Burr data acquisition system and image processing method suitable for automobile stamping parts

Abstract

The application provides a burr data acquisition system and image processing method suitable for automobile stamping parts, and relates to the technical field of image processing, and comprises the following steps: acquiring preset angle images of a to-be-detected stamping part under multiple different preset visual angles; extracting target region images of a to-be-measured edge section from each preset angle image respectively to generate a target region image sequence; for any physical position point on the to-be-measured edge section, all target region images in the target region image sequence are traversed, and local image feature saliency of the physical position point in each target region image is calculated respectively; the target region image with the highest local image feature saliency is selected as an analysis source image for burr identification of the physical position point; based on the analysis source image allocated for each physical position point on the to-be-measured edge section, a burr region on the to-be-measured edge section is identified, geometric dimension quantification is performed, and a burr data result is output. The method can improve detection efficiency and precision.

Description

Technical Field

[0001] This application relates to the field of image processing technology, specifically to a method for acquiring and processing burr data for automotive stamping parts. Background Technology

[0002] In the automotive manufacturing industry, stamped parts are key components forming the body-in-white, and the quality of their edge burrs directly affects the precision of subsequent welding and final assembly processes, as well as the safety and aesthetics of the entire vehicle. Therefore, high-precision burr inspection of stamped parts is a crucial step in the production process.

[0003] Traditional testing methods mainly rely on manual visual inspection and contact measurement. These methods are not only inefficient and labor-intensive, but also susceptible to the subjective experience of the inspectors, which can easily lead to missed detections and misjudgments. The results are inconsistent and cannot meet the stringent quality control requirements of modern intelligent manufacturing.

[0004] To improve the level of automation in inspection, some machine vision-based inspection devices have been proposed in the existing technology. For example, utility model patent CN222125112U discloses a burr inspection device for stamped parts, which includes a worktable, a rotating mechanism, a fixing mechanism, and a CCD inspection camera. By driving the rotating disk with a servo motor to rotate the workpiece, the CCD camera can capture images of the workpiece from different angles, thus initially replacing manual inspection and improving work efficiency.

[0005] However, such devices only solve the problem of automating image acquisition, and their detection algorithms are usually relatively simple, often relying solely on threshold segmentation or edge extraction based on images from a single viewpoint. Due to the complex structure and varied edge directions of stamped parts, as well as the small size and irregular shape of burrs, the detection algorithm is easily rendered ineffective under a single viewpoint due to factors such as unsatisfactory lighting conditions, insufficient contrast between burrs and the background, or obstruction by the workpiece's own structure, resulting in missed detections or substandard measurement accuracy. Summary of the Invention

[0006] In view of the above-mentioned defects or deficiencies in the prior art, this application aims to provide a method for burr data acquisition and image processing applicable to automotive stamping parts, so as to improve the detection accuracy of burr data of stamping parts; the method includes the following steps:

[0007] Acquire images of the stamping part to be inspected at preset angles from multiple preset viewpoints;

[0008] The target region image of the edge segment to be tested of the stamping part to be tested is extracted from each of the preset angle images, and a target region image sequence is generated; wherein, the edge segment to be tested is formed by connecting a number of physical location points;

[0009] For any physical location point on the edge segment to be tested, traverse all target region images in the target region image sequence and calculate the local image feature saliency of the physical location point in each target region image;

[0010] The target region image with the highest local image feature saliency is selected as the source image for spur identification of the physical location point;

[0011] Based on the analysis source image assigned to each physical location point on the edge segment to be tested, the burr region on the edge segment to be tested is identified;

[0012] The geometric dimensions of the burr region are quantized, and the burr data results are output.

[0013] According to the technical solution provided in this application, after identifying the burr region on the edge segment to be tested based on the analysis source image assigned to each physical location point on the edge segment to be tested, the method further includes the following steps:

[0014] Based on the size and location information of all identified burr areas, a burr distribution map corresponding to the stamping part to be inspected is generated;

[0015] The historical defect pattern library is retrieved and traversed to obtain the defect cause corresponding to the burr distribution pattern of the stamping part to be inspected; wherein, the historical defect pattern library includes a variety of burr distribution patterns and their corresponding defect causes, and the defect causes include mold wear type or mold failure type.

[0016] Based on the defect cause corresponding to the burr distribution pattern of the stamping part to be inspected, a corresponding mold maintenance prompt is generated and sent to the production management terminal.

[0017] According to the technical solution provided in this application, the method also includes constructing a historical defect pattern library;

[0018] The construction of the historical defect pattern library includes the following steps:

[0019] The data on burrs in the stamped part is correlated with the mold identity data of the mold used in the production of the stamped part; the burr data includes the burr distribution map.

[0020] The mold identification data associated with the mold used is linked to its maintenance record data;

[0021] Based on the temporal correlation between the burr data and the maintenance record data, the correspondence between the burr distribution map and the mold wear type or the mold failure type is obtained, so as to obtain the historical defect pattern library.

[0022] According to the technical solution provided in this application, the step of obtaining the correspondence between the burr distribution map and the mold wear type or the mold failure type based on the time-series correlation between the burr data and the maintenance record data includes the following steps:

[0023] Based on the maintenance time in the maintenance record data, determine the set of stamped parts produced before and after the maintenance;

[0024] Comparative analysis of the burr distribution patterns of each stamped part in the stamped parts set produced before and after maintenance;

[0025] Determine whether the frequency or severity of any burr distribution pattern corresponding to the burr distribution pattern in the stamped parts produced after repair has decreased in the set.

[0026] If so, then establish the correspondence between the burr distribution pattern corresponding to the burr distribution map and the mold wear type or mold fault type resolved in this maintenance operation.

