Wiring terminal full appearance detection system and method based on annular track platform

Abstract

The invention discloses a wiring terminal full appearance detection system and method based on an annular track platform, and the system comprises a track which is provided with a transparent material conveying chain plate segment and is used for the uniform motion transmission of a wiring terminal; the multi-station visual detection device comprises a plurality of visual camera groups and an object distance adjusting assembly, the plurality of visual camera groups are respectively arranged at different stations of the track and are used for carrying out image acquisition on six front-view main surfaces and a set surface of the wiring terminal, and the object distance adjusting assembly is connected with the visual camera groups and is used for carrying out image acquisition on the set surface of the wiring terminal. The object distance adjusting assembly is used for adjusting the object distance or / and the position of the linear scanning camera; the AI visual inspection module comprises an image marking unit used for carrying out defect classification and marking on the collected images; the model training unit is used for carrying out iterative training on the marked image based on a small sample learning algorithm to generate a multi-station detection model; and the real-time detection unit is used for deploying a multi-station detection model to execute component statistics, defect detection and defective product rejection.

Description

Technical Field

[0001] The present invention relates to the field of visual inspection, and in particular to a system and method for inspecting the full appearance of connection terminals based on an annular track platform. Background Art

[0002] Currently, conventional visual inspection machines face numerous technical bottlenecks for long PIN terminal products (terminals with more than 11 pins and a length greater than 55mm). Due to the unique dimensions of long PIN terminals, conventional vibrating plates struggle to achieve stable discharge, and traditional glass turntables are incompatible with the diverse shapes and sizes of long PIN terminal products. Furthermore, area scan cameras suffer from insufficient accuracy during dynamic inspection, making it difficult to detect even the smallest defects. This has led to a long reliance on manual visual inspection for long wiring harness terminal products.

[0003] In the terminal block industry, visual defect inspection primarily relies on manual labor. This method not only requires significant labor costs but also leads to visual fatigue due to long hours of work. This can lead to undetected defective products entering the market, severely impacting product quality and equipment reliability. Traditional visual inspection technology has significant limitations, only detecting specific defects within a fixed range, making it difficult to fully replace manual inspection. Existing inspection systems lack universality for terminal blocks of varying lengths, colors, and types, resulting in a high rate of defective products.

[0004] Existing solutions for multi-faceted inspection face significant shortcomings. Flipping inspection reduces efficiency, while transparent transmission systems often rely on glass tray sorting. This tray-like structure struggles to adapt to the diverse inspection needs of unusually shaped products, particularly those with long pins. During glass tray inspection, the area array camera's limited installation position leads to severe image distortion, resulting in incomplete detection of defects in key areas like clip openings and clip indentations. Furthermore, it's difficult to ensure accurate dimensional measurement. These issues severely restrict the automation and accuracy of terminal block visual inspection.

[0005] In addition, for products with different sizes, such as width or height differences greater than 10mm and products with different structures, due to the focusing problem of industrial cameras, existing solutions cannot obtain cleaning inspection images for such products with different sizes, resulting in poor equipment compatibility and can only detect products within a certain width and height range. However, there are many types and specifications of terminal block products, resulting in poor compatibility of inspection lines. Summary of the Invention

[0006] The present invention addresses the shortcomings of existing visual inspection systems in the prior art, such as poor compatibility with long PIN terminal products, severe image distortion, and inability to achieve multi-angle detection without blind spots, resulting in reliance on manual labor and low efficiency. A full-appearance inspection system and method for terminal blocks based on a circular track platform are provided.

[0007] In order to solve the above technical problems, the present invention is solved by the following technical solutions: a terminal block full appearance inspection system based on a ring track platform, comprising: The track is equipped with a transparent material transmission chain plate segment, which is used for uniform motion transmission of the terminal blocks; A multi-station visual inspection device, comprising a plurality of visual camera groups and an object distance adjustment assembly, wherein the plurality of visual camera groups are respectively arranged at different stations of the track and are used to capture images of the six main front-view surfaces and the setting surface of the terminal block; the object distance adjustment assembly is connected to the visual camera groups and is used to adjust the object distance and / or position of the visual camera groups according to the terminal block to be inspected; The material guide station adjustment device includes a material guide rail and a material guide control motor for driving the material guide rail to move. The material guide rail is controlled by the material guide control motor to drive the connection terminals of different specifications to be arranged relative to the multi-station monitoring device; AI visual inspection module, including: Image annotation unit, used to classify and annotate defects in the collected images; The model training unit iteratively trains the labeled images based on a small sample learning algorithm to generate a multi-station detection model; A real-time inspection unit deploys the multi-station inspection model to perform component counting, defect detection, and rejection of defective products.

[0008] By adopting the above technical solution and the collaborative design of track transport, multi-station vision equipment, and AI modules, we achieve comprehensive inspection of terminal blocks. The track's uniform speed ensures motion stability, while the multi-camera array covers six main surfaces and inclined surfaces, combined with dynamic height adjustment to accommodate products of varying sizes. The AI ​​module utilizes small sample sizes for rapid training of high-precision models, significantly improving inspection efficiency and accuracy while reducing manual intervention.

[0009] The present invention is further configured such that: the visual camera group includes a plurality of linear scanning camera groups.

[0010] The plurality of linear scan camera groups include at least six main surface detection camera groups and at least two tilt detection camera groups, which are: Six main surface inspection camera groups are provided and are respectively arranged opposite to the six front views of the terminal block, for inspecting the six surfaces of the terminal block; The first tilt detection camera group is used to detect planar defects on the side of the clamp opening and the step side of the terminal block; The second tilt inspection camera group is used to detect scratches, missing materials and burrs on the surface of the terminal blocks.

[0011] By employing this technical solution, the six main camera groups and two oblique camera groups ensure comprehensive coverage of front views and side details such as clamp openings and step surfaces. This classification-based inspection strategy reduces the data processing load on individual cameras, improves system response speed, and avoids missed inspections due to blind spots.

[0012] The present invention is further configured such that the depth of field range of the special lens is determined by the following parameters: The calculation formula for the depth of field near boundary distance L1 is: , Where f is the focal length of the lens, N is the aperture factor, d is the diameter of the permissible circle of confusion, and δ is the object distance.

