Visual inspection system and method based on aluminum plate size and appearance surface defects

Abstract

The invention discloses a visual detection system and method based on aluminum plate size and appearance surface defects, and the system comprises a mobile conveying mechanism, an image collection module, a control module, an upper computer, a data management module and an automatic sorting module. And the upper computer internally comprises an algorithm module, an image processing module, a size measurement module and a defect detection module. According to the invention, multi-modal image acquisition, intelligent algorithm fusion and automatic sorting are utilized to realize rapid and accurate detection of aluminum plate size errors and surface defects, and the production quality and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation detection technology, and in particular to a visual detection system and method based on aluminum plate size and appearance surface defects. Background Art

[0002] Aluminum plates are key materials widely used in the fields of construction, automobiles, electronics, etc. Their dimensional accuracy (such as thickness, length, flatness) and surface quality (such as scratches, oxidation spots, dents, coating defects) directly affect product performance. At present, aluminum plate inspection mainly relies on manual sampling or contact measurement tools, which have the following problems: Low efficiency: Manual visual inspection is slow and cannot meet the high-speed inspection requirements of the production line. Insufficient accuracy: Contact measurement is easily affected by the operator's subjective influence and cannot detect subtle surface defects. High missed detection rate: Under complex lighting conditions, it is difficult for humans to find tiny defects (such as microcracks and shallow dents). Lack of data management: Lack of real-time detection data recording and analysis makes it difficult to achieve quality traceability.

[0003] Although existing machine vision inspection technology has been applied, it still has limitations in the following aspects:

[0004] The fast scanning and splicing accuracy of large-size aluminum plates is insufficient; the recognition accuracy of highly reflective surface defects (such as aluminum plate oxidation spots) is low; the collaborative detection capability of multiple types of defects (geometric defects and surface defects) is weak. Patent CN113034445A discloses a multi-scale connected image defogging algorithm based on UNet3+. The patent uses a monocular camera to measure the size. When detecting aluminum plates, there will be a problem of edge positioning deviation caused by reflection. Therefore, there is an urgent need for an efficient, high-precision, fully automatic aluminum plate detection system and method. Summary of the invention

[0005] The purpose of the present invention is to provide a visual inspection system and method based on the size and appearance surface defects of aluminum plates to solve the problems in the above-mentioned background technology.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A visual inspection system based on aluminum plate size and appearance surface defects includes a mobile conveying mechanism, an image acquisition module, a control module, a host computer, a data management module and an automatic sorting module. The host computer includes an algorithm module, an image processing module, a size measurement module and a defect detection module.

[0008] The mobile conveying mechanism is used for continuous conveying of aluminum plates, on which a QR code / RFID is provided, and the mobile conveyor can obtain the QR code / RFID information and upload the information to the host computer;

[0009] The image acquisition module is used to acquire the image of the aluminum plate to be tested, and includes a high-resolution camera, a multi-spectral light source system and a 3D structured light sensor, so as to obtain the three-dimensional geometric information of the aluminum plate;

[0010] The control module is electrically connected to the mobile conveying mechanism, the image acquisition module, the host computer, the automatic sorting module, and the data management module;

[0011] The image processing module is used to process the photos taken by the high-resolution camera and transmit the processed information to the size measurement module and the defect detection module;

[0012] The dimension measurement module is used to calculate the length, width and edge straightness of the aluminum plate based on the information transmitted by the image processing module;

[0013] The defect detection module is used to capture scratches, pits, and coating shedding on the surface of the aluminum plate. The defect detection module has a built-in deep learning model to complete defect classification and positioning;

[0014] The defect detection module calculates the defect area, defect depth and defect distribution density by quantitatively analyzing the defects;

[0015] The data management module is connected to the host computer and is a real-time database for storing detection data;

[0016] The automated sorting module comprises a sorting mechanism for sorting the aluminum plates into qualified products, returned products, and scrap products according to the detection results of the defect detection module.

[0017] Furthermore, the mobile conveying mechanism is provided with a code scanning mechanism and a laser rangefinder. The code scanning mechanism is arranged on the end side of the mobile conveying mechanism and is used to scan the QR code / RFID of the aluminum plate. The information obtained after scanning is transmitted to the host computer. The laser rangefinder is used to measure the thickness, width and length of the aluminum plate.

