Ai-based automatic correction system for laser welding positions on spacer grid

Abstract

The present invention relates to an AI-based automatic correction system for laser welding positions on spacer grid, the system automatically correcting laser welding positions through deep learning. The AI-based automatic correction system for laser welding positions on spacer grid according to the present invention comprises: a vision unit (100) which learns image data collected through a camera to determine welding positions; a CNC unit (200) which indicates, to a welding head, the coordinates of welding positions based on the data received from the vision unit; and a welding unit (300) which corrects welding positions within a sealed chamber according to commands received from the CNC unit. The AI-based automatic correction system, for a laser welding position of a spacer grid, can automatically correct the coordinate values of laser welding positions on spacer grid, which supports nuclear fuel rods, via deep learning on the basis of images of the corresponding welding positions themselves (base material) visible in the images.

Description

AI-based automatic grid laser welding position correction system

[0001] The present invention relates to an AI-based grid laser welding position automatic correction system, and relates to a system that automatically corrects the laser welding position through deep learning.

[0002] Conventional laser welding position correction technology corrects the position by storing the relative coordinates of the welding point using a reference marker on the welding fixture.

[0003] At this time, it is necessary to machine a reference marker through hole in the welding fixture and assemble the pin. This pin is a crucial pin for reading the reference point of the welding part. It is sensitive to light intensity and can be damaged during handling, leading to misrecognition and welding at a misaligned welding location.

[0004] In addition, since the coordinate values ​​for the location of the welding part centered on the reference point of the pin must be input into the CNC, there was an inconvenience in that the worker had to modify the coordinate values ​​according to the recognition of the pin, as it was a semi-automatic welding concept.

[0005] In reality, the external weld joint is manually adjusted by the worker and cannot be applied to all types of support grids for supporting nuclear fuel rods.

[0006] [Prior Art Literature]

[0007] [Patent Document]

[0008] (Patent Document 1) Republic of Korea Patent Publication No. 10-2023-0091245 (June 23, 2023)

[0009] According to the present invention, a support grid for supporting a nuclear fuel rod is laser welded, and an AI-based support grid laser welding position automatic correction system for automatically correcting the welding position during the welding process is provided.

[0010] The AI-based grid laser welding position automatic correction system according to the present invention includes a vision unit (100) that learns image data collected through a camera to determine a welding position, a CNC unit (200) that instructs a welding head on the coordinates of a welding position using data received from the vision unit, and a welding unit (300) that corrects the welding position within a closed chamber according to a command received from the CNC unit.

[0011] The vision unit (100) further includes an adjustment module (110) that monitors the welding process in real time and enables adjustment through feedback according to mode selection.

[0012] The welding unit (300) includes a laser generator (310) that generates a laser light source, a focal length conversion unit (320) that converts the focal length of the laser, a beam control unit (330) that corrects the welding position to the coordinates of the welding position according to a command received from the CNC unit, a collimation lens (340) that parallels the laser beam, and a focusing lens (350) that focuses the laser beam on a specific work point.

[0013] Additionally, the welded portion (300) further includes a cover slide (360) that protects against contaminants generated during the welding process.

[0014] When an image for which a welding location cannot be recognized is generated among the image data collected through the camera, the vision unit (100) repeatedly learns the corresponding location through deep learning to recognize the welding location.

[0015] And the vision unit (100) automatically corrects the coordinate values ​​based on the result of recognizing the welding location through deep learning based on the image of the corresponding welding location base material shown in the video.

[0016] According to the present invention, there is an effect of automatically correcting the coordinate values ​​of the laser welding position of the support grid for supporting the nuclear fuel rod through deep learning based on the image of the corresponding welding position itself (parent material) shown in the video.

[0017] FIG. 1 is a configuration diagram of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0018] FIG. 2 illustrates a welding part configuration of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0019] FIG. 3 illustrates an interface of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0020] FIG. 4 illustrates an external weld learning image of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0021] FIG. 5 illustrates a learning image of a cross-point weld joint of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0022] FIG. 6 illustrates an interface map of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0023] An AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention relates to an AI-based support grid laser welding position automatic correction system for automatically correcting the welding position through deep learning during the welding process of laser welding a support grid for supporting a nuclear fuel rod.

