Method and system for detecting rubberizing and tearing

Abstract

The invention relates to the technical field of blade battery rubberizing and tearing detection, in particular to a rubberizing and tearing detection method and system. The six-axis mechanical arm moves to a glue tearing position, and a clamping jaw of the glue tearing device clamps the protruding part of the A-face protective film to tear off; detecting the adhesive tearing quality of the side A, collecting an image through a lower camera, detecting adhesive tearing defects through a Blob algorithm and an improved YOLOv5 algorithm tool, feeding back adhesive tearing NG when defects exceeding an allowable range exist, and otherwise, feeding back OK; the six-axis manipulator attaches the double-faced adhesive tape to a preset area of the blade battery, and the conveying device conveys the blade battery to a B-side adhesive tearing position; a clamping jaw of the adhesive tearing device clamps the protruding part of the protective film on the B surface to tear off; the adhesive tearing detection is the same as that of the A surface, if no defect exists, the center of the workpiece is positioned through template matching, and then the edge distance between the double-faced adhesive tape and the four edges of the battery is measured through a caliper tool.

Description

Technical Field

[0001] This application relates to the field of detection technology for adhesive adhesion and peeling of blade batteries, specifically a method and system for detecting adhesive adhesion and peeling. Background Technology

[0002] In the production of blade batteries, applying and removing double-sided tape are crucial steps. Traditionally, these steps are usually done manually. However, manual application and removal of tape suffers from low efficiency, reliance on experience for accuracy, and uncontrollable skill levels, leading to problems such as misalignment, skewing, and other issues during the tape application and removal process. Currently, with the development of automation, robotic arms are used to replace manual application and removal of tape, solving the problems of low efficiency and reliance on experience and skill.

[0003] In actual operation, existing technologies are prone to positional shifts when the robotic arm grasps double-sided tape due to various factors, resulting in misalignment of the adhesive application position with the blade battery. Furthermore, when removing the adhesive, residual adhesive film can affect the subsequent bonding quality with other components.

[0004] In view of the above, in order to overcome the above technical problems, the present invention designs a method and system for detecting adhesive application and removal, which solves the above technical problems. Summary of the Invention

[0005] To address the aforementioned issues, this application proposes a detection method and system for adhesive application and removal, which solves the technical problems of low efficiency, poor accuracy of adhesive application position, and difficulty in identifying small target defects in the prior art. By combining an automated mechanical structure with an improved visual detection algorithm, the method achieves fully automated execution and high-precision detection of adhesive application and removal for blade batteries.

[0006] A method for detecting adhesive residue removal includes the following steps:

[0007] Step 1: A vacuum suction cup-type material handling device consisting of eight suction nozzles arranged in a 2*4 array at the end of a six-axis robotic arm picks up the double-sided tape with the protective film still attached. The suction pressure is detected by a vacuum pressure sensor. If the pressure is less than a preset threshold, the material is picked up again. After picking up the material, the suction nozzles are driven to rotate inward repeatedly at a preset angle to create wrinkles in the protective film.

[0008] Step 2: Remove the adhesive from side A. The six-axis robotic arm moves to the adhesive removal position, and the adhesive removal device grips the protruding part of the protective film on side A and removes it.

[0009] Step 3: Inspect the adhesive removal quality of side A. Take images with the lower camera and use the Blob algorithm to filter areas larger than the set grayscale threshold. If such areas are detected, report "Adhesive Removal NG". Then, call the YOLOv5 algorithm tool, which is a deep learning improvement based on a custom encapsulation of VisionPro, to detect other defects with a pixel area smaller than 32*32 that cannot be detected by the Blob algorithm. If there are defects that exceed the allowable range, report "Adhesive Removal NG"; otherwise, report "OK".

[0010] Step 4: The six-axis robot attaches double-sided tape to the preset area of ​​the blade battery, and the conveyor transports the blade battery to the adhesive removal position on side B.

[0011] Step 5: Remove the adhesive from side B. The adhesive removal device uses its grippers to hold and remove the protruding part of the protective film on side B.