[0027] According to the technical solution provided in this application, determining whether the frequency or severity of any burr distribution pattern corresponding to the burr distribution pattern in the stamped parts set produced after repair has decreased includes the following steps:

[0028] Calculate the first occurrence frequency and first average severity of the first burr distribution pattern in the stamped parts set before repair;

[0029] Calculate the second occurrence frequency and second average severity of the first burr distribution pattern in the set of stamped parts after repair;

[0030] If the rate of decrease of the second occurrence frequency relative to the first occurrence frequency is greater than the first preset threshold, and / or the rate of decrease of the second average severity relative to the first average severity is greater than the second preset threshold, then it is determined that the occurrence frequency or severity of any of the burr distribution patterns corresponding to the burr distribution patterns has decreased.

[0031] According to the technical solution provided in this application, the step of acquiring images of the stamping part to be inspected at preset angles under multiple preset viewpoints includes the following steps:

[0032] Obtain the stamping type of the stamping part to be tested, and use it as the target stamping type;

[0033] The acquisition angle database is retrieved and traversed to obtain a set of preset viewing angles corresponding to the target stamping part type. The set of preset viewing angles includes multiple preset viewing angles. The acquisition angle database includes multiple stamping part types and the set of preset viewing angles corresponding to each stamping part type.

[0034] The servo motor drives the rotating disk and the stamping part to be inspected fixed on it to rotate to each preset viewing angle. After pausing at each angle, the CCD inspection camera is triggered to acquire an image, which is used as the preset angle image under the preset viewing angle.

[0035] According to the technical solution provided in this application, the method further includes constructing the acquisition angle database;

[0036] The construction of the collected angle database includes the following steps:

[0037] Select multiple test samples corresponding to the stamping part type to ensure that the test samples cover the common burr conditions of the stamping part type;

[0038] A test sample is fixed and controlled to rotate at a constant speed around an axis, while the CCD detection camera continuously captures images to obtain a continuous video stream of one revolution of the test sample.

[0039] The continuous video stream is processed by frame segmentation to obtain multiple high-definition images corresponding to uniformly distributed angles, which constitute the full-view image set of the test sample.

[0040] Obtain the pixel-level segmentation mask of the spiky region in each of the high-definition images in the full-view image set;

[0041] Based on the segmentation mask, the average contrast of the spiky regions in each of the high-definition images is calculated;

[0042] The high-definition images with an average contrast higher than a third preset threshold are selected, and their corresponding rotation angles are used as valid candidate viewpoints for the test samples.

[0043] Based on the effective candidate viewpoints of each test sample corresponding to the stamping part type, the preset viewpoint set for the stamping part type is obtained;

[0044] Repeat the above steps to obtain the preset viewing angle set corresponding to each of the stamping part types, so as to obtain the acquisition angle database.

[0045] According to the technical solution provided in this application, obtaining the preset view set for the stamping part type based on the effective candidate view for each test sample corresponding to the stamping part type includes the following steps:

[0046] Summarize the effective candidate viewpoints corresponding to all test samples under the aforementioned stamping part type to form an initial set containing multiple effective candidate viewpoints;

[0047] The frequency of occurrence of each of the effective candidate viewpoints in the initial set is counted;

[0048] The valid candidate viewpoints that appear more frequently than a preset frequency threshold are selected as the preset viewpoints corresponding to the stamping part type, so as to form the preset viewpoint set for the stamping part type.

[0049] According to the technical solution provided in this application, the step of calculating the local image feature saliency of the physical location point in each target region image includes the following steps:

[0050] Extract multi-dimensional image features from the image of each target region corresponding to the region where the physical location point is located;

[0051] The multi-dimensional image features are fused and calculated to generate a quantitative score value that represents the visual salience of the physical location point in each of the target region images;

[0052] The saliency of local image features in each target region image is obtained based on the quantitative score of the visual saliency of the physical location point in each target region image.

[0053] According to the technical solution provided in this application, the step of extracting multi-dimensional image features corresponding to the region where the physical location point is located from the images of each target region includes the following steps:

[0054] For each target region image, a local image patch of a predetermined size is extracted with the physical location point as the center;

[0055] Calculate the gradient magnitude map and gradient direction histogram as the gradient features of the local image patch;

[0056] Local binary pattern features are extracted using the LBP operator and used as texture features for the local image patch;

[0057] The gradient features and the texture features are used as the multi-dimensional image features.

[0058] Compared with the prior art, the beneficial effects of this application are as follows:

[0059] I. Improved Detection Rate and Reliability of Burr Detection: A multi-view image sequence and intelligent source image selection strategy are employed. For each physical location point on the edge segment to be detected, the system calculates the saliency of its local image features in images from different viewpoints and matches it with the optimal viewpoint (source image) that best clearly presents the features of that point. This method effectively overcomes the problem of missed detection caused by occlusion, uneven lighting, and insufficient contrast. Even if a burr is difficult to observe from most angles, as long as there is an angle from which its features are visible, the system can capture it, thus achieving a high detection rate and improving the reliability and consistency of the detection results.

[0060] II. Achieved High-Precision Quantification of Burr Geometric Dimensions: Under a single viewpoint, due to perspective distortion, blurred edge pixels, and other factors, the measurement of burr height and length often suffers from significant errors. This application analyzes the image with the highest saliency of local image features for each measurement point, meaning that pixel-level positioning and measurement are performed under conditions of the clearest edges and strongest contrast. This is equivalent to automatically selecting the optimal angle for each minute measurement task, thereby greatly improving the accuracy of geometric dimension quantification and providing an extremely accurate data foundation for process improvement and quality assessment.

