Real-time detection method for seal size deviation based on machine vision

Abstract

The present application relates to the field of machine vision detection, and particularly relates to a sealing size deviation real-time detection method based on machine vision, which comprises the following steps: firstly, obtaining an image in an uncompressed state and establishing an arc length coordinate; then, applying two-stage compression load according to a detection formula, collecting contact information, edge appearance information and displacement information to form a response sequence; subsequently, determining a local cross-section state sequence, and performing time consistency check and space consistency check; and finally, outputting a size deviation detection result and a process adjustment instruction. The present application can improve the accuracy of abnormal identification, the stability of position correspondence, and the timeliness of process correction.

Description

Technical Field

[0001] This invention relates to the field of machine vision inspection, and more specifically to a machine vision-based method for real-time detection of dimensional deviations in sealing components. Background Technology

[0002] The dimensional stability of seals directly affects product assembly consistency, sealing reliability, and batch manufacturing quality. Timely detection of dimensional anomalies helps reduce leakage risks, minimize rework and scrap, and improve production line cycle time. Existing detection methods mostly rely on static visual measurement, spot checks, or single image interpretation, which usually only reflect local geometric dimensions and cannot simultaneously take into account changes in pressure contact, edge deformation, and recovery status. This leads to insufficient identification of hidden anomalies such as flanges, eccentricity, and local hardening, and there is no stable correlation between the location of anomalies and process deviations. Summary of the Invention

[0003] This invention provides a machine vision-based method for real-time detection of seal dimensional deviations, which at least solves the problem of how to accurately locate, reliably identify, and achieve process linkage adjustment of dimensional deviations of seals under multi-stage pressure conditions.

[0004] This invention provides a machine vision-based method for real-time detection of dimensional deviations in sealing components, the method comprising: Obtain an image of the uncompressed state of the seal to be tested, extract the inner and outer contours, determine the centerline, and establish arc length coordinates; The test formula corresponding to the seal to be tested is called, and the first compressive load and the second compressive load are applied in sequence. Contact information is obtained by total internal reflection contact imaging, edge morphology information is obtained by dark field imaging, speckle images are acquired and displacement information is determined by digital image correlation to form a response sequence. Determine the local cross-sectional state sequence corresponding to each arc length position based on the arc length coordinates and the response sequence; Based on the response sequence and local cross-sectional state sequence, perform time consistency checks and spatial consistency checks, output dimensional deviation detection results, and output process adjustment instructions based on the dimensional deviation detection results.

[0005] In one possible implementation, determining the centerline and establishing arc length coordinates includes: determining the centerline based on the midline between the inner and outer contours; and accumulating the arc length along the centerline to obtain the arc length coordinates.

[0006] In one possible implementation, the detection formula includes at least a first compressive load, a second compressive load, an image acquisition timing sequence, and a judgment threshold; forming a response sequence includes: performing time-series mapping of contact information, edge morphology information, and displacement information according to the uncompressed state, the first compressive load application state, the second compressive load application state, and the released state to form a response sequence.

[0007] In one possible implementation, contact information is obtained through total internal reflection contact imaging, including: lateral illumination of the transparent bearing surface carrying the seal to be tested; acquisition of reflection change images of the transparent bearing surface; and determination of the contact band width and contact band center position at each arc length position based on the reflection change images.

[0008] In one possible implementation, edge morphology information is obtained through dark-field imaging, including: acquiring dark-field images under a first compressive load state and a second compressive load state; determining the edge slope distribution at each arc length position based on the dark-field images to obtain edge morphology information.

[0009] In one possible implementation, speckle images are acquired and displacement information is determined by digital image correlation, including: acquiring speckle images in an uncompressed state, a first compressive load state, a second compressive load state, and a post-release state; performing digital image correlation calculations on the speckle images to determine the displacement change at each arc length position and the residual displacement after release, thereby obtaining displacement information.

[0010] In one possible implementation, the local cross-sectional state sequence corresponding to each arc length position is determined based on the arc length coordinates and the response sequence, including: performing positional mapping on the response sequence based on the arc length coordinates; extracting the local width, contact zone width, contact zone center offset, edge slope change, displacement change, and residual displacement after release for each arc length position based on the position-mapped response sequence; and determining the local cross-sectional state sequence based on the local width, contact zone width, contact zone center offset, edge slope change, displacement change, and residual displacement after release.

[0011] In one possible implementation, the local cross-sectional state sequence includes at least equivalent width deviation, contact deviation, anomaly type, and anomaly confidence; the anomaly type is used to indicate at least one of the following anomalies: material shortage, edge turning, cross-sectional eccentricity, local collapse, or local hardening.

[0012] In one possible implementation, temporal consistency checks and spatial consistency checks are performed, including: performing a temporal comparison of contact information, edge morphology information, and displacement information at the same arc length position in the response sequence to complete the temporal consistency check; performing a continuity analysis of the local cross-sectional state sequences at adjacent arc length positions to complete the spatial consistency check, and identifying continuous deviation regions and local abrupt change regions.

