A data decision based intelligent pipe welding method and system

Abstract

The application provides a kind of method and system for pipeline intelligent welding based on data decision, it is related to the technical field of intelligent pipeline welding, including in response to welding request, the groove image of the pipeline to be welded is acquired by active vision sensor, based on the groove image, the groove topographic feature parameter of the pipeline to be welded is determined, based on the groove topographic feature parameter, the corresponding basic layer arrangement rule and reference process parameter are determined from the preset layer arrangement database, and combined with the groove topographic feature parameter, the welding process scheme matched with the current groove is generated, based on the welding process scheme, the welding operation is executed in the groove of the pipeline to be welded, solve the technical problems of global planning ability loss, local lag of correction strategy and lack of precise control in the whole process in the prior art, with the effect of improving the stability and reliability of pipeline welding quality.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent pipeline welding, and in particular to an intelligent pipeline welding method and system based on data decision-making. Background Technology

[0002] Pipeline welding, especially in high-end equipment manufacturing fields such as nuclear power and chemical industry, is a key process to ensure the integrity of pressure boundaries and structural safety. Among them, the welding quality of thick-walled pipelines is directly related to the long-term safe operation of the entire plant. Traditionally, this type of pipeline welding relies heavily on the experience and technical level of skilled welders. The setting of welding parameters and the selection of layer layout are mostly based on manual judgment, which has inherent limitations such as large quality fluctuations, low efficiency, and difficulty in digital control.

[0003] Therefore, in order to improve the level of welding automation, existing technologies have introduced welding assistance solutions using vision sensing. For example, patent CN119319350A proposes a pipe welding device and method based on vision sensors, which uses a 2D camera to identify the weld position and guide the robot to weld, thereby improving positioning accuracy. Another example is patent CN117245175A, which attempts to combine active and passive vision technologies to correct weld offset by establishing a bevel model and comparing it.

[0004] However, these solutions mostly focus on the initial positioning of the weld or local tracking and correction during the welding process. Therefore, they cannot solve the problem of differences in bevel form and assembly gap caused by the variety of specifications and fluctuations in processing accuracy of thick-walled pipelines in nuclear power plants. The system is difficult to adaptively allocate suitable, globally optimal multi-layer and multi-pass welding schemes for the diverse actual bevels, resulting in poor process adaptability. At the same time, even if deviations can be detected during the welding process, the correction strategy is often local and lagging, lacking predictive planning and precise control capabilities based on a global process database. This leads to insufficient control over welding stress and deformation caused by large welding heat input, making it difficult to meet the almost stringent stability and reliability requirements for welding quality in the nuclear power field. Summary of the Invention

[0005] In view of the above-mentioned shortcomings in the existing technology, the purpose of this invention is to provide a data-driven intelligent pipeline welding method, which has the characteristics of improving the stability and reliability of pipeline welding quality.

[0006] The above-mentioned objective of this invention is achieved through the following technical solution:

[0007] A data-driven intelligent pipeline welding method includes:

[0008] In response to a welding request, an active vision sensor acquires an image of the bevel of the pipe to be welded.

[0009] Based on the bevel image, the bevel morphology characteristic parameters of the pipe to be welded are determined;

[0010] Based on the bevel morphology feature parameters, the corresponding basic layer layout rules and benchmark process parameters are determined from the preset layer layout database, and a welding process scheme matching the current bevel is generated by combining the bevel morphology feature parameters.

[0011] Based on the aforementioned welding process, welding operations are performed within the bevel of the pipe to be welded.