[0061] Third, the adaptive capability and intelligence level of the detection system are improved: Compared with the simple rotating imaging devices in the background technology, the technical solution of this application embodies a higher level of intelligence. The system no longer passively acquires images, but actively evaluates image quality and makes decisions. It can adaptively and dynamically select the most suitable data source for analysis based on the specific characteristics of different locations. This "point-to-point" optimization strategy enables the system to cope with various complex structural stamping parts and changing on-site conditions, and its generalization ability and robustness are far superior to traditional visual inspection systems with fixed algorithms.

[0062] IV. High Efficiency While Improving Detection Quality: Although this application processes multiple images, it focuses on key areas through target region image extraction and uses intelligent algorithms to quickly calculate saliency instead of processing the entire image, effectively controlling the computational cost of the entire analysis process. Compared to the soaring subsequent repair costs due to missed detections or the need for repeated detections due to insufficient accuracy, the increased computational cost of this solution is negligible. It successfully achieves the optimal balance between "detection accuracy" and "efficiency," truly realizing high-quality, high-efficiency automated intelligent detection. Attached Figure Description

[0063] Figure 1 A flowchart illustrating the steps of the method for acquiring and processing burr data for automotive stamping parts provided in this application. Detailed Implementation

[0064] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0065] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0066] Example 1

[0067] As mentioned in the background section, in view of the problems in the prior art, this application proposes a method for burr data acquisition and image processing applicable to automotive stamping parts, such as... Figure 1 As shown, it includes the following steps:

[0068] S1. Acquire images of the stamping part to be inspected at preset angles from multiple preset viewpoints;

[0069] S2. Extract the target region image of the edge segment to be tested of the stamping part to be tested from each of the preset angle images, and generate a target region image sequence; wherein, the edge segment to be tested is formed by connecting a number of physical location points;

[0070] S3. For any physical location point on the edge segment to be tested, traverse all target region images in the target region image sequence and calculate the local image feature saliency of the physical location point in each target region image.

[0071] S4. Select the target region image with the highest local image feature saliency as the source image for spur identification of the physical location point;

[0072] S5. Based on the analysis source image assigned to each physical location point on the edge segment to be tested, identify the burr region on the edge segment to be tested;

[0073] S6. Quantize the geometric dimensions of the burr area and output the burr data results.

[0074] Specifically, the stamped parts to be inspected refer to automotive metal parts formed by stamping with molds, such as car doors, hoods, and brackets. Stamped parts coming off the production line or samples for inspection are placed and fixed on the inspection fixture. The preset viewing angle refers to a pre-set specific camera angle used to photograph the stamped parts. These angles are not arbitrarily chosen but determined after extensive experimentation and analysis to best highlight the edge features (especially burrs) of a certain type of stamped part. The implementation involves controlling a programmable rotating platform (such as a servo motor-driven turntable) to precisely rotate the workpiece to these preset angles. Preset angle images are high-definition digital images captured by an industrial camera (such as a CCD camera) at each preset viewing angle. During implementation, a stable and uniform lighting environment must be ensured to avoid shadows and reflections interfering with image quality. The edge segment to be inspected refers to the contour edge portion of the stamped part that needs to be inspected for burrs. It may be a continuous edge or a segmented edge. Physically, this edge segment is composed of countless continuous physical location points connected together. In digital image processing, these physical location points correspond to the set of pixel coordinates in the image. In practice, the edge to be tested can be located from the original image using image segmentation and edge extraction algorithms (such as the Canny operator). Target region image: To reduce computation and focus on key areas, the entire image at the preset angle is not processed. Instead, a strip-shaped image of appropriate width is cropped centered on the extracted edge segment to be tested. This series of images, cropped from images at different angles and containing the same physical edge, constitutes the target region image sequence. Physical location points: These refer to specific, tiny points on the edge segment to be tested. In practice, the extracted edge lines can be densely sampled to obtain a series of discrete coordinate points representing the edge. Local image feature saliency: This is a core quantitative indicator used to evaluate whether the features of a region (i.e., a local image patch) near a certain physical location point in an image at a specific angle are clear, obvious, and easily recognized by the algorithm. Higher saliency means that the edges, textures, gradients, and other features at that point are more prominent in the image, making it easier to determine whether there are burrs at that point. Source Image Analysis: For each point on the edge, its saliency is calculated across all angles of the target region image. The image with the highest saliency score is selected as the best evidence image for determining whether a burr exists at that point. Burr Region Recognition: After assigning the best source image for each point, burr recognition algorithms (e.g., edge anomaly detection, template matching, machine learning classifiers, etc.) are executed on these images to accurately locate the burr region. Geometric Size Quantization: For the identified burr pixel region, the actual physical dimensions of the burr, such as length, height, and area, are calculated through camera calibration and pixel equivalent conversion.Burr data results: The final output structured data usually includes information such as the location, size and quantity of burrs, which can be used to generate inspection reports or perform statistical analysis.

[0075] In a preferred embodiment, after identifying the burr region on the edge segment under test based on the analysis source image assigned to each physical location point on the edge segment under test, the method further includes the following step:

[0076] Based on the size and location information of all identified burr areas, a burr distribution map corresponding to the stamping part to be inspected is generated;

[0077] The historical defect pattern library is retrieved and traversed to obtain the defect cause corresponding to the burr distribution pattern of the stamping part to be inspected; wherein, the historical defect pattern library includes a variety of burr distribution patterns and their corresponding defect causes, and the defect causes include mold wear type or mold failure type.

[0078] Based on the defect cause corresponding to the burr distribution pattern of the stamping part to be inspected, a corresponding mold maintenance prompt is generated and sent to the production management terminal.