[0013] In one possible implementation, a process adjustment instruction is output based on the dimensional deviation detection result, including: outputting a rejection instruction when the dimensional deviation detection result corresponds to a local abrupt change region; and outputting an upstream process adjustment instruction when the dimensional deviation detection result corresponds to a continuous deviation region.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By establishing arc-length coordinates, a unified positional correspondence of data from different detection stages was achieved. Through a combination of two-stage compressive loads, contact imaging, dark-field imaging, and displacement calculation, joint characterization of static dimensions, compressive contact, and deformation recovery was realized. By employing temporal and spatial consistency checks, the distinction between continuous deviations and local abrupt changes was achieved, and process adjustment commands could be output simultaneously. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram illustrating the establishment of the center line in a specific embodiment of the present invention; Figure 3 This is a local width distribution diagram at the arc length position in a specific embodiment of the present invention; Figure 4 This is a diagram showing the width distribution of the dual-stage compression contact band in a specific embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the determination of abnormal regions in a specific embodiment of the present invention. Detailed Implementation

[0016] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0017] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0019] Machine vision is a technology that uses optical imaging devices to acquire target images and combines image processing, feature extraction, and rule determination to achieve target state recognition, size measurement, and position analysis. This technology features non-contact operation, fast response, good repeatability, and ease of online integration, and can convert the appearance information of the object under test into calculable and comparable data results. In this invention, machine vision is not only used to acquire the contour information of the seal, but also further used to acquire contact information, edge morphology information, and displacement information during the pressure process, thereby providing a unified data foundation for real-time detection of seal dimensional deviations and subsequent process adjustments.

[0020] like Figure 1 As shown, a machine vision-based real-time detection method for seal dimensional deviations includes: Obtain an image of the uncompressed state of the seal to be tested, extract the inner and outer contours, determine the centerline, and establish arc length coordinates; In one embodiment, after the seal under test enters the detection position, it is first kept in an uncompressed state and an image of the uncompressed state is acquired. Subsequently, the uncompressed state image is denoised, brightness corrected, and edge enhanced, and then the inner and outer contours are extracted. The inner and outer contours together characterize the boundary range of the seal in its natural state, and the centerline running through the entire seal is determined accordingly. Arc length is accumulated along the extension direction of the centerline to establish arc length coordinates, transforming the discrete boundary of the entire seal from a planar image into a continuous position sequence. All subsequent detection results obtained at different compression stages are correlated using the arc length coordinates as a unified position reference, thereby ensuring accurate comparison of the same position across different stages.

[0021] Determine the centerline and establish arc length coordinates, including: determining the centerline based on the midline between the inner and outer contours; and accumulating the arc length along the centerline to obtain the arc length coordinates.

[0022] In one embodiment, the centerline is established by centering the inner and outer contours. A new limitation is that the centerline is not directly fitted from either the inner or outer contour, but rather determined based on the midline between the inner and outer contours. This method ensures the reference path is always located in the middle of the seal's cross-sectional width, preventing unilateral traction on the position reference caused by local burrs, edge collapse, or slight deformation on one side of the boundary. Specifically, the uncompressed image is first binarized to separate the main seal area, and then the inner and outer contour point sets are extracted separately.

[0023] After extraction, the inner and outer contour points are sequentially paired according to the continuous extension direction of the contour, ensuring that each inner contour position corresponds to an outer contour position. Pairing can be based on adjacent sampling points in the image, combined with local shortest distance relationships or the same circumferential order relationship. After pairing, for each pair of corresponding inner and outer contour points, the midpoint between the two points is taken, and all midpoints are connected sequentially according to the contour direction to form an initial median line. Since edge extraction may result in local jagged edges, a few jump points, or individual gaps, the initial median line often exhibits short-distance fluctuations. Therefore, smoothing is necessary. Smoothing can be performed using methods such as averaging adjacent points, sliding window correction, or interpolation to eliminate the impact of isolated outliers on path continuity while maintaining the true direction of corner positions and curvature change positions.

[0024] After smoothing, a centerline is obtained for position calibration. The arc length coordinates are established starting from a preset starting point on the centerline. From this starting point, the distance between adjacent sampling points is calculated segment by segment along the centerline, and these distances are continuously accumulated to obtain the cumulative length value corresponding to each sampling position. Once the cumulative length value corresponds one-to-one with the position on the centerline, the arc length coordinates of the entire seal are formed. To facilitate subsequent positional analysis, the centerline can be resampled at fixed intervals, dividing the entire path into continuous arc length segments. In this way, each arc length segment has a clear length range and sequential position, and subsequent contact changes, edge morphology changes, and displacement changes can all be mapped to the corresponding arc length segment. Using this processing method, even if the angle of the seal under test in the image changes slightly, or the overall position shifts slightly, the data collected at different stages can still correspond around the same center path, thus ensuring that the positional reference remains consistent in subsequent analysis processes.

[0025] The test formula corresponding to the seal to be tested is called, and the first compressive load and the second compressive load are applied in sequence. Contact information is obtained by total internal reflection contact imaging, edge morphology information is obtained by dark field imaging, speckle images are acquired and displacement information is determined by digital image correlation to form a response sequence. In one embodiment, after establishing the arc length coordinates, the test formula corresponding to the seal under test is first read, and then a first compressive load and a second compressive load are applied to the seal under test sequentially according to the test formula. Contact information, edge morphology information, and displacement information are acquired in the uncompressed state, the two-stage compressive state, and the released state, respectively. These three types of information are then organized according to the arc length position and the acquisition sequence to form a response sequence. The response sequence is used to characterize the contact changes, edge changes, and displacement changes at the same arc length position under different stress stages, providing a continuous and unified data foundation for subsequent local cross-sectional state judgment.

[0026] The detection formula includes at least a first compressive load, a second compressive load, an image acquisition time sequence, and a judgment threshold; forming a response sequence includes: according to the uncompressed state, the first compressive load state, the second compressive load state, and the released state, performing time-series correspondence on contact information, edge morphology information, and displacement information to form a response sequence.