[0012] By adopting the above technical solution, the bevel image is quickly acquired and the morphological feature parameters are extracted by an active vision sensor. Then, based on the bevel feature parameters, the welding process scheme is intelligently matched from a preset database to adapt to changes in different bevel specifications and assembly gaps, effectively improving the adaptability and flexibility of the welding process. Finally, by executing the optimized welding process scheme, key parameters such as heat input and layer arrangement during the welding process are precisely controlled, effectively reducing the occurrence of defects such as poor weld formation and incomplete penetration, and significantly improving the stability and reliability of welding quality. It is especially suitable for high-standard welding scenarios such as thick-walled pipelines in nuclear power plants. Through data-driven decision-making throughout the entire process, precise control of welding quality and traceability of the process are achieved.

[0013] Preferably, determining the bevel morphology characteristic parameters of the pipe to be welded based on the bevel image includes:

[0014] The bevel image is subjected to region interest extraction to obtain the target region image;

[0015] The target region image is binarized and denoised to obtain a denoised image;

[0016] Edge detection and skeleton extraction are performed on the denoised image to obtain the center contour line of the bevel;

[0017] Based on the center contour line of the bevel, the bevel morphology characteristic parameters of the pipe to be welded are calculated through coordinate transformation.

[0018] By adopting the above technical solution, the ROI region is first extracted from the bevel image, focusing on key feature areas. Then, adaptive threshold binarization and denoising are performed on the target area image to enhance the contrast between the bevel and the background. Next, edge detection and skeleton extraction are performed on the denoised image to accurately obtain the center contour line of the bevel, effectively eliminating noise interference and preserving key geometric features. Finally, based on the center contour line of the bevel, the accurate bevel morphology feature parameters are calculated through coordinate transformation, establishing an accurate mapping relationship from two-dimensional image to three-dimensional space measurement, which can meet the accurate measurement needs of bevel size under different working conditions.

[0019] Preferably, the step of determining the corresponding basic layer layout rules and benchmark process parameters from a preset layer layout database based on the bevel morphology feature parameters, and generating a welding process scheme matching the current bevel in combination with the bevel morphology feature parameters, includes:

[0020] The bevel morphology feature parameters are matched with the bevel types in the preset layer layout database to determine the successfully matched bevel types.

[0021] Based on the successfully matched bevel type, the corresponding basic layer layout rules and benchmark process parameters are retrieved from the preset layer layout database.

[0022] Based on the basic layer layout rules, the bevel morphology characteristic parameters and the benchmark process parameters, the number of weld beads required for each layer of the current bevel, the weld bevel layout position and the corresponding welding process parameters are determined.

[0023] Based on the number of weld beads required for each layer, the weld bead arrangement position, and the welding process parameters, a welding process scheme matching the current bevel is generated.

[0024] By adopting the above technical solution, the automatic identification and process adaptation of different specifications of bevels are realized by intelligently matching the bevel morphology feature parameters with the preset database. Based on the matching results, the corresponding basic layer layout rules and benchmark process parameters are called. The experience accumulation of the preset process database is fully utilized. Combined with specific bevel features and process rules, the number of weld passes, the layout position and the corresponding process parameters of each layer are accurately calculated. The precise mapping from bevel morphology to welding process is realized, and finally a welding process scheme that is completely matched with the current bevel is generated.

[0025] Preferably, the step of performing welding operations within the bevel of the pipe to be welded based on the welding process scheme includes:

[0026] Based on the aforementioned welding process scheme, the current weld bead is welded within the bevel of the pipe to be welded;

[0027] An image of the current weld bead is acquired using a passive vision sensor;

[0028] The image of the current weld bead is input into a preset image segmentation model to obtain contour information;

[0029] Based on the contour information, a three-dimensional reconstruction is performed to obtain a three-dimensional model of the current weld bead;

[0030] The three-dimensional model is compared with the preset weld bead model to determine the real-time deviation.

[0031] Based on the real-time deviation, the welding process plan is updated until the deposition of each layer in the welding process plan is completed within the bevel of the pipe to be welded, forming a weld.