[0024] In addition, the AI-based grid laser welding position automatic correction system according to this embodiment can minimize welding position errors through image recognition and perform high-quality welding.

[0025] The support grid according to this embodiment is intended to support the nuclear fuel rods. The AI-based support grid laser welding position automatic correction system accurately learns 70 welding points for the PLUS7 intermediate support grid and performs automatic correction. This intermediate support grid, which is part of the nuclear fuel assembly, ensures that the nuclear fuel rods are accurately positioned within the nuclear reactor. This support grid provides appropriate spacing between the nuclear fuel rods, optimizes heat transfer, and allows coolant to flow around the fuel rods.

[0026] The position of the PLUS7 intermediate support grid can be manually corrected for each welding point, including internal tabs, sleeves, intersections, corner tacks, corners, and external welds. 70 points of the weld are fully automated through deep learning to collect and learn weld reference point images, preventing human error.

[0027] Distinctive position welding point manual compensation PLUS7 intermediate support grid tunnel tap 4020 sleeve 4020 intersection 41014 corner tack 88 corner 44 external 1204 series 62270

[0028] In addition, in the AI-based grid laser welding position automatic correction system of this embodiment, deep learning (AI) analyzes the image of the point to be welded to identify the exact position of each contact point, and based on this, automatically adjusts the coordinates so that the robot welding system can perform welding at the exact position.

[0029] An AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention will be described with reference to the attached drawings.

[0030] Figure 1 is a configuration diagram of an AI-based support grid laser welding position automatic correction system (10) according to one embodiment of the present invention.

[0031] As illustrated in Fig. 1, the AI-based support grid laser welding position automatic correction system (10) includes a vision unit (100), a CNC unit (200), and a welding unit (300).

[0032] The vision unit (100) is a configuration that learns image data collected through a camera to determine a welding location.

[0033] This vision unit (100) repeatedly learns the corresponding location through deep learning in order to recognize the welding location when an image for which the welding location cannot be recognized is generated among the image data collected through the camera.

[0034] In addition, the vision unit (100) automatically corrects the coordinate values ​​based on the result of recognizing the welding location through deep learning based on the image of the corresponding welding location base material shown in the video.

[0035] In addition, the vision unit (100) further includes an adjustment module (110) that monitors the welding process in real time and enables adjustment through feedback according to mode selection.

[0036] This adjustment module (110) includes a Light & Controller that sets appropriate lighting.

[0037] The vision unit according to this embodiment is configured for automatic welding position correction of the PLUS7 intermediate grid. This vision unit uses a camera and other image acquisition equipment to collect images of 70 welding points on the PLUS7 intermediate grid. These images include the specific location of each welding point and its surrounding environment.

[0038] The vision unit standardizes the collected images through processes such as resizing, cropping, and rotating, and labels each image with the location of the corresponding welding point. These labels are the targets that the model must predict during the learning process. For learning, a powerful deep learning architecture for image processing, such as a convolutional neural network (CNN), is used to effectively recognize and classify image features. The vision unit performs learning based on the given images and labels, minimizing the loss function and finding the optimal weights. Based on the image of the corresponding welding location parent material as seen in the image, the coordinate values ​​are automatically corrected based on the results of deep learning-based welding location recognition.

[0039] The vision unit collects images of the welding location base material through cameras or other image collection equipment, recognizes the welding location using a deep learning model based on this image data, and transmits this information to the CNC unit.

[0040] The CNC section (200) is configured to instruct the welding robot on the coordinates of the welding position using data received from the vision section.

[0041] The CNC section receives the recognition results (coordinate information) from the vision section and indicates the welding position coordinates, and the CNC section handles the correction of the coordinate values. This is to place the workpiece in the correct position and move the welder precisely to that position.

[0042] The welding unit (300) corrects the welding position within a closed chamber according to a command received from the CNC unit.

