Single-column rotary turntable type weld leakage detection repair welding system and application

Abstract

This invention discloses a single-column rotating turntable weld leakage detection and repair system and its application, relating to the field of weld leakage detection technology. The single-column rotating turntable weld leakage detection and repair system mainly includes a machine vision motion system and a detection system. The machine vision motion system ensures that the detection system can scan the target weld. The detection system is used to detect the target weld. The machine vision motion system includes a control subsystem and a motion subsystem. The control subsystem is used for movement control, laser repair welding control, working condition control, alarm, and signal docking. The motion subsystem includes a vision platform, a workpiece platform, a spraying structure, and a support structure. A laser welding machine is installed on the spraying structure. Implementing the single-column rotating turntable weld leakage detection and repair system and its application provided by this invention can improve the accuracy and efficiency of weld leakage detection in cylindrical steel pipes.

Description

Technical Field

[0001] This invention relates to the field of weld leakage detection technology, and more specifically, to a single-column rotating turntable type weld leakage detection and repair system and its application. Background Technology

[0002] Weld leakage refers to the leakage of the internal medium (such as gas or liquid) from the weld seam during the welding process of cylindrical steel pipes due to various reasons that result in an incomplete weld. Weld leakage not only leads to resource waste but can also cause environmental pollution and, in extreme cases, safety accidents. With increasingly stringent industrial standards and heightened environmental awareness, the requirements for welding quality are becoming more demanding, and the tolerance for weld leakage is decreasing. Therefore, current requirements for addressing weld leakage in cylindrical steel pipe welding are extremely stringent.

[0003] Currently, commonly used methods for detecting weld leaks include X-ray inspection, ultrasonic testing, magnetic particle inspection, and visual inspection. However, X-ray inspection is suitable for detecting smaller defects, while ultrasonic testing is more suitable for detecting deep defects. X-ray inspection equipment is expensive, and there are radiation safety requirements for operators. Ultrasonic testing may be difficult to detect welds with complex geometries and materials with high sound absorption. Magnetic particle inspection can only detect surface defects and is insensitive to deep defects. Visual inspection heavily relies on the experience of the inspector, resulting in low efficiency. Therefore, all commonly used existing methods for detecting weld leaks have limitations, especially for detecting weld spirals and upper and lower weld joints, which may affect the accuracy or efficiency of the detection.

[0004] How to improve the accuracy and efficiency of weld leak detection in cylindrical steel pipes without damaging the welded structure is an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a single-column rotating turntable type weld leakage detection and repair system and its application, which can improve the accuracy and efficiency of detecting leaks in welds of cylindrical steel pipes.

[0006] This invention provides a single-column rotating turntable type weld leakage detection and repair system, including a machine vision motion system and a detection system; the machine vision motion system is used to ensure that the detection system can scan the target weld; the detection system is used to detect the target weld; the machine vision motion system includes a control subsystem and a motion subsystem, the control subsystem is used for movement control, laser repair welding control, working condition control, alarm and signal docking; the motion subsystem includes a vision platform, a workpiece platform, a spraying structure and a support structure; a laser welding machine is installed on the spraying structure.

[0007] The present invention also provides an application of the above-described single-column rotating turntable weld leakage detection and repair system, which is applied to the detection of weld leakage in cylindrical steel pipes.

[0008] The single-column rotating turntable weld leakage detection and repair system and its application provided by this invention have the following beneficial effects: This invention uses modern image recognition technology and deep learning technology to detect weld leaks, and also utilizes colorimetric recognition technology to detect weld leaks, aiming to complement deep learning technology and achieve the requirement of zero missed detections. This invention also utilizes laser welding technology to quickly repair leak locations, improving the accuracy and efficiency of detecting weld leaks in cylindrical steel pipes. Attached Figure Description