[0032] By adopting the above technical solution, the images of the welded bead are acquired in real time by a passive vision sensor and input into a neural network model for accurate segmentation. This effectively overcomes the interference of welding arc light and the influence of background noise. Based on the contour information, a precise three-dimensional model of the current weld bead is generated through three-dimensional reconstruction, realizing the digital representation of the weld bead morphology. By comparing the actual weld bead model with the preset ideal model to calculate the real-time deviation, the forming deviation and positional offset in the welding process can be detected in a timely manner. Based on the real-time deviation, the welding process scheme is dynamically adjusted to form a closed loop, which effectively corrects the weld bead offset problem caused by factors such as thermal deformation and parameter fluctuations, ensuring the forming accuracy and overall quality consistency of multi-layer and multi-pass welding.

[0033] Preferably, the step of welding the current weld bead within the bevel of the pipe to be welded based on the welding process scheme includes:

[0034] Based on the weld bead arrangement position, calculate the spatial position coordinates of each weld bead in the current layer;

[0035] Based on the spatial coordinates of each weld bead, the welding path of the welding gun is generated;

[0036] Based on the welding path and the welding process parameters of the welding process scheme, welding materials are deposited in the bevel of the pipe to be welded to form a weld bead.

[0037] By adopting the above technical solution and calculating the spatial coordinate information through the arrangement of weld beads, the movement accuracy and positional accuracy of the welding torch within the complex bevel are ensured. Finally, by combining the welding process parameters, welding materials are deposited within the bevel, resulting in a high-quality weld bead with precise geometry and good interlayer fusion.

[0038] Preferably, the step of performing three-dimensional reconstruction based on the contour information to obtain a three-dimensional model of the current weld bead includes:

[0039] Obtain the acquisition time series of the passive vision sensor;

[0040] Based on the acquired time series and the contour information, the temporal position data of the weld contour is determined;

[0041] Based on the preset camera calibration parameters and the time-series position data, the three-dimensional point cloud data of the weld contour is obtained;

[0042] The three-dimensional point cloud data is filtered and fitted with a surface to obtain the three-dimensional model of the current weld bead.

[0043] By adopting the above technical solution, a precise correspondence between weld contour and timestamp was established by acquiring the acquisition time series of the passive vision sensor, providing time dimension information for dynamically tracking weld morphology changes. Based on the acquisition time series and contour information, the temporal position data of the weld contour was determined, realizing continuous digital recording of the weld growth process. By combining preset camera calibration parameters and temporal position data, the two-dimensional image information was accurately converted into three-dimensional point cloud data of the weld contour, completing the precise mapping from image space to three-dimensional physical space. Finally, the three-dimensional point cloud data was filtered and surface fitted to effectively eliminate measurement noise and abnormal point interference, generating an accurate and smooth three-dimensional weld model.

[0044] Preferably, the step of comparing the three-dimensional model with the preset weld bead model to determine the real-time deviation includes:

[0045] Geometric registration is performed between the current weld bead's 3D model and the preset weld bead model to obtain the registered model;

[0046] The Euclidean distance between corresponding points in the horizontal and depth directions of the registered model and the preset weld bead model is calculated and used as the real-time deviation.

[0047] By adopting the above technical solution, the actual welding result and the ideal design model are spatially aligned by accurately geometrically registering the current 3D model of the weld bead with the preset weld bead model. Then, the Euclidean distance between corresponding points in the horizontal and depth directions of the registered model and the preset model is calculated as the real-time deviation, realizing the quantitative evaluation of the weld bead position deviation. It can accurately identify the multi-dimensional deviation of the weld bead in terms of width, height and center position. This deviation calculation method based on three-dimensional spatial distance has higher accuracy and comprehensiveness than the traditional two-dimensional image comparison, effectively ensuring the forming accuracy and quality consistency of multi-layer and multi-pass welding.

[0048] The second objective of this invention is to provide a data-driven intelligent pipeline welding system that improves the stability and reliability of pipeline welding quality.