[0027] In one embodiment, the test formula is used to define the pressure application method, image acquisition method, and result judgment boundaries of the seal under test during the testing process. The added limitation is that the testing process does not directly employ fixed pressing and fixed imaging. Instead, the corresponding test formula is first invoked based on the specifications, material state, and cross-sectional dimensions of the seal under test, and then multi-stage acquisition is completed under the constraints of the test formula. This setup ensures that different types of seals maintain consistent processing logic within the same testing procedure, while matching specific parameters to their respective characteristics, avoiding result distortion due to insufficient or excessive pressure, or inconsistent acquisition timing.

[0028] In practice, the testing formula includes at least a first compression load, a second compression load, an image acquisition sequence, and a judgment threshold. The first compression load establishes the initial compression state, allowing the seal under test to begin stable contact with the bearing surface. The second compression load establishes a further compression state, making local contact changes, edge deformation, and displacement changes more fully apparent. The image acquisition sequence defines the order of image acquisition at each stage, ensuring a one-to-one correspondence between images in the uncompressed state, the state under the first compression load, the state under the second compression load, and the state after release. The judgment threshold defines the effective range of contact band width, edge morphology changes, and displacement changes during subsequent analysis, providing boundary conditions for anomaly identification. In actual testing, the initial image is first acquired in the uncompressed state to record the natural state of the seal under test. Then, the first compression load is applied, and contact images, dark field images, and speckle images are acquired after the load stabilizes. Subsequently, the second compression load is applied, and acquisition is repeated after the load stabilizes. After the acquisition under the second compression load is completed, the compression is released, and images are acquired again during the recovery process of the seal under test. Thus, image data in all four states are completely preserved.

[0029] Next, according to the image acquisition sequence, the contact information, edge morphology information, and displacement information corresponding to the same arc length position are arranged chronologically, and the information in each state is bound to the corresponding position in the arc length coordinates. In this way, the three types of information—uncompressed state, first compressive load state, second compressive load state, and post-release state—are organized into a continuously changing record over time. Repeating the same process for each arc length position throughout the entire circumference yields a complete response sequence. The response sequence not only preserves the changes between different states but also ensures that data from the same arc length position under different states can be directly compared, providing a continuous basis for subsequent judgment of whether there are abnormal changes at that position.

[0030] Contact information is obtained through total internal reflection contact imaging, including: lateral illumination of the transparent bearing surface carrying the seal under test; acquisition of reflection change images of the transparent bearing surface; and determination of the contact band width and contact band center position at each arc length position based on the reflection change images.

[0031] In one embodiment, contact information is acquired through total internal reflection contact imaging. A new limitation is that the contact information is not obtained by directly observing the external boundary, but rather extracted using the difference in reflection state before and after the seal under test contacts the transparent bearing surface. This setup directly reflects the location and extent of the actual contact area, ensuring a clear physical correspondence between the contact band width and the center position of the contact band, avoiding deviations caused by simply inferring the contact state based on changes in the outer contour.

[0032] In practice, the seal under test is first placed on a transparent bearing surface, and then lateral illumination is applied to the side of the transparent bearing surface, allowing light to propagate along its interior. When the seal under test is in an uncompressed state or only in weak contact, the reflection state of the transparent bearing surface remains relatively stable. When the first and second compressive loads are applied sequentially, a more pronounced contact is formed between the seal and the transparent bearing surface at a localized location, causing a change in the reflection path at the contact point. This change is then reflected as a brightness or reflection variation area in the acquired image. Subsequently, an image of the reflection variation on the transparent bearing surface is acquired. To ensure the effectiveness of the reflection variation image, a background image under non-contact conditions can be acquired first, followed by a working image under pressure. Background subtraction is used to remove interference from the inherent texture of the transparent bearing surface and ambient light. After obtaining the reflection variation image, each position in the image is mapped according to arc length coordinates to identify the contact area boundary near each arc length position.

[0033] The width of the contact area on the local cross-section corresponds to the contact band width. The contact band width characterizes the range of effective contact formed at that location under the current load condition. The center position of the contact area on the local cross-section corresponds to the center position of the contact band. The center position of the contact band characterizes whether the contact area has shifted relative to the centerline. To make the results more comparable, the same acquisition and analysis can be performed under both the first and second compressive load conditions to obtain the contact band width and center position under each level of compression. If a relatively narrow contact band has formed at the same arc length location under the first compressive load, but the contact band width still shows limited growth under the second compressive load, it indicates insufficient contact expansion at that location during compression. If the center position of the contact band deviates from the centerline under both levels of compression, it indicates a persistent contact shift at that location. Thus, the contact information obtained through total internal reflection contact imaging can directly reflect the contact state at each arc length location and provide a basis for determining the type of anomaly when combined with edge morphology and displacement information.

[0034] Edge morphology information is obtained through dark-field imaging, including: acquiring dark-field images under the first and second compressive load conditions; determining the edge slope distribution at each arc length position based on the dark-field images to obtain edge morphology information.

[0035] In one embodiment, edge morphology information is acquired through dark-field imaging. A new limitation is that the edge morphology information is not directly read from ordinary bright-field images. Instead, dark-field images are acquired under both first and second compressive load conditions, and the edge slope distribution is determined based on the brightness variations in the dark-field images. This setup makes edge warping, edge collapse, flanging, and local protrusions more easily visible in the images, thus compensating for the deficiency that contact information can only reflect the contact result and is difficult to distinguish the source of edge anomalies independently.

