Photovoltaic cell screen comprehensive detection device

By designing a comprehensive detection device for photovoltaic cell network versions that integrate multiple measurement stations, the problems of low quality detection efficiency and inconsistent standards of the network version are solved, and the precise positioning and automated detection of the network version are realized, which improves detection efficiency and data management. convenience.

CN117054446BActive Publication Date: 2024-11-05MAI MU ZHI NENG KE JI (NAN JING) YOU XIAN GONG SI

Patent Information

Application Number
CN202311027048.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-08-16
Publication Date
2024-11-05
Estimated Expiration
2043-08-16

AI Technical Summary

Technical Problem

In the prior art, screen quality inspection relies on manual inspection, low efficiency and inconsistent standards, making it difficult to meet the needs of large-scale production and manufacturing, especially in the stages of thin film measurement, geometric measurement and defect identification and positioning, lacking automation, standardization and integration. detection method.

Method used

A photovoltaic cell screen integrated detection device was designed, integrating multiple measurement stations, including stages, tension and thickness measurement stations, dimension measurement stations and defect detection stations, and using thickness sensors, tension sensors, and detection The camera and preset marker image matching analysis algorithm and defect detection model realize accurate positioning and multi-station detection of the screen.

Benefits of technology

It improves the efficiency of screen quality inspection, realizes accurate positioning and calibration of screens on different workstations, replaces manual inspection, facilitates production data management, and improves the automation and standardization level of inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117054446B_ABST
    Figure CN117054446B_ABST
Patent Text Reader

Abstract

The present invention discloses a photovoltaic cell screen comprehensive detection device, which belongs to the field of quality detection. The device integrates multiple measurement stations, can replace manual detection, improve the efficiency of solar screen quality detection, and facilitate production data management. The screen is positioned in film measurement, tension measurement, screen geometry measurement and defect identification, and ensures accurate positioning and calibration of its own position during large-scale movement on a translation stage.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of quality detection, and in particular to a photovoltaic cell sheet screen comprehensive detection device. Background Art

[0002] The screen is the core equipment of screen printing. It is often used to make templates for solar cells and other components. It is widely used in the field of electronic component manufacturing. Since the screen is the reference template for printing, its quality directly determines the quality of user product production. In the large-scale production process, screen quality defects will lead to batch product quality problems, causing great economic and time cost losses to manufacturers. Therefore, early screen quality inspection is very important.

[0003] At present, manual inspection is commonly used for the inspection of screen quality, such as: using mechanical thickness gauge to inspect screen thickness, using handheld tension gauge to inspect screen tension, using two-dimensional equipment to inspect screen geometric parameters, and inspecting screen defects by naked eyes. All of the above inspections need to be completed by workers at different workstations, with low inspection efficiency, inconsistent standards, and greatly affected by human factors, making it difficult to meet the production and manufacturing requirements of the screen manufacturing industry. Therefore, the screen manufacturing industry is in urgent need of automated, standardized, and integrated screen quality inspection methods and equipment. Summary of the invention

[0004] The present invention provides a photovoltaic cell sheet screen comprehensive detection device, which solves the problem of accurate positioning and calibration of the screen's own position at different workstations during the large-scale movement of the screen on a translation stage in the film measurement, screen geometry measurement and defect identification and positioning stages.

[0005] The embodiment of the present invention provides a photovoltaic cell sheet screen comprehensive detection device, comprising:

[0006] Integrated machine body;

[0007] The stage is used to place the screen to be tested and to move the screen to be tested through the motion axis;

[0008] A tension and thickness measuring station is arranged on the integrated machine, wherein a thickness sensor and a tension sensor are arranged on the tension and thickness measuring station to measure the tension and thickness of the screen to be tested;

[0009] A dimension measuring station is arranged on the integrated machine, wherein a first detection camera is arranged on the dimension measuring station, and a first detection image of the screen to be tested is collected by the first detection camera;

[0010] A defect detection station is arranged on the integrated machine, wherein the defect detection station is provided with a second detection camera, and a second detection image of the screen to be tested is collected by the second detection camera;

[0011] A processing module, wherein the processing module is connected to the thickness sensor, the tension sensor, the first detection camera and the second detection camera, and is used to obtain the detection results of the thickness sensor and the tension sensor, and to control the movement of the stage so that the screen to be tested enters the tension and thickness detection station, the size measurement station and the defect detection station respectively, and uses a preset landmark point image matching analysis algorithm to detect the first detection image to obtain the geometric dimensions of the grid lines of the screen to be tested, and uses a pre-trained defect detection model to detect the second detection image to obtain the defect detection result of the screen to be tested.

[0012] In one embodiment of the present invention, the stage is a glass plate, and the device further comprises:

[0013] A lower light source, the lower light source is arranged below the glass plate;

[0014] An upper light source is disposed at the positions of the first detection camera and the second detection camera respectively, and is used to enhance the light brightness when the first detection camera and the second detection camera collect the image of the screen to be tested.

[0015] In one embodiment of the present invention, it also includes:

[0016] A plurality of driving motors are used to control the movement of the stage, the thickness sensor, the tension sensor, the first detection camera and / or the second detection camera.

[0017] In one embodiment of the present invention, the device further comprises:

[0018] An alarm device is provided on the integrated machine body and is used for sounding an alarm when an alarm condition is met.