[0049] The second objective of this invention is achieved through the following technical solution:

[0050] A data-driven intelligent pipeline welding system includes:

[0051] The vision perception module is used to respond to welding requests and acquire bevel images of the pipe to be welded through an active vision sensor;

[0052] The feature processing module is used to determine the bevel morphology feature parameters of the pipe to be welded based on the bevel image.

[0053] The intelligent decision-making module is used to determine the corresponding basic layer layout rules and benchmark process parameters from the preset layer layout database based on the bevel morphology feature parameters, and generate a welding process scheme that matches the current bevel by combining the bevel morphology feature parameters.

[0054] The welding execution module is used to perform welding operations within the bevel of the pipe to be welded, based on the welding process scheme.

[0055] By adopting the above technical solution, the visual perception module quickly acquires bevel images through an active visual sensor, providing accurate initial data input for the system. The feature processing module accurately extracts bevel morphology feature parameters based on the bevel images, realizing the digital representation of welding features. The intelligent decision-making module intelligently matches welding process schemes from a preset database based on feature parameters, forming a data-driven process decision-making mechanism. The welding execution module accurately executes the optimized welding process scheme, realizing precise control of the welding process. The collaborative work of each module constructs a complete intelligent welding closed loop, effectively solving the problems of poor process consistency and insufficient adaptability caused by traditional welding relying on manual experience, and improving the stability and reliability of thick-walled pipe welding quality.

[0056] The third objective of this invention is to provide an electronic device that improves the stability and reliability of pipeline welding quality.

[0057] The above-mentioned objective three of this invention is achieved through the following technical solution:

[0058] An electronic device includes a memory and a processor, wherein the memory stores a computer program capable of being loaded by the processor and executing any of the above-described data-based intelligent pipe welding methods.

[0059] The fourth objective of this invention is to provide a computer storage medium capable of storing corresponding programs, which facilitates the improvement of the stability and reliability of pipeline welding quality.

[0060] The fourth objective of this invention is achieved through the following technical solution:

[0061] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed any of the above-described data-based intelligent pipe welding methods.

[0062] In summary, the present invention has at least one of the following beneficial technical effects:

[0063] 1. This invention enables the welding process to adapt to different bevel specifications and assembly gaps, and through data-driven precise control of welding parameters and layer arrangement, it effectively improves the stability, reliability and controllability of welding quality throughout the entire process.

[0064] 2. This invention effectively improves the recognition accuracy and anti-interference ability of bevel contour by combining algorithms of regional interest extraction, binarization, edge detection and skeleton extraction, and obtains accurate three-dimensional morphological parameters by means of coordinate transformation, providing key dimensional basis for adaptive welding;

[0065] 3. This invention performs dynamic three-dimensional digitization of the formed weld bead, and achieves comprehensive and accurate perception of the weld morphology by accurately converting time-series two-dimensional images into high-quality three-dimensional models. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of the steps of a data-driven intelligent pipeline welding method provided in Embodiment 1 of the present invention.

[0067] Figure 2 This is an active visual image schematic diagram of another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0068] Figure 3 This is a schematic diagram of the ROI extracted from another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0069] Figure 4 This is a binarized active visual image of another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0070] Figure 5 This is an active visual image after corrosion expansion treatment, representing another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0071] Figure 6 This is an active visual image after edge detection processing, provided in Embodiment 2 of the present invention, representing another data-driven intelligent pipeline welding method.

[0072] Figure 7 This is an active visual image after skeleton extraction of another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0073] Figure 8 This is a schematic diagram of the UNet model training sample set for another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0074] Figure 9 A schematic diagram of the process correction of another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0075] Figure 10This is a flowchart illustrating another data-driven intelligent pipeline welding method provided in Embodiment 2 of the present invention.

[0076] Figure 11 This is a structural block diagram of a data-driven intelligent pipeline welding system provided in Embodiment 3 of the present invention. Detailed Implementation

[0077] This invention provides a data-driven intelligent pipeline welding method and system to address the technical problems in existing technologies, such as lack of global planning capabilities, local lag in correction strategies, and insufficient precise control throughout the entire process. It effectively improves the stability and reliability of pipeline welding quality.