[0012] Step 6: Inspect the quality of peeling and applying adhesive on side B. Use a camera to capture images. First, use the Blob algorithm and the improved YOLOv5 algorithm to detect peeling defects. If there are no defects, use template matching to locate the center of the workpiece. Then, use calipers to measure the distance between the double-sided tape and the four sides of the battery. If the distance exceeds the preset range, it will be reported that the adhesive application is NG. Otherwise, it will be reported that the adhesive application and peeling are OK.

[0013] The improved YOLOv5 algorithm tool adds a multi-scale dilated convolution module to the shallow feature layer, adds a CAM module before feature fusion, and uses the Focal_EIOU loss function as the bounding box localization loss function.

[0014] Preferably, the multi-scale dilated convolution module uses convolution kernels of three sizes, namely 1*1, 3*3, and 5*5, with dilation ratios of 1, 3, and 5, respectively. After multi-scale extraction of shallow feature maps, the feature extraction results are merged through a weighted fusion strategy.

[0015] Preferably, the expression for the Focal_EIOU loss function is as follows: , ;in, Intersection over Union (IoU) is a metric used to measure the degree of overlap between predicted bounding boxes and ground truth bounding boxes in the detected target. A parameter indicating the degree of outlier suppression; Indicates IOU loss; Indicates distance loss; Indicates positional loss; This represents the center distance loss term; Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance; This represents the length of the diagonal of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box. and This represents the aspect ratio loss term; and Indicates the width and height of the prediction box; and Indicates the actual width and height of the bounding box; and This represents the width and height of the smallest bounding rectangle that includes the predicted bounding box and the ground truth bounding box.

[0016] Preferably, the template matching uses the CogPMAlignTool based on the VisionPro PatMax algorithm to locate the center of the workpiece by extracting key feature points of the template and searching for matches in the real-time image.

[0017] Preferably, the caliper tool is the CogCaliperTool tool based on edge detection technology, which calculates distance information by identifying edges or edge pairs in the image and measures the distance between the double-sided tape and the four sides of the battery.

[0018] Preferably, the preset angle of rotation of the suction nozzle is 15°~30°, and the preset pressure threshold of the vacuum pressure sensor is -80kPa~60kPa.

[0019] Preferably, the training process of the improved YOLOv5 algorithm tool includes: collecting at least 10,000 sample images of small target defects with residual protective film fragments and local un-torn areas, and containing pixel areas ranging from 1×1 to 32×32 pixels; dividing them into training set, validation set, and test set in a 7:2:1 ratio; using the Adam optimizer, setting the batch size to 16, the initial learning rate to 0.001, and iterating for at least 300 rounds until the detection accuracy of the test set is not less than 99% and the false negative rate is not higher than 0.5%.

[0020] Preferably, the allowable range of the predicted edge distance for the adhesive application position accuracy is ±0.1mm, and the repeatability of the six-axis robot is not less than ±0.02mm.

[0021] A detection system for adhesive application and removal, comprising an automated execution module, a vision inspection module, and a control and communication module, for implementing the detection method for adhesive application and removal.

[0022] The automated execution module includes a six-axis robot, a vacuum suction cup picking device, a glue-tearing device, and a conveying device; the vacuum suction cup picking device is installed at the end of the six-axis robot and consists of eight suction nozzles arranged in a 2*4 array and a vacuum pressure sensor; the glue-tearing device is equipped with grippers for holding the protective film; the conveying device is used to transport the blade battery between various workstations.

[0023] The visual inspection module includes an upper camera, a lower camera, an image acquisition card, and an industrial computer. The upper camera is installed directly above the battery adhesive inspection station, and the lower camera is installed directly below the A-side adhesive removal inspection station. The industrial computer is pre-installed with VisionPro vision software, which integrates the Blob algorithm module, the improved YOLOv5 algorithm module, the CogPMAlignTool template matching module, and the CogCaliperTool caliper tool module.