[0009] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a block diagram of the single-column rotating turntable weld leakage detection and repair system provided by the present invention. Figure 2 This is a schematic diagram of the inspection site and weld type provided by the present invention; Figure 3 This is a schematic diagram of the overall layout of the testing equipment provided by the present invention; Figure 4 This is a schematic diagram of the vertical slide deployment scheme provided by the present invention; Figure 5 This is a schematic diagram of the vertical slide table drive and connection provided by the present invention; Figure 6 This is a schematic diagram of the rotating platform deployment scheme provided by the present invention; Figure 7 This is a schematic diagram of the relationship between frames in weld seam image acquisition provided by the present invention; Figure 8 This is a schematic diagram illustrating the practical problems that can be solved by different network models provided by this invention; Figure 9 This is a block diagram of the leakage detection module provided by the present invention; Figure 10 This is a schematic diagram illustrating different strategies for different leakage patterns provided by the present invention; Figure 11 This is a block diagram of the reinforcement learning scheme provided by the present invention; Figure 12 This is a schematic diagram illustrating an example verification provided by the present invention; Figure 13 This is a schematic diagram of the motion system for repair welding provided by the present invention; Figure 14 This is a schematic diagram of the motion control system for repair welding provided by the present invention; Figure 15This is a schematic diagram of the walking control and defect marking control provided by the present invention. Detailed Implementation