[0036] In practice, under the first compressive load condition, low-angle illumination is provided from the oblique direction of the edge of the seal under test, making the edge area more sensitive to light reflection and scattering. Edges in a smooth state typically show a gradual change in brightness in the dark-field image; where there are local uplifts or flared edges, the sudden change in local slope will create a more obvious area of ​​concentrated light and dark in the dark-field image. After acquiring the dark-field image under the first compressive load condition, the same acquisition process is repeated under the second compressive load condition. This allows for the simultaneous preservation of edge morphology changes under both mild and further compressive conditions. After obtaining the dark-field image, the edge local regions corresponding to each arc length position are divided using arc length coordinates, and the brightness change trend within each local region is analyzed. If the brightness change from the inside to the outside is continuous and gradual, it indicates a relatively smooth edge transition at that position; if the brightness increases or decreases rapidly over a short distance, it indicates a significant change in the edge slope at that position. This edge change based on local brightness changes is mapped to an edge slope distribution, and edge morphology information is established using arc length positions as indices. The distribution of edge slope does not require the establishment of a complex three-dimensional model; it only needs to reflect the relative degree of undulation and direction of change of the edge under pressure.

[0037] Furthermore, the edge slope distribution at the same arc length under the first and second compressive load conditions can be compared. If a significant abrupt change occurs at a certain location under mild compression and persists under further compression, then a stable edge anomaly can be considered present at that location. If a significant change only occurs at a certain location under the second compressive load condition, then edge instability is considered to have only been exposed under greater compression. The edge morphology information obtained in this way complements the contact information. When the contact zone width decreases but the edge slope distribution remains gentle, it usually leans towards insufficient contact; when the contact zone width decreases and abrupt changes occur simultaneously, it more easily indicates that local edge deformation participates in the contact change process. In this way, edge morphology information can provide more specific morphological basis for subsequent local cross-sectional state analysis.

[0038] Acquire speckle images and determine displacement information through digital image correlation, including: acquiring speckle images in the uncompressed state, the first compressive load state, the second compressive load state, and the state after release; performing digital image correlation calculations on the speckle images to determine the displacement change at each arc length position and the residual displacement after release, thereby obtaining displacement information.

[0039] In one embodiment, speckle images are acquired and displacement information is determined using digital image correlation. The added limitation is that the displacement information is not obtained by direct comparison of boundary positions. Instead, speckle images are acquired separately in the uncompressed state, the first compressive load state, the second compressive load state, and the released state. The displacement change at each arc length position and the residual displacement after release are then calculated using digital image correlation. This setup continuously reflects the displacement change path of the same location during compression and decompression, allowing the displacement information to characterize both the compression response amplitude and the recovery status.

[0040] In practice, a stable and identifiable speckle texture is formed on the surface of the seal under test. The speckle texture can be formed through a preset texture layer or by projection, as long as the local texture remains identifiable under different conditions. First, speckle images are acquired in the uncompressed state as an initial reference. Then, speckle images are acquired under the first and second compressive load conditions to record local displacement changes during the compression process. After the compression is released, speckle images are acquired in the post-release state to record local residual changes during the recovery phase. After acquiring images for all four states, the local image regions corresponding to each arc length position are divided according to arc length coordinates. Using the local image region in the uncompressed state as a reference region, digital image correlation calculations are performed between the corresponding regions in the first, second, and post-release states and the reference region. By comparing the relative offset of the speckle texture under different states, the displacement change at each arc length position during the compression process is determined. The displacement change corresponding to the first compressive load state reflects the initial compression response; the displacement change corresponding to the second compressive load state reflects the further compression response; the offset that the released state still retains relative to the uncompressed state is the residual displacement after release.

[0041] The displacement change is used to characterize the response of a location under compression conditions, while the residual displacement after release characterizes the recovery of that location after the compression is relieved. If the displacement change at a certain arc length location is small under both levels of compression, and the contact information shows that the contact band width is also small, it usually indicates insufficient compression response at that location. If the displacement change is large, but the residual displacement after release is still significant, it usually indicates insufficient recovery at that location. The displacement information is obtained by organizing the displacement changes and residual displacements after release at each arc length location according to time and location. This displacement information, along with the contact information and edge morphology information at the same arc length location, is written into the response sequence. This ensures that the response sequence includes not only contact and edge changes but also the complete displacement change process, providing continuous and comparable data for subsequent local cross-sectional condition assessment.

[0042] Determine the local cross-sectional state sequence corresponding to each arc length position based on the arc length coordinates and the response sequence; In one embodiment, after establishing the arc length coordinates and forming a response sequence, the arc length coordinates are used as a unified position reference to map contact information, edge morphology information, and displacement information from the uncompressed state, the first compressive load state, the second compressive load state, and the post-release state to the corresponding arc length positions. Subsequently, for each arc length position, local width, contact zone width, contact zone center offset, edge slope change, displacement change, and post-release residual displacement are extracted. Then, based on the combination of these results, the local cross-sectional state is determined and arranged sequentially along the arc length direction to form a local cross-sectional state sequence. In this way, multiple responses collected at different stages are transformed into continuous position state results, facilitating subsequent anomaly identification and consistency judgment.

[0043] The local cross-sectional state sequence corresponding to each arc length position is determined based on the arc length coordinates and the response sequence, including: matching the response sequence to the position based on the arc length coordinates; extracting the local width, contact zone width, contact zone center offset, edge slope change, displacement change, and residual displacement after release for each arc length position based on the position-matched response sequence; and determining the local cross-sectional state sequence based on the local width, contact zone width, contact zone center offset, edge slope change, displacement change, and residual displacement after release.