[0079] Specifically, the burr distribution map is a structured data file or visualization. It integrates the size (e.g., large, medium, small) and location information of all identified burrs (e.g., located at the Cth millimeter of segment B on the edge of workpiece A). In implementation, the location and severity level of burrs can be marked on a schematic diagram of a stamped part using markers of different colors and sizes, thus forming an intuitive "lesion map," i.e., the burr distribution map. The historical defect pattern library is a database storing a large amount of historical case data. Each record contains a typical burr distribution map and a verified defect cause leading to this map. In implementation, this library can be a relational database or knowledge graph, supporting efficient map matching and querying. Defect causes refer to the types of problems existing in the mold, mainly including: Mold wear types: such as "upper die edge radius wear," "lower die clearance uniform wear," etc., which are usually gradual. Mold failure types: such as "punch chipping," "guide post jamming," "spring failure," etc., which are usually sudden. Mold maintenance prompt: Automatically generated text information based on matching results, such as: "Alert: The current workpiece burr distribution pattern (code: M-Pattern-07) matches the 'severe wear at the lower die corner' pattern in the historical database with a 92% match rate. It is recommended to immediately inspect and repair the lower die corner of mold D-052." This information is sent to production management terminals, such as MES (Manufacturing Execution System) screens, engineers' mobile terminals, or Kanban systems, via the factory network (e.g., MQTT, HTTP protocol).

[0080] This implementation method enables predictive maintenance: it elevates the assessment of inspection results from "post-production judgment of workpiece quality" to "pre-production warning of mold health status," allowing intervention before the mold is completely damaged or a large number of scrap products are produced, reducing unplanned downtime. It also supports decision support: providing maintenance personnel with clear fault clues, narrowing the scope of troubleshooting, and improving maintenance efficiency and accuracy. This approach digitizes and standardizes experienced technicians' expertise ("you can tell where the mold is damaged just by looking at burrs") and embeds it into the system, preventing knowledge loss.

[0081] In a preferred embodiment, the method further includes constructing a historical defect pattern library;

[0082] The construction of the historical defect pattern library includes the following steps:

[0083] The data on burrs in the stamped part is correlated with the mold identity data of the mold used in the production of the stamped part; the burr data includes the burr distribution map.

[0084] The mold identification data associated with the mold used is linked to its maintenance record data;

[0085] Based on the temporal correlation between the burr data and the maintenance record data, the correspondence between the burr distribution map and the mold wear type or the mold failure type is obtained, so as to obtain the historical defect pattern library.

[0086] Specifically, the system builds a historical defect pattern database: This is an offline, continuous learning process, typically led by a system administrator or data engineer. Mold identification data: Information uniquely identifying a mold, such as mold number, mold system number, production line, and type of part produced. During implementation, each mold should have an RFID tag or QR code, automatically scanned and collected during production and inspection. Maintenance record data: Electronic records of each mold maintenance and repair. This includes: maintenance time, maintenance personnel, maintenance operation performed (e.g., "grinding the upper die cutting edge," "replacing the guide sleeve"), replaced parts, and description of the discovered fault (e.g., "cutting edge chipped 3mm"). This data is usually entered by maintenance personnel via their terminals or integrated with a CMMS (Computerized Maintenance Management System). Time-series association: This is crucial for building the knowledge base. During implementation, the system needs to establish a unified timeline: when a stamped part is produced, its timestamp and the mold ID used are recorded. When this stamped part is inspected, its burr data (including burr distribution patterns) is linked to the mold ID and timestamp that produced it. The maintenance records of the mold also have their timestamps. The system can use the mold ID and timestamp to query "how the burr condition of all workpieces produced by this mold changed before and after a certain maintenance".

[0087] Furthermore, the step of obtaining the correspondence between the burr distribution map and the mold wear type or the mold failure type based on the time-series correlation between the burr data and the maintenance record data includes the following steps:

[0088] Based on the maintenance time in the maintenance record data, determine the set of stamped parts produced before and after the maintenance;

[0089] Comparative analysis of the burr distribution patterns of each stamped part in the stamped parts set produced before and after maintenance;

[0090] Determine whether the frequency or severity of any burr distribution pattern corresponding to the burr distribution pattern in the stamped parts produced after repair has decreased in the set.

[0091] If so, then establish the correspondence between the burr distribution pattern corresponding to the burr distribution map and the mold wear type or mold fault type resolved in this maintenance operation.

[0092] Specifically, the system determines the set of stamped parts: Based on the time point T_maintenance of a maintenance event in the maintenance record, the system automatically performs the following queries: Before maintenance set: All stamped parts produced by the mold within the time period [T_maintenance - ΔT, T_maintenance). After maintenance set: All stamped parts produced by the mold within the time period [T_maintenance, T_maintenance + ΔT]. ΔT is a preset time window that can be adjusted according to the production cycle. The system compares and analyzes the burr distribution patterns: It performs cluster analysis or pattern matching on the burr distribution patterns of all workpieces in both the before and after maintenance sets, and counts the occurrence of each recurring burr distribution pattern (i.e., the first burr distribution pattern). The system then determines whether the pattern has improved: For each burr distribution pattern that appeared before maintenance, it checks whether its frequency or severity has significantly decreased after maintenance. Establish a correspondence: If a certain pattern (e.g., "continuous long burrs in the middle of the edge") almost disappears after maintenance, and the maintenance record shows that the operation of this maintenance was "replacing the broken pressure plate spring", then the system can establish a correspondence: the defect cause corresponding to the burr distribution pattern (continuous long burrs in the middle of the edge) (mold failure type: broken pressure plate spring).

[0093] This implementation method automates the construction of a knowledge base: it avoids manually and subjectively inputting fault mode correspondences, instead using a data-driven approach to automatically mine and verify these correspondences from historical data, ensuring the objectivity and accuracy of the knowledge base. Furthermore, this knowledge base is continuously evolving: as production data accumulates, the system can continuously discover new fault modes and enrich the pattern library, making it increasingly complete and intelligent.