[0013] By adopting the above technical solution and depth of field calculation formula, the depth of field range of the camera can be precisely controlled so that planar defects with a drop of 3mm to 10mm can be clearly imaged, avoiding image blur caused by object distance error and enhancing detection reliability.

[0014] The present invention is further configured such that: the visual camera group further includes a special lens group, and the special lens group is configured to be compatible with clear imaging of planar defects with a height difference of 3mm to 10mm.

[0015] By adopting the above technical solution, the special lens group supports the imaging of planar defects with different height differences, expanding the system's ability to capture tiny defects such as burrs and missing materials. At the same time, it is compatible with various sizes of terminal blocks, reducing the frequency of hardware replacement.

[0016] The present invention is further configured such that the object distance adjustment component sets the line scan rate VC of the linear scan camera by the following formula: ， Where Hc is the number of pixels per line of the line scan camera, Vo is the target operating rate, and Lo is the field of view width; The field of view width Lo is dynamically adjusted by the ratio of the focal length f to the camera sensor size to accommodate detection of different sizes.

[0017] By adopting the above technical solution, the scanning rate is accurately matched with the target motion rate, the impact of motion blur on image quality is eliminated, and no hardware modification is required when adapting to products of different sizes, thereby improving system flexibility.

[0018] The present invention is further configured as follows: the track adopts a circular track, and the circular track of the track adopts a plane circular track or an upper and lower circular track. When the track adopts a plane circular track, the connecting terminal is arranged to move in a circular manner at a uniform speed along the track; when the track adopts an upper and lower circular track, the connecting terminal is located on one side of the track and moves in a linear manner at a uniform speed.

[0019] By adopting the above technical solutions, the circular track can be designed horizontally or vertically to meet the diverse production line layout requirements. The horizontal track supports continuous circular transportation, which facilitates the cyclic detection of incorrectly marked terminal blocks. The vertical track saves space and maintains detection stability. Both can achieve efficient and continuous operation and reduce downtime.

[0020] The present invention is further configured as follows: the model training unit cooperates with an incremental learning algorithm, the model training unit is configured with a general defect database, and the multi-station detection model is pre-trained with a database containing general defect types before training, and the database contains several sample images of common types of defects.

[0021] By adopting the above technical solutions, pre-training and incremental learning algorithms based on a general defect database enable the detection model to quickly adapt to new defect types, reduce dependence on small sample data, and maintain high accuracy in the long term through continuous iterative optimization.

[0022] The present invention is further configured as follows: the conveying chain plate section adopts a transparent material portion, and the transparent material adopts high-transmittance optical glass. The transparent material portion of the conveying chain plate section is configured with a plurality of magnifying lens groups, and the plurality of magnifying lens groups are arranged at even intervals and the intervals are set the same as the gaps between the plurality of clip openings on the wiring terminal. The conveying chain plate section is configured with a blocking device. When the wiring terminal is placed on the conveying chain plate section made of transparent material, the blocking device drives the plurality of clip openings on the wiring terminal to be aligned with the plurality of magnifying lens groups. When the plurality of clip openings on the wiring terminal are aligned with the plurality of magnifying lens groups, the wiring terminal is released.

[0023] By adopting the above technical solution, a magnifying lens group is embedded in the transparent conveyor chain plate segment, and its spacing is aligned with the clamp opening to magnify the details of key areas; high-transmittance optical glass reduces light interference, ensures image acquisition clarity, and improves defect recognition accuracy.

[0024] An AI visual inspection method for terminal block appearance includes the following steps: Step S1: The track transports the terminal to be inspected at a constant speed, wherein the track is equipped with a transparent material conveying chain plate segment; Step S2: driving the wiring terminals of different specifications to the specified positions of the conveying chain plate section through the material guide rail and the material guide control motor, and the conveying chain plate section drives the wiring terminals to be in a stable operating state; Step S3: Automatically adjust the height direction of the visual camera group through the object distance adjustment component, so that the lens of the visual camera group automatically focuses on the object distance to be compatible with the detection of terminal blocks of different sizes and types; Step S4: six main surface inspection camera groups, two tilt inspection camera groups, and a special lens group are used to perform a 360° scan of the terminal block to be inspected without blind spots. The special lens is configured to clearly image planar defects with a drop height of 3 mm to 10 mm. Step S5: Classify and annotate the collected defect images, including multiple missing parts, functional differences, and appearance defect types, and confirm the qualified terminal blocks; Step S6: Based on the small sample annotated data and the transfer learning algorithm, the defect image is iteratively trained for a preset number of times to generate a multi-station detection model, wherein the initial model is pre-trained on a database containing common defects; Step S7: Deploy the trained inspection model to the inspection system to perform component statistics, defect detection, and rejection of defective products, and verify the inspection function through actual trial operation; Step S8: During the trial operation phase, the test results are manually reviewed, missed defects are marked, and the test model is iteratively updated. Terminal blocks that are judged by the system to be defective but are actually qualified are corrected for misjudgment, and problematic terminal blocks that were missed by the system are returned to the S5 marking training phase until the detection accuracy exceeds the preset standard.

[0025] By adopting the above technical solutions, a full-process automated inspection method from conveying, scanning to model iteration is implemented, combined with manual review and closed-loop optimization during the trial operation phase, to gradually improve the system accuracy and ensure that the preset quality standards are met before stable production. The object distance adjustment component dynamically adapts to terminal blocks of different sizes, expands the system's applicable scenarios, avoids hardware adjustments caused by changes in product specifications, and enhances the versatility and cost-effectiveness of the inspection system.

[0026] Due to the adoption of the above technical solution, the present invention has significant technical effects: through the synergy of track transmission, multi-station visual inspection device and AI visual inspection module, automatic detection of multi-faceted defects of terminal blocks is realized, which has the advantages of improving detection efficiency and accuracy, realizing multi-faceted detection without blind spots, and adapting to products of different sizes. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the detection system in the embodiment; Figure 2 Schematic diagram of the upper and lower annular track structures in the embodiment; Figure 3 Schematic diagram of the position of the main surface detection camera group in the visual camera group in the embodiment; Figure 4 2. It is a schematic diagram of the planar annular track structure in the embodiment; Figure 5 Schematic diagram of the position of the main surface detection camera group in the visual camera group of the planar circular track in the embodiment.

