 of the pre-lamination stage, the area A2 of the main pressing stage, and the area A3 of the exhaust and shaping stage; the center coordinates C1 of the pre-lamination stage, C2 of the main pressing stage, and C3 of the exhaust and shaping stage; and the corresponding process parameter sets P1, P2, and P3. Through this temporal feature sequence, the evolution of a certain area during the coating process can be comprehensively reflected.

[0071] After venting and shaping, the workpiece enters the edge-trimming stage. The edge-trimming mechanism presses, wraps, or shapes the edges of the membrane material. For products with high curvature edges, the edge-trimming stage is where edge-rebound bubbles are most likely to occur. This is because internal stress accumulates in the membrane material in areas of high curvature, and new cavities may form after trimming due to springback or insufficient local adhesion. Therefore, the edge-trimming stage is set as an independent re-inspection stage, rather than being the final judgment after venting and shaping.

[0072] During the edge finishing stage, the image acquisition unit prioritizes capturing images of the aforementioned area to be finished and verified. The control unit simultaneously reads process parameters such as edge finishing pressure, edge finishing angle, edge finishing speed, number of edge re-pressing times, and edge temperature, and merges them with the aforementioned time-series feature sequence.

[0073] The control unit extracts abnormal areas from the re-inspection images during the edge-trimming stage and compares them with those from the previous stages. If an area exhibits abnormalities in the pre-bonding, main pressing, or venting and shaping stages, and its size has significantly decreased during the venting and shaping stage but increases again during the edge-trimming stage, preferably, the increase can be set to exceed a preset edge rebound threshold, which is set to 10-30%. In this case, the area is identified as an edge rebound type bubble. The control unit can also adjust the edge rebound risk based on the distance of the area from the edge, the main axis direction of the area, the edge-trimming pressure fluctuation, and the edge temperature.

[0074] If a region is identified as a candidate region for stable residual bubbles, the control unit will process the flow diversion based on the region it is located in. For a candidate region for stable residual bubbles located in a flat central region and whose area is lower than the upper limit of the central region's repair trigger, it is preferable to perform a local repressurization first; for a candidate region for stable residual bubbles located in a curvature transition region, it is preferable to perform a combination of local repressurization and supplementary exhaust.

[0075] Local repressurization refers to applying pressure to a specific point or a small area of ​​the target abnormal region using a local repressurization mechanism or a localized pressing mechanism. The control unit can determine the repressurization path, direction, pressure, number of repressurizations, and dwell time based on the location, area, aspect ratio, and risk level of the abnormal region. For stable residual bubble areas with small areas and clear boundaries, localized repressurization is preferred; for elongated abnormal regions, short-stroke repressurization along their long axis is preferred; for abnormal regions near the edge, directional repressurization towards the nearest edge is preferred.

[0076] The supplementary venting refers to re-performing the venting and rolling action around the abnormal area to force the residual air under the membrane to be discharged towards the edge along the shortest path. The control unit automatically plans the venting direction and path based on the location, size, and distance from the edge of the abnormal area. For abnormal areas close to the edge, the venting direction is preferably set towards the nearest edge; for abnormal areas located off-center, the venting direction is preferably set to the optimal path between the extension direction of the main pressing direction and the direction of the nearest edge.

[0077] In this embodiment, any abnormal area that has undergone partial repressurization or supplemental venting must enter a post-processing re-inspection state. The post-processing re-inspection is triggered immediately after the corresponding processing action is completed by the image acquisition unit. The control unit compares the images of the same area before and after processing to confirm the processing effect. If the post-processing re-inspection result shows that the abnormal area has disappeared, the control unit updates the status of that area to "processed and passed." If the abnormal area has not completely disappeared but its area has significantly shrunk, the control unit can decide whether to allow it to proceed to the next process for continued observation, or directly perform a second processing step, based on the reduction ratio, remaining area, and risk level. If the post-processing re-inspection result shows that the abnormal area has essentially no change, the control unit determines that the first processing was invalid and classifies it as unqualified.