[0078] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0079] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this disclosure.

[0080] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0081] Example 1:

[0082] Please see Figure 1 The present invention provides a data-driven intelligent pipeline welding method, comprising:

[0083] Step 101: Respond to the welding request and acquire the bevel image of the pipe to be welded using an active vision sensor.

[0084] Pipes to be welded refer to thick-walled metal pipes that require circumferential butt welding in high-standard industrial settings such as nuclear power and chemical industries. Typical characteristics include significant wall thickness, materials primarily austenitic stainless steel or low-alloy high-strength steel, and butt joints requiring specific V-shaped or U-shaped bevel designs. In this embodiment, pipes to be welded specifically refer to those that have been assembled and positioned but have not yet begun welding.

[0085] Active vision sensors refer to vision inspection devices that integrate structured light projection and image acquisition. Their core consists of a blue laser and a high-resolution industrial camera, which project structured light of a specific pattern onto the bevel surface and acquire images of deformed light stripes.

[0086] A bevel image refers to a raw grayscale image containing structured light stripes acquired by an active vision sensor. The image clearly records the geometric deformation of the laser stripes on the bevel surface, reflecting the cross-sectional contour features of the bevel.

[0087] It should be noted that when the welding system receives a welding request signal from the host computer, it first activates the active vision sensor installed in front of the welding torch. The blue laser inside the sensor projects a linear structured light beam onto the bevel area of ​​the pipe to be welded. Simultaneously, the industrial camera acquires bevel images with preset exposure parameters. These exposure parameters include, but are not limited to, exposure time, gain, and frame rate, and specific exposure parameters can be set as needed.

[0088] It is worth mentioning that the blue laser wavelength used in this embodiment has a stronger ability to resist arc light interference, and the specific wavelength of 450nm is effectively offset from the welding arc light spectrum, which can improve the image signal-to-noise ratio.

[0089] In practice, the active vision sensor is connected to the industrial control computer via a gigabit Ethernet interface to ensure the real-time and stable transmission of image data. The sensor is precisely calibrated so that its light plane is at a fixed angle to the welding torch axis, ensuring that the acquired bevel image can completely cover the characteristic area of ​​the V-shaped or U-shaped bevel. During the image acquisition process, the spatial pose information of the sensor is also recorded simultaneously.

[0090] Preferably, the fixed included angle is in the range of 30°-45°.

[0091] It should be noted that before the formal acquisition, the bevel surface will be pre-processed and inspected. If impurities such as rust, oil, or spatter that affect the imaging quality are detected on the bevel surface, the cleaning station will be automatically triggered to clean the welding torch and sensor, and the operator will be prompted to grind the bevel to ensure that the quality of the acquired bevel image meets the feature extraction requirements.

[0092] The acquired raw bevel image will be temporarily stored in the industrial control computer's memory in 16-bit grayscale format. The image resolution is fixed at 2048×1536 pixels, and the grayscale value of each pixel ranges from 0 to 65535. This high dynamic range image format preserves sufficient detail information.

[0093] In this embodiment of the invention, in response to a welding request, an active vision sensor acquires an image of the bevel of the pipe to be welded.

[0094] Step 102: Based on the bevel image, determine the bevel morphology characteristic parameters of the pipe to be welded.

[0095] Bevel morphology features refer to the quantified geometric features extracted from bevel images using image processing algorithms, including but not limited to key dimensional parameters such as bevel angle, blunt edge thickness, bevel depth, and bevel width.