[0010] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] Figure 1 A schematic diagram of a single-column rotating turntable weld leakage detection and repair system according to this embodiment is shown. In this embodiment, the single-column rotating turntable weld leakage detection and repair system includes: A machine vision motion system and an inspection system are provided. The machine vision motion system ensures that the inspection system can scan the target weld. The inspection system is used to inspect the target weld. The machine vision motion system includes a control subsystem and a motion subsystem. The control subsystem is used for movement control, laser welding control, working condition control, alarm, and signal docking. The motion subsystem includes a vision platform, a workpiece platform, a spraying structure, and a support structure. In one exemplary embodiment, a laser welding machine is provided on the sprayed structure; In one exemplary embodiment, the movement control controls the movement of the detection equipment, enabling it to perform detection on the weld seam along a predetermined trajectory; the laser repair welding control controls the laser equipment to perform repair welding operations on detected leaks, achieving automated repair; the operating condition control controls various operating conditions during the detection process to simulate the actual working environment and improve the realism of the detection, including water injection and pressure holding; the alarm system issues an alarm signal when a serious leak is detected or the system malfunctions, reminding the operator to handle the situation; and the signal interface interacts with the detection system to achieve coordinated operation of all parts. In one exemplary embodiment, the vision platform is used to mount a vision inspection device, providing the viewing angle and image acquisition function required for inspection; the workpiece platform is used to place the workpiece to be inspected, providing fixation and support to ensure the stability of the workpiece during inspection; the spraying structure is used to repair and protect the target weld; and the support structure provides stable support for the entire motion subsystem. In one exemplary embodiment, the support structure includes a column, a vertical sliding platform, and a forward / reverse mechanism; the column is mechanically fixed to the forward / reverse mechanism and completes horizontal displacement with the forward / reverse mechanism; the vertical sliding platform is connected to the column via a precision guide rail, and the vertical sliding platform is driven by a servo motor to move up and down with a rack and pinion; the vertical sliding platform is used to mount the vision platform and the spraying structure, and drives the vision platform and the spraying structure to move vertically; In one exemplary embodiment, the workpiece platform is a rotating platform, which is supported by a thrust bearing platform on a bracket. A high-power motor drives a vertical reversing reducer via a coupling, and finally the rotating platform rotates via a gear set. A rotary joint is connected directly below the rotating platform for injecting water into the workpiece to check for leaks. The rotating platform is also equipped with a water tank and a hydraulic system. The water tank is used to store water, and the hydraulic system is used to drive hydraulic actuators. In one exemplary embodiment, the detection system includes a trajectory information processing module, a leakage detection module, a human-machine interface, and a signal docking module. The trajectory information processing module is responsible for processing the motion trajectory information of the detection system to ensure that the detection process can cover the entire weld area without missing any areas. The leakage detection module includes a leakage detection model, a reinforcement learning model, and a minor defect model. The leakage detection model is used to identify leakage points in the weld. The reinforcement learning model uses reinforcement learning technology to continuously optimize the detection model, improving the accuracy and efficiency of the detection. The minor defect model is specifically used to detect some small leakage defects in the weld, improving the sensitivity of the detection. The human-machine interface provides a human-machine interaction interface, facilitating operators to set parameters, perform detection operations, and view detection reports. The signal docking module is responsible for signal transmission and interaction with the control subsystem to ensure the coordinated operation of the entire system. In one exemplary embodiment, the inspection report includes an overview, details, and review recommendations. The overview includes the number of leaks and the inspection time. The number of leaks represents the total number of leak points in the weld, helping to assess the overall leakage situation of the weld. The inspection time records the time spent on the entire inspection process, facilitating the analysis of inspection efficiency. The details include the location and severity of leaks. The location of leaks precisely indicates the specific location of each leak point on the weld, facilitating subsequent repair work. The severity of leaks is used to assess the severity of the leaks and determine the priority of repairs. The review recommendations include manual verification and confidence level. The manual verification is used to recommend that complex or suspected leak points be manually re-inspected and confirmed to improve the accuracy of the inspection. The confidence level indicates the reliability of the inspection results, helping users judge the credibility of the inspection results. In one exemplary embodiment, the leakage detection model includes traditional image processing and a convolutional neural network; the traditional image processing includes image preprocessing, morphological processing, determining the leakage type, searching for the leakage location, and providing leakage parameters; the convolutional neural network is a YOLOv5 model trained using a leakage image dataset. As an exemplary embodiment, the leakage detection model determines the leakage type after image preprocessing and morphological processing. Different evaluation methods can be adopted for different leakage types. For imprint-type leakage, the leakage point can be predicted by finding the imprint terminal, and the size of the leakage opening can be determined by imprint width analysis, and finally the leakage parameters are obtained. Convolutional neural networks, by determining the network structure and training the network to obtain leakage parameters; The output of the leakage detection model is a two-dimensional vector (whether there is leakage) and a leakage parameter vector (including leakage location parameters, crack size estimate, and classification confidence value); the prediction results are weighted (the default is the recommended weight), which provides an interface for modification to facilitate application in more engineering environments; In one exemplary embodiment, the reinforcement learning model is a DQN model, which is used to iteratively update the classifier based on the negative evaluation given by the network when mislabeling or missing labels is found during manual detection, so as to achieve continuous learning of the system. In one exemplary embodiment, the minor defect model includes recording the location of the minor defect pattern, controlling a high-definition camera to re-move to the defect location, taking pictures of it under different lighting conditions, and performing high-dimensional recognition based on light intensity to improve the recognition effect of minor defects. If the two-dimensional vector values ​​output by the leakage detection model are similar, a manual verification mark is added to the corresponding detection location, and the manual verification result is returned to the reinforcement learning model. As an exemplary embodiment, the minor defect model, when the two-dimensional vector values ​​output by the leakage detection model are similar, should be considered minor defects and handled using minor defect processing methods. The minor defect model includes re-identification processing and annotation processing. The location of the minor defect pattern is recorded, and a high-definition camera is moved back to the defect location for different lighting conditions to perform high-dimensional recognition based on light intensity, thereby improving the identification effect of minor defects. Subsequently, if similar two-dimensional vector values ​​are still generated, a manual verification mark is added to that location. The manual verification result is returned to DQN, allowing the device to enhance its learning. The overall principle for minor defect processing is: while ensuring accurate detection, mark suspected leaks as much as possible to ensure foolproof leakage detection. In one exemplary embodiment, the leakage detection module further includes a colorimetric recognition submodule, which is used to detect weld leakage and complements the leakage detection model, reinforcement learning model, and subtle defect model. The colorimetric recognition submodule is configured to: set a red channel threshold, multiplier parameter, and input / output path; traverse the images in the input folder to ensure that the image files are complete and valid; extract the red-dominant pixel region based on the threshold and multiplier relationship of the red channel; identify the reddest point of the red region and its red channel proportion through connectivity analysis; annotate the red characteristic information on the original image and save the result image to a specified folder; and output the total processing time.

[0012] In some embodiments, the above-described single-column rotating turntable weld leakage detection and repair system can also be implemented in the following ways.