[0019] In one embodiment of the present invention, the alarm device is an audio / video alarm module, which is used to emit sound and / or light to alarm according to the current alarm situation when the alarm condition is met.

[0020] In one embodiment of the present invention, detecting the first detection image using a preset landmark point image matching analysis algorithm includes:

[0021] Dividing the plurality of first detection images into a plurality of image sequences, and making each image sequence equivalent to a standard halftone image;

[0022] Comparing a plurality of standard halftone images with a reference halftone image to obtain pixel coordinates of a plurality of positioning points in the halftone image to be tested;

[0023] The pixel coordinates of the plurality of positioning points are converted into world coordinates according to a preset coordinate conversion relationship to obtain the current position information of the screen to be tested;

[0024] Based on the current position information of the screen to be tested, the screen to be tested is moved to the detection positions of the thickness sensor and the tension sensor to obtain the thickness and tension of the screen to be tested;

[0025] Based on the current position information of the screen to be tested, the screen to be tested is moved to the size measurement station, and the coordinates of the screen to be tested relative to the world coordinate system after being moved from the tension and thickness measurement station to the size measurement station are obtained by positioning the positioning points of the screen to be tested multiple times and converting the pixel coordinates of the positioning points into world coordinates multiple times;

[0026] Based on the current position information of the screen to be tested, the stage is moved to four preset positions, and the width of the grid lines of the screen to be tested in the first detection image of each preset position is identified;

[0027] Control the stage to move to the position of the uppermost grid line of the screen to be tested, scan from the uppermost grid line to the lowermost grid line, measure the distance between the uppermost grid line and the lowermost grid line, and obtain the geometrical dimensions of the grid lines of the screen to be tested;

[0028] Controlling the stage to move to the defect detection station, positioning the positioning points of the screen to be tested multiple times, and converting the pixel coordinates of the positioning points into world coordinates multiple times, to obtain the coordinates of the screen to be tested relative to the world coordinate system after it is moved from the size measurement station to the defect detection station;

[0029] The screen to be tested is divided into a plurality of areas, and the second test image is tested using a preset defect position detection algorithm to determine whether the screen to be tested has defects and the defect positions.

[0030] In one embodiment of the present invention, the step of scanning from the uppermost grid line to the lowermost grid line, measuring the distance between the uppermost grid line and the lowermost grid line, and obtaining the geometric dimensions of the grid line of the screen to be tested includes:

[0031] a) moving the stage so that the screen to be tested enters the measurement field 1, obtaining the pixel coordinates y11 of the position of the uppermost grid line and the pixel coordinates y12 of the position of the lowermost grid line according to the collected first detection image analysis, converting y11 and y12 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate y1 of the current stage;

[0032] b) moving the stage downward, and the movement amount is smaller than the measurement area, obtaining the pixel coordinates y21 of the position of the uppermost grid line and the pixel coordinates y22 of the position of the lowermost grid line according to the collected first detection image analysis, converting y21 and y22 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate mark y2 of the current stage;

[0033] c) calculating the movement amount of the stage Δd1=y2-y1, calculating the translation amount of the two first detection images Δd1'=y21-y12, and when the difference between Δd1 and Δd1' is less than the set accuracy threshold, executing step d), otherwise, returning to step a) and moving again;

[0034] d) Repeating step b) and step c) until the bottom grid line is observed in the measurement area, obtaining the position pixel coordinates yn1 of the top grid line and the position pixel coordinates yn2 of the bottom grid line according to the collected first detection image analysis, converting yn1 and yn2 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate mark yn of the current stage, and calculating the movement amount Δdn=yn-yn-1 of the stage;

[0035] e) Calculate the geometric dimensions of the grid lines of the screen to be tested:

[0036]

[0037] In one embodiment of the present invention, when the second inspection image of the screen to be tested is collected by the second inspection camera, multiple screen images to be tested are collected by staggered method and then fused and calculated to obtain a super-resolution second inspection image, wherein optical frequency comb technology is used to accurately measure and locate the staggered information during the staggered collection.

[0038] In one embodiment of the present invention, before using a pre-trained defect detection model to detect the second detection image to obtain the defect detection result of the screen to be tested, the method further includes:

[0039] Acquire defective and non-defective screen images, crop and downsample the defective and non-defective screen images respectively, automatically enhance the defective and non-defective screen images, and mark the defective screen image;

[0040] A defect detection model is constructed, and the defect detection model is trained using the defective and non-defective screen images and their labels until the loss function of the defect detection model meets the training termination condition, thereby obtaining a trained defect detection model, wherein the input of the defect detection model is the screen image, and the output is the defect position and defect type of the screen image.

[0041] In one embodiment of the present invention, the defect detection model includes an Input input end, a Backbone network, a Neck feature fusion layer and a Head detection head, wherein the Input input end adopts Mosaic data enhancement technology to merge multiple screen images into one screen image; the Backbone network includes an improved C3 module and an SPPF module, the C3 module is a residual structure for feature extraction; the Neck feature fusion layer adopts an FPN feature pyramid and a PAN path aggregation network structure, comprehensively utilizes feature information of multiple scales, and fuses the features extracted by the Backbone network; the Head detection head parses the feature maps of different scales output by the Neck feature fusion layer, divides the feature maps of each scale into grid units, sets multiple prior boxes for each grid cell to predict multiple bounding boxes, and eliminates bounding boxes with overlap greater than a preset value by a non-maximum suppression method to obtain the defect position and defect type of the screen.