[0096] In this embodiment of the invention, the bevel image is subjected to ROI region extraction, adaptive threshold binarization, Sobel operator edge detection to obtain sub-pixel precision bevel boundaries, skeleton extraction algorithm to refine the double-pixel width edge into a single-pixel width center contour line, and cubic spline interpolation to smooth the contour line. Finally, the two-dimensional pixel coordinates are converted into three-dimensional world coordinates through a preset coordinate transformation model, thereby obtaining the bevel angle, blunt edge thickness, bevel depth, and bevel width parameters of the pipe to be welded.

[0097] Step 103: Based on the bevel morphology feature parameters, determine the corresponding basic layer layout rules and benchmark process parameters from the preset layer layout database, and generate a welding process scheme that matches the current bevel by combining the bevel morphology feature parameters.

[0098] The preset layer layout database refers to the process knowledge base built through a large number of welding process experiments in the early stage. It is stored in the form of a relational database. The core data table includes, but is not limited to, fields such as groove type, groove size range, number of layers, weld layout pattern, and welding process parameters, which record the mapping relationship between different groove morphologies and the optimal welding scheme.

[0099] A welding process plan refers to a complete set of work instructions generated to complete a specific bevel welding operation. This includes, but is not limited to, key parameters such as: the current number of welding layers, the number of weld passes per layer, the coordinates of the weld pass layout, welding current, arc voltage, welding speed, and wire feed speed.

[0100] It is worth mentioning that the construction of the layer layout database in this embodiment adopts a data-driven method based on a large number of welding experiments. In specific implementation:

[0101] Based on the actual needs of nuclear power plant sites, a full-factor welding test scheme was designed for typical pipe specifications and bevel forms in the nuclear power field. The test scope comprehensively covers two typical materials, carbon steel and stainless steel, and includes two types of standard bevels, V-shaped and U-shaped. The experimental design covers three key stages: filler weld, capping weld and assembly gap. By precisely controlling different combinations of process parameters such as current, welding speed and wire feed speed, the influence of these parameters on weld formation quality is studied.

[0102] In the filler weld test, automated welding was performed in all positions for pipes of different specifications according to set parameter combinations. Key indicators such as layer arrangement, penetration depth, and weld width were recorded in detail under different parameters. The cap weld test, based on the optimization results of the filler weld, further fine-tuned the process parameters to obtain the best surface finish quality. In particular, the assembly gap test, by changing the groove width while keeping other parameters basically constant, specifically studied the influence mechanism of groove size fluctuations on weld formation. Throughout all tests, high-precision sensors were used to collect process parameters such as welding current, voltage, and heat input in real time, and the geometric dimensions of the weld were accurately measured using a laser vision system.

[0103] During the experimental data collection phase, the complete combination of process parameters, the corresponding layer layout scheme, and the final weld formation effect of each group of experiments were recorded. Statistical methods and data mining techniques were used to analyze the massive amount of collected data, and a quantitative relationship model between welding parameters and layer layout was established. By summarizing the patterns of a large amount of experimental data under different bevel forms, a knowledge base of the correspondence between bevel, team gap, and process data was formed.

[0104] Based on the analysis results, a relational database model was used to construct the architecture of the layer layout database. The database stores key information fields, including but not limited to bevel type, bevel size range, number of layers, weld bead layout order, overlap rate, and corresponding welding process parameters. By establishing the relationships between these fields, the structured storage and rapid retrieval capabilities of the data are ensured.

[0105] It should be noted that valid data obtained from welding tests need to be cleaned and preprocessed before being entered into the database, and appropriate indexes should be created to optimize query efficiency.

[0106] It is worth mentioning that, to verify the accuracy and usability of the database, rigorous internal and external testing was conducted during its construction. Internal testing employed automated testing tools to comprehensively examine the database's query functionality, data integrity, and response speed. External verification involved reproducing the welding process using typical welding cases and comparing the actual welding results with the layer layout strategy recommended by the database. Based on the test results and user feedback, the database was continuously optimized and adjusted, including data structure optimization, query logic optimization, and error correction, ultimately forming a complete process database covering the parameter combinations required for multi-layer, multi-pass welding path planning.