[0044] In one embodiment, the determination of the local cross-sectional state sequence is based on the processing order of "position correspondence first, feature extraction then, and result merging." The added limitation is that instead of directly making a global judgment on the entire response sequence, the position correspondence is first completed according to the arc length coordinates, and then a local state record is formed for each arc length position. This setting ensures that the response at the same position under different compression stages is always compared around the same reference, avoiding misalignment caused by overall translation, slight rotation, or changes in image boundaries.

[0045] In specific processing, the contact information, edge morphology information, and displacement information in the response sequence are first expanded according to the acquisition stage, and then the information of each stage is bound to the position number in the arc length coordinate. After binding, all data of the same arc length position in the four states are organized into a set of local responses. Subsequently, for each set of local responses, the local width, contact zone width, contact zone center offset, edge slope change, displacement change, and residual displacement after release are extracted. The local width is determined by the distance between the inner and outer contours at the corresponding arc length position in the uncompressed state, and is used to reflect the cross-sectional width of the position in its natural state. The contact zone width is determined by the contact area width under the first and second compressive load conditions, and is used to reflect the range of effective contact formed at this position under compression. The contact zone center offset is determined by the relative deviation between the contact zone center position and the centerline at the same arc length position, and is used to reflect whether the contact area has shifted to one side. The edge slope change is determined by comparing the difference in edge slope distribution at the same position under the first and second compressive load conditions, and is used to reflect the degree of change in edge morphology under continued compression.

[0046] The displacement change is determined by comparing the speckle image shift at the same location under uncompressed and two-stage compression conditions, reflecting the displacement response amplitude at that location during compression. The residual displacement after release is determined by comparing the speckle image shift at the same location under uncompressed and released conditions, reflecting the recovery after load removal. After all features are extracted, multiple features corresponding to the same arc length are merged into a single local state record and arranged sequentially along the arc length direction. In this way, the entire seal is transformed from discrete image information into a continuous sequence of local cross-sectional states, allowing subsequent anomaly classification and adjacent position comparisons to be directly based on this unified state record structure.

[0047] The local cross-sectional state sequence includes at least equivalent width deviation, contact deviation, anomaly type, and anomaly confidence level; the anomaly type is used to indicate at least one of the following anomalies: material shortage, edge turning, cross-sectional eccentricity, local collapse, or local hardening.

[0048] In one embodiment, the local cross-sectional state sequence is further defined to include at least equivalent width deviation, contact deviation, anomaly type, and anomaly confidence. The added limitation is that the local cross-sectional state sequence not only retains the basic measurement results but also provides the merged state judgment results. This setting can compress the multiple local features extracted in the previous stage into a state description that is easier to interpret and compare, and provide a direct basis for subsequent consistency checks.

[0049] In specific processing, the equivalent width deviation is first determined based on the local width. The equivalent width deviation does not correspond to a single instantaneous geometric dimension, but is determined based on the local width under uncompressed conditions, combined with the changes in the contact band width under both levels of compression. It characterizes the overall deviation degree in the width direction at that location. The contact deviation is determined jointly based on the contact band width and the contact band center offset. If the contact band width is too small, or the contact band center is significantly deviated from the centerline, the contact deviation will increase accordingly. The anomaly type is judged based on the combined characteristics of the equivalent width deviation, contact deviation, edge slope change, displacement change, and residual displacement after release. If the local width is too small, the contact band width remains small, and the displacement change is also small, then the location is closer to a material shortage. If the contact band center offset is large, and the edge slope change is concentrated on the same side, then the location is closer to a flange. If the local width change is not significant, but the contact band center is consistently biased to the same side, then the location is closer to cross-sectional eccentricity. If the local width suddenly decreases over a short distance, the contact zone width decreases synchronously, and the changes at adjacent locations are small, then this location is closer to a local collapse. If the local width is basically normal, but the displacement change is small and the residual displacement after release is large, then this location is closer to local hardening.

[0050] Anomaly confidence is used to characterize the reliability of anomaly type identification. It is determined by the consistency of features at the same arc length position across multiple compression stages and the continuity of changes between adjacent arc length positions. If the trend of change at the same position is stable across different stages and forms continuous features with adjacent positions, the anomaly confidence is high; conversely, if local features fluctuate significantly or the differences between positions are unstable, the anomaly confidence decreases accordingly. In this way, the local cross-sectional state sequence not only reflects the state results at each arc length position but also forms a continuous and comparable basis for anomaly distribution across the entire circumference.

[0051] Based on the response sequence and local cross-sectional state sequence, perform time consistency checks and spatial consistency checks, output dimensional deviation detection results, and output process adjustment instructions based on the dimensional deviation detection results.

[0052] In one embodiment, after forming the response sequence and the local cross-sectional state sequence, a temporal consistency check is first performed around the same arc length position, and then a spatial consistency check is performed around adjacent arc length positions. The temporal consistency check is used to confirm whether the change process at the same position is continuous in the unpressurized state, the two-stage pressurized state, and the post-release state. The spatial consistency check is used to confirm whether the abnormal distribution between adjacent positions is continuous. Based on the results of the two types of checks, the dimensional deviation detection results are determined, and the abnormal positions are divided into continuous deviation regions or local abrupt change regions. Subsequently, corresponding process adjustment instructions are output according to different region types, so that the detection results can be directly converted into subsequent handling actions.

[0053] Temporal and spatial consistency checks are performed, including: temporal comparison of contact information, edge morphology information, and displacement information at the same arc length position in the response sequence to complete the temporal consistency check; and continuity analysis of the local cross-sectional state sequence at adjacent arc length positions to complete the spatial consistency check, and to identify continuous deviation regions and local abrupt change regions.