[0042] The photovoltaic cell screen comprehensive detection device of the embodiment of the present invention integrates multiple measurement stations, can replace manual detection, improve the efficiency of solar screen quality detection, and facilitate production data management. It solves the problem of accurate positioning and calibration of the screen's own position at different stations during the large-scale movement of the screen on the translation stage during the film measurement, screen geometry measurement and defect identification and positioning stages.

[0043] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0045] Figure 1 A schematic diagram of a framework of a photovoltaic cell sheet screen comprehensive detection device provided according to an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the structure of a photovoltaic cell sheet screen comprehensive detection device provided according to an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of a photovoltaic cell sheet screen comprehensive detection device provided according to an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of positioning marking points of a screen to be tested according to an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of a deformation continuity constraint provided according to an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of digital correlation of time series marking points provided according to an embodiment of the present invention;

[0051] Figure 7 A schematic diagram of a world coordinate system provided according to an embodiment of the present invention;

[0052] Figure 8 A schematic diagram of a screen provided according to an embodiment of the present invention in world coordinates;

[0053] Fig. 9 A schematic diagram of super-resolution optical imaging according to an embodiment of the present invention;

[0054] Fig.10 A schematic diagram of a moving step length provided according to an embodiment of the present invention;

[0055] Fig.11 The figure is a schematic diagram of a process for detecting a screen to be tested according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0057] Figure 1 The present invention is a schematic diagram of a photovoltaic cell sheet screen comprehensive detection device provided according to an embodiment of the present invention.

[0058] like Figure 1 As shown, the photovoltaic cell sheet screen comprehensive detection device includes:

[0059] Integrated machine body 10;

[0060] Stage( Figure 1 ), used to place the screen to be tested, and to drive the screen to be tested to move through the motion axis, wherein a pneumatic clamping device is also provided on the stage for fixing the screen to be tested;

[0061] A tension and thickness measuring station 101 is provided on the integrated machine 10, and a thickness sensor and a tension sensor are provided on the tension and thickness measuring station 101 for measuring the tension and thickness of the screen to be tested;

[0062] A dimension measuring station 102 is provided on the integrated machine, and a first detection camera is provided on the dimension measuring station 102, and a first detection image of the screen to be tested is collected by the first detection camera;

[0063] A defect detection station 103 is provided on the integrated machine, and the defect detection station 103 is provided with a second detection camera, and a second detection image of the screen to be tested is collected by the second detection camera;

[0064] The processing module 20 is connected to the thickness sensor, the tension sensor, the first detection camera and the second detection camera, and is used to obtain the detection results of the thickness sensor and the tension sensor, and control the movement of the stage so that the screen to be tested enters the tension and thickness detection station 101, the size measurement station 102 and the defect detection station 103 respectively, and uses a preset landmark point image matching analysis algorithm to detect the first detection image to obtain the geometric dimensions of the grid lines of the screen to be tested, and uses a pre-trained defect detection model to detect the second detection image to obtain the defect detection result of the screen to be tested.

[0065] like Figure 2 and Figure 3 As shown, a high-precision motion axis is provided on the integrated machine body 10, and the thickness sensor, tension sensor, first detection camera and second detection camera can be provided on the high-precision motion axis, and can realize movement in the X, Y and Z directions, so as to adjust the relative position with the screen to be tested. In addition, the stage is also provided on the high-precision motion axis, and the stage is transported by the motion axis together with the screen to be tested to the tension and thickness detection station, the size measurement station and the defect detection station.

[0066] In an embodiment of the present invention, the photovoltaic cell screen integrated detection device also includes multiple drive motors, which can drive the components on the high-precision motion axis to move, wherein one drive motor can drive one component or multiple components, and there is no specific limitation on this.

[0067] In an embodiment of the present invention, the photovoltaic cell screen comprehensive detection device also includes an illumination light source, wherein the illumination light source is divided into an upper light source and a lower light source. When the stage is transparent, the lower light source is arranged under the glass plate, and the upper light source is a ring light, which is respectively arranged at the positions of the first detection camera and the second detection camera, and is used to enhance the light brightness when the first detection camera and the second detection camera collect the image of the screen to be tested. In actual work, the lower light source backlight mode is mainly used.

[0068] Furthermore, in order to realize data transmission, the integrated machine body is also provided with multiple IO interfaces, through which the detection information of multiple stations is sent to the processing module. In the actual implementation process, multiple IOs can also be arranged at switch buttons, detection product presence or absence sensing, alarm, etc.

[0069] In an embodiment of the present invention, an alarm device is also provided on the integrated machine, which is used to sound an alarm when an alarm condition is met. For example, when the stage is stuck during movement, or a sensor fails, or a camera fails, the alarm module can sound an alarm to remind the user to handle it in time.

[0070] Specifically, the alarm module can be an audio / visual alarm module, which emits sound and / or light to alarm according to the current alarm situation when the alarm condition is met. It should be noted that the alarm can be issued only by sound or light, or by a combination of sound and light, without limitation, and can be adjusted according to actual conditions by those skilled in the art.

[0071] In addition to the alarm device, the photovoltaic cell sheet and screen comprehensive detection device of the embodiment of the present invention is also provided with a double-button pneumatic and emergency stop button, etc., to ensure the safety of the detection process.