[0024] The control and communication module includes a PLC controller and an S7 communication unit; the PLC controller is electrically connected to the automation execution module and the vision inspection module respectively, and is used to receive sensor signals and send control commands; the S7 communication unit realizes high-speed data interaction between the PLC controller and the vision inspection module.

[0025] Preferably, when the visual inspection module reports an NG (Not Acceptable) error in peeling or applying adhesive, it simultaneously sends defect information to the PLC controller. The PLC controller then controls the conveying device to transport the corresponding blade battery to a preset defective temporary storage station and records the battery's production number and defect type, thus forming a traceable defect ledger.

[0026] The beneficial effects of this invention are as follows:

[0027] 1. This invention utilizes automated execution modules such as a six-axis robotic arm and a vacuum material handling device, combined with the real-time response capability of a vision inspection module, to achieve fully automated operation from double-sided tape picking and protective film removal to blade battery attachment. Each process can be completed without human intervention. Simultaneously, when the vision inspection system detects defects, it instantly feeds back the results to the PLC controller. The PLC controller quickly adjusts the equipment's operating status, avoiding process interruptions or delays caused by manual judgment and operation, effectively reducing waiting time between processes, and ensuring a smooth and continuous tape application and removal production process, thus improving operational efficiency.

[0028] 2. This invention addresses the problem of traditional detection methods' difficulty in identifying small-target defects. The improved YOLOv5 algorithm in the system accurately captures defect features with small pixel areas by adding a multi-scale dilated convolution module, a CAM module, and a Focal_EIOU loss function to the shallow feature layer. Simultaneously, the Blob algorithm can quickly identify obvious defects such as large-area protective film residue, achieving comprehensive coverage of defects at different scales. Furthermore, the combined use of a template matching tool and a caliper tool accurately locates the workpiece center and measures the adhesive edge distance, ensuring that the adhesive application position meets process requirements. This constructs a quality inspection barrier from both the thoroughness of adhesive removal and the accuracy of adhesive application, effectively reducing the generation of unqualified products due to defect omissions or adhesive application deviations. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of the visual inspection system of the present invention;

[0031] Figure 2 This is a diagram of the improved YOLOv5 model of the present invention;

[0032] Figure 3 This is a flowchart illustrating the process of the adhesive peeling and applying system of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0034] The application principle of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0035] Example 1:

[0036] like Figure 1-3 As shown, a method for detecting adhesive tape adhesion and peeling includes the following steps:

[0037] Step 1: Automatic feeding and pre-treatment of double-sided tape. A vacuum suction cup picking device is used to perform the picking operation. The picking device consists of eight suction nozzles at the end of a six-axis robot. The eight suction nozzles are arranged in a 2*4 array on the same plane to directly pick up the double-sided tape with the protective film still on from the material tray. During the picking process, the suction pressure is detected in real time by the vacuum pressure sensor integrated on the picking device. When the pressure value is less than the preset threshold, the insufficient picking pressure signal is fed back to the PLC controller, and the PLC controls the robot to re-execute the picking action. After picking up the double-sided tape, the PLC controls the six-axis robot to drive the suction nozzles to rotate inward repeatedly at a preset angle, so that the protective film of the double-sided tape is uniformly wrinkled, reducing the resistance to subsequent peeling.

[0038] Step 2: Peel the double-sided tape from side A. The six-axis robotic arm carries the adsorbed double-sided tape to the preset peeling position. The grippers of the peeling device precisely hold the protruding part of the protective film on side A of the double-sided tape and peel off the protective film on side A evenly along the preset trajectory to complete the peeling of side A.

[0039] Step 3: Inspect the quality of peeling off side A. After peeling off side A, the six-axis robot moves to the lower camera's image capture position. The PLC controller sends an image capture command to the vision inspection system, causing the lower camera of the vision inspection system to acquire a real-time image of side A of the double-sided tape. The inspection is completed according to the following process:

[0040] Step 3.1: Preprocess the image using the Blob algorithm, including grayscale conversion, noise reduction and binarization, and filter out image areas with a grayscale value greater than the set threshold. These areas correspond to large areas of residual protective film that have not been removed.