[0013] Inspection site and weld type, such as Figure 2 As shown; this relates to an automatic weld leakage detection and repair device; the weld is scanned using machine vision (the field of view of each frame is initially set to...). This system detects leaks using modern image processing and feature recognition principles. After detection, laser welding technology is used to automatically repair the leak. Steel pipe specifications: diameter... to ,length In this embodiment, in order to detect every location of the weld, the single-column rotating turntable weld leakage detection and repair system includes: 1. Machine vision motion system: This system enables industrial camera equipment to scan the weld seam and sample images at each location for subsequent leak identification; the machine vision motion system is a motion system that carries the industrial camera equipment and moves it along the weld seam. 2. Weld Leakage and Leakage Location Detection: We use modern image processing and feature recognition theories to detect weld leaks in weld images acquired by machine vision. This ensures accurate detection while marking suspected leaks as much as possible to guarantee foolproof leak detection. At the same time, colorimetric recognition technology is also used to detect weld leaks, in order to complement deep learning technology. 3. Control system: Provides comprehensive control over various working conditions required for weld inspection, machine vision motion, laser weld repair control, and information transmission; In another embodiment, the single-column rotating turntable weld leakage detection and repair system also includes a laser repair welding system: using hook welding technology, under the control of a CNC system, the leaking weld is repaired.

[0014] In this embodiment, the specific solution for the single-column rotating turntable type weld leakage detection and repair system includes: 1. Overall Technical Solution like Figure 1 The diagram shown is a block diagram of a single-column rotating turntable weld leakage detection and repair system. 2. Machine Vision Motion System Solution To ensure that the image acquisition equipment can scan the weld seam, the following motion scheme is designed, such as... Figure 3 The diagram shows the overall layout of the testing equipment. Supporting columns are designed between the floor and ceiling, with vertical sliding platforms (vertical sliding platforms) mounted on the columns. Cameras and laser welding heads are installed on the vertical sliding platforms. The sliding platforms drive image acquisition equipment to photograph the surface of the workpiece weld seam in conjunction with the rotating platform. Simultaneously, the sliding platforms also drive the laser welding heads to perform repair welding on the weld seam at the embroidery area. A rotating platform is placed on the floor, and its rotation is achieved by a high-power motor and reducer. A rotary joint is installed at the bottom of the equipment to inject water and the required high-pressure air into the workpiece. A steel pipe installation structure is designed on the rotating platform to ensure a seal during water injection, prevent leakage, and maintain pressure.

[0015] 2.1 Vertical Slide Section A vertical slide table is mounted on a column, driving the high-definition camera and laser welding system to move vertically up and down. The column, through a horizontal drive system and reverse forward and backward movement, adjusts the distance between the camera / laser welding head and the steel cylinder to accommodate the inspection of steel cylinders of different diameters. Figure 4 The diagram shows a vertical slide deployment scheme. The vertical slide is connected to the support column via two sets of precision guide rails, and is driven by a servo motor with a rack and pinion mechanism, causing the slide to move up and down. Figure 5 The diagram shown is a schematic of the vertical slide table drive and connection.

[0016] 2.2 Rotary Platform Section The rotating platform is supported by a thrust bearing platform on a bracket. A high-power motor drives a vertical directional reducer via a coupling, and ultimately the platform rotates through a gear set. A rotary joint is connected directly below the platform to inject water into the workpiece for leak detection. Figure 6 The diagram shown is a schematic of the rotating platform deployment scheme.

[0017] 3. Weld leakage and leakage location detection plan 3.1 Overall Approach to Weld Leakage Detection Through the coordinated use of mechanical, control, and algorithm systems, the field of view is focused on the weld to be inspected, employing a universal adjustable focal length of 0-5m and automatic focusing. Considering the system's real-time requirements, the algorithm must process at least one frame per centimeter in both hardware and software. Based on a detection speed of 70 meters per hour, this means processing one frame per second. Figure 7The diagram shows the relationship between frames in weld seam image acquisition. In weld seam leakage detection, three main practical issues are considered: how to determine leakage, how to improve model accuracy when encountering entirely new problems, and how to handle minor defects. Three main models are established for these purposes: a leakage detection model, a reinforcement learning model, and a minor defect classification model. Figure 8 The diagram shows the practical problems that different network models can solve.