[0072] In the above embodiment, both the dimension measurement station and the defect detection station are provided with detection cameras, and two detection cameras of different models may be selected according to different functions to be implemented.

[0073] In the embodiment of the present invention, the processing module can realize the control of the entire photovoltaic cell screen integrated detection device, including but not limited to controlling the movement of each component and processing the data. The processing module can be Figure 3 The display screen shown has multiple algorithms integrated inside.

[0074] The solar screen quality comprehensive detection solution proposed in the embodiment of the present invention integrates multiple measurement stations, can replace manual detection, improve the efficiency of solar screen quality detection, and facilitate production data management. It solves the problem of accurate positioning and calibration of the screen's own position at different stations during the large-scale movement of the screen on the translation stage during the film measurement, screen geometry measurement and defect identification and positioning stages.

[0075] In an embodiment of the present invention, detecting the first detection image using a preset landmark point image matching analysis algorithm includes:

[0076] Dividing the plurality of first detection images into a plurality of image sequences, and making each image sequence equivalent to a standard halftone image;

[0077] Compare multiple standard halftone images with reference halftone images to obtain pixel coordinates of multiple positioning points in the halftone image to be tested;

[0078] According to the preset coordinate conversion relationship, the pixel coordinates of multiple positioning points are converted into world coordinates to obtain the current position information of the screen to be tested;

[0079] Based on the current position information of the screen to be tested, the screen to be tested is moved to the detection position of the thickness sensor and the tension sensor to obtain the thickness and tension of the screen to be tested;

[0080] The screen to be tested is moved to the dimension measurement station based on the current position information of the screen to be tested, and the coordinates of the screen to be tested relative to the world coordinate system after being moved from the tension and thickness measurement station to the dimension measurement station are obtained by positioning the positioning points of the screen to be tested multiple times and converting the pixel coordinates of the positioning points into the world coordinates multiple times;

[0081] The stage is moved to four preset positions based on the current position information of the screen to be tested, and the width of the grid lines of the screen to be tested in the first detection image of each preset position is identified;

[0082] Control the stage to move to the position of the uppermost grid line of the screen to be tested, scan from the uppermost grid line to the lowermost grid line, measure the distance between the uppermost grid line and the lowermost grid line, and obtain the geometric size (PT value) of the grid line of the screen to be tested;

[0083] Control the stage to move to the defect detection station, locate the positioning points of the screen to be tested multiple times, and convert the pixel coordinates of the positioning points and the world coordinates multiple times to obtain the coordinates of the screen to be tested relative to the world coordinate system after it moves from the size measurement station to the defect detection station;

[0084] The screen to be tested is divided into a plurality of areas, and a preset defect position detection algorithm is used to detect the second detection image to determine whether the screen to be tested has defects and the defect positions.

[0085] The process is described in detail below through a specific embodiment.

[0086] Step 1: Place the screen to be tested on the stage and fix it to ensure that the four marking points on the screen to be tested for positioning are not blocked.

[0087] Step 2: The stage carries the screen to be tested into the tension and thickness detection station, uses the first detection camera to capture the screen image, and identifies the pixel coordinates of the four positioning points such as Figure 4 The embodiment of the present invention introduces deformation continuity constraints in the process of landmark image matching analysis to improve the landmark positioning accuracy. The principle is shown in FIG. Figure 5 and Figure 6 As shown in the figure, the basic idea is to use a shorter image sequence to produce a high-quality image to compare with the reference image, so as to more accurately identify the position of the landmark image.

[0088] The image positioning analysis equation with time continuity introduced is as follows:

[0089]

[0090] Among them, f(x) represents the gray value of the reference image, g t Represents the grayscale value of the deformed image. In fact, g t is a special “average” of the 2m+1 deformed images, so the equation on the right is valid. If U(X,t)=0, that is, the displacement between the 2m+1 images and the reference image is zero, then these average equivalent images G i The noise of (X+0) is reduced If U(X,t)≠0, that is, there is a displacement between the equivalent image and the reference image, by compensating for the displacement, it can still be regarded as the average of the original 2m+1 images. Then, these average equivalent images G i The noise of (X+U(X,t)) is also reduced Therefore, it can be inferred that the averaged image G i The quality of (X+U(X,t)) is higher than that of the original image.

[0091] Considering that solid deformation is continuous in time, the displacement function U(X,t) of the marker point is a continuous function of time. Performing a Taylor expansion on the function U(X,t), it can be expressed as:

[0092]

[0093] Among them, Δt is the time interval between two images.

[0094] For the digital image correlation of a marker point, the deformation of the marker point itself is not considered, only the rigid body displacement of the marker point is considered. Therefore, its displacement can be simplified as:

[0095]

[0096] in,

[0097]

[0098] Finally, the grayscale invariant assumption is used to express the landmark digital correlation matching algorithm considering time continuity, which can be expressed as:

[0099]

[0100] This is similar to the solution process of traditional DIC, and the Newton-Raphson method can be used to solve α in this equation.

[0101] Through optimization solution, μ and v, i.e., the lateral and longitudinal displacement values ​​of the marked point, are obtained.