[0041] Step 3.2: If the Blob algorithm detects a large area of ​​residual material, the vision inspection system sends a "A-side adhesive tearing NG" signal to the PLC via the S7 communication protocol. The PLC then controls the equipment to stop and waits for manual assistance.

[0042] Step 3.3: If the Blob algorithm does not detect large-area residues, start the small target defect detection process. The visual inspection system calls the YOLOv5 algorithm tool, which is a deep learning improvement based on a custom encapsulation of VisionPro, to detect small target defects in the image with a pixel area of ​​less than 32*32, including tiny adhesive residues and protective film fragments.

[0043] Step 3.4: If the small target defect detection results do not show any defects exceeding the allowable number or size, the feedback is "A-side adhesive removal NG"; if no defects are detected, including large-area residues and small target defects, the feedback is "A-side adhesive removal OK", and proceed to the next adhesive application step.

[0044] Step 4: The six-axis robot precisely applies double-sided tape to the preset adhesive application area of ​​the blade battery according to the preset adhesive application coordinates. During the application process, the application pressure is adjusted by the pressure feedback of the robot to ensure that the double-sided tape adheres tightly to the battery surface without bubbles or wrinkles. After the tape application is completed, the blade battery is transported to the adhesive removal position on side B by a conveyor device.

[0045] Step 5: Peel the B side of the double-sided tape. The gripper of the peeling device holds the protruding distribution of the protective film on the B side of the double-sided tape and peels off the protective film on the B side at a uniform speed along the preset trajectory to complete the peeling of the B side.

[0046] Step 6: Conduct a comprehensive inspection of double-sided tape B and the quality of the adhesive application. After the tape on side B is peeled off, the conveyor moves the blade battery to the upper camera's photographing position. The PLC controller sends a photographing command to the vision inspection system. The upper camera of the vision inspection system captures real-time images of the motor adhesive application area. The inspection is completed according to the following process:

[0047] Step 6.1: Perform adhesive removal quality inspection. First, use the Blob algorithm to filter out large residual areas larger than the set grayscale threshold. Then, use the improved YOLOv5 algorithm tool to detect small target defects with a pixel area smaller than 32*32. If any type of defect exists, the feedback is "B-side adhesive removal NG".

[0048] Step 6.2: If the adhesive peeling quality inspection passes, perform an adhesive application position accuracy inspection. The vision inspection system uses the CogPMAlignTool template matching tool based on VisionPro's PatMax algorithm to automatically extract key feature points from the template image and search for these feature points in the real-time image. Through precise affine transformation, it accurately positions and adjusts the posture of the target object to achieve precise positioning of the workpiece center. Then, it calls the CogCaliperTool caliper tool to find the edge features of the four sides of the double-sided tape and the four sides of the battery based on edge detection technology, and measures the corresponding edge distances of the four sides of the double-sided tape and the four sides of the battery.

[0049] Step 6.3: If the measured edge distance values ​​are all within the preset allowable range, the vision inspection system will send a "sticking and peeling OK" message to the PLC, completing the sticking and peeling inspection process for the battery; if the edge distance exceeds the preset range, a "sticking NG" message will be sent, and the PLC will control the equipment to stop, waiting for manual processing.

[0050] The improved YOLOv5 algorithm is used for small target defect detection as follows:

[0051] This includes the encapsulation of the algorithm model. The improved YOLOv5 deep learning algorithm model is encapsulated using the VisionPro custom tool to form a small target defect detection module that can be directly called. This module works in conjunction with the Blob algorithm module, template matching module, and caliper tool module to form a complete visual inspection function.

[0052] This includes defining defects in small targets, clarifying that defects in small targets are feature images with a pixel area less than 32*32. Since the original YOLOv5 network requires five max pooling layers from the input image to the feature vector of the fully connected layer, a single point on the final feature vector corresponds to a 32*32 pixel area of ​​the input image. It does not pay much attention to the extreme cases of target scale, so the convergence direction of the model training is more inclined towards medium and large scale targets. Small targets will perform poorly in the recognition results due to insufficient feature extraction, imbalanced training data, and other factors, so targeted improvements are needed.