[0043] These welds are configured to weld at a calibrated welding position on a loading table for aligning and fixing the workpiece within a closed chamber according to commands received from the CNC unit.

[0044] The AI-based grid laser welding position automatic correction system according to this embodiment minimizes contamination that may occur during the welding process on a loading table for aligning and fixing a workpiece so that it can be performed in a sealed chamber so that it can be applied to a grid type for supporting nuclear fuel rods, performs welding based on precise image analysis results provided by a vision unit, and monitors the welding process in real time to enable adjustment through feedback.

[0045] FIG. 2 is a drawing for explaining a welding part configuration (300) of an AI-based support grid laser welding position automatic correction system according to one embodiment of the present invention.

[0046] As illustrated in FIG. 2, the welding part (300) according to the present embodiment is a laser welding device, and includes a laser generator (310), a focal length conversion part (320), a beam adjustment part (330), a collimation lens (340), a focusing lens (350), and a cover slide (360).

[0047] The laser generator (310) is a configuration that generates a laser light source.

[0048] The focal length conversion unit (320) converts the focal length of the laser from 300 mm to 400 mm, and the extension of the focal length allows the laser to be applied to a wider area, thereby expanding the working range.

[0049] The beam spot size of the laser beam increases from 1.13 mm to 1.33 mm depending on the converted focal length. This means that the width of the bead during welding increases as the spot size increases.

[0050] The beam control unit (330) is configured to correct the welding position to the coordinates of the welding position according to the command received from the CNC unit. It precisely adjusts the path of the laser beam and controls the direction and position of the beam.

[0051] The collimation lens (340) collimates the laser beam. This prevents the beam from diverging until it reaches the target material, enabling more precise welding. The collimation lens has a focal length of 180 mm.

[0052] A focusing lens (350) is used to precisely focus the laser beam on a specific work point. The focal length of the focusing lens is 400 mm.

[0053] The laser beam generated from the laser generator is steered by a beam control unit, aligned parallel by a collimating lens, and then precisely focused on the target welding point by a focusing lens. Through this process, the laser beam precisely reaches the surface of the target material, performing welding.

[0054] The welded portion (300) according to the present embodiment uses the FLW-d50-L model of IPG Photonics, and the core size of the process fiber is 0.6 mm. The calculated laser spot size is 0.6 mm × (400 mm / 180 mm) = 1.33 mm, which indicates that the width of the beam increases in proportion to the focal length.

[0055] As shown in Fig. 2, it can be confirmed that the welded part (300) is equipped with a camera and is protected by a cover slide (360).

[0056] The cover slide (360) protects the lens from dust, spatter, and contaminants in the working environment. This is to prevent metal spatter generated during the welding process, which can damage the focusing lens. This cover slide maximizes the transmittance of the laser beam without altering its path, thereby performing its protective function without degrading laser performance.

[0057] The vision component of the AI-based grid laser welding position automatic correction system according to this embodiment is connected to the CNC via LAN. Its image recognition function collects images of the workpiece via a camera. The collected image data is processed through deep learning to determine the precise welding location. This component is also connected to other system components via LAN, including the Light & Controller, to provide optimal image capture conditions.

[0058] In relation to the vision department, the process of determining the exact location to weld using deep learning is explained as follows.

[0059] First, the vision unit collects a large number of images of the workpiece under various welding working conditions through a camera.

[0060] Each collected image is labeled with the exact locations to be welded (e.g., reference points, edges, joining points, etc.). The labeled data constitutes a training dataset, which is used to train the model to identify welding locations. For example, techniques such as data augmentation can be used to ensure robust model training under various lighting, angle, and background conditions.

[0061] The vision unit uses the collected training dataset to train a deep learning model, such as a convolutional neural network (CNN). The CNN recognizes visual patterns in images and, through feature extraction, learns the information necessary to determine welding locations.

[0062] And to evaluate the accuracy of the model, the model is tested using a separate validation dataset, and at this stage, measures can be taken to optimize the model's performance and prevent overfitting.