[0102] Step 3: According to the pre-calibrated imaging parameters, convert this coordinate into the world coordinate system (the coordinate system that controls the movement of the detection system) to determine the current position of the screen. First, define several coordinate systems:

[0103] The coordinate system of the stage (translation stage) is the world coordinate system, and the coordinate system of the camera on each inspection station is the local coordinate system, that is, the local coordinate system of the screen itself, such as Figure 7 When the inspection starts, the stage moves to the zero point of the world coordinate system, which is Figure 7 After that, the stage is controlled to move to the tension and thickness detection station, and the pixel coordinates (x k ,y k ) is the local coordinate, and its coordinate relative to the world coordinate system can be expressed as:

[0104]

[0105] Among them, d 1 x and d 1y is the amount of stage movement. In this way, the conversion relationship between the tension and thickness detection station and the world coordinate system can be established. Because the thickness sensor and tension sensor are installed based on the position of the world coordinate system, the position of the thickness sensor and tension sensor relative to the screen itself is also obtained.

[0106] Step 4: Give a moving instruction according to the position in step 3 to move the screen to the thickness detection position, and use an optical thickness measuring device to complete the thickness measurement (this step can be repeated multiple times).

[0107] Step 5: Give a moving instruction according to the position in step 3 to move the screen to the tension detection position, and use the tension measuring device to complete the tension measurement (this step can be repeated multiple times).

[0108] Step 6: Give a movement command based on the position in step 3 to move the screen to the geometric parameter detection station. Repeat steps 2 and 3 to obtain the coordinates of the screen relative to the world coordinate system after completing a large-scale movement.

[0109] Step 7: Give movement instructions according to the position in step 3, and control the translation stage to move to four preset positions. Collect an image at each position, identify the width of the grid lines in the image and record this data.

[0110] In this inspection condition, in order to ensure the object plane resolution of the measurement, the first inspection camera needs to divide the screen into several areas (such as 9 areas). In this process, it is necessary to ensure that the observation area meets the preset requirements. The preset position of the observation area is the relative position relative to the screen, that is, the known Figure 8If you want to move the screen to a preset position, you need to transform this coordinate into world coordinate. The method can be referred to step 3.

[0111] Step 8: Control the stage to move to the top grid line position of the screen to be tested, scan from top to bottom, and measure the distance between the top and bottom grid lines, which is the PT value data. The specific method of calculating the PH value data is introduced in the following steps.

[0112] Step 9: Control the stage to move the screen to be tested to the defect detection station. Repeat steps 2 and 3 to obtain the coordinates of the screen to be tested relative to the world coordinate system after completing the large-scale movement.

[0113] Step 10: Divide the screen to be tested into several small areas, collect images of each small area with an ultra-high resolution digital camera, and determine whether the image of the area has defects and the defect location based on the deep learning algorithm, and record them. The specific steps of the defect location detection method of dislocation super-resolution imaging and deep learning are introduced in the following steps.

[0114] In step 8, 2. the PT value detection method of image tracking and displacement sensing data dual verification includes the following steps:

[0115] a) moving the stage so that the screen to be tested enters the measurement field of view 1, obtaining the pixel coordinates y11 of the uppermost grid line and y12 of the lowermost grid line according to the acquired first detection image analysis, converting y11 and y12 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate y1 of the current stage;

[0116] b) moving the stage downward, and the movement amount is smaller than the measurement area, obtaining the pixel coordinates y21 of the position of the uppermost grid line and the pixel coordinates y22 of the position of the lowermost grid line according to the acquired first detection image analysis, converting y21 and y22 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate mark y2 of the current stage;

[0117] c) calculating the movement amount of the stage Δd1=y2-y1, calculating the translation amount of the two first detection images Δd1'=y21-y12, and when the difference between Δd1 and Δd1' is less than the set accuracy threshold, executing step d), otherwise, returning to step a) and moving again;

[0118] d) Repeat step b) and step c) until the bottom grid line is observed in the measurement area, obtain the position pixel coordinates yn1 of the top grid line and the position pixel coordinates yn2 of the bottom grid line according to the acquired first detection image analysis, convert yn1 and yn2 into world coordinates according to a preset coordinate conversion relationship, record the world coordinate mark yn of the current stage, and calculate the movement amount Δdn=yn-yn-1 of the stage;

[0119] e) Calculate the geometric dimensions of the grid lines to be tested:

[0120]

[0121] In step 10, the screen defect detection in the present invention relies on super-resolution optical imaging. Super-resolution imaging uses staggered acquisition of multiple images and then performs image fusion calculation to obtain a super-resolution image. Staggered acquisition relies on high-precision positioning control. The stage high-precision positioning uses optical frequency comb technology to accurately measure and locate the staggered information. The basic measurement principle is as follows: Fig. 9 shown.

[0122] The repetition rate of the ultrafast laser is locked to the atomic clock; the reference pulse and measurement pulse are perpendicular to each other; the reference pulse and measurement pulse returned to the PBS enter the BCC system; the servo controller controls the laser cavity length based on the BCC signal feedback, changes the pulse sequence interval, and locks the BCC signal to zero - at this time, the flight time of the measurement pulse is locked to an integer multiple of the pulse interval.