[0018] Furthermore, the high-resolution camera continuously scans the conveyed aluminum plate to acquire images, the multi-spectral light source system uses a combination of annular LED light source and polarized light to suppress reflection interference on the surface of the aluminum plate, and the 3D structured light sensor is arranged next to the high-resolution camera.

[0019] Furthermore, the size measurement module adopts an edge extraction algorithm and performs calculations through sub-pixel edge detection technology to improve the accuracy of edge detection, and fits the reference plane through 3D point cloud data to calculate the flatness of the aluminum plate, that is, the deviation between the surface of the aluminum plate and the reference plane.

[0020] Furthermore, the defect detection module uses multi-scale image fusion to enhance the global contrast and local detail features of defects, and uses industrial-grade defect datasets for transfer learning, combined with data enhancement to improve model generalization.

[0021] Furthermore, the data management module also includes a quality traceability system for obtaining data of a QR code bound to the aluminum plate batch information to generate a test report.

[0022] Furthermore, the automated sorting module is also designed with an alarm feedback system for pushing the defect type and location to the production line control terminal in real time.

[0023] Also included is a method for detecting size and surface defects of an aluminum plate based on machine vision, comprising the following steps:

[0024] In the first step, the aluminum plate is transported by a mobile conveying mechanism, and the scanning mechanism scans the QR code / RFID of the aluminum plate, and the information obtained after scanning is transmitted to the host computer;

[0025] In the second step, after the scanning is completed, the laser rangefinder is operated to measure the thickness of the aluminum plate. At the same time, the multi-spectral light source system of the image acquisition module starts working, and the high-resolution camera and 3D structured light sensor are operated at the same time to complete the 3D scanning and taking pictures of the aluminum plate. After taking pictures, the image information obtained is transmitted to the image processing module for image processing, and the processed information is transmitted to the size measurement module and the defect detection module;

[0026] In the third step, the dimension measurement module obtains the dimension and flatness results of the aluminum plate after receiving the information;

[0027] In the fourth step, the defect measurement module obtains the type, location and area of ​​the defect after receiving the information;

[0028] Step 5: All the inspection data from Step 3 and Step 4 are transmitted to the data management module through the host computer, and are associated with the QR code of the aluminum plate to generate a quality inspection report;

[0029] The sixth step is to upload the information of the quality inspection report to the host computer, which then transmits it to the control module, and finally instructs the automated sorting module to sort the aluminum plates into three categories: qualified products, returned products, and scrap products.

[0030] Furthermore, when laser ranging is used in the second step, multiple groups of point data are measured and the average value is finally calculated.

[0031] Furthermore, in the fourth step of defect measurement, multi-scale image fusion is used to confirm the surface and internal defect types, the defect classification and positioning are further completed through the deep learning model, and the defect area is calculated through defect quantification analysis.

[0032] Beneficial effects:

[0033] The present invention utilizes multimodal image acquisition, intelligent algorithm fusion and automated sorting to achieve rapid and accurate detection of aluminum plate dimensional errors and surface defects, thereby improving production quality and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a structural diagram of the visual inspection system based on the size and appearance surface defects of the aluminum plate in the present invention;

[0035] Figure 2 It is the schematic diagram of the flatness measurement principle in the present invention;

[0036] Figure 3 It is a flow chart of the aluminum plate size and surface defect detection method based on machine vision in the present invention.

[0037] In the figure: 1. Mobile conveying mechanism; 2. Image acquisition module; 3. Control module; 4. Data management module; 5. Automatic sorting module; 6. Code scanning mechanism; 7. Laser rangefinder; 8. High-resolution camera; 9. 3D structured light sensor; 10. Multi-spectral light source system; 11. Algorithm module; 12. Image processing module; 13. Defect detection module; 14. Dimension measurement module. DETAILED DESCRIPTION

[0038] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0039] Example

[0040] This embodiment provides a visual inspection system based on the size and appearance of aluminum plate surface defects, such as Figure 1 As shown, it includes a mobile conveying mechanism 1, an image acquisition module 2, a control module 3, a host computer, a data management module 4 and an automatic sorting module 5, and the host computer includes an algorithm module 11, an image processing module 12, a dimension measurement module 14 and a defect detection module 13.