[0063] Once trained, the model analyzes images of the workpiece in real time in the vision unit. The learning model analyzes images captured during the actual welding operation to identify the precise welding location. The identified welding location is transmitted to the CNC unit in coordinate form, and the CNC unit adjusts the movements of the robotic welder based on this information. The robot then moves to the determined coordinates and performs precise welding.

[0064] The CNC is the central control unit for the entire welding system, issuing commands to the welding robot based on data provided by the vision unit. The CNC also coordinates and controls the operation of the robot and other system components through the I / O unit.

[0065] A chamber is a sealed space where welding work takes place, protecting the interior from the external environment, including sparks and smoke that may occur during welding.

[0066] FIG. 3 illustrates an interface of an AI-based automatic support grid laser welding position correction system according to one embodiment of the present invention. As illustrated in FIG. 3, in the AI-based automatic support grid laser welding position correction system according to the present embodiment, the interface of the vision unit is used to operate and monitor a deep learning-based automatic laser welding position alignment system.

[0067] The interface components and functions of Figure 3 are described as follows.

[0068] The MAIN tab is the tab that takes you back to the main page of the system.

[0069] The PLC tab takes you to pages related to programmable logic controller (PLC) settings.

[0070] The ADMIN PAGE tab is a tab that allows you to go to the system administrator page, where you can manipulate functions related to administrator settings and permissions.

[0071] The GRID TYPE selection section allows you to select the type of grid to weld, for example, you can select options such as "Train Stream" or "Radius Machining (Border Processing)".

[0072] Image + Line Position is a section that shows the image currently captured by the camera and the position of the line to be welded.

[0073] The inspection image classification results show the results of the deep learning algorithm classifying the image, and indicate the accuracy of the welding location as a percentage.

[0074] The READY, AUTO, and EXIT buttons indicate the system readiness status and are used to start operation in automatic mode or shut down the system.

[0075] Process Result and Detail Result are counters that indicate the total result of the welding process and the status of each welding result (OK, NG, RE-CYCLE).

[0076] X,Y coordinate results and detection results details show the X,Y coordinate results of the welding location and the detailed inspection results for it.

[0077] System Status indicates the current operating mode of the system and the status of deep learning, camera, and PLC.

[0078] The RESET, Guide Line, and Manual buttons are used to reset system settings, set guidelines, and switch to manual operation mode.

[0079] Figure 4 illustrates an external weld learning image of an AI-based grid laser welding position automatic correction system according to one embodiment of the present invention. Figure 4 is for identifying an area requiring welding.

[0080] The image in Figure 4 illustrates how the welding system identifies the location of external welds. External welds are weld points located outside the support grid, which may form the perimeter of the structure or connect to other structures. The green lines or boxes in the image indicate areas to be welded, as identified by the deep learning algorithm, while numbers or labels indicate the exact locations of the welds.

[0081] Figure 5 illustrates a learning image of a cross-point weld joint of an AI-based grid laser welding position automatic correction system according to one embodiment of the present invention. Figure 5 shows the state before welding, and illustrates position correction before welding.

[0082] Regarding the intersection weld learning image in Figure 5, intersection welds refer to welding operations at points where multiple support grids meet. These points are crucial for structural strength and stability and are key points where multiple structural elements are connected. In the intersection weld learning image, the green areas indicate points to be welded, and these locations are precisely identified and displayed by the algorithm.

[0083] As illustrated in Figures 4 and 5, it is visually demonstrated how the AI-based system accurately identifies different types of welding points and determines the appropriate welding method and path for each.

[0084] Figure 6 illustrates an interface map of an AI-based automatic grid laser welding position correction system according to one embodiment of the present invention. The interface map depicted in Figure 6 illustrates communication between the two systems and includes information on what data is transmitted and in what format. This interface map is essential for understanding data flow and protocols during system integration. Here, Vision refers to a system that identifies welding locations using a camera, and CNC refers to a control system that receives this data and issues welding commands to the robot.

[0085] A heartbeat (flicker) is a signal sent at regular intervals to check the connection status between systems. This signal indicates that communication is active.

[0086] Program Mode is the part that sets the CNC's program mode, and can be selected from Auto, Manual, and Bypass.