[0078] For edge-rebound bubbles, the control unit preferably employs a specialized edge treatment strategy instead of the standard central area repressing method. This edge treatment strategy includes at least one of the following: secondary edge trimming, directional edge repressing, and auxiliary edge heating. Secondary edge trimming refers to the edge trimming mechanism performing a second pressing and shaping action on the target edge area; directional edge repressing refers to directional repressing of local anomalies along the normal or tangential direction of the glass edge; and auxiliary edge heating refers to short-term heating of the edge area within permissible process limits to reduce the localized rebound tendency of the film material and improve edge conformability.

[0079] After edge-rebound bubbles undergo edge treatment, the control unit immediately triggers an edge confirmation re-inspection. If the re-inspection shows that the edge abnormality has disappeared, or its area is lower than the release threshold corresponding to the high-curvature edge area, the control unit updates the status of that area to "processed and passed." If the bubble still exists after the re-inspection, the control unit decides whether to continue edge treatment or directly determine it as a final non-conformity based on the remaining area, rebound trend, and number of processing operations already performed.

[0080] To avoid uncontrolled repetitive processing of the same area, this embodiment preferably sets a maximum number of processing times and a maximum number of re-inspections for each suspected area. The maximum number of processing times limits the number of times actions such as local repressurization, supplementary venting, and secondary edge tightening are repeated; the maximum number of re-inspections limits the number of times the system repeatedly observes the same area. If an area has reached the maximum number of processing times but still does not meet the release conditions, the control unit preferably directly determines it as ultimately unqualified; if an area has reached the maximum number of re-inspections but its status is still uncertain, the control unit preferably processes or determines it according to the higher-risk category and no longer retains the observation status for a long time.

[0081] For pseudo-defects, this embodiment preferably uses the following rules for exclusion: If an abnormal area significantly shifts in position and exhibits obvious contour changes under different lighting angles, it is preferably identified as a reflection artifact; if an abnormal area has an unclosed boundary and is elongated, and highly consistent with the film texture direction, it is preferably identified as a glue texture or wrinkle pseudo-defect; if an abnormal area is only noticeable under one polarization state and significantly attenuates under another polarization state, it is preferably identified as an optical reflection pseudo-defect; if an abnormal area does not have a reasonable evolutionary relationship between different process stages and is unrelated to pressing, venting, or edge finishing actions, it is preferably identified as a pseudo-defect. For pseudo-defects, the control unit does not trigger rework processing, but preferably records its location and type for subsequent optimization of lighting and judgment parameters.

[0082] In this embodiment, the release of a workpiece is not determined based on a single abnormal area, but rather on the combined final state of all abnormal areas of the workpiece. Therefore, the control unit establishes a workpiece-level status table for each piece of curved glass. This table includes at least the final category of all suspected areas on the workpiece, whether they have been processed, whether they passed processing, whether there are any pending abnormalities, and whether there are any abnormalities that prohibit release. If all suspected areas are ultimately eliminated, passed processing, or are pseudo-defects, the workpiece enters the release state; if any abnormal area has a final state of failing stable residual bubbles or failing edge-rebound bubbles, the workpiece enters the unqualified state.

[0083] To further improve the certainty of the final judgment, the control unit preferably includes release confirmation logic. This release confirmation logic includes at least the following checks: checking whether all abnormal areas have completed their state transition; checking whether there are any abnormal areas that have reached the maximum number of processing attempts but still failed to pass; checking whether there are any high-risk edge abnormalities marked as prohibited from release; and checking whether the final stage image acquisition and final state determination are complete. Only when all the above checks meet the release conditions will the control unit output a release command to the feeder station or buffer station.

[0084] For defective workpieces, the control unit preferably outputs a defective identifier and sends the location, category, risk level, and processing history of the corresponding abnormal area to the sorting or rework station. This allows rework personnel or equipment to directly obtain information about the problem area without having to perform a full search of the entire workpiece, improving rework efficiency and targeting.

[0085] In a further preferred embodiment, the control unit can also statistically analyze regional quality trend information based on the detection results of multiple workpieces. For example, when multiple consecutive workpieces develop edge-rebound bubbles in the same high-curvature edge area, the system can automatically increase the re-inspection frequency of subsequent workpieces in that area, or trigger edge strengthening treatment in advance; when a large number of workpieces in a batch exhibit stable residual bubbles in the central area, the system can prompt adjustments to the main pressing speed, venting pressure, or membrane tension. While such trend optimization does not affect the basic judgment logic of a single workpiece, it helps to further improve the overall line yield.