[0123] Assuming the repetition frequency of the laser is fr, the time interval of its output pulse sequence is t r , expressed as:

[0124] l is the optical path in the laser cavity, c is the speed of light

[0125] The round-trip flight time T of the pulse in the measured distance rt for:

[0126] L is the distance to be measured, n is the refractive index of the space medium

[0127] After the laser phase is locked (the BCC signal is locked to zero), the round trip time of the pulse to be measured and the time interval of the pulse sequence satisfy:

[0128] T rt =m·t r m is a positive integer

[0129] Thus:

[0130] In the formula, m and f n One-to-one relationship

[0131] Let the two adjacent repetition frequencies that can achieve phase locking be f r1 and f r2 ,but:

[0132]

[0133] Move the stage so that the shooting area of ​​the ultra-high resolution digital camera is located at the upper left corner of the screen to be tested, and then move the stage in a small step size to collect images. The moving step size should be sub-pixel size, for example, less than the actual physical size corresponding to 1 pixel. The collected images are fused, and the image resolution will be increased by n times after fusion, where n is determined by the number of steps moved within a pixel. Fig.10 As shown in the figure, A, B, C, and D represent the four-times misalignment imaging process, and H represents the high-resolution image reconstructed after four times of misalignment imaging. The image resolution can be increased by 4 times. The mathematical expression of the super-resolution process can be expressed as follows:

[0134]

[0135] In one embodiment of the present invention, before using the pre-trained defect detection model to detect the second detection image to obtain the defect detection result of the screen to be tested, the method further includes:

[0136] Acquire defective and non-defective screen images, crop and downsample the defective and non-defective screen images respectively, automatically enhance the defective and non-defective screen images, and mark the defective screen image;

[0137] A defect detection model is constructed, and defective and non-defective screen images and their labels are used to train the defect detection model until the loss function of the defect detection model meets the training termination condition, thereby obtaining a trained defect detection model, wherein the input of the defect detection model is the screen image, and the output is the defect position and defect type of the screen image.

[0138] In one embodiment of the present invention, a defect detection model includes an Input input end, a Backbone network, a Neck feature fusion layer and a Head detection head, wherein the Input input end adopts Mosaic data enhancement technology to merge multiple screen images into one screen image; the Backbone network includes an improved C3 module and an SPPF module, the C3 module is a residual structure for feature extraction; the Neck feature fusion layer adopts an FPN feature pyramid and a PAN path aggregation network structure, comprehensively utilizes feature information of multiple scales, and fuses the features extracted by the Backbone network; the Head detection head parses the feature maps of different scales output by the Neck feature fusion layer, divides the feature maps of each scale into grid units, sets multiple prior boxes for each grid cell to predict multiple bounding boxes, and eliminates bounding boxes with overlap greater than a preset value by a non-maximum suppression method to obtain the defect position and defect type of the screen.

[0139] Specifically, the defect recognition method based on the deep learning framework includes the following steps:

[0140] Dataset preparation: The dataset used in this method contains defective and normal screen images, where the defect types are various, including fiber filaments, hole blockage, wire abnormality, gap, pinhole, dirt, etc. The original image resolution is 15000 pixels × 10000 pixels, and the minimum defect size to be detected is required to be around 5 pixels × 5 pixels.

[0141] In the process of making the dataset, the original high-resolution images cannot be directly used for model training, so a cropping and downsampling strategy is adopted. First, the original image is cropped into 512 low-resolution images with a size of 640 pixels × 640 pixels, and then added to the dataset.

[0142] Using YOLOv5's AutoAugment effectively enriches the data set and introduces more data changes during the training process, making the model better adapted to target detection tasks in various scenarios, helping to improve the robustness and generalization ability of the model, and improving the accuracy and reliability of defect detection.

[0143] LabelImg annotation is used, following the VOC2007 dataset format. K-means++ algorithm is used to cluster the prior box. In the initialization stage, a target box with a larger area is selected as the first cluster center, and then the cosine similarity metric is used to gradually select the target box with the smallest similarity to the current cluster center as the new cluster center. Subsequently, in the iterative process, the target box with the smallest similarity metric to the current cluster center is selected as the new cluster center to better capture the defect characteristics of each cluster, thereby generating a prior box size that is more suitable for the defect size and shape.

[0144] In addition, in view of the characteristics of small-sized defects, the three loss functions of YOLOv5 have been modified to better adapt to the detection of small targets. Specifically, the weights and calculation methods of positioning loss, confidence loss, and classification loss have been adjusted to improve the model's detection performance for small-sized defects, making it more accurate and robust.

[0145] Model training: Considering the real-time detection of this task and the complexity of defect detection, YOLOv5m was selected as the defect detection model for training. The YOLOv5m model consists of four main parts: Input, Backbone network, Neck network layer and Head detection head.

[0146] The input uses Mosaic data enhancement technology, randomly selecting 4 pictures from the same batch of samples, and cropping, scaling and merging them into one picture, thereby enriching the data set and improving the training speed; the Backbone network includes an improved C3 module and an SPPF module. The improved C3 module uses a residual structure to solve the gradient explosion and gradient disappearance problems in back propagation; the Neck feature fusion layer mainly fuses the features extracted by the backbone network, adopts the structure of the FPN feature pyramid and the PAN path aggregation network, and comprehensively utilizes feature information at multiple scales; the Head detection head parses the feature maps of different scales output by the feature fusion layer, and trains them through loss functions and non-maximum suppression. Finally, 4 feature map outputs of different scales y1, y2, y3, and y4 are obtained to achieve multi-scale prediction.