[0053] Network structure improvement methods include feature extraction optimization, feature fusion enhancement, and detection head optimization, among which:

[0054] Feature extraction optimization involves adding multi-size dilated convolutional modules to the shallow feature layers of the YOLOv5 backbone network. Convolutional kernels of three sizes (1*1, 3*3, and 5*5) with dilation ratios of 1, 3, and 5 are used to perform multi-scale feature extraction on the shallow feature maps. By utilizing the rich spatial information of small targets in the shallow network, the receptive field of the convolutional kernel is increased without changing the scale of the feature map through dilated convolution, ensuring more complete feature capture of small target defects. In particular, the wrinkles produced by the six-axis robotic arm driving the nozzle to rotate inward repeatedly, which improves the ease of peeling off the adhesive, pose a risk of small target defects.

[0055] Feature fusion enhancement merges the feature maps extracted by multi-scale dilated convolution through a weighted fusion strategy to strengthen the feature representation of small target defects; at the same time, a CAM (channel attention mechanism) module is added to the front end of feature fusion to enhance sparse small target features and improve the generalization effect of feature fusion.

[0056] like Figure 2 As shown, the detection head optimization is based on the YOLOv5 backbone network outputting three feature layers of sizes: 80×80, 40×40, and 20×20. YOLOHead1, YOLOHead2, and YOLOHead3 are output according to feature fusion methods such as upsampling and downsampling, representing effective feature maps containing shallow features and semantic information at small, medium, and large scales. The CAM module, representing sparse feature extraction, is added to the front end of the feature fusion process, making small target feature extraction more generalized. Detect represents the model's prediction phase, including the calculation and representation of output results such as the difference between the ground truth annotation and the model's prediction, and the confidence score of the predicted class. In the prediction phase, the pixel proportion of local features is amplified to improve the detection rate of small targets. In the prediction result evaluation phase, an improved Focal_EIOU function is added to improve the localization accuracy of the predicted bounding box during model parameter iteration updates. YOLOHead1 is specifically used for small target defect detection, improving the detection rate of small target defects by amplifying the pixel proportion of local features.

[0057] The Focal_EIOU loss function is used as the bounding box localization loss function, replacing the original YOLOv5 CIOU loss function:

[0058] The EIOU loss function calculates the difference in width and height between the predicted bounding box and the ground truth bounding box separately, while minimizing the aspect ratio difference. Compared with the CIOU and DIOU loss functions, it has a faster convergence speed and higher positioning accuracy, making it suitable for small target defects with high positioning accuracy requirements.

[0059] The Focal function addresses the imbalance of training samples through weight allocation, utilizes the distribution pattern of long-tailed data to distinguish between high-quality and low-quality prior boxes, and increases the contribution of prediction boxes with smaller regression errors to the loss result.

[0060] Combining the two, the expression for the Focal_EIOU loss function is as follows:

[0061]

[0062] in, Intersection over Union (IoU) is a metric used to measure the degree of overlap between predicted bounding boxes and ground truth bounding boxes in the detected target. A parameter indicating the degree of outlier suppression; Indicates IOU loss; Indicates distance loss; Indicates positional loss;

[0063]

[0064] in, This represents the center distance loss term; Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance; This represents the length of the diagonal of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box. and This represents the aspect ratio loss term; and Indicates the width and height of the prediction box; and Indicates the actual width and height of the bounding box; and This represents the width and height of the smallest bounding rectangle that includes the predicted bounding box and the ground truth bounding box;

[0065] The training process for the improved YOLOv5 algorithm module is as follows: At least 10,000 sample images containing small target defects (pixel area 1×1 to 32×32) are collected. The LabelImg annotation tool is used to label the defect location and category in VOC format. The labeled samples are divided into training, validation, and test sets in a 7:2:1 ratio. The PyTorch framework is used for model training, with a batch size of 16, an initial learning rate of 0.001, and a cosine annealing decay strategy. The Adam optimizer is used, with a weight decay coefficient of 0.0005. The training is iterated for at least 300 epochs, with model performance evaluated on the validation set every 10 epochs. Training is stopped and the model is exported when the small target defect detection accuracy on the test set is no less than 99% and the false negative rate is no more than 0.5%. The model is then encapsulated using VisionPro's custom tool interface.