[0018] 3.2 Basic Principles of Leakage Detection 3.2.1 Leakage Detection Model In leakage detection, a comprehensive classification model combining traditional morphological classification with CNN is established based on machine learning theory. The model outputs a two-dimensional vector (whether leakage occurs) and a leakage parameter vector (including leakage location parameters, crack size estimates, and classification confidence values). Training the convolutional neural network requires sufficient training image samples, with a large sample size and comprehensive coverage. However, in practical engineering applications, complete coverage is impossible, thus requiring a traditional classifier as a supplement. The leakage classifier design is as follows: Figure 9 The diagram shows the block diagram of the leakage detection module. The solution consists of two parts: traditional image processing and convolutional neural networks. The combination method involves predicting the structure and assigning weights (the default is the recommended weights). These weights are provided with an interface for modification, facilitating application in more engineering environments. Traditional image processing, after image preprocessing and morphological processing, determines the leakage type. Different leakage types can be evaluated using different methods. For example, for imprint-type leakage, the method of finding imprint terminals can predict the leakage point, and the imprint width analysis can determine the size of the leakage opening, etc. Finally, leakage parameters are obtained. The convolutional neural network method, by determining the network structure and training the network supercomputer, obtains leakage parameters. This embodiment uses the widely used YOLOv5 model, such as... Figure 10 The diagram illustrates different strategies for different types of leakage.

[0019] 3.2.2 Reinforcement Learning Model With the increasing use of leakage detection systems, the number of image samples is growing, and new leakage features may emerge in these new samples. To ensure the system can reliably detect these new features, the leakage detection model must be reinforced using these new samples. However, with traditional deep learning algorithms, once the deep learning model is determined, the network parameters must be redefined when new samples are added, which is inconvenient for engineers to manage the algorithm. To avoid retraining after adding new samples, this paper introduces the concept of reinforcement learning and builds an algorithm for automatically updating the model based on this. The algorithm is based on the action value function (…). ):

[0020] Build (abbreviation) If, during manual inspection, the network is found to have mislabeled or missed labels, the manual calibration result will receive a negative evaluation. Based on this, iterative processes are performed to develop an evolved classifier, enabling the system to continuously learn, such as... Figure 11 The diagram shown is a block diagram of a reinforcement learning scheme.

[0021] 3.2.3 Handling Minor Leaks When the two-dimensional vector values ​​output by the leakage detection model are similar, they should be considered minor defects and handled using methods for minor defects. Minor defect handling involves re-identification and annotation. The location of the minor defect pattern is recorded, and a high-definition camera is moved back to the defect location for imaging under different lighting conditions to perform high-dimensional identification based on light intensity, thereby improving the identification effect of minor defects. If similar two-dimensional vector values ​​are still found thereafter, a manual verification mark is added to that location. The manual verification results are then returned. The equipment can be enhanced to learn; the handling of minor defects follows these principles: while ensuring accurate detection, suspected leaks should be marked as much as possible to ensure that leak detection is foolproof.

[0022] 4. Weld Leakage Detection Technology Based on Image RGB Color Space Feature Values 4.1 Principle Explanation By thresholding and proportionally judging the red channel in the RGB space, regions dominated by red can be efficiently identified. This method relies on the following rules: (1) the red channel value must be higher than the set threshold (e.g., 200) to ensure that the red intensity of the target area is sufficiently prominent; (2) the red channel must be significantly higher than the green and blue channels to satisfy the visual perception characteristics of red in the human eye; by setting the multiple relationship between the red channel and the green and blue channels, regions dominated by red can be accurately extracted.

[0023] 4.2 Comparison with HSV and LAB spaces The characteristics of different color spaces are shown in Table 1; Table 1: Comparison of Characteristics of Different Color Spaces

[0024] Based on the requirements of the recognition task, this embodiment operates directly in the RGB space, which can efficiently meet the extraction of red features in the weld area.