[0147] For each input image, it is divided into grid cells on its feature map at each scale, and 3 prior boxes are set for each cell to predict 3 bounding boxes. The output dimension is S×S×(3×(4+1+M)), where S is the number of grid cells, and 4+1+M is the dimension of bounding box related information, representing bounding box, defect type, and defect type, respectively. Finally, ceramic substrate defect detection is achieved.

[0148] The loss function during training consists of three parts: bounding box error, confidence error, and category error. The bounding box error uses the mean square error, and the confidence error and category error use the cross entropy error. The feature map is divided into S×S cells, each cell predicts 3 bounding boxes, and the loss function is:

[0149] LOSS=LOSS reg +LOSS conf +LOSS cls

[0150]

[0151]

[0152]

[0153] Among them, LOSS is the total loss function, LOSS reg is the positioning loss function, LOSS conf is the confidence loss function, LOSS cls is the classification loss function. Indicates that there is an object in the jth prediction box in the i-th grid in the prediction output; S 2 b represents that there are S×S cells on each feature layer, B=3, and there are 3 prediction boxes in each cell; b ij , is the predicted box and the real box; IoU, ρ 2 、c 2 They represent the intersection-and-union ratio of two boxes, the Euclidean distance of the center points, and the diagonal distance of the minimum closure area; v is the consistency of the aspect ratio; C ij is the confidence level; α, λ noobj is the weight coefficient; represents that there is no target in the jth prediction box in the i-th grid; P ij (c) represents the probability value of the target belonging to each category.

[0154] According to the above-mentioned data set, model, and loss function, the defect detection model of the present invention can be trained.

[0155] Defect detection: For an original 15000 pixels × 10000 pixels image to be inspected, it is first cropped into 530 images of equal size of 640 pixels × 640 pixels, and they are simultaneously input into the model for defect detection.

[0156] After the model completes the detection, it is necessary to perform post-processing methods such as size filtering, bounding box filtering, spatial relationship analysis filtering, etc. to further filter and optimize the results. Size filtering is to exclude too small or too large defect areas by setting an appropriate threshold to reduce the false detection rate; bounding box filtering is to use a bounding box to limit the defect area, and judge whether the defect is reasonable based on the overlap between the bounding box and the image boundary, thereby further reducing false detection; spatial relationship analysis filtering is to use the spatial relationship between defect areas, such as distance, relative position, etc., to judge the rationality of the defect and exclude those defect areas that do not conform to the actual situation or are incoherent in space.

[0157] like Fig.11 As shown, the detection process of the network board to be tested during the actual execution process is demonstrated.

[0158] The photovoltaic cell screen comprehensive detection device proposed in the embodiment of the present invention integrates multiple measurement stations, can replace manual detection, improve the efficiency of solar screen quality detection, and facilitate production data management. It solves the problem of accurate positioning and calibration of the screen's own position at different stations during the large-scale movement of the screen on the translation stage during the film measurement, screen geometry measurement and defect identification and positioning stages.

[0159] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0160] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