[0087] A program is a program of work to be performed. Here, the code for a specific program or task will be transmitted.

[0088] Result #1 is the section that conveys information about the welding result, indicating the status such as OK, NG, Re-Cycle, etc.

[0089] Locate x #1, Locate y #1 transmit the coordinate data of the welding position to the CNC. This is necessary to enable the robot to perform welding at the exact position.

[0090] The flow chart on the right shows the communication process between the CNC and the Vision system, and includes the steps of checking the equipment On status, selecting the equipment mode, entering product information, starting the equipment operation, starting the vision system, and correcting the welding result information.

[0091] This shows the sequential steps in which the CNC receives information from the Vision system, processes it, and adjusts the welding process based on the results.

[0092] In this embodiment, the AI-based automatic grid laser welding position correction system uses deep learning to classify welding images, enabling welding locations to be identified with a high accuracy of 99%. This significantly improves the efficiency and accuracy of the welding process.

[0093] Additionally, the AI-based grid laser welding position automatic correction system has completed training on 5,000 image data sets for PLUS7 intermediate grid welding, and can confirm that welding is performed at the correct location.

[0094] According to this embodiment, coordinate values ​​can be automatically corrected based on an image of the welding location itself (parent material) as seen in the video. There is no need to process reference marker penetration holes in the welding fixture or assemble the plate, and even external welding, which was not possible with conventional techniques, is fully automated.

[0095] In addition, since there are tolerances in the support grid products and welding fixtures, the recognition rate cannot be 100%, but by introducing the concept of defective (NG), if recognition is incorrect at the welding location, welding does not proceed, eliminating the possibility of non-conforming products.

[0096] If an image fails to be recognized above, deep learning can be used to accurately identify the location. The more deep learning is performed, the closer we can get to our goal of 100% recognition. Regardless of the product type, if we properly set the reference image for the welded joint, the system can be applied to all types. With a single click of a button, workers can perform auxiliary work until two products are produced, significantly improving work efficiency and productivity.

[0097] [Explanation of symbols]

[0098] 100: Vision Department

[0099] 110: Adjustment module

[0100] 200: CNC department

[0101] 300: Welding

[0102] 310: Laser generator

[0103] 320: Focal length conversion unit

[0104] 330: Beam control unit

[0105] 340: Collimation lens

[0106] 350: Focusing lens

[0107] 360: Cover Slide

Claims

1. A vision unit (100) that learns image data collected through a camera to determine a welding location; A CNC unit (200) that instructs the coordinates of the welding position at the welding part using data received from the above vision unit, and An AI-based support grid laser welding position automatic correction system characterized by including a welding unit (300) that performs welding at a corrected welding position on a loading table for aligning and fixing a workpiece within a closed chamber according to a command received from the CNC unit.

2. In paragraph 1, The above vision part (100) An AI-based grid laser welding position automatic correction system characterized by further including an adjustment module (110) that monitors the welding process in real time and enables adjustment through feedback according to mode selection.

3. In paragraph 1, The above welding part (300) A laser generator (310) that generates a laser light source; A focal length conversion unit (320) that converts the focal length of the laser; A beam control unit (330) that corrects the welding position to the coordinates of the welding position according to the command received from the CNC unit; A collimation lens (340) that parallels the laser beam; An AI-based grid laser welding position automatic correction system characterized by including a focusing lens (350) that focuses a laser beam on a specific work point.

4. In paragraph 1, The above welding part (300) An AI-based support grid laser welding position automatic correction system characterized by further including a cover slide (360) for protection from contaminants generated during the welding process.

5. In paragraph 1, The above vision unit (100) is an AI-based grid laser welding position automatic correction system characterized in that, when an image for which a welding position cannot be recognized occurs among image data collected through a camera, the corresponding position is repeatedly learned through deep learning to recognize the welding position.

6. In paragraph 1, The above vision unit (100) is an AI-based support grid laser welding position automatic correction system characterized in that it automatically corrects the coordinate values ​​according to the welding position recognition result through deep learning based on the image of the corresponding welding position base material shown in the video.

Images