[0041] The mobile conveying mechanism 1 is used for continuous conveying of aluminum plates, and specifically can be roller conveying, belt conveying, chain conveying, etc. A QR code / RFID is provided on the aluminum plate, and a code scanning mechanism 6 and a laser rangefinder 7 are provided on the mobile conveying mechanism 1. The code scanning mechanism 6 is provided on the end side of the mobile conveying mechanism 1 and is used to scan the QR code / RFID of the aluminum plate. The information obtained after scanning is transmitted to the host computer, and the laser rangefinder 7 is used to measure the thickness, width and length of the aluminum plate.

[0042] The image acquisition module 2 is used to acquire the image of the aluminum plate to be tested, including a high-resolution camera 8, which is arranged above the transmission direction of the aluminum plate to realize continuous scanning imaging; there is also a multi-spectral light source system 10 arranged at the 45° side direction of the transmission direction, which adopts a combination of annular LED light source and polarized light to suppress the reflection interference on the surface of the aluminum plate; and there is also a 3D structured light sensor 9, which is arranged next to the high-resolution camera 8, and emits a beam of specifically coded structured light to illuminate the surface of the object, and then the high-resolution camera 8 captures the light reflected from the surface of the object, and uses the algorithm module 11 to decode and reconstruct it, so as to obtain the three-dimensional geometric information of the object, which is finally used for the thickness and flatness measurement of the aluminum plate.

[0043] The control module 3 is electrically connected to the mobile conveying mechanism 1, the image acquisition module 2, the host computer, the automatic sorting module 5, and the data management module 4. The control module 3 is mainly arranged in a PLC controller, and the PLC controller is connected to the host computer.

[0044] The algorithm module 11 is a core module, including an image processing module 12, a dimension measurement module 14 and a defect detection module 13, and is arranged in a host computer.

[0045] The image processing module 12 is used to perform special processing on the photos taken by the high-resolution camera 8 and transmit the processed information to the size measurement module 14 and the defect detection module 13 .

[0046] The size measurement module 14 uses an edge extraction algorithm and sub-pixel edge detection technology to perform more detailed calculations below the pixel level, thereby improving the accuracy of edge detection. Based on the extracted edge information, the module can accurately calculate the length, width and straightness of the edge of the aluminum plate. The reference plane can be fitted through the 3D point cloud data, and based on the reference plane, the flatness of the aluminum plate, that is, the deviation between the surface of the aluminum plate and the reference plane, can be calculated. The thickness measurement combines the laser rangefinder 7 with structured light data to achieve non-contact thickness detection.

[0047] The defect detection module 13 uses multi-scale image fusion to capture visible defects such as scratches, pits, and coating shedding on the aluminum plate surface by fusing visible light images, and relies on high-resolution imaging technology to improve detail clarity. Near-infrared images penetrate the surface coating or oxide layer to detect abnormal oxide layer thickness (such as local over-thickness or over-thinness) and internal structural defects (such as bubbles and inclusions).

[0048] Specifically, in the defect detection module 13, through fusion algorithms, image registration, feature alignment and multi-scale fusion (such as wavelet transform or pyramid fusion), the image information of the two modalities is complementarily combined to enhance the global contrast and local detail features of the defects and reduce missed detections and false detections.

[0049] Specifically, in the defect detection module 13, a deep learning model is also built in, using an improved YOLOv7 network, and introducing an attention mechanism (such as CBAM or SE module) based on YOLOv7 to enhance the feature extraction capability of tiny defects (such as fine scratches); through adaptive anchor frame adjustment, the positioning accuracy of irregular defects (such as oxidation spots) is improved. At the same time, defect classification (scratches, oxidation spots, coating shedding, etc.) and positioning (boundary box annotation) are completed, and the defect type, location and confidence are output.