Claims

1. A photovoltaic cell screen comprehensive detection device, characterized in that: include: Integrated machine body; The stage is used to place the screen to be tested and to move the screen to be tested through the motion axis; A tension and thickness measuring station is arranged on the integrated machine, wherein a thickness sensor and a tension sensor are arranged on the tension and thickness measuring station to measure the tension and thickness of the screen to be tested; A dimension measuring station is arranged on the integrated machine, wherein a first detection camera is arranged on the dimension measuring station, and a first detection image of the screen to be tested is collected by the first detection camera; A defect detection station is arranged on the integrated machine, and the defect detection station is provided with a second detection camera, and a second detection image of the screen to be tested is collected by the second detection camera, wherein when the second detection image of the screen to be tested is collected by the second detection camera, multiple screen images to be tested are collected by staggered manner, and then the multiple screen images are fused and calculated to obtain a super-resolution second detection image, wherein the optical frequency comb technology is used to accurately measure and locate the staggered information during the staggered collection; A processing module, the processing module is connected to the thickness sensor, the tension sensor, the first detection camera and the second detection camera, and is used to obtain the detection results of the thickness sensor and the tension sensor, and control the movement of the stage so that the screen to be tested enters the tension and thickness detection station, the size measurement station and the defect detection station respectively, and uses a preset mark point image matching analysis algorithm to detect the first detection image to obtain the geometric size of the grid line of the screen to be tested, and uses a pre-trained defect detection model to detect the second detection image to obtain the defect detection result of the screen to be tested; wherein, using the preset mark point image matching analysis algorithm to detect the first detection image includes: Divide the plurality of first detection images into a plurality of image sequences, and equate each image sequence to a standard halftone image; introduce deformation continuity constraints in the equivalence process, and use an image sequence to equate a standard halftone image to compare with the reference halftone image to identify the position of the marker point image; Comparing a plurality of standard halftone images with a reference halftone image to obtain pixel coordinates of a plurality of positioning points in the halftone image to be tested; The pixel coordinates of the plurality of positioning points are converted into world coordinates according to a preset coordinate conversion relationship to obtain the current position information of the screen to be tested; Based on the current position information of the screen to be tested, the screen to be tested is moved to the detection positions of the thickness sensor and the tension sensor to obtain the thickness and tension of the screen to be tested; Based on the current position information of the screen to be tested, the screen to be tested is moved to the size measurement station, and the coordinates of the screen to be tested relative to the world coordinate system after being moved from the tension and thickness measurement station to the size measurement station are obtained by positioning the positioning points of the screen to be tested multiple times and converting the pixel coordinates of the positioning points into world coordinates multiple times; Based on the current position information of the screen to be tested, the stage is moved to four preset positions, and the width of the grid lines of the screen to be tested in the first detection image of each preset position is identified; Control the stage to move to the position of the uppermost grid line of the screen to be tested, scan from the uppermost grid line to the lowermost grid line, measure the distance between the uppermost grid line and the lowermost grid line, and obtain the geometrical dimensions of the grid lines of the screen to be tested; Controlling the stage to move to the defect detection station, positioning the positioning points of the screen to be tested multiple times, and converting the pixel coordinates of the positioning points into world coordinates multiple times, to obtain the coordinates of the screen to be tested relative to the world coordinate system after it is moved from the size measurement station to the defect detection station; Divide the screen to be tested into multiple areas, and use a preset defect position detection algorithm to detect the second detection image to determine whether the screen to be tested has defects and the defect positions; The integrated machine body is provided with a high-precision motion axis, and the thickness sensor, the tension sensor, the first detection camera and the second detection camera are arranged on the high-precision motion axis to realize movement in the X, Y and Z directions so as to adjust the relative position with the screen to be tested; The stage is a glass plate, and the device further comprises: A lower light source, the lower light source is arranged below the glass plate; An upper light source, the upper light source is respectively arranged at the positions of the first detection camera and the second detection camera, and is used to enhance the light brightness when the first detection camera and the second detection camera collect the image of the to-be-tested screen; The device also includes: A plurality of driving motors, used for controlling the movement of the stage, the thickness sensor, the tension sensor, the first detection camera and / or the second detection camera; An alarm device, which is arranged on the integrated machine body and is used to sound an alarm when an alarm condition is met; The alarm device is an audio / visual alarm module, which is used to emit sound and / or light to alarm according to the current alarm situation when the alarm condition is met; The step of scanning from the uppermost grid line to the lowermost grid line, measuring the distance between the uppermost grid line and the lowermost grid line, and obtaining the geometrical dimensions of the grid line of the screen to be tested comprises: a) moving the stage so that the screen to be tested enters the measurement field 1, obtaining the pixel coordinates y11 of the position of the uppermost grid line and the pixel coordinates y12 of the position of the lowermost grid line according to the collected first detection image analysis, converting y11 and y12 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate y1 of the current stage; b) moving the stage downward, and the movement amount is smaller than the measurement area, obtaining the pixel coordinates y21 of the position of the uppermost grid line and the pixel coordinates y22 of the position of the lowermost grid line according to the collected first detection image analysis, converting y21 and y22 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate mark y2 of the current stage; c) calculating the movement amount of the stage Δd1=y2-y1, calculating the translation amount of the two first detection images Δd1'=y21-y12, and when the difference between Δd1 and Δd1' is less than the set accuracy threshold, executing step d), otherwise, returning to step a) and moving again; d) Repeating step b) and step c) until the bottom grid line is observed in the measurement area, obtaining the position pixel coordinates yn1 of the top grid line and the position pixel coordinates yn2 of the bottom grid line according to the collected first detection image analysis, converting yn1 and yn2 into world coordinates according to a preset coordinate conversion relationship, and recording the world coordinate mark yn of the current stage, and calculating the movement amount Δdn=yn-yn-1 of the stage; e) Calculate the geometric dimensions of the grid lines of the screen to be tested: Before using the pre-trained defect detection model to detect the second detection image to obtain the defect detection result of the screen to be tested, the method further includes: Acquire defective and non-defective screen images, crop and downsample the defective and non-defective screen images respectively, automatically enhance the defective and non-defective screen images, and mark the defective screen image; Constructing a defect detection model, using the defective and non-defective screen images and their labels, training the defect detection model until the loss function of the defect detection model satisfies the training termination condition, and obtaining a trained defect detection model, wherein the input of the defect detection model is the screen image, and the output is the defect position and defect type of the screen image; The defect detection model includes an Input input end, a Backbone network, a Neck feature fusion layer and a Head detection head, wherein the Input input end adopts Mosaic data enhancement technology to merge multiple screen images into one screen image; the Backbone network includes an improved C3 module and an SPPF module, and the C3 module is a residual structure for feature extraction; the Neck feature fusion layer adopts an FPN feature pyramid and a PAN path aggregation network structure, and comprehensively utilizes feature information of multiple scales to fuse the features extracted by the Backbone network; the Head detection head parses feature maps of different scales output by the Neck feature fusion layer, divides grid units on the feature map of each scale, sets multiple prior boxes for each grid cell to predict multiple bounding boxes, and removes bounding boxes with overlap greater than a preset value by a non-maximum suppression method to obtain the defect position and defect type of the screen.

Citation Information

Patent Citations

  • Automatic optical detection equipment and method

    CN104535583A

  • Train operation fault automatic detection system and method based on binocular stereoscopic vision

    CN106600581A

  • Fixed solution continuous judgment-based quasi-static scene positioning precision optimizing method

    CN108121001A

  • Multi-sensor combined complex multi-structure micro-part automatic micro-assembling equipment

    CN109595239A

  • Screen quality automatic detection method and system

    CN116465315A

Cited By

  • A high-precision online detection device for printing screen board defects

    CN122689648A