[0025] 4.3 Overall Code Structure 4.3.1 Process Overview The main flow of the code is as follows: (1) Initialization and parameter setting: Set the red channel threshold, multiplier parameter and input / output path. (2) Image loading and inspection: Traverse the images in the input folder to ensure that the image files are complete and valid. (3) Red pixel extraction and analysis: Extract the red-dominant pixel region based on the threshold and multiplier relationship of the red channel. (4) Connectivity detection and feature calculation: Identify the reddest point of the red region and its red channel ratio through connectivity analysis. (5) Feature annotation and result saving: Annotate the red feature information on the original image and save the result image to the specified folder. (6) Running time statistics: Output the total processing time.

[0026] 4.3.2 Programming Table 2 shows the complete code and step-by-step annotations: Table 2: Complete Code and Step-by-Step Annotations

[0027] like Figure 12 The diagram shown is a sample verification diagram.

[0028] 5. Laser repair welding of leaks in PCCP steel cylinder welds Laser welding replaces manual welding, achieving the same technical requirements while saving manpower. It produces no smoke or intense light, significantly benefiting environmental protection and worker health. The laser welding system consists of a motion control system and a laser welding system. The motion control system directs the laser welding torch along a predetermined trajectory to repair the designated weld seam.

[0029] 5.1 Musculoskeletal System like Figure 13 The diagram shows a motion system for repair welding. Under the control of the motion system, the welding torch can move in the X and Y directions, forming any planar trajectory. Under the control of the laser welding system, it can weld the set weld seam.

[0030] 5.2 Motion Control System like Figure 14 The diagram shows a motion control system for weld repair. Based on digital control technology, the motion control system is designed. Following the set weld seam trajectory, the servo system moves under the control of the main controller, simultaneously controlling the laser welding system to perform the welding, thus achieving the designed weld seam.

[0031] like Figure 15The diagram shows the walking control and defect marking control; it is a video monitoring system with camera capture as its core. It consists of two parts: a host computer and a slave computer. The host computer mainly comprises a human-machine interface, capable of video display, command sending, and data feedback. The slave computer includes four modules: a camera capture module, an image data acquisition and processing module, a neural network prediction module, and a motor control module. Through pre-defined rules, this system can achieve automatic tracking of the moving vehicle by the camera without human intervention. Observers can also send relevant commands through the host computer's display interface to directly obtain various information about the intelligent vehicle.

[0032] Based on the overall architecture of the video surveillance system, this embodiment focuses on the design of a neural network predictor, elucidating the advantages of using an error backpropagation neural network to predict the vehicle's position. Simulation experiments of the neural network predictor are conducted in Matlab, and further related practical hardware operations are discussed. The final simulation results of this embodiment compare the vehicle's path with the camera's tracking trajectory, showing that the neural network can perfectly predict the vehicle's position at the next moment and guide the camera for accurate real-time tracking. Under simulated real-world sampling frequencies, the system's error can be controlled within a reasonable range.

[0033] 5.3. Laser welding system: A mature laser welding system available on the market is adopted.

[0034] This embodiment provides an application of the single-column rotating turntable weld leakage detection and repair system described above, which is applied to the detection of weld leakage in cylindrical steel pipes.

[0035] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A single-column rotating turntable type weld leakage detection and repair system, characterized in that, It includes a machine vision motion system and a detection system; the machine vision motion system is used to ensure that the detection system can scan the target weld; the detection system is used to detect the target weld; the machine vision motion system includes a control subsystem and a motion subsystem, the control subsystem is used for movement control, laser welding control, working condition control, alarm and signal docking; the motion subsystem includes a vision platform, a workpiece platform, a spraying structure and a support structure; a laser welding machine is installed on the spraying structure.

2. The single-column rotating turntable type weld leakage detection and repair system according to claim 1, characterized in that, The movement control controls the movement of the detection equipment, enabling it to perform detection on the weld seam along a predetermined trajectory. The laser repair welding control controls the laser equipment to perform repair welding operations on detected leaks, achieving automated repair. The operating condition control controls various operating conditions during the detection process to simulate the actual working environment and improve the realism of the detection. These operating conditions include water injection and pressure holding. The alarm system issues an alarm signal when a serious leak is detected or the system malfunctions, alerting the operator to take action. The signal interface interacts with the detection system to achieve coordinated operation of all parts.