[0050] In addition, the defect detection module 13 also uses industrial-grade defect data sets for transfer learning, and combines data enhancement (such as random occlusion and brightness perturbation) to improve the generalization of the model and adapt to complex lighting and surface reflection scenes.

[0051] The defect detection module 13 also calculates the defect area by quantitatively analyzing the defects, and calculates the projected area of ​​the defect in the two-dimensional image based on the segmentation network (such as U-Net) or pixel-level statistics; and estimates the actual physical size of the defect by combining multi-view images or 3D point cloud data. The defect depth (3D data) is calculated by quantitative analysis of the defects, and the comprehensive impact of the defects on the overall quality of the aluminum plate is evaluated by spatial distribution statistics (such as defect spacing and regional density). The defect detection module 13 calculates the defect distribution density by quantitative analysis of the defects, and measures the depth information of the defects (such as pit depth and scratch depression amount) by combining the 3D point cloud generated by structured light or laser scanning; and provides data support for process improvement (such as adjusting the rolling pressure or coating process) by quantifying the three-dimensional morphological characteristics of the defects.

[0052] The data management module 4 is connected to the host computer and is a real-time database for storing test data (dimensional parameters, defect information, time, etc.); the data management module 4 is also provided with a quality traceability system, which generates a test report by binding the aluminum plate batch information through a QR code / RFID.

[0053] The automatic sorting module 5 includes a sorting mechanism for sorting the aluminum plates into qualified products, returned products, and scrapped products according to the detection results of the defect detection module 13. The automatic sorting module 5 is also designed with an alarm feedback system for pushing the defect type and location to the production line control terminal in real time.

[0054] This embodiment also provides a method for detecting aluminum plate size and surface defects based on machine vision, such as Figure 2 As shown, the following steps are included:

[0055] In the first step S1, the aluminum plate is transported by the mobile conveying mechanism 1, and the scanning mechanism 6 scans the QR code / RFID of the aluminum plate, and the information obtained after the scanning is transmitted to the host computer.

[0056] In the second step S2, after the scanning is completed, the laser rangefinder 7 is operated to measure the thickness of the aluminum plate, and at the same time, the multi-spectral light source system 10 of the image acquisition module 2 starts to work, and at the same time, the high-resolution camera 8 and the 3D structured light sensor 9 are operated to complete the 3D scanning and photographing of the aluminum plate. After the photographing, the image information obtained is transmitted to the image processing module 12 for image processing, and the processed information is transmitted to the size measurement module 14 and the defect detection module 12. When laser ranging is used, multiple groups of point data are measured and the average value is finally calculated.

[0057] In the third step S3, the size measurement module 14 obtains the size and flatness results of the aluminum plate after receiving the information.

[0058] Specifically, the size measurement module 14 uses an edge extraction algorithm to obtain the result. The edge extraction algorithm uses the Robert algorithm based on the first-order derivative: the gradient of the image is calculated using local differences, and the edge is detected by calculating the first-order differences of the image in the horizontal and vertical directions. Specifically, it uses two 2×2 templates to perform convolution operations on the image respectively to obtain the approximate gradient values ​​in the horizontal and vertical directions, and then calculates the square root of the sum of the squares of the two gradients to obtain the edge strength.

[0059] The calculation formula is:

[0060]

[0061] Among them G x (x,y) and G y (x, y) are the horizontal and vertical gradients of the image at (x, y), respectively.

[0062] Flatness measurement is based on the point group (x, y, z) collected on the measured surface, using the three-point plane formula A x +B y +C z +D=0, that is, the Chebyshev method fits into a reference plane, such as Figure 3 As shown, this plane is called the Chebyshev plane. Then use two planes parallel to the Chebyshev plane to "clamp" the collected point group. At this time, the area clamped by the two planes is the so-called minimum area. The width D of this area is the geometric error of the flatness of the measured part, that is, the measured value of the flatness.

[0063] Then any point P(x 0 ,y 0 ,z 0 ) to A x +B y +C z +D=0 distance is:

[0064]

[0065] Finally, the flatness is obtained by subtracting the minimum distance from the maximum distance.

[0066] In the fourth step S4, the defect measurement module 13 obtains the type, location and area of ​​the defect after receiving the information.