[0054] In one embodiment, temporal consistency checks and spatial consistency checks are performed consecutively within the same processing stage. This step is further defined as first verifying the single-point change process, and then determining the overall distribution pattern. Using this processing sequence, it is possible to first confirm whether the change in the position of a single arc length is real, and then determine whether the change has spatial continuity, thereby avoiding misjudging sporadic disturbances as stable anomalies.

[0055] In specific processing, the contact information, edge morphology information, and displacement information of each arc length position in the response sequence are read one by one using arc length coordinates as an index. A time-series comparison is then performed according to the uncompressed state, the first compressive load state, the second compressive load state, and the post-release state. During the comparison, the focus is not on a single value at a particular moment, but rather on whether the direction of change at the same position is consistent, the magnitude of change is matched, and the recovery process is reasonable across the four states. If the contact band width increases under the first compressive load state, continues to increase under the second compressive load state, and then decreases under the post-release state, this change is generally consistent. If the contact band width suddenly decreases under the first compressive load state and fluctuates irregularly under the second compressive load state, the time-series stability of that position is poor. Edge morphology information and displacement information are compared in the same way. If the change in edge slope gradually increases under both levels of compression and weakens in the post-release stage, it indicates that the edge morphology change corresponds to the compression process. If the change in displacement is significant during the compression stage, but the residual displacement is small after release, it indicates that the recovery process is relatively normal.

[0056] After completing the time series comparison at the same arc length position, the equivalent width deviation, contact deviation, anomaly type, and anomaly confidence level already formed in the local cross-sectional state sequence are used to determine whether the state result at that position is consistent with the response change process. If the local cross-sectional state sequence shows a significant anomaly at that position, but the change process in the response sequence lacks continuity, the confidence level of the anomaly judgment at that position is reduced; if the local cross-sectional state sequence and the response sequence are consistent in the direction and degree of change, the anomaly judgment result at that position is retained. After completing the time consistency check for all arc length positions, a continuity analysis is performed on the local cross-sectional state sequences of adjacent arc length positions. The continuity analysis focuses on comparing whether the changes in equivalent width deviation, contact deviation, anomaly type, and anomaly confidence level at adjacent positions are stable. If multiple adjacent arc length positions continuously show an increase in contact deviation, similar equivalent width deviation, and consistent anomaly type in the same direction, then that segment of position is classified as a continuous deviation region; if a certain arc length position suddenly shows a significant anomaly compared to the positions before and after it, and the changes in the positions before and after it are small, then that position is classified as a local abrupt change region. By first performing a temporal consistency check and then a spatial consistency check, the dimensional deviation detection results can simultaneously possess temporal and spatial basis. The resulting dimensional deviation detection results not only indicate the presence of anomalies but also provide the distribution pattern of the anomalies across the entire circle, offering a direct classification basis for subsequent process adjustments.

[0057] Based on the dimensional deviation detection results, process adjustment instructions are output, including: outputting a rejection instruction when the dimensional deviation detection results correspond to a local abrupt change region; and outputting an upstream process adjustment instruction when the dimensional deviation detection results correspond to a continuous deviation region.

[0058] In one embodiment, when outputting process adjustment instructions based on dimensional deviation detection results, the process is further limited to handling abnormal areas separately, rather than applying the same treatment method to all abnormal locations. This approach allows for different processing paths for localized short defects and continuous deviations, avoiding over-rejection or under-treatment in practical applications.

[0059] In specific processing, the continuous deviation areas and local abrupt change areas identified in the dimensional deviation detection results are first read. Then, the final handling conclusion is formed by combining the anomaly type and anomaly confidence level corresponding to each type of area. If the dimensional deviation detection result corresponds to a local abrupt change area, a rejection instruction is output first. These local abrupt change areas are usually manifested as a sudden increase in contact deviation over a short distance, a sudden concentration of edge morphology changes, and a displacement change process that is significantly inconsistent with adjacent positions. They often correspond to local collapse, local flanging, local damage, or instantaneous forming anomalies. For these types of areas, even if the anomaly range is small, it can easily have a concentrated impact on the local sealing capability. Therefore, the current seal to be tested is marked as an object that will not enter the next process, and a rejection instruction is output. The rejection instruction can be used directly to perform sorting, or it can be used first for anomaly recording and then completed in subsequent cycles. If the dimensional deviation detection result corresponds to a continuous deviation area, an upstream process adjustment instruction is output. Continuous deviation regions typically manifest as consistent deviations occurring along multiple adjacent arc lengths. Examples include a continuously smaller contact band width, a contact band center consistently biased to the same side, and equivalent width deviations remaining close over a longer arc segment. These often correspond to continuous offsets in preceding molding, coating, bonding, or feeding processes. For such regions, simply removing the current seal under test only addresses the individual part's result and cannot eliminate the source of the deviation. Therefore, further upstream process adjustment instructions are issued.

[0060] Upstream process adjustment commands can correspond to corrections such as molding position, feeding position, pressing path, or verification of previous process parameters. If the degree of abnormality corresponding to the continuous deviation area is minor, upstream process adjustment takes precedence while retaining the inspection record of the current seal under test. If the degree of abnormality corresponding to the continuous deviation area exceeds the preset boundary, the current seal under test can be prevented from proceeding to the next process while simultaneously outputting the upstream process adjustment command. By mapping local abrupt change areas and continuous deviation areas to rejection commands and upstream process adjustment commands, respectively, dimensional deviation detection results can be directly converted into executable processing actions. This allows for timely interception of single-piece abnormalities and source correction of persistent deviations, forming a closed loop between the inspection process and subsequent process handling.