[0107] It is worth mentioning that, in order to improve query efficiency, the database has also established an index with keywords such as bevel type and bevel width to ensure that the optimal welding process solution can be matched and retrieved in a short time.

[0108] In this embodiment of the invention, a preset layer layout database is first constructed, and the bevel morphology characteristic parameters are input into the database for querying to obtain the optimal welding process scheme.

[0109] Step 104: Based on the welding process plan, perform welding operations within the bevel of the pipe to be welded.

[0110] Welding operation refers to the automated process of controlling the welding system to complete the deposition of welding materials according to the process plan, including multiple coordinated actions such as welding torch movement control, welding energy control, and wire feeding control.

[0111] It is worth mentioning that, in practice, after each welding pass is completed, the system will automatically detect the interpass temperature using an infrared thermometer installed behind the welding torch. When the temperature exceeds the 150-250℃ range specified in the process plan, the system will automatically pause welding and start the cooling device. Welding will continue only after the temperature returns to the permissible range, in order to avoid welding defects caused by uncontrolled heat input.

[0112] It should be noted that key parameters during the welding process (including actual current, voltage, welding speed, interpass temperature, etc.) are collected and recorded in the process database in real time. This data is not only used for monitoring the current welding process, but also for real-time assessment and traceability of welding quality by comparing it with the target values ​​in the process plan.

[0113] It is worth mentioning that after welding is completed, the system can prompt the operator with audible and visual signals, and automatically generate a welding record report by combining the actual welding parameters and quality inspection data. The whole process realizes the precise conversion from process plan to high-quality weld formation, so as to ensure the reliability and consistency of thick-walled pipe welding.

[0114] In this embodiment of the invention, the system first calls the weld bead arrangement position and welding process parameters of the first layer in the welding process plan to perform welding. After the welding of this layer is completed, the system automatically calls the parameters of the next layer to continue the process until all layers are welded.

[0115] Example 2:

[0116] Please see Figures 2-10 Another data-driven intelligent pipeline welding method provided by the present invention includes:

[0117] Step 201: Respond to the welding request and acquire the bevel image of the pipe to be welded using an active vision sensor.

[0118] The specific implementation process of step 201 is similar to that of step 101, and will not be repeated here.

[0119] In this embodiment of the invention, in response to a welding request, an active vision sensor acquires an image of the bevel of the pipe to be welded.

[0120] Step 202: Based on the bevel image, determine the bevel morphology characteristic parameters of the pipe to be welded.

[0121] Preferably, step 202 may include the following sub-steps:

[0122] S11. Extract regional interests from the bevel image to obtain the target region image.

[0123] The target area image refers to a rectangular sub-image extracted from the original bevel image based on a preset pixel range or dynamically identified structured light position. This image excludes background interference areas unrelated to the bevel features in the original image, allowing subsequent image processing computational resources to be concentrated on the key area containing the bevel outline.

[0124] It should be noted that regional interest extraction can be performed on the beveling image in two ways:

[0125] The first method is the ROI extraction method based on a fixed pixel range. The processing area is directly defined by a preset pixel coordinate range. This method has the advantages of simple code implementation, fast processing speed, and convenient subsequent coordinate calculation. It is particularly suitable for working environments where the structured light position is relatively fixed.

[0126] The second method is a dynamic ROI extraction method based on grayscale features. By identifying the structured light stripe with the highest grayscale value in the image, the processing area is dynamically delineated with it as the center. This method can adapt to small changes in the position of the structured light and achieve more accurate ROI region division, but it will increase the algorithm complexity and running time.

[0127] In practical applications, the system can automatically select the appropriate ROI extraction strategy according to the welding process requirements. When the structured light position fluctuation is detected to exceed the preset threshold, it can also automatically switch to dynamic ROI extraction mode to ensure that accurate bevel feature parameters can be obtained