[0066] This includes data interaction methods. The vision inspection system and the PLC controller communicate via the S7 communication protocol. The S7 communication protocol is a dedicated communication protocol for Siemens S7 series PLCs, which supports high-speed data transmission between the PLC and PC workstations and external inspection systems. This ensures real-time interaction of inspection results and equipment status information, and guarantees the reliability and response speed of the system's collaborative control.

[0067] Example 2:

[0068] The system for implementing the above-mentioned adhesive application and peeling detection method includes an automated execution module, a vision inspection module, and a control and communication module, wherein:

[0069] The automated execution module includes a six-axis robot, a vacuum suction cup picking device, a glue-tearing device, and a conveying device. The vacuum suction cup picking device is installed at the end of the six-axis robot and consists of eight suction nozzles arranged in a 2*4 array and a vacuum pressure sensor. The glue-tearing device is equipped with grippers to hold the protective film. The conveying device is used to transport the blade battery between various workstations.

[0070] The vision inspection module includes an upper camera, a lower camera, an image acquisition card, and an industrial computer. The upper camera is installed directly above the battery adhesive inspection station, and the lower camera is installed directly below the A-side adhesive removal inspection station. The camera resolution is no less than 20 megapixels. The industrial computer comes pre-installed with VisionPro vision software, which integrates the Blob algorithm module, the improved YOLOv5 algorithm module, the CogPMAlignTool template matching module, and the CogCaliperTool caliper tool module.

[0071] The control and communication module includes a PLC controller and an S7 communication unit. The PLC controller is electrically connected to the automation execution module and the vision inspection module respectively, and is used to receive sensor signals and send control commands. The S7 communication unit enables high-speed data interaction between the PLC controller and the vision inspection module.

Claims

1. A method for detecting a taping and tearing, characterized by: The method comprises the following steps: Step 1: a vacuum chuck type material taking device composed of eight suction nozzles arranged in a 2*4 array at the end of a six-axis manipulator is used to suck double-sided adhesive tape without protective film, a vacuum pressure sensor is used to detect the adsorption pressure, and when the pressure is less than a preset threshold, the material is taken again; after suction, the suction nozzle is driven to rotate inward by a preset angle to make the protective film wrinkle; Step 2: A-side adhesive tape is torn, the six-axis manipulator moves to the adhesive tape tearing position, and the clamping jaw of the adhesive tape tearing device clamps and tears the protruding part of the A-side protective film; Step 3: A-side adhesive tape tearing quality is detected, an image is collected by a lower camera, a region larger than a set gray threshold is screened through a Blob algorithm, and the region is fed back as tearing NG when detected; Then a deep learning improved YOLOv5 algorithm tool based on VisionPro is called to detect other defects with a pixel area less than 32*32 and unable to be detected by the Blob algorithm, and feed back as tearing NG when the defects exist beyond the allowed range, otherwise feed back as OK; Step 4: the six-axis manipulator attaches the double-sided adhesive tape to the preset area of the blade battery, and a conveying device conveys the blade battery to the B-side adhesive tape tearing position; Step 5: B-side adhesive tape is torn, the clamping jaw of the adhesive tape tearing device clamps and tears the protruding part of the B-side protective film; Step 6: B-side adhesive tape tearing and adhesive tape attaching quality are detected, an image is collected by an upper camera, tearing defects are detected through the Blob algorithm and the improved YOLOv5 algorithm tool, the workpiece center is located through template matching after no defects are detected, and the double-sided adhesive tape and the four edges of the battery are measured by a caliper tool, and the adhesive tape attaching is fed back as NG when the edge margin exceeds the preset range, otherwise the adhesive tape tearing and adhesive tape attaching are fed back as OK. The network structure of the improved YOLOv5 algorithm tool adds a multi-scale dilated convolution module in a shallow feature layer and adds a CAM module before feature fusion, and adopts a Focal_EIOU loss function as a boundary box positioning loss function.