[0086] As can be seen from the above, this embodiment does not use a single-moment photograph to determine whether an item is qualified or not. Instead, it implements a closed-loop control throughout the entire process of "initial inspection—re-inspection—classification and judgment—processing—re-re-inspection—final judgment" for suspected bubble areas. This method can distinguish between transient anomalies and real residues, as well as identify edge-rebound bubbles that regenerate after edge finishing. Furthermore, through post-processing re-inspection and workpiece-level final judgment mechanisms, it ensures that the final release conclusion is based on complete, clear, and verifiable technical logic. Example

[0087] This embodiment provides a system architecture corresponding to the above method, such as... Figure 3 As shown. The system mainly includes the following modules.

[0088] Curvature information acquisition module: used to acquire the surface curvature information of the curved glass to be coated, and establish the workpiece coordinate system and surface area mapping relationship; Area division module: used to divide the surface of the curved glass into detection areas based on the surface curvature information; Image acquisition module: used to acquire images of the detection area at different process stages during the coating process of the curved glass coating equipment; Control module: the control module is connected to the curvature information acquisition module, the area division module and the image acquisition module respectively, and is used to: suppress reflection and extract abnormal areas from the images acquired by the image acquisition module to obtain suspected bubble areas; and process the suspected bubble areas... The system performs a re-inspection of the same area in subsequent process stages to obtain the change characteristics of the suspected bubble area between different process stages, and simultaneously obtains the equipment operating parameters of the corresponding process stages; based on the change characteristics, the suspected bubble area is classified and judged to distinguish between transient removable bubbles, stable residual bubbles, edge rebound bubbles, and false defects; according to the classification and judgment results, the curved glass coating equipment is controlled to perform corresponding processing actions, and a post-processing re-inspection is performed after the processing actions are performed; the state of the suspected bubble area is updated according to the post-processing re-inspection results, and based on the final state of all suspected bubble areas on the workpiece, a release judgment or a non-conformance judgment is performed on the curved glass. Example

[0089] The invention can also be implemented as a computer program product on a computing device and storage medium.

[0090] An electronic device, such as a general-purpose computer, server, or dedicated image processing device, includes at least one processor, a memory, and necessary interface means. Computer program instructions are pre-stored in the memory, and when the program is loaded and executed by the processor, it causes the electronic device to perform the steps described in Embodiment 1.

[0091] A computer-readable storage medium stores a computer program (e.g., in the form of a program mentioned in the above-described electronic device). When the program is read and executed by a processor, the steps of the method described in Embodiment 1 are also implemented. The storage medium may be a non-volatile storage device, such as ROM, flash drive, portable hard drive, optical disc, etc.

[0092] It should be noted that the above-described embodiments of the electronic devices and storage media are not essential independent components of the present invention, but rather equivalent extensions of the methods and systems of the present invention. Based on the content of this invention, those skilled in the art can easily write corresponding software programs and deploy them on existing computer devices to implement the methods and functions of the present invention. The scope of protection of this invention covers equivalent solutions of such software and hardware implementations.

[0093] Those skilled in the art should also understand that the specific technical features in the above embodiments can be combined or replaced as needed to form different solutions. Any such solutions falling within the principles and spirit of this invention will be protected by the claims of this invention. Contents not described in detail in this specification are well-known to those skilled in the art.

[0094] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. An AI-based intelligent detection method for bubble defects in curved glass coating processes, characterized in that, The method, applied to curved glass coating equipment, includes the following steps: S1. Position the curved glass to be coated on the positioning fixture, and use a vacuum adsorption platform to adsorb and fix the curved glass, obtain the surface curvature information of the curved glass, and establish the workpiece coordinate system and surface area mapping relationship. S2. Divide the surface of the curved glass into detection areas based on the surface curvature information; S3. During the coating process of the curved glass coating equipment, images of the detection area are acquired at different process stages, and reflection suppression and abnormal area extraction are performed on the acquired images to obtain suspected bubble areas. S4. The suspected bubble area is re-inspected in the same area in subsequent process stages to obtain the change characteristics of the suspected bubble area between different process stages, and the equipment operating parameters of the corresponding process stages are obtained simultaneously. S5. Classify and determine the suspected bubble area based on the change characteristics to distinguish between transient eliminable bubbles, stable residual bubbles, edge rebound type bubbles and pseudo defects. S6. Based on the classification judgment result, control the curved glass coating equipment to perform corresponding processing actions, and after the processing actions are performed, process and re-inspect the corresponding area; S7. Update the status of the suspected bubble area based on the re-inspection results after processing, and perform a release or non-conformance judgment on the curved glass based on the final status of all suspected bubble areas on the workpiece.

2. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 1, characterized in that, Obtaining the surface curvature information of the curved glass includes: obtaining the contour data of the curved glass surface, or calling the pre-stored curved template data, and establishing the transformation relationship between the equipment coordinate system and the workpiece coordinate system in combination with the reference position information of the positioning fixture, and dividing the curved glass surface into at least two regions among the high curvature edge region, the curvature transition region and the gentle center region.

3. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 2, characterized in that, Different detection rules are used for different detection areas. The detection rules include at least one or more of the following: defect area release threshold, re-inspection priority, image acquisition frequency, and processing priority.

4. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 1, characterized in that, The different process stages include a pre-bonding stage, a main pressing stage, an venting and shaping stage, and an edge-trimming stage; wherein, the pre-bonding stage is used to generate an initial suspected bubble area, the main pressing stage is used to determine whether the suspected bubble area migrates or shrinks as the pressing progresses, the venting and shaping stage is used to make a main judgment on the suspected bubble area, and the edge-trimming stage is used to identify edge-rebound bubbles.

5. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 4, characterized in that, The classification and determination of the suspected bubble area based on the change characteristics and the equipment operating parameters includes: when the suspected bubble area shows a natural disappearance trend in the main pressing stage or the exhaust shaping stage, and disappears in the exhaust shaping stage, it is determined to be a transiently removable bubble.

6. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 5, characterized in that, The classification and determination of the suspected bubble region based on the change characteristics and the equipment operating parameters includes: when the suspected bubble region exists continuously in the main pressing stage and the exhaust shaping stage, it is determined to be a stable residual bubble.

7. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 1, characterized in that, The classification and determination of the suspected bubble region based on the change characteristics and the equipment operating parameters includes: when the suspected bubble region is located in the high curvature edge region, and shrinks during the main pressing stage or the exhaust shaping stage, and then increases again during the edge closing stage, it is determined to be an edge rebound type bubble.

8. The AI-based intelligent detection method for bubble defects in curved glass coating process according to claim 7, characterized in that, For edge-rebound bubbles, the processing steps include at least one of secondary edge trimming, directional re-pressing of the edge, and auxiliary heating of the edge; after the processing steps are performed, a post-processing re-inspection is conducted; When the re-inspection results after processing show that the abnormal area still exists and the preset processing limit has been reached, the suspected bubble area will be updated to the final unqualified state.

9. A smart detection system for bubble defects in curved glass coating for implementing the method according to any one of claims 1 to 8, characterized in that, include: Curvature information acquisition module: used to acquire the surface curvature information of the curved glass to be coated, and to establish the workpiece coordinate system and surface area mapping relationship; Region division module: used to divide the detection region of the curved glass surface according to the surface curvature information; Image acquisition module: used to acquire images of the detection area at different stages during the coating process of the curved glass coating equipment; Control Module: The control module is connected to the curvature information acquisition module, the region division module, and the image acquisition module, and is used for: suppressing reflection and extracting abnormal regions from the images acquired by the image acquisition module to obtain suspected bubble regions; re-inspecting the same region in subsequent process stages to obtain the change characteristics of the suspected bubble regions between different process stages, and simultaneously acquiring the equipment operating parameters of the corresponding process stages; classifying the suspected bubble regions according to the change characteristics to distinguish between transient removable bubbles, stable residual bubbles, edge rebound bubbles, and false defects; controlling the curved glass coating equipment to perform corresponding processing actions according to the classification results, and performing post-processing re-inspection after the processing actions are performed; updating the state of the suspected bubble regions according to the post-processing re-inspection results, and performing release or non-conformity judgment on the curved glass based on the final state of all suspected bubble regions on the workpiece.

Images