[0028] The names of the parts indicated by the numerical labels in the above drawings are as follows: 1. Track; 11. Conveyor chain plate segment; 2. Visual camera group; 3. Object distance adjustment component; 201. Top-view visual camera; 202. Top-view visual camera; 203. Left-view visual camera; 204. Right-view visual camera; 205. Forward-view visual camera; 206. Rear-view visual camera; 4. Material guide rail; 5. Material guide control motor. DETAILED DESCRIPTION

[0029] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0030] Example: Existing technologies limit the compatibility and accuracy of traditional inspection equipment for long PIN terminals. Manual visual inspection is often used, but this is inefficient and prone to fatigue, leading to missed detections. Traditional visual inspection technology can only detect defects within a fixed range and lacks multi-angle coverage. The glass turntable structure struggles to adapt to terminal products of varying shapes and sizes, and image distortion during dynamic inspection directly impacts the accuracy of identifying subtle defects such as clip openings and indentations.

[0031] To solve the above problems, it is necessary to build an inspection system that can adapt to products of multiple specifications and achieve blind-angle scanning and intelligent defect recognition. The key lies in designing an adjustable imaging device to eliminate detection blind spots, combining a dynamic parameter adjustment mechanism to be compatible with inspection requirements of different sizes, and establishing an iterative intelligent model to improve the generalization ability of defect recognition.

[0032] This application proposes a system solution including a track 1, a multi-station visual inspection device and an AI visual inspection module. The track 1 uses a transparent material conveyor chain segment 11 to achieve uniform speed transportation. The multi-station visual inspection device has a visual camera group 2 and an object distance adjustment component 3. The visual camera group 2 is equipped with multiple groups of cameras to complete scanning of six main surfaces and specific inclined surfaces. The AI ​​module generates a dynamic detection model through labeling training.

[0033] Track 1 adopts a conveying structure with a continuous transmission function, specifically a ring conveyor belt structure. The transparent material segment can realize transmission imaging of the bottom features of the product. The multi-station visual inspection device includes multiple imaging units, such as a combination of a linear scanning camera and an oblique angle camera, and changes the shooting angle and object distance through a mechanical adjustment mechanism. The object distance adjustment component 3 in this embodiment adopts a servo motor. The object distance adjustment component 3 drives the visual camera group 2 to move through its servo motor, thereby adjusting the relative object distance and / or relative position of the visual camera group 2 relative to the tested terminal. The AI ​​visual inspection module has a built-in image processing algorithm, such as a classification model based on a convolutional neural network, which adapts to different defect characteristics through a transfer learning mechanism.

[0034] The system includes a multi-station servo automatic focus object distance device, including inspection stations for station 1, station 2, and station 3, to achieve automatic focus of the lens on the object distance to achieve compatibility with products of different heights, widths, and structures, and obtain clear inspection images to meet the inspection needs of products of different specifications.

[0035] The terminal blocks to be inspected are moved at a constant speed via transparent conveyor chains. Multiple cameras simultaneously collect image data of each surface at different workstations. The object distance adjustment component 3 automatically adjusts the camera focal length according to the product size to ensure image clarity. The collected images are annotated and input into the training model. Through small sample learning, the system quickly generates inspection rules applicable to the current product. During the real-time inspection stage, the system simultaneously performs dimensional measurement, surface defect recognition and classification, and triggers the rejection mechanism to separate unqualified products.

[0036] Traditional inspection equipment cannot cover multi-angle features due to its fixed workstation layout. However, this solution achieves full surface coverage through collaborative scanning of multiple cameras. Existing technology relies on manual flipping operations. This solution uses a transparent transmission structure to synchronously obtain bottom features. Conventional vision systems need to be remodeled for different products. This solution reduces repeated debugging time through adaptive parameter adjustment and model migration.

[0037] The present application realizes the automated full inspection of long PIN terminal products, eliminates the visual blind spots of manual inspection, and the multi-angle imaging system effectively identifies three-dimensional characteristic defects such as the inward depression of the clip opening and surface scratches. The intelligent model iteration mechanism reduces the misjudgment rate under small sample data. The transparent conveying structure and dynamic adjustment components improve the compatibility of the equipment with products of different sizes. The visual camera group 2 includes several linear scanning camera groups, and the several linear scanning camera groups include at least six main surface detection camera groups and at least two tilt detection camera groups 207. The main surface detection camera groups are provided with six groups and are respectively arranged opposite to the six front views of the terminal block, for detecting the six surfaces of the terminal block; the first tilt detection camera group is used to detect the planar defects on the side and step side of the clip opening of the terminal block; the second tilt detection camera group is used to detect scratches, missing materials and burr defects on the surface of the terminal block.

[0038] A linear scan camera set is an optical device that captures images by line-by-line scanning. Specifically, it can be implemented using a CMOS or CCD sensor in conjunction with a linear light source. It is suitable for continuous image capture in dynamic transmission scenarios. A main surface inspection camera set is an optical acquisition unit arranged in six orthogonal directions of the object to be inspected. For example, a fixed focal length lens can be used in conjunction with a vertical mounting bracket. Figure 1 , ensuring that the six main viewing surface images are covered without blind spots, the first inclined detection camera group refers to a detection device aimed at the target area at a non-orthogonal angle, such as a camera installed at an angle of 30° to 45°, which can capture structural defects on the side wall of the clamp opening and the side of the step. The second inclined detection camera group refers to a detection device designed for subtle surface defects, such as a ring light source or coaxial light source to enhance the imaging contrast of scratches and missing materials.

[0039] See also Figures 2 to 5 , six groups of main surface inspection cameras are arranged along the transmission direction of track 11 respectively. The six groups of main surface inspection cameras include a top-down visual camera 201, an upward visual camera 202, a left-view visual camera 203, a right-view visual camera 204, a front-view visual camera 205, and a rear-view visual camera 206, covering the six main viewing surfaces of the terminal block, namely, up and down, left and right, and front and back, to ensure that each main surface is captured independently. The first tilted inspection camera group is aimed at the side of the clamp opening and the step side at a preset angle, and the plane projection overlap is eliminated through tilted imaging, so that the side depression or step misalignment defects are clearly visible. The second tilted inspection camera group adopts low-angle lighting to make subtle defects such as surface scratches and burrs form high-contrast features in the image. Combined with the high-resolution characteristics of linear scanning, stable identification of tiny defects can be achieved.