[0094] Furthermore, determining whether the frequency or severity of any burr distribution pattern corresponding to a burr distribution pattern decreases in the set of stamped parts produced after repair includes the following steps:

[0095] Calculate the first occurrence frequency and first average severity of the first burr distribution pattern in the stamped parts set before repair;

[0096] Calculate the second occurrence frequency and second average severity of the first burr distribution pattern in the set of stamped parts after repair;

[0097] If the rate of decrease of the second occurrence frequency relative to the first occurrence frequency is greater than the first preset threshold, and / or the rate of decrease of the second average severity relative to the first average severity is greater than the second preset threshold, then it is determined that the occurrence frequency or severity of any of the burr distribution patterns corresponding to the burr distribution patterns has decreased.

[0098] Specifically, the pre-repair indicators are calculated as follows: For a certain burr distribution pattern to be examined (the first burr distribution pattern), the following are calculated in the pre-repair stamping parts set: First frequency of occurrence = (number of parts exhibiting this pattern) / (total number of parts in the pre-repair set); First average severity: The severity of this pattern on each part needs to be quantified first (for example, the total area of ​​burrs under this pattern, maximum height, etc. can be used). Then, the average severity of all parts exhibiting this pattern is calculated. The post-repair indicators are calculated as follows: Similarly, the second frequency of occurrence and the second average severity of the same pattern are calculated in the post-repair set. Thresholds are set and judged: A first preset threshold (e.g., 30%) and a second preset threshold (e.g., 25%) are set. The frequency reduction rate is calculated as: (first frequency of occurrence - second frequency of occurrence) / first frequency of occurrence; the severity reduction rate is calculated as: (first average severity - second average severity) / first average severity. If the frequency reduction rate > 30%, or the severity reduction rate > 25%, or both rates exceed their respective thresholds, the system automatically determines that the burr distribution pattern has been effectively improved after repair.

[0099] In a preferred embodiment, acquiring images of the stamping part to be inspected at preset angles from multiple preset viewpoints includes the following steps:

[0100] Obtain the stamping type of the stamping part to be tested, and use it as the target stamping type;

[0101] The acquisition angle database is retrieved and traversed to obtain a set of preset viewing angles corresponding to the target stamping part type. The set of preset viewing angles includes multiple preset viewing angles. The acquisition angle database includes multiple stamping part types and the set of preset viewing angles corresponding to each stamping part type.

[0102] The servo motor drives the rotating disk and the stamping part to be inspected fixed on it to rotate to each preset viewing angle. After pausing at each angle, the CCD inspection camera is triggered to acquire an image, which is used as the preset angle image under the preset viewing angle.

[0103] Specifically, the stamping type of the stamped part to be inspected refers to the category or model of the stamped part, such as "left front door inner panel," "engine hood outer panel," and "B-pillar reinforcement panel." Different types of workpieces have drastically different geometries, edge orientations, and locations prone to burrs. At the inspection station, the target stamping part type can be automatically obtained by scanning the QR code / RFID tag attached to the workpiece or by using a visual recognition algorithm to perform preliminary classification. Acquisition Angle Database: This is a core database that stores the mapping relationship between various stamping part types and their corresponding preset viewing angle sets. Implementation: This database can be a simple table structure, such as an SQL database table containing two columns: stamping part type and optimal angle set, such as [0, 45, 90, 135, 180, 225, 270, 315]. Preset Viewing Angle Set: For a specific type of stamping part, a set of camera observation angles (usually in degrees) that, after optimization calculations, best capture all potential burr features. For example, for the "left front door inner panel," its preset viewing angle set might be [0°, 15°, 30°, 90°, 180°, 270°]. This set ensures that burrs at its complex flanges and bends are clearly visible at at least one angle. Servo motor and turntable: forming a high-precision rotary positioning system. Implementation: The servo motor receives angle commands from the host computer and drives the turntable connected to its axis to rotate. The stamped part to be inspected is securely mounted at the center of the turntable using a customized chemical fixture, ensuring that its rotation axis coincides with the turntable axis to avoid large offsets during imaging. The CCD inspection camera is a high-resolution, high-sensitivity industrial area array camera, fixedly mounted directly above one side of the turntable, its field of view covering the workpiece on the turntable. The lighting system (such as ring LED lights, coaxial lights, etc.) must ensure uniform illumination and no significant reflection at the workpiece edges at any angle.

[0104] This implementation avoids the massive amount of redundant images and data processing pressure caused by 360° indiscriminate all-angle acquisition, capturing only the most valuable angles, greatly improving inspection efficiency. The optimal viewing angle is "tailor-made" for different workpieces, ensuring that even on the most complex geometries, critical edges are covered by at least one high-quality angle, laying a solid foundation for subsequent high-precision inspection. The entire process requires no manual intervention to adjust the angles; the system automatically adapts to different workpiece types, making it ideal for modern flexible manufacturing scenarios involving mixed-line production and small-batch, multi-variety production.

[0105] Furthermore, the method also includes constructing the acquisition angle database;

[0106] The construction of the collected angle database includes the following steps:

[0107] Select multiple test samples corresponding to the stamping part type to ensure that the test samples cover the common burr conditions of the stamping part type;

[0108] A test sample is fixed and controlled to rotate at a constant speed around an axis, while the CCD detection camera continuously captures images to obtain a continuous video stream of one revolution of the test sample.

[0109] The continuous video stream is processed by frame segmentation to obtain multiple high-definition images corresponding to uniformly distributed angles, which constitute the full-view image set of the test sample.

[0110] Obtain the pixel-level segmentation mask of the spiky region in each of the high-definition images in the full-view image set;

[0111] Based on the segmentation mask, the average contrast of the spiky regions in each of the high-definition images is calculated;

[0112] The high-definition images with an average contrast higher than a third preset threshold are selected, and their corresponding rotation angles are used as valid candidate viewpoints for the test samples.