[0067] First, multi-scale image fusion is used to confirm the surface and internal defect types:

[0068] When there are scratches or pits on the surface of the aluminum plate, the reflection direction and intensity of light on the surface will change because these defects change the surface flatness. Scratches will cause diffuse reflection or scattering of light, which contrasts with the mirror reflection of the surrounding normal surface, and appears as a difference in brightness or grayscale in the image. Pits will appear as darker areas in the image due to light obstruction and multiple internal reflections, so the defect type is determined by the change in image brightness.

[0069] When near-infrared light is irradiated to the interface of media with different refractive indices, such as the interface between the oxide layer and the air, and the interface between the oxide layer and the aluminum plate, interference occurs. Since there is an optical path difference between the reflected light on the upper and lower surfaces of the oxide layer, when the optical path difference meets certain conditions, the two beams of reflected light will interfere constructively or destructively, thereby forming light and dark interference fringes in the near-infrared image. Changes in the thickness of the oxide layer will cause the optical path difference to change, and then change the spacing, shape, and intensity of the interference fringes. By analyzing these changes in the interference fringes, the thickness of the oxide layer can be calculated, and then whether the thickness of the oxide layer is abnormal can be detected.

[0070] When near-infrared light encounters defects such as bubbles or inclusions inside the aluminum plate, the light will scatter due to the difference in optical properties (such as refractive index, scattering coefficient, etc.) between the defects and the aluminum plate matrix. Bubbles will scatter light in all directions, resulting in changes in the light intensity distribution of the bubble area in the near-infrared image, which usually appears as a brighter or darker area than the surrounding area. Due to its own optical properties, the inclusions will also scatter and absorb the near-infrared light to varying degrees, causing it to present a different grayscale or color feature from the matrix in the image, so that it can be identified.

[0071] Specifically, through the deep learning model, defect classification (scratches, oxidation spots, coating peeling, etc.) and positioning (bounding box annotation) are further completed, and the defect type, location and confidence are output.

[0072] Specifically, the defect area is calculated through defect quantification analysis, and the projected area of ​​the defect in the two-dimensional image is calculated based on the segmentation network (such as U-Net) or pixel-level statistics. The actual physical size of the defect is estimated by combining multi-view images or 3D point cloud data, and the comprehensive impact of the defect on the overall quality of the aluminum plate is evaluated through spatial distribution statistics (such as defect spacing and area density).

[0073] In the fifth step S5, all the inspection data of the third and fourth steps are transmitted to the data management module 4 through the host computer, and are associated with the QR code of the aluminum plate to generate a quality inspection report.

[0074] In the sixth step S6, the information of the quality inspection report is uploaded to the host computer, which then transmits it to the control module 3, and finally instructs the automatic sorting module 5 to sort the aluminum plates into three categories: qualified products, returned products, and waste products.

[0075] Although the embodiments of the present invention are described in the specification, these embodiments are only for reference and should not limit the protection scope of the present invention. Various omissions, substitutions and changes within the scope of the present invention should be included in the protection scope of the present invention.

Claims

1. A visual inspection system based on aluminum plate size and surface defects, characterized in that: It includes a mobile conveying mechanism, an image acquisition module, a control module, a host computer, a data management module and an automatic sorting module. The host computer includes an algorithm module, an image processing module, a size measurement module and a defect detection module. The mobile conveying mechanism is used for continuous conveying of aluminum plates, on which a QR code / RFID is provided, and the mobile conveyor can obtain the QR code / RFID information and upload the information to the host computer; The image acquisition module is used to acquire the image of the aluminum plate to be tested, and includes a high-resolution camera, a multi-spectral light source system and a 3D structured light sensor, so as to obtain the three-dimensional geometric information of the aluminum plate; The control module is electrically connected to the mobile conveying mechanism, the image acquisition module, the host computer, the automatic sorting module, and the data management module; The image processing module is used to process the photos taken by the high-resolution camera and transmit the processed information to the size measurement module and the defect detection module; The dimension measurement module is used to calculate the length, width and edge straightness of the aluminum plate based on the information transmitted by the image processing module; The defect detection module is used to capture scratches, pits, and coating shedding on the surface of the aluminum plate. The defect detection module has a built-in deep learning model to complete defect classification and positioning; The defect detection module calculates the defect area, defect depth and defect distribution density by quantitatively analyzing the defects; The data management module is connected to the host computer and is a real-time database for storing detection data; The automated sorting module comprises a sorting mechanism for sorting the aluminum plates into qualified products, returned products, and scrap products according to the detection results of the defect detection module.