[0040] Existing inspection solutions typically only deploy single-angle area array cameras, which cannot cover multi-angle inspection requirements and require manual flipping or complex mechanical structures to adjust the viewing angle. However, this solution achieves full coverage inspection of the six main surfaces and inclined surfaces through the coordinated arrangement of multiple groups of linear scan cameras without flipping, while eliminating measurement errors caused by image distortion caused by traditional area array cameras.

[0041] It solves the problem of incomplete coverage of multi-angle defects by traditional visual inspection systems, especially for inclined angle defects such as the side wall of the clip opening, the side of the step and surface scratches, to achieve blind spot detection, reduce missed inspections due to angle of view limitations, reduce the need for manual re-inspection, and improve inspection efficiency and accuracy. The depth of field range of the special lens is determined by the following parameters: The calculation formula for the depth of field near limit distance L1 is:

[0042] , f is the focal length of the lens, N is the aperture coefficient, d is the diameter of the permissible circle of confusion, and δ is the object distance.

[0043] The depth of field near limit distance L1 refers to the closest object distance boundary at which the lens can clearly image under specific parameters. It can be achieved by adjusting the focal length, aperture coefficient and allowable confusion circle diameter parameters, such as selecting different optical lens combinations or adjusting sensor parameters. The allowable confusion circle diameter d refers to the maximum diameter of the acceptable blurred light spot on the imaging plane, which can be controlled by setting the pixel size of the image sensor or the image processing algorithm. The object distance δ refers to the actual distance between the object being measured and the optical center of the lens. By dynamically adjusting the object distance, it can adapt to detection targets with different height differences.

[0044] The depth of field is calculated based on the lens' focal length, aperture factor, and allowable circle of confusion diameter. Adjusting these parameters expands the effective depth of field. During inspection, when height differences exist on the surface of the terminal block to be inspected, the depth of field's near-limit distance, L1, is dynamically calculated. Combined with the adjustment of the object distance δ, this ensures that the lens maintains clear imaging within the range of planar defects with height differences between 3mm and 10mm. For example, using a large aperture factor increases the depth of field, while reducing the allowable circle of confusion improves image sharpness.

[0045] Traditional inspection systems rely on lens configurations with fixed depth of field, making it difficult to detect planar defects with different height differences, resulting in some defects being blurred due to exceeding the depth of field range. This solution dynamically calculates the depth of field parameters and adjusts the lens configuration to enable the imaging system to cover a larger height difference range, eliminating the imaging blur problem caused by changes in the surface height of the object being measured. This solves the problem of defect imaging blur caused by surface height differences of the object being measured in traditional inspections, ensuring that planar defects at different heights can be clearly imaged, thereby improving inspection accuracy and the system's compatibility with products of different specifications. Visual camera group 2 includes a special lens group, which is configured to be compatible with clear imaging of planar defects with height differences of 3 mm to 10 mm.

[0046] A special lens group refers to a lens combination with a large depth of field optical design. This can be achieved by adjusting the aperture coefficient or using a high-resolution sensor to make the depth of field cover the height changes of the target detection area, ensuring that defects in different plane positions remain clear in the same imaging plane. Compatible height difference means that the lens can accommodate a vertical drop of 3 mm to 10 mm on the surface of the object being measured while maintaining focus. This can be achieved by optimizing the matching relationship between object distance and focal length to avoid blurred imaging due to surface height differences.

[0047] During the inspection process, when there are height differences on the terminal surface, a special lens group extends the depth of field so that defects on different planes (such as the stepped structure on the side of the clip opening or surface scratches) remain clear in the image. This is achieved through pre-set depth of field parameters: the combination of lens focal length and object distance ensures that within the height difference range of 3 mm to 10 mm, the light from all target areas can be accurately focused on the sensor. As a result, the details of defects on different planes can be captured without frequent adjustment of the camera height, solving the problem of local defocus caused by height differences in traditional inspection.

[0048] Traditional inspection systems use fixed depth-of-field lenses, which can only clearly image a single plane. When there are height differences in the objects being inspected, the camera position must be adjusted multiple times or the images must be taken in stages, resulting in low inspection efficiency and the risk of missing defects. However, this solution uses the depth-of-field expansion capability of a special lens group to cover a wider height range in a single shot, significantly reducing the number of inspection steps and improving the defect capture rate.

[0049] It effectively solves the problem of imaging blur caused by height differences on the surface of the object being measured, allowing defects in different plane positions, such as step defects on the side of the clamp opening and surface scratches, to be clearly identified, reducing the missed detection rate and improving compatibility with multiple types of products. The object distance adjustment component 3 sets the line scan rate Vc of the linear scan camera using the following formula: Vc=(Hc×Vo) / Lo, where Hc is the number of pixels per line of the linear scan camera, Vo is the target operating rate, and Lo is the field of view width; the field of view width Lo is dynamically adjusted by the ratio of the focal length f to the camera sensor size to accommodate inspections of different sizes.

[0050] Hc refers to the number of pixels that a linear scan camera can capture in a single scan line. This can be achieved using a linear array sensor with a fixed number of pixels, and its value is determined by the camera hardware parameters. Vo refers to the movement speed of the terminal block during transmission on track 1, which can be achieved by adjusting the drive rate of the conveyor chain plate through a variable frequency motor or servo control system. Lo refers to the actual detection area width corresponding to a single scan line of the camera, which can be achieved by adjusting the focal length of the lens or replacing sensors of different sizes. This parameter is positively correlated with the ratio of focal length to sensor size. Dynamic adjustment of Lo means recalculating the field of view width by changing the configuration of the focal length or sensor size based on the actual size of the terminal block to be inspected, thereby matching the inspection requirements of products of different specifications.