[0113] Based on the effective candidate viewpoints of each test sample corresponding to the stamping part type, the preset viewpoint set for the stamping part type is obtained;

[0114] Repeat the above steps to obtain the preset viewing angle set corresponding to each of the stamping part types, so as to obtain the acquisition angle database.

[0115] Specifically, the test sample refers to the physical sample of the stamped part used for the perspective optimization experiment. A carefully selected batch of samples is needed to represent all common burr states of this type of workpiece. For example, it should include burr-free good products, products with small, medium, and severe burrs, and samples with burrs appearing in different locations. The larger the sample size and the more comprehensive the coverage, the more reliable the final angle set. Continuous video stream and frame processing: A test sample is fixed on a turntable, which is set to rotate continuously at a constant speed (e.g., 1° / second). Simultaneously, a CCD camera is set to continuously capture video at a high frame rate (e.g., 30fps) to obtain a video of the workpiece rotating one revolution. Then, a video frame segmentation algorithm (e.g., extracting frames at fixed rotation angle intervals) is used to decompose the video into a series of high-definition images, for example, extracting one frame every 2° rotation, ultimately obtaining 180 images, constituting the full-view image set of the sample. Pixel-level segmentation mask: A binary image where white pixels represent pixels belonging to the burr area, and black pixels represent the background. Obtaining the mask is a prerequisite for calculating contrast. This can be achieved in several ways: Manual annotation: Professionals manually delineate the spur areas in each image using image annotation tools. This offers the highest accuracy but is time-consuming. Semi-automatic algorithms: Initial segmentation is performed using traditional image segmentation algorithms (such as thresholding and edge detection) or pre-trained machine learning models (such as U-Net), followed by manual verification and correction. This is a practical solution that balances efficiency and accuracy. Average contrast: This metric quantifies the difference in brightness or color between the spur area and the surrounding background area at a specific angle. The greater the difference, the higher the contrast, and the more clearly the spurs are visible. The calculation formula is usually: Average contrast = (Average gray value of the spur area - Average gray value of the background area) / (Average gray value of the spur area + Average gray value of the background area). Using the obtained segmentation mask, the spur and background areas can be easily separated, and their respective average gray values ​​can be calculated. Effective candidate viewpoints: For a single test sample, all rotation angles corresponding to images whose average contrast is higher than a third preset threshold (e.g., 0.4) are considered "good angles" for detecting spurs in that sample. Preset viewpoint set: After statistical screening of all valid candidate viewpoints of this type, a set of the most robust viewpoints is finally determined for the inspection of all workpieces of this type.

[0116] In a preferred embodiment, obtaining the preset view set for the stamping part type based on the effective candidate viewpoints of each test sample corresponding to the stamping part type includes the following steps:

[0117] Summarize the effective candidate viewpoints corresponding to all test samples under the aforementioned stamping part type to form an initial set containing multiple effective candidate viewpoints;

[0118] The frequency of occurrence of each of the effective candidate viewpoints in the initial set is counted;

[0119] The valid candidate viewpoints that appear more frequently than a preset frequency threshold are selected as the preset viewpoints corresponding to the stamping part type, so as to form the preset viewpoint set for the stamping part type.

[0120] Specifically, the initial set: When constructing a preset viewpoint set for a certain stamping part type (e.g., "Type_A"), it is first necessary to summarize the effective candidate viewpoints of all test samples under that type. For example, suppose 10 test samples are selected for "Type_A", and each sample yields 10-20 effective candidate viewpoints. All the effective candidate viewpoints from these 10 samples are placed into an empty set to obtain the initial set. It may contain a large number of repeated angle values ​​(e.g., 0°, 5°, 10°, 0°, 355°, 0°, 90°...). Frequency of occurrence: This refers to the proportion of times a specific candidate viewpoint (e.g., 0°) appears in the entire initial set. All angles in the initial set are normalized (e.g., mapped to the 0-360° range). Statistical counting is performed. For example, if the initial set contains 500 viewpoint data points, and 0° (or the range 359.5°-0.5°) appears 150 times, then the frequency of occurrence of the candidate viewpoint 0° = 150 / 500 = 30%. A preset frequency threshold is used to filter out perspectives that occur frequently enough to be considered generally valid. This threshold needs to be set based on the sample size and requirements. For example, a preset frequency threshold of 15% can be set. This means that an angle must be considered valid in at least 15% of the test sample data to be ultimately selected. Filtering and composition: Iterate through each unique candidate perspective in the initial set. If its calculated frequency of occurrence is higher than 15%, then that perspective is selected as a preset perspective for that stamping part type. The set of all selected preset perspectives is the final preset perspective set.

[0121] The following describes the implementation steps: Data aggregation: The system reads the list of valid candidate viewing angles for all test samples under model "Type_A" in the database and merges them into a large list. The 360-degree range is divided into several smaller intervals, for example, one interval per degree. The large list is traversed, and each angle value is classified into the corresponding smaller interval. The number of angles in each interval is counted. Frequency calculation: For each angle interval, such as 0°-1°, its frequency is calculated: Frequency = (number of angles in the interval) / (total length of the large list). Threshold filtering: The frequency of each interval is compared with a preset frequency threshold (such as 15%). The center angles represented by all intervals with a frequency greater than 15% are retained (such as 0.5°, 90.5°, 180.5°, 270.5°). Set output: The retained angle values ​​are rounded to the nearest integer degree and sorted after deduplication to finally obtain the preset viewing angle set for "Type_A", for example, {0, 90, 180, 270}.