[0061] In one specific embodiment, a fluorosilicone rubber annular seal used for sealing an electrochemical device was selected as the test object. This seal has an outer diameter of 118 mm, an inner diameter of 98 mm, a nominal width of 10.00 mm, and a centerline circumference of approximately 360 mm. During testing, raw images were first acquired under unpressurized conditions. The image resolution was 2448 x 2048, with one sampling position corresponding to each millimeter of arc length, resulting in a total of 361 arc length positions. Subsequently, the inner and outer contours were extracted according to each arc length position, and the midline between them was used as the centerline. Arc length accumulation was then performed along the centerline. This method of obtaining arc length coordinates eliminates the influence of placement angle and ensures that data from different stages subsequently fall on the same positional reference. Figure 2 As shown, the outer contour, inner contour, and center line are simultaneously calibrated in the same image, with the starting point set at the right reference position, and the arc length direction continuously unfolded along the center line. Figure 2 The process of establishing the center line in the data serves as the basis for all subsequent data.

[0062] In this embodiment, the detection formula sets the first compressive load to 0.35 mm and the second compressive load to 0.65 mm; the lower limit of the contact band width under the second compressive load is set to 2.10 mm, the contact center offset warning value is 0.25 mm, and the residual displacement warning value after release is 0.08 mm. First, images of the uncompressed state are acquired. Then, after the first and second compressive loads stabilize, contact images, dark field images, and speckle images are acquired respectively. Finally, a recovery image is acquired 20 milliseconds after release. The data from the four stages are correlated according to the arc length position and acquisition sequence to form a response sequence. The average local width of the entire circle obtained from the uncompressed state image is 10.01 mm, which meets the nominal requirement, but it continuously declines in the range of 85 mm to 110 mm, with a minimum value of 9.51 mm, lower than the lower limit of 9.80 mm. Figure 3 As shown, the measured local width fluctuates around 10.00 mm along the arc length. The nominal width is represented by a dashed line, with the upper and lower limits represented by different line types to facilitate differentiation of the allowable width range. A distinct continuous concave area is formed between 85 mm and 110 mm, corresponding to the material shortage area A, where the lowest point is close to 9.51 mm, already below the lower limit. Except for the material shortage area A, the measured local widths at other arc length locations are generally within the allowable range, with only minor normal fluctuations. Figure 3 It intuitively reflects the local width distribution of the seal under test within the entire circumference, as well as the continuous distribution characteristics of the material shortage area A in the arc length direction.

[0063] To further determine whether the width change has transformed into an effective contact anomaly, this embodiment calculates the load contact increment rate at the same arc length position. The calculation expression is as follows: in, For local width, The contact band width under the first compressive load. This refers to the contact band width under the second compressive load. The load contact increment rate is given. In the normal area, the average contact band width under the first compressive load is 1.93 mm, and under the second compressive load, the average contact band width is 2.77 mm, with an average load contact increment rate of 0.084. In the material shortage area A, the average contact band width decreases to 1.19 mm under the first compressive load, and to only 1.58 mm under the second compressive load, with the average load contact increment rate dropping to 0.041. Taking the 90 mm position as an example, the local width is 9.54 mm, the contact band width under the first compressive load is 1.19 mm, and under the second compressive load, the contact band width is 1.56 mm, with a load contact increment rate of only 0.038, indicating that even with continued pressure, contact expansion at this position remains insufficient.

[0064] This embodiment further analyzes non-width-type anomalies by combining edge morphology and displacement changes. The average local width in the 212 mm to 228 mm range is 10.10 mm, which does not superficially exceed the lower limit of width. However, the average contact band width under the first compressive load is only 1.46 mm, and under the second compressive load, the average contact band width is only 1.89 mm, both lower than the normal area. At the same time, the average contact center offset in this range reaches 0.41 mm, exceeding the warning value of 0.25 mm, and the average edge slope change reaches 0.427, significantly higher than the level of about 0.1 in the normal area. The contact center offset at the 220 mm position reaches 0.418 mm, indicating that the contact band is shifted to one side. Combined with the unilateral bright and dark concentration changes in the dark field image, it can be determined that this area belongs to the edge-flanging-dominated anomaly, rather than a simple material shortage. The average local width in the 300 mm to 318 mm range is 9.90 mm, close to normal. However, under the second compressive load, the average contact band width is 2.23 mm, which, although higher than the lower limit, shows a decrease in displacement from 0.46 mm in the normal area to 0.32 mm. After release, the residual displacement increases from 0.03 mm in the normal area to 0.10 mm. Taking the 308 mm position as an example, the residual displacement reaches 0.101 mm after release, indicating insufficient post-compression recovery in this area, consistent with localized hardening characteristics.

[0065] like Figure 4 As shown, the orange curve represents the contact band width under the first compressive load, the blue curve represents the contact band width under the second compressive load, and the red dashed line represents the lower limit of contact. Abnormal zones A, B, and C all exhibit varying degrees of insufficient contact band compression, with zone A showing the most significant decrease. Zone B is lower than the normal zone under both levels of compression, and zone C is lower under the first compressive load but has not recovered to the normal level under the second compressive load. Figure 4 This indicates that relying solely on the uncompressed width cannot cover all anomalies; contact changes under dual-level compression must be taken into account in the judgment.