[0051] As the terminal blocks move along track 1, the line scan rate Vc is automatically calculated and generated according to a formula to ensure that at the target operating rate Vo, the pixel acquisition of each scan line can fully cover the field of view width Lo, avoiding image stretching or compression due to rate mismatch. For example, when inspecting larger terminal blocks, the Lo value can be increased by increasing the focal length or replacing a larger sensor, allowing a single scan line to cover a wider inspection area. At the same time, Vc is adjusted to maintain image acquisition accuracy. As a result, terminal blocks of different sizes do not need to be physically adjusted during transmission; compatible inspection can be achieved simply through dynamic parameter configuration. Traditional line scan cameras use a fixed rate for image acquisition. When product size or transmission speed changes, image distortion or missed inspection areas are prone to occur. However, this solution dynamically calculates the line scan rate and field of view width to match the camera parameters to product size and transmission speed in real time, avoiding image distortion or detection blind spots caused by fixed parameters.

[0052] The automatic feeding system also includes a material guide station adjustment device for adjusting the position of various products on the inspection machine and a stable running trajectory. The material guide station adjustment device includes a material guide rail 4 and a material guide control motor 5 for driving the material guide rail 4. The material guide control motor 5 adopts a servo motor with an extended structure. The material guide rail 4 is in the shape of a strip plate and is directly connected to the telescopic movement end of the material guide control motor 5 by screws. The material guide rail 4 is controlled by the material guide control motor 5. The material guide station adjustment device is used to push the material guide control motor 5 to the corresponding position before the terminal enters the visual inspection. This can reduce the subsequent camera's photo inspection of the terminal and drive the terminal of different specifications to the fixed position of the track's conveying chain plate section 11, which is convenient for relative setting with the camera on the multi-station inspection device, thereby facilitating the overall adjustment of the camera and image acquisition.

[0053] This application solves the technical problem that the traditional detection system cannot adapt to different sizes of terminal blocks due to fixed parameters, and realizes the adaptive matching of line scanning rate and product size, thereby improving the detection accuracy and reducing the complexity of system adjustment. Track 1 adopts circular track 1, and circular track 1 adopts plane circular track 1 or upper and lower circular track 1, combined with Figure 4 When the track 1 adopts a plane ring track 1, the terminal is set along the track 1 in a ring-shaped uniform motion. Figure 2 When the track 1 adopts an upper and lower ring track 1, the terminal is located on one side of the track 1 and is arranged to move in a straight line at a uniform speed.

[0054] See also Figure 4 The plane circular track 1 refers to a closed-loop track 1 structure arranged in the horizontal direction. Specifically, a circular transmission path can be constructed using metal or high-strength composite materials to achieve this, so that the terminal maintains a uniform speed along the closed-loop path to avoid distortion of the detection image due to sudden changes in direction. Figure 2The upper and lower circular tracks 1 refer to a closed-loop track 1 structure arranged in the vertical direction. Specifically, it can be achieved by using a laterally installed guide rail and a drive component to make the terminal block move linearly along the single-sided track 1 surface, which is suitable for the stable transmission of long-sized products. The circular uniform motion setting means that the terminal block moves at a constant speed along the circular path in the horizontal track 1. Specifically, it can be achieved by controlling the linear speed of the conveyor chain plate through the linkage of the servo motor and the encoder to ensure that there is no speed fluctuation in the motion trajectory during image acquisition. The linear uniform motion setting means that the terminal block moves at a constant speed along the linear path in the vertical track 1. Specifically, it can be achieved through the cooperation of the synchronous belt drive and the speed feedback module to avoid detection blind spots caused by changes in the direction of movement.

[0055] The flat circular track 1 allows the terminal blocks to move continuously in a circular motion through a closed-loop transmission path. Multiple sets of visual inspection devices are distributed circumferentially along the track 1, thereby capturing images of the product from multiple angles. The upper and lower circular tracks 1 guide the terminal blocks to move in a straight line through the single-side track 1 surface, which is suitable for products with longer lengths, such as long PIN terminals with a length exceeding 55 mm, to avoid transmission jams caused by size limitations of traditional disc structures. In the horizontal track 1, the uniform circular motion can be combined with the synchronous triggering mechanism of the multi-station camera to achieve 360-degree detection without blind spots; in the vertical track 1, the linear motion can simplify the product positioning process, and the bottom surface image can be directly obtained through the transparent transmission chain plate segment 11 on the single-side track 1 surface without flipping the product.

[0056] Traditional glass disk structures are not compatible with long PIN terminals or special-shaped products due to their fixed size, and disk rotation can easily cause image distortion. This solution uses a horizontal or vertical circular track 1 to replace the disk structure. It adapts to different product sizes through circular or linear motion paths and eliminates image distortion caused by rotational motion. For example, the horizontal track 1 is suitable for multi-faceted inspection of conventional-sized products, while the vertical track 1 solves the stability issues of long PIN terminals through linear transmission, thereby achieving wide compatibility of inspection equipment.

[0057] This application solves the problem that the traditional disc structure cannot adapt to the transmission of long-sized or special-shaped terminal blocks. The horizontal and vertical track 1 configurations meet the inspection needs of diversified products, while avoiding image distortion caused by rotational motion, ensuring accurate identification of clip opening and indentation defects. In addition, the linear uniform motion setting can reduce the risk of positional offset of long PIN terminals during transmission, and improve the reliability of dimensional measurement and plane defect detection. The model training unit adopts an incremental learning algorithm. The model training unit is configured with a general defect database. Before training the multi-station detection model, it is pre-trained with a database containing general defect types. The database contains several sample images of common types of defects.

[0058] An incremental learning algorithm refers to a learning method that continuously receives new data and gradually updates model parameters. This can be achieved using an online training framework, which enables the model to adapt to new defect types without retraining existing knowledge. A general defect database refers to a standardized data set that pre-collects a variety of common defect samples. It can be constructed using an industrial standard defect library, such as typical defect images such as scratches on metal surfaces and missing materials on plastic parts. Multi-station inspection model pre-training refers to basic training of the initial model using a general database. Specifically, transfer learning technology can be used to transfer general feature extraction capabilities to specific inspection tasks.

[0059] During the model training phase, pre-training is first performed based on a general defect database to enable the model to master basic defect recognition capabilities. Subsequently, incremental learning is initiated through small sample annotated data, and the model parameters are gradually optimized using newly collected terminal defect images. When missed defects are found during the trial operation of the detection system, the newly annotated defect samples are input into the model for iterative updating while retaining the original detection capabilities. A dynamic weight adjustment strategy is adopted during the training process to balance the impact of new and old samples on the model.