[0122] In a preferred embodiment, calculating the local image feature saliency of the physical location point in each of the target region images includes the following steps:

[0123] Extract multi-dimensional image features from the image of each target region corresponding to the region where the physical location point is located;

[0124] The multi-dimensional image features are fused and calculated to generate a quantitative score value that represents the visual salience of the physical location point in each of the target region images;

[0125] The saliency of local image features in each target region image is obtained based on the quantitative score of the visual saliency of the physical location point in each target region image.

[0126] Specifically, multi-dimensional image features refer to numerical vectors extracted from an image that describe different attributes (such as shape and texture) of local regions. Common dimensions include: gradient features, which describe edge strength and direction; and texture features, which describe the roughness and regularity of a region. A higher quantization score indicates that the visual features of the local region where the current physical location is located are clearer and more prominent in the specific target area image, making it more conducive to burr recognition.

[0127] The following describes the implementation steps: Input: A physical location point P_i on the edge segment to be tested, and an image Img_j from the target region image sequence. Local image patch extraction: On image Img_j, using the pixel coordinates of point P_i as the center, extract a patch of a fixed size (e.g., 32 pixels). A square image patch Patch_ij (32 pixels) is extracted. Multi-dimensional feature extraction: Image features of each dimension are calculated for this small image patch Patch_ij. The feature vectors of each dimension are input into a pre-defined fusion model (e.g., a weighted summation formula), and the model outputs a final quantized score value, Score_ij. This value represents the local image feature saliency of point P_i in image Img_j. The steps are repeated for each image Img_j in the sequence to obtain a set of scores [Score_i1, Score_i2, ..., Score_iN].

[0128] Further, the step of extracting multi-dimensional image features corresponding to the region where the physical location point is located from each of the target region images includes the following steps:

[0129] For each target region image, a local image patch of a predetermined size is extracted with the physical location point as the center;

[0130] Calculate the gradient magnitude map and gradient direction histogram as the gradient features of the local image patch;

[0131] Local binary pattern features are extracted using the LBP operator and used as texture features for the local image patch;

[0132] The gradient features and the texture features are used as the multi-dimensional image features.

[0133] Specifically, a local image patch is a small region of the image captured centered on a physical location point, serving as the basic unit for feature calculation. Implementation: The size needs to be carefully balanced; too small a patch may contain insufficient information, while too large a patch may introduce excessive irrelevant background interference. Typically, 16 is chosen. 16, 32 32 or 64 64 pixels. Gradient magnitude map: describes the edge intensity at each pixel in the image. Implementation: Convolution is performed in the X and Y directions using the Sobel operator to obtain gradient components Gx and Gy. The gradient magnitude of each pixel is calculated using the formula: Magnitude = sqrt(Gx) 2 + Gy 2The calculated amplitude map is the gradient feature. Strikes typically appear as prominent edges, thus generating high gradient amplitudes at their locations. Gradient Orientation Histogram (HOG): This describes the statistical characteristics of the distribution of edge directions within an image patch. Implementation: Based on the calculation of Gx and Gy, the direction of each pixel is calculated: Theta = arctan2(Gy, Gx). Then, the 0-360° direction range is divided into several bins, and the number of gradient directions of all pixels within the image patch falling into each bin is counted, forming a histogram. This variant of HOG is also a gradient feature and is useful for describing the morphology of strikes.

[0134] Specifically, the LBP operator (Local Binary Pattern) is a powerful texture descriptor. For each pixel in an image patch, it is associated with its 3x3 matrix. The eight pixels within a 3-neighborhood are compared. If a neighboring pixel value is greater than the center pixel value, that position is marked as 1; otherwise, it is marked as 0. This results in an 8-bit binary number (usually read in clockwise order), which is converted to decimal; this is the LBP code for the center pixel. Finally, a histogram of the LBP codes for all pixels within the entire image patch is calculated; this histogram represents the texture feature of the image patch. The surface texture of burr areas often differs from that of smooth substrate materials, and LBP features effectively capture this difference.

[0135] This implementation offers strong feature complementarity: gradient features are sensitive to edges and shapes, effectively capturing the "protrusion" morphology of burrs; texture features are sensitive to changes in surface material, effectively distinguishing the rough surface of burrs from the smooth surface of the workpiece. The combination of these two features almost covers the two most crucial visual attributes of burrs. High computational efficiency: Sobel and LBP are both classic operators that are computationally lightweight and extremely efficient, making them ideal for large-scale use in real-time scenarios such as industrial inspection, ensuring high-speed system operation.

[0136] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for burr data acquisition and image processing applicable to automotive stamping parts, characterized in that, Includes the following steps: Acquire images of the stamping part to be inspected at preset angles from multiple preset viewpoints; The target region image of the edge segment to be tested of the stamping part to be tested is extracted from each of the preset angle images, and a target region image sequence is generated; wherein, the edge segment to be tested is formed by connecting a number of physical location points; For any physical location point on the edge segment to be tested, traverse all target region images in the target region image sequence and calculate the local image feature saliency of the physical location point in each target region image; The target region image with the highest local image feature saliency is selected as the source image for spur identification of the physical location point; Based on the analysis source image assigned to each physical location point on the edge segment to be tested, the burr region on the edge segment to be tested is identified; The geometric dimensions of the burr region are quantized, and the burr data results are output. After identifying the burr region on the edge segment under test based on the analysis source image assigned to each physical location point on the edge segment under test, the method further includes the following steps: Based on the size and location information of all identified burr areas, a burr distribution map corresponding to the stamping part to be inspected is generated; The historical defect pattern library is retrieved and traversed to obtain the defect cause corresponding to the burr distribution pattern of the stamping part to be inspected; wherein, the historical defect pattern library includes a variety of burr distribution patterns and their corresponding defect causes, and the defect causes include mold wear type or mold failure type. Based on the defect cause corresponding to the burr distribution map of the stamping part to be inspected, a corresponding mold maintenance prompt is generated and the mold maintenance prompt is sent to the production management terminal. This method also includes building a library of historical defect patterns; The construction of the historical defect pattern library includes the following steps: The data on burrs in the stamped part is correlated with the mold identity data of the mold used in the production of the stamped part; the burr data includes the burr distribution map. The mold identification data associated with the mold used is linked to its maintenance record data; Based on the temporal correlation between the burr data and the maintenance record data, the correspondence between the burr distribution map and the mold wear type or the mold failure type is obtained, so as to obtain the historical defect pattern library.