2. The method of claim 1, wherein: The multi-scale dilated convolution module adopts 1*1, 3*3 and 5*5 convolution kernels with dilated rates of 1, 3 and 5 respectively to perform multi-scale extraction on a shallow feature map, and then combines the feature extraction results through a weighted fusion strategy.

3. The method of claim 1, wherein: The expression of the Focal_EIOU loss function is as follows: , ; wherein, represents the intersection over union, an index for measuring the degree of overlap between the predicted box and the real box in the detection target; represents a parameter for controlling the degree of abnormal value suppression; represents the IOU loss; represents the distance loss; represents the orientation loss; represents the center distance loss term; represents the Euclidean distance between the center of the predicted box and the center of the real box ; represents the diagonal length of the minimum enclosing rectangle surrounding the predicted box and the real box; and represent the aspect ratio loss term; and represent the width and height of the predicted box; and represent the width and height of the real box; and represent the width and height of the minimum enclosing rectangle including the predicted box and the real box.

4. The method of claim 1, wherein: The template matching adopts a CogPMAlignTool tool based on the PatMax algorithm of VisionPro, locates the workpiece center by extracting key feature points of a template and searching for matching in a real-time image.

5. The method of claim 1, wherein: The caliper tool is a CogCaliperTool tool based on edge detection technology, measures the edge margin of the double-sided adhesive tape and the four edges of the battery by identifying edges or edge pairs in an image.

6. The method of claim 1, wherein: The preset angle of the suction nozzle rotation is 15°-30°, and the preset pressure threshold of the vacuum pressure sensor is -80kPa-60kPa.

7. The method of claim 1, wherein: The training process of the improved YOLOv5 algorithm tool includes: collecting at least 10,000 small target defect sample images with residual protective film fragments and local unpeeling areas, and containing pixel areas of 1*1 to 32*32, dividing them into a training set, a validation set and a test set in a ratio of 7:2:1, using the Adam optimizer, setting the batchsize to 16, the initial learning rate to 0.001, and iteratively training for not less than 300 rounds.

8. The method of claim 1, wherein: The allowed range of the prediction margin of the position accuracy of the adhesive tape is ±0.1mm, and the repeatability of the six-axis robot is not less than ±0.02mm.

9. A detection system for the detection method of the adhesive tape and the tearing of the adhesive tape according to any one of claims 1-8, comprising an automatic execution module, a visual detection module and a control and communication module, characterized in that: The automatic execution module comprises a six-axis robot, a vacuum suction cup type material taking device, a tearing device and a conveying device; the vacuum suction cup type material taking device is installed at the end of the six-axis robot and is composed of eight suction nozzles arranged in a 2*4 array and a vacuum pressure sensor; the tearing device is provided with a clamping jaw for clamping the protective film; the conveying device is used for conveying the blade battery between the stations; The visual detection module comprises an upper camera, a lower camera, an image acquisition card and an industrial computer; the upper camera is installed directly above the battery adhesive tape detection station, and the lower camera is installed directly below the A-face tearing detection station; the industrial computer is pre-installed with VisionPro visual software, integrated with Blob algorithm module, improved YOLOv5 algorithm module, CogPMAlignTool template matching module and CogCaliperTool caliper tool module; The control and communication module comprises a PLC controller and an S7 communication unit; the PLC controller is electrically connected with the automatic execution module and the visual detection module respectively, and is used for receiving sensor signals and sending control instructions; the S7 communication unit realizes high-speed data interaction between the PLC controller and the visual detection module.

10. The system for detecting the presence or absence of adhesive and / or the presence or absence of a tear according to claim 9, wherein: When the visual detection module feeds back tearing NG or adhesive tape NG, it sends defect information to the PLC controller at the same time, the PLC controller controls the conveying device to convey the corresponding blade battery to the preset defective product temporary storage station, and records the production number and defect type of the battery, forming a traceable defect account.

Images