[0060] Traditional methods rely on a fixed defect library to train static models, which cannot effectively respond to newly emerging defect types or product variants and require frequent model retraining. This solution, however, achieves dynamic model evolution through incremental learning and, combined with the broad feature base provided by a general database, significantly improves the ability to adapt to unknown defects.

[0061] The present application realizes the autonomous optimization of the detection model in continuous operation, effectively reduces the risk of model failure due to product iteration or new defect types, and reduces the dependence on large amounts of labeled data through the synergy of pre-training and incremental learning. At the same time, it ensures that the detection system can quickly adapt to the differences in products of different production batches, improves the coverage and long-term stability of defect detection, and the transmission chain plate segment 11 adopts a transparent material part. The transparent material adopts high-transmittance optical glass. The transparent material part of the transmission chain plate segment 11 is equipped with a number of magnifying lens groups. The number of magnifying lens groups are evenly spaced and the spacing is the same as the gap between the openings of the clips on the terminal block. The chain plate segment 11 is equipped with a blocking device. When the terminal is placed on the transmission chain plate segment 11 made of transparent material, the blocking device drives the several clip openings on the terminal to align with the several magnifying lens groups. When the several clip openings on the terminal are aligned with the several magnifying lens groups, the terminal is released. The transparent material uses high-transmittance optical glass, which refers to a transparent material with low light loss characteristics. Specifically, it can be achieved by using optical glass with a refractive index in the range of 1.5 to 1.6. Its function is to provide a distortion-free light transmission channel for visual inspection. The magnifying lens group refers to a lens array with a magnifying function. Specifically, a convex lens with a focal length in the range of 5mm to 15mm can be used. The lens combination is used to locally magnify the opening area of ​​the clamp to improve the accuracy of defect recognition. The uniform spacing setting means that the spacing between the lens groups is consistent with the gap between the clamp openings. It can be achieved through mold forming or mechanical processing. Its function is to provide a consistent positioning reference for different types of terminal blocks. The blocking device refers to a physical structure that limits the movement of the workpiece, specifically including a blocking block and a blocking pusher. The blocking block is used to physically limit one end of the workpiece, and the blocking pusher is used to drive the blocking block to move. It can be achieved by using a retractable baffle or a pneumatic push rod. Its function is to force the position of the terminal to be aligned with the lens group before detection. The terminal When the clamp is placed on the conveyor chain plate section 11 made of transparent material, the blocking device structure will temporarily restrict its movement, forcing the position of the clamp opening to be aligned with the center area of ​​the magnifying lens group. When the alignment is completed, the blocking structure automatically releases the restriction, allowing the terminal to continue to move at a constant speed along the track 1. During this process, the magnifying lens group amplifies the image of the clamp opening area, allowing the visual inspection system to clearly capture subtle defects in the area, such as clamp indentation or burrs. Since the spacing of the lens group is consistent with the gap of the clamp opening, under the forced positioning action of the blocking structure, each clamp opening can correspond to a magnifying lens, avoiding image blur or missed detection problems caused by misalignment.

[0062] Traditional glass disc detection devices cannot adapt to terminal blocks of different shapes, and because the disc surface is fixed, image distortion is easily generated when detecting long PIN terminals. However, this solution uses a linear transmission chain plate segment 11 to replace the disc structure, and cooperates with the blocking positioning mechanism through an adjustable magnifying lens group to solve the positioning problem during dynamic detection of long wiring harness terminals and eliminate the image distortion problem caused by structural limitations.

[0063] This application realizes precise imaging detection of the clip opening area, effectively solving the problem of missed detection of clip sunken defects caused by positioning deviation in traditional transparent disk structures. At the same time, it is compatible with the detection needs of terminal blocks of different sizes and shapes, significantly reducing the proportion of manual re-inspection.

[0064] An AI visual inspection method for terminal block appearance includes the following steps: Step S1: Track 1 transports the terminal to be inspected at a constant speed, and the track 1 is equipped with a transparent material conveying chain plate segment 11; Step S2: The material guide rail controls the material guide motor to drive the wiring terminals of different specifications to the specified positions of the conveying chain plate segment 11, and the conveying chain plate segment 11 drives the wiring terminals to be in a stable operating state; Step S3: Automatically adjust the height direction of the visual camera group through the object distance adjustment component 3, so that the lens of the visual camera group automatically focuses on the object distance to be compatible with the detection of terminal blocks of different sizes and types; Step S4: six main surface inspection camera groups, two tilt inspection camera groups, and a special lens group are used to perform a 360° scan of the terminal block to be inspected without blind spots. The special lens is configured to clearly image planar defects with a drop height of 3 mm to 10 mm. Step S5: Classify and annotate the collected defect images, including multiple missing parts, functional differences, and appearance defect types, and confirm the qualified terminal blocks; Step S6: Based on the small sample annotated data and the transfer learning algorithm, the defect image is iteratively trained for a preset number of times to generate a multi-station detection model, wherein the initial model is pre-trained on a database containing common defects; Step S7: Deploy the trained inspection model to the inspection system to perform component statistics, defect detection, and rejection of defective products, and verify the inspection function through actual trial operation; Step S8: During the trial operation phase, the test results are manually reviewed, missed defects are marked, and the test model is iteratively updated. Terminal blocks that are judged by the system to be defective but are actually qualified are corrected for misjudgment, and problematic terminal blocks that were missed by the system are returned to the S5 marking training phase until the detection accuracy exceeds the preset standard.

[0065] Track 1 transports the terminals to be inspected at a constant speed, and a conveyor chain plate section 11 made of transparent material is installed on Track 1. The terminals to be inspected are scanned 360° without blind spots using six main surface inspection camera groups, two tilt inspection camera groups, and a special lens group. The special lens is configured to clearly image planar defects with a drop of 3mm to 10mm. The collected defect images are classified and annotated, including types of missing parts, functional differences, and appearance defects, and qualified terminals are confirmed. Based on small sample annotation data and a transfer learning algorithm, the defect images are iteratively trained a preset number of times to generate a multi-station inspection model, where the initial model is pre-trained on a database containing common defects. The trained inspection model is deployed to the inspection system to perform component statistics, defect detection, and rejection of defective products, and the inspection function is verified through actual trial operation. During the trial operation phase, the inspection results are manually reviewed, missed defects are annotated, and the inspection model is iteratively updated. Terminals that are judged as defective by the system but are actually qualified are corrected, and problematic terminals that were missed by the system are returned to the annotation training phase until the inspection accuracy exceeds the preset standard.