2. The method for burr data acquisition and image processing applicable to automotive stamping parts according to claim 1, characterized in that: The method for obtaining the correspondence between the burr distribution map and the mold wear type or the mold failure type based on the time-series correlation between the burr data and the maintenance record data includes the following steps: Based on the maintenance time in the maintenance record data, determine the set of stamped parts produced before and after the maintenance; Comparative analysis of the burr distribution patterns of each stamped part in the stamped parts set produced before and after maintenance; Determine whether the frequency or severity of any burr distribution pattern corresponding to the burr distribution pattern in the stamped parts produced after repair has decreased in the set. If so, then establish the correspondence between the burr distribution map corresponding to the burr distribution and the mold wear type or mold fault type resolved in this maintenance operation.

3. The method for burr data acquisition and image processing applicable to automotive stamping parts according to claim 2, characterized in that: The determination of whether the frequency or severity of any burr distribution pattern corresponding to the burr distribution pattern in the stamped parts produced after repair has decreased includes the following steps: Calculate the first occurrence frequency and first average severity of the first burr distribution pattern in the stamped parts set before repair; Calculate the second occurrence frequency and second average severity of the first burr distribution pattern in the set of stamped parts after repair; If the rate of decrease of the second occurrence frequency relative to the first occurrence frequency is greater than the first preset threshold, and / or the rate of decrease of the second average severity relative to the first average severity is greater than the second preset threshold, then it is determined that the occurrence frequency or severity of any of the burr distribution patterns corresponding to the burr distribution patterns has decreased.

4. The method for burr data acquisition and image processing applicable to automotive stamping parts according to claim 1, characterized in that: The process of acquiring images of the stamping part to be inspected at preset angles from multiple preset viewpoints includes the following steps: Obtain the stamping type of the stamping part to be tested, and use it as the target stamping type; The acquisition angle database is retrieved and traversed to obtain a set of preset viewing angles corresponding to the target stamping part type. The set of preset viewing angles includes multiple preset viewing angles. The acquisition angle database includes multiple stamping part types and the set of preset viewing angles corresponding to each stamping part type. The servo motor drives the rotating disk and the stamping part to be inspected fixed on it to rotate to each preset viewing angle. After pausing at each angle, the CCD inspection camera is triggered to acquire an image, which is used as the preset angle image under the preset viewing angle.

5. The method for burr data acquisition and image processing applicable to automotive stamping parts according to claim 4, characterized in that: The method also includes constructing the acquisition angle database; The construction of the acquisition angle database includes the following steps: Select multiple test samples corresponding to the stamping part type to ensure that the test samples cover the common burr conditions of the stamping part type; A test sample is fixed and controlled to rotate at a constant speed around an axis, while the CCD detection camera continuously captures images to obtain a continuous video stream of one revolution of the test sample. The continuous video stream is processed by frame segmentation to obtain multiple high-definition images corresponding to uniformly distributed angles, which constitute the full-view image set of the test sample; Obtain the pixel-level segmentation mask of the spiky region in each of the high-definition images in the full-view image set; Based on the segmentation mask, the average contrast of the spiky regions in each of the high-definition images is calculated; The high-definition images with an average contrast higher than a third preset threshold are selected, and their corresponding rotation angles are used as the effective candidate viewpoints of the test samples. Based on the effective candidate viewpoints of each test sample corresponding to the stamping part type, the preset viewpoint set for the stamping part type is obtained; Repeat the above steps to obtain the preset viewing angle set corresponding to each of the stamping part types, so as to obtain the acquisition angle database.

6. The method for burr data acquisition and image processing applicable to automotive stamping parts according to claim 5, characterized in that: The process of obtaining the preset view set for each test sample corresponding to the stamping part type based on the effective candidate viewpoints includes the following steps: Summarize the effective candidate viewpoints corresponding to all test samples under the aforementioned stamping part type to form an initial set containing multiple effective candidate viewpoints; The frequency of occurrence of each of the effective candidate viewpoints in the initial set is counted; The valid candidate viewpoints that appear more frequently than a preset frequency threshold are selected as the preset viewpoints corresponding to the stamping part type, so as to form the preset viewpoint set for the stamping part type.

7. The method for burr data acquisition and image processing applicable to automotive stamping parts according to claim 1, characterized in that: The step of calculating the saliency of local image features of the physical location point in each of the target region images includes the following steps: Extract multi-dimensional image features from the image of each target region corresponding to the region where the physical location point is located; The multi-dimensional image features are fused and calculated to generate a quantitative score value that represents the visual salience of the physical location point in each of the target region images; The saliency of local image features in each target region image is obtained based on the quantitative score of the visual saliency of the physical location point in each target region image.

8. The method for burr data acquisition and image processing of automotive stamping parts according to claim 7, characterized in that: The step of extracting multi-dimensional image features from each target region image corresponding to the region where the physical location point is located includes the following steps: For each target region image, a local image patch of a predetermined size is extracted with the physical location point as the center; Calculate the gradient magnitude map and gradient direction histogram as the gradient features of the local image patch; Local binary pattern features are extracted using the LBP operator and used as texture features for the local image patch; The gradient features and the texture features are used as the multi-dimensional image features.

Images