[0066] After completing time and space consistency checks, this embodiment identified 62 abnormal sampling points, accounting for 17.17% of all sampling points, and aggregated them into 3 abnormal regions. Specifically, the 85 mm to 110 mm range was identified as a continuous deviation region, corresponding to insufficient material region A; the 212 mm to 228 mm range was identified as a local abrupt change region, corresponding to flanged region B; and the 300 mm to 318 mm range was identified as a continuous deviation region, corresponding to hardened region C. If only the lower limit of the unpressurized width is used for judgment, only 26 abnormal sampling points can be identified, and the 212 mm to 228 mm and 300 mm to 318 mm ranges would be missed. Using the process of this invention, not only can an additional 36 abnormal sampling points that would otherwise be easily missed be identified, but the causes of the abnormalities can also be distinguished. Finally, a rejection command is output for flanged region B, an upstream forming position correction command is output for insufficient material region A, and an upstream vulcanization parameter verification command is output for hardened region C. In this way, the test results not only indicate whether there is an abnormality, but also directly indicate "where the abnormality is located, what category it belongs to, and how to handle it." For example... Figure 5 As shown, the entire sealing element is mapped as a ring distribution diagram. The red area on the right is the material shortage area A, the orange area in the lower left is the flange area B, the blue area in the upper left is the hardened area C, and the green dashed line corresponds to the center line. Figure 5 The distribution of the three types of anomalies on the entire circle is clearly displayed, making it easy to match the detection results with the actual process locations.

[0067] As can be seen from the above specific implementation process, this invention unifies the analysis of unpressurized width, changes in dual-level contact, edge morphology changes, and post-compression recovery under the same arc length coordinate system. This allows for the separate identification of three types of problems—insufficient material, flared edges, and excessive hardness—that cannot be distinguished by simple width detection. For the seal under test in this embodiment, the average width of the entire circumference is close to the nominal value, yet three abnormal areas of different natures were still accurately detected. This demonstrates that this method can detect both obvious geometric anomalies and hidden defects where the surface width is basically normal but the pressure response is abnormal, exhibiting high accuracy and feasibility.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0069] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for real-time detection of dimensional deviations in sealing components based on machine vision, characterized in that, The method includes: Obtain an image of the uncompressed state of the seal to be tested, extract the inner and outer contours, determine the centerline, and establish arc length coordinates; The test formula corresponding to the seal to be tested is called, and the first compressive load and the second compressive load are applied in sequence. Contact information is obtained by total internal reflection contact imaging, edge morphology information is obtained by dark field imaging, speckle images are acquired and displacement information is determined by digital image correlation to form a response sequence. The local cross-sectional state sequence corresponding to each arc length position is determined based on the arc length coordinates and the response sequence. Based on the response sequence and the local cross-sectional state sequence, perform time consistency checks and spatial consistency checks, output dimensional deviation detection results, and output process adjustment instructions based on the dimensional deviation detection results.

2. The method according to claim 1, characterized in that, The process of determining the centerline and establishing arc length coordinates includes: The center line is determined based on the midline between the inner contour and the outer contour; The arc length coordinates are obtained by accumulating the arc length along the center line.

3. The method according to claim 1, characterized in that, The detection formula includes at least a first compressive load, a second compressive load, an image acquisition timing sequence, and a judgment threshold; The formation of the response sequence includes: performing a time-series mapping of the contact information, the edge morphology information, and the displacement information according to the uncompressed state, the first compressive load state, the second compressive load state, and the released state, to form the response sequence.

4. The method according to claim 1, characterized in that, The method of acquiring contact information through total internal reflection contact imaging includes: Laterally illuminate the transparent bearing surface supporting the seal to be tested; Acquire images showing changes in reflection from the transparent bearing surface; The width of the contact strip and the center position of the contact strip at each arc length position are determined based on the reflection change image.

5. The method according to claim 1, characterized in that, The acquisition of edge topography information through dark-field imaging includes: Dark field images were acquired under the first compressive load condition and the second compressive load condition. The edge slope distribution at each arc length position is determined based on the dark field image to obtain the edge morphology information.

6. The method according to claim 3, characterized in that, The acquisition of speckle images and determination of displacement information through digital image correlation include: Speckle images were acquired in the uncompressed state, the first compressive load state, the second compressive load state, and the post-release state. Digital image correlation calculations are performed on the speckle image to determine the displacement change and residual displacement at each arc length position, thereby obtaining the displacement information.

7. The method according to claim 1, characterized in that, The step of determining the local cross-sectional state sequence corresponding to each arc length position based on the arc length coordinates and the response sequence includes: The response sequence is positionally mapped according to the arc length coordinates; Based on the response sequence after position correspondence, extract the local width, contact zone width, contact zone center offset, edge slope change, displacement change, and residual displacement after release for each arc length position; The local cross-sectional state sequence is determined based on the local width, the contact strip width, the contact strip center offset, the edge slope change, the displacement change, and the residual displacement after release.

8. The method according to claim 7, characterized in that, The local cross-sectional state sequence includes at least the equivalent width deviation, contact deviation, anomaly type, and anomaly confidence level; The anomaly type is used to indicate at least one of the following anomalies: material shortage, edge turning, cross-sectional eccentricity, local collapse, or local hardening.

9. The method according to claim 1, characterized in that, The aforementioned temporal consistency checks and spatial consistency checks include: The contact information, edge morphology information, and displacement information at the same arc length position in the response sequence are compared in time sequence to complete the time consistency check. A continuity analysis is performed on the local cross-sectional state sequence at adjacent arc length positions to complete the spatial consistency check and determine the continuous deviation region and the local abrupt change region.

10. The method according to claim 9, characterized in that, The step of outputting process adjustment instructions based on the dimensional deviation detection results includes: If the size deviation detection result corresponds to the local abrupt change region, a rejection instruction is output; If the dimensional deviation detection result corresponds to the continuous deviation region, an upstream process adjustment command is output.

Images