2. The visual inspection system based on aluminum plate size and appearance surface defects according to claim 1 is characterized by: The mobile conveying mechanism is provided with a code scanning mechanism and a laser rangefinder. The code scanning mechanism is arranged on the end side of the mobile conveying mechanism and is used to scan the QR code / RFID of the aluminum plate. The information obtained after scanning is transmitted to the host computer. The laser rangefinder is used to measure the thickness, width and length of the aluminum plate.

3. The visual inspection system based on aluminum plate size and appearance surface defects according to claim 1 is characterized by: The high-resolution camera continuously scans the conveyed aluminum plate to obtain images, the multi-spectral light source system uses a ring-shaped LED light source and a polarized light combination to suppress the reflection interference on the surface of the aluminum plate, and the 3D structured light sensor is arranged next to the high-resolution camera.

4. The visual inspection system based on aluminum plate size and appearance surface defects according to claim 1 is characterized in that: The size measurement module adopts an edge extraction algorithm and performs calculations through sub-pixel edge detection technology to improve the accuracy of edge detection, and fits the reference plane through 3D point cloud data to calculate the flatness of the aluminum plate, that is, the deviation between the surface of the aluminum plate and the reference plane.

5. The visual inspection system based on aluminum plate size and appearance surface defects according to claim 1 is characterized by: The defect detection module uses multi-scale image fusion to enhance the global contrast and local detail features of defects, and uses industrial-grade defect datasets for transfer learning, combined with data enhancement to improve model generalization.

6. The visual inspection system based on aluminum plate size and appearance surface defects according to claim 1 is characterized by: The data management module also includes a quality traceability system for obtaining data of a QR code bound to the batch information of the aluminum plate to generate a test report.

7. The visual inspection system based on aluminum plate size and appearance surface defects according to claim 1 is characterized by: The automated sorting module is also designed with an alarm feedback system for pushing defect types and locations to the production line control terminal in real time.

8. A method for detecting aluminum plate size and surface defects based on machine vision, characterized in that: The following steps are involved: In the first step, the aluminum plate is transported by a mobile conveying mechanism, and the scanning mechanism scans the QR code / RFID of the aluminum plate, and the information obtained after scanning is transmitted to the host computer; In the second step, after the scanning is completed, the laser rangefinder is operated to measure the thickness of the aluminum plate. At the same time, the multi-spectral light source system of the image acquisition module starts working, and the high-resolution camera and 3D structured light sensor are operated at the same time to complete the 3D scanning and taking pictures of the aluminum plate. After taking pictures, the image information obtained is transmitted to the image processing module for image processing, and the processed information is transmitted to the size measurement module and the defect detection module; In the third step, the dimension measurement module obtains the dimension and flatness results of the aluminum plate after receiving the information; In the fourth step, the defect measurement module obtains the type, location and area of ​​the defect after receiving the information; Step 5: All the inspection data from Step 3 and Step 4 are transmitted to the data management module through the host computer, and are associated with the QR code of the aluminum plate to generate a quality inspection report; The sixth step is to upload the information of the quality inspection report to the host computer, which then transmits it to the control module, and finally instructs the automated sorting module to sort the aluminum plates into three categories: qualified products, returned products, and scrap products.

9. The method for detecting aluminum plate size and surface defects based on machine vision according to claim 8, characterized in that: When laser ranging is used in the second step, multiple groups of point data are measured and the average value is finally calculated.

10. The method for detecting aluminum plate size and surface defects based on machine vision according to claim 8, characterized in that: In the fourth step of defect measurement, multi-scale image fusion is used to confirm the surface and internal defect types, the defect classification and positioning are further completed through the deep learning model, and the defect area is calculated through defect quantification analysis.

Images