[0066] The transparent material conveyor chain plate segment 11 refers to a conveying component made of high-transmittance optical glass, which can be specifically implemented by a structure in which the surface is covered with a magnifying lens group. The spacing of the magnifying lens group is consistent with the gap between the terminal clip openings, so that the clip opening defects are magnified by the lens and captured by the camera. The special lens group refers to a lens with a depth of field adjustment function that is compatible with a height difference of 3mm to 10mm. This can be achieved by adjusting the focal length and aperture coefficient to ensure clear imaging of defects under different plane drops. Small sample annotation data refers to a technical means for model training based on a small number of annotated defect samples. Specifically, the general defect features in the pre-trained model can be transferred to the current task through the transfer learning algorithm, reducing dependence on a large amount of annotated data. Manual review in the trial operation stage refers to the process of comparing and verifying the system judgment results with the actual detection results. Specifically, it can be achieved by manually confirming missed or misjudged samples and re-annotating them to achieve dynamic optimization of model parameters.

[0067] The terminal blocks to be inspected are transported at a constant speed via transparent conveyor chains to avoid abnormal image acquisition due to vibration or positional displacement. Six groups of main surface inspection cameras and two groups of tilted inspection cameras scan the terminal surface from different angles. The special lens group dynamically adjusts the depth of field to ensure the imaging quality of different drop defects such as the side of the clip opening, the step side, and surface scratches. After the collected images are classified and labeled, a transfer learning algorithm is used to iteratively train the pre-trained model, enabling the detection model to quickly adapt to new defect types. After the model is deployed, the model parameters are continuously optimized through real-time detection and trial operation feedback combined with manual review data to ultimately achieve the preset detection accuracy requirements.

[0068] In some specific embodiments, the magnifying lens group of the transparent conveyor chain plate segment 11 can be replaced with a lens array of different magnifications to meet the detection requirements of clamp openings of different sizes. The manual review data during the trial operation phase can be synchronized to multiple detection systems through the cloud, realizing one-click deployment of model updates.

[0069] Traditional inspection methods rely on manual vision or single-angle area array cameras, which cannot cover the multiple surface defects of long PIN terminals and are easily affected by image distortion. This method achieves 360° detection without blind spots through the combination of a multi-angle linear scanning camera and a special lens group, eliminating image distortion. Existing technologies use static models with fixed defect libraries, which are difficult to deal with new types of defects. However, this method uses transfer learning and dynamic iteration mechanisms to enable the model to have continuous learning capabilities, reducing the missed detection rate.

[0070] This application solves the problems of low coverage and high manual dependence in multi-surface defect detection of long PIN terminals. The transparent conveyor chain plate is combined with the magnifying lens group to avoid the decrease in detection efficiency caused by traditional flipping operations. Dynamic depth of field adjustment is combined with multi-angle scanning to ensure the imaging consistency of defects with different height differences. The model training method based on small sample transfer learning can quickly build a high-precision detection model with limited labeled data, reducing dependence on manual experience. The closed-loop optimization mechanism in the trial operation stage enables the system to adapt to the detection needs of terminal products of different sizes and colors, reducing the risk of defective products flowing out.

[0071] In the AI ​​visual inspection method for the appearance of terminal blocks, the object distance and spatial position of the camera are adjusted through the object distance adjustment component 3 to be compatible with the inspection of terminal blocks of different sizes and types.

[0072] The object distance adjustment component 3 refers to a device that dynamically adjusts the vertical distance between the camera and the detection object through a mechanical structure. Specifically, it can be achieved by using a servo motor to drive a precision screw mechanism, and the camera position is accurately adjusted by controlling the rotation amount of the motor. The object distance refers to the distance from the optical center of the lens to the surface of the object to be detected. Specifically, it can be achieved by adjusting the lifting range of the camera mounting bracket. For example, the object distance adjustment range is set to 50-200mm, so that the focal length can adapt to the detection plane with a height difference of 3-10mm. Spatial position adjustment refers to the three-dimensional coordinate adjustment of the camera in the horizontal direction. Specifically, it can be achieved by using a camera gimbal with an XY slide rail. For example, a translation mechanism with a stroke of ±15cm is configured to ensure that the camera can be aligned with the detection area of ​​terminal blocks of different sizes.

[0073] When inspecting terminal blocks of different batches or models, the operator can input product specifications and parameters, and the control program automatically calculates the required object distance compensation and camera installation angle. The servo motor drives the object distance adjustment component 3 to move the linear scan camera to a preset height. At the same time, the XY slide rail adjusts the camera's horizontal position according to the product length. For example, when inspecting long PIN terminals with a length of 80mm, the camera group moves along the extension direction of track 1 to a detection position covering the entire length of the terminal. At the same time, the object distance is adjusted to the optimal imaging distance that matches the terminal surface height. During the adjustment process, the laser ranging sensor provides real-time feedback of position data to ensure that the overlapping field of view of multiple camera groups completely covers the inspection target.

[0074] Traditional inspection equipment uses fixed-mounted camera groups, and when the size of the inspection object exceeds the preset range, the machine must be shut down and the hardware must be replaced. However, this solution uses a programmable height adjustment mechanism to enable the same inspection system to adapt to various terminal products with a length of 55-120mm and a height of 5-15mm. The existing glass turntable inspection method cannot adjust the inspection angle due to structural limitations. This solution uses three-dimensional spatial position adjustment to enable the tilted inspection camera group to optimize the shooting angle for the characteristic parts of products with different shapes.

[0075] This application realizes the rapid changeover detection of terminal products with multiple specifications, eliminates the detection blind spots caused by hardware fixation of traditional equipment, and increases the detection rate of defects such as inward-facing clip openings and scratches on irregular surfaces by more than 40%. At the same time, it shortens the product changeover time from 2 hours with traditional methods to less than 10 minutes, significantly improving the generalization capability of the detection system and the flexibility of the production line.