3. The single-column rotating turntable type weld leakage detection and repair system according to claim 1, characterized in that, The vision platform is used to mount the vision inspection equipment, providing the required viewing angle and image acquisition function for inspection; the workpiece platform is used to place the workpiece to be inspected, providing fixation and support to ensure the stability of the workpiece during the inspection process. The spraying structure is used to repair and protect the target weld; the support structure provides stable support for the entire motion subsystem.

4. The single-column rotating turntable type weld leakage detection and repair system according to claim 1, characterized in that, The workpiece platform is a rotating platform, which is supported by a thrust bearing platform on a bracket. A high-power motor drives a vertical reversing reducer through a coupling, and finally drives the rotating platform to rotate through a gear set. A rotary joint is connected directly below the rotating platform for injecting water into the workpiece to check for leaks. The rotating platform is also equipped with a water tank and a hydraulic system. The water tank is used to store water, and the hydraulic system is used to drive hydraulic actuators.

5. The single-column rotating turntable type weld leakage detection and repair system according to claim 1, characterized in that, The detection system includes a trajectory information processing module, a leakage detection module, a human-machine interface, and a signal docking module. The trajectory information processing module processes the motion trajectory information of the detection system to ensure that the detection process covers the entire weld area without any missed detections. The leakage detection module includes a leakage detection model, a reinforcement learning model, and a minor defect model. The leakage detection model is used to identify leakage points in the weld. The reinforcement learning model continuously optimizes the detection model using reinforcement learning techniques, improving the accuracy and efficiency of detection. The minor defect model is specifically designed to detect small leakage defects in the weld, improving detection sensitivity. The human-machine interface provides a user interface for operators to set parameters, perform detection operations, and view detection reports. The signal docking module is responsible for signal transmission and interaction with the control subsystem, ensuring the coordinated operation of the entire system.

6. The single-column rotating turntable type weld leakage detection and repair system according to claim 5, characterized in that, The leakage detection model includes traditional image processing and convolutional neural networks; the traditional image processing includes image preprocessing, morphological processing, determining the leakage type, searching for the leakage location, and providing leakage parameters; the convolutional neural network is a YOLOv5 model trained using a leakage image dataset.

7. The single-column rotating turntable type weld leakage detection and repair system according to claim 5, characterized in that, The reinforcement learning model is a DQN model, which is used to iteratively update the classifier based on the negative evaluations given by the network when mislabeling or missing labels are found during manual inspection, so as to achieve continuous learning of the system.

8. The single-column rotating turntable type weld leakage detection and repair system according to claim 5, characterized in that, The subtle defect model includes recording the location of subtle defect patterns, controlling a high-definition camera to reposition to the defect location, taking pictures of it under different lighting conditions, and performing high-dimensional recognition based on light intensity to improve the recognition effect of subtle defects. If the two-dimensional vector values ​​output by the leakage detection model are similar, a manual verification mark is added to the corresponding detection location, and the manual verification result is returned to the reinforcement learning model.

9. The single-column rotating turntable type weld leakage detection and repair system according to claim 5, characterized in that, The leakage detection module also includes a colorimetric recognition submodule, which is used to detect weld leakage and complements the leakage detection model, reinforcement learning model, and subtle defect model. The colorimetric recognition submodule is configured to: set the red channel threshold, multiplier parameter, and input / output path; traverse the images in the input folder to ensure that the image files are complete and valid; and extract the red-dominant pixel region based on the threshold and multiplier relationship of the red channel. Through connectivity analysis, the reddest point in the red region and the proportion of its red channel are identified; Mark the red feature information on the original image and save the result image to the specified folder; output the total processing time.

10. An application of the single-column rotating turntable weld leakage detection and repair system as described in any one of claims 1 to 9, characterized in that, It is used for detecting leaks in welded seams of cylindrical steel pipes.

Images