Claims

1. A terminal block full appearance inspection system based on a ring track platform, characterized in that: include: The track (1) is provided with a transparent material transmission chain plate segment (11) for transmitting the terminal blocks in a uniform motion; A multi-station visual inspection device comprises a plurality of visual camera groups (2) and an object distance adjustment component (3), wherein the plurality of visual camera groups (2) are respectively arranged at a plurality of stations of the track (1) and are used to collect images of six front-view main surfaces and a setting surface of the terminal block, and the visual camera groups (2) are all connected to the object distance adjustment component (3), and the object distance adjustment component (3) is used to adjust the object distance or / and position of the visual camera group (2) relative to the terminal block according to the terminal block to be inspected; The material guide station adjustment device comprises a material guide rail (4) and a material guide control motor (5) for driving the material guide rail (4) to move, wherein the material guide rail (4) is controlled by the material guide control motor (5) to move and is used to drive the connection terminals of different specifications to be arranged relative to the multi-station monitoring device; AI visual inspection module, including: Image annotation unit, used to classify and annotate defects in the collected images; The model training unit iteratively trains the labeled images based on a small sample learning algorithm to generate a multi-station detection model; A real-time inspection unit deploys the multi-station inspection model to perform component counting, defect detection, and rejection of defective products.

2. The terminal block full appearance inspection system based on the circular track platform according to claim 1 is characterized in that: The visual camera group (2) includes several linear scanning camera groups. The plurality of linear scan camera groups include at least six main surface detection camera groups and at least two tilt detection camera groups, which are: Six main surface inspection camera groups are provided and are respectively arranged opposite to the six front views of the terminal block, for inspecting the six surfaces of the terminal block; The first tilt detection camera group is used to detect planar defects on the side of the clamp opening and the step side of the terminal block; The second tilt detection camera group is used to detect scratches, missing materials and burrs on the surface of the terminal blocks.

3. The terminal block full appearance inspection system based on the circular track platform according to claim 2, characterized in that: The depth of field range of the special lens is determined by the following parameters: The calculation formula for the depth of field near boundary distance L1 is: ， Where f is the focal length of the lens, N is the aperture coefficient, d is the diameter of the permissible circle of confusion, and δ is the object distance.

4. The terminal block full appearance inspection system based on the circular track platform according to claim 2, characterized in that: The visual camera group (2) further comprises a special lens group, wherein the special lens group is configured to be compatible with clear imaging of planar defects with a height difference of 3 mm to 10 mm.

5. The terminal block full appearance inspection system based on the circular track platform according to claim 2, characterized in that: The object distance adjustment component (3) sets the line scan rate Vc of the linear scan camera by the following formula: ， Where Hc is the number of pixels per line of the line scan camera, Vo is the target operating rate, and Lo is the field of view width; The field of view width Lo is dynamically adjusted by the ratio of the focal length f to the camera sensor size to accommodate detection of different sizes.

6. The terminal block full appearance inspection system based on the circular track platform according to claim 1, characterized in that: The track (1) adopts an annular track, and the annular track of the track (1) adopts a planar annular track or an up-down annular track. When the track (1) adopts a planar annular track, the connecting terminal is arranged along the track (1) in a circular uniform motion. When the track (1) adopts an up-down annular track, the connecting terminal is located on one side of the track and is arranged in a linear uniform motion.

7. The terminal block full appearance inspection system based on a circular track platform according to claim 1, characterized in that: The model training unit cooperates with an incremental learning algorithm, and the model training unit is configured with a general defect database. Before training the multi-station detection model, it is pre-trained with a database containing general defect types, and the database contains several sample images of common types of defects.

8. The terminal block full appearance inspection system based on the circular track platform according to claim 2, characterized in that: The transmission chain plate section (11) is made of a transparent material portion, and the transparent material is made of high-transmittance optical glass. The transparent material portion of the transmission chain plate section (11) is provided with a plurality of magnifying lens groups, and the plurality of magnifying lens groups are evenly spaced and the spacing is the same as the gap between the plurality of clip openings on the connection terminal. The transmission chain plate section (11) is provided with a blocking device. When the connection terminal is placed on the transmission chain plate section (11) made of the transparent material, the blocking device drives the plurality of clip openings on the connection terminal to be aligned with the plurality of magnifying lens groups. When the plurality of clip openings on the connection terminal are aligned with the plurality of magnifying lens groups, the connection terminal is released.

9. A method for inspecting the full appearance of terminal blocks based on a circular track platform, according to a system for inspecting the full appearance of terminal blocks based on a circular track (1) platform as claimed in any one of claims 1 to 8, characterized in that: The following steps are involved: Step S1: The track (1) transports the terminal to be inspected at a constant speed, wherein the track (1) is provided with a transparent material conveying chain plate segment (11); Step S2: driving the wiring terminals of different specifications to the specified positions of the conveying chain plate section (11) through the material guide rail and the material guide control motor, and the conveying chain plate section (11) drives the wiring terminals to be in a stable operating state; Step S3: Automatically adjust the height direction of the visual camera group through the object distance adjustment component (3), so that the lens of the visual camera group automatically focuses on the object distance to be compatible with the detection of terminal blocks of different sizes and types; Step S4: six main surface inspection camera groups, two tilt inspection camera groups, and a special lens group are used to perform a 360° scan of the terminal block to be inspected without blind spots. The special lens is configured to clearly image planar defects with a drop height of 3 mm to 10 mm. Step S5: Classify and annotate the collected defect images, including multiple missing parts, functional differences, and appearance defect types, and confirm the qualified terminal blocks; Step S6: Based on the small sample annotated data and the transfer learning algorithm, the defect image is iteratively trained for a preset number of times to generate a multi-station detection model, wherein the initial model is pre-trained on a database containing common defects; Step S7: Deploy the trained inspection model to the inspection system to perform component statistics, defect detection, and rejection of defective products, and verify the inspection function through actual trial operation; Step S8: During the trial operation phase, the test results are manually reviewed, missed defects are marked, and the test model is iteratively updated. Terminal blocks that are judged by the system to be defective but are actually qualified are corrected for misjudgment, and problematic terminal blocks that were missed by the system are returned to the S5 marking training phase until the detection accuracy exceeds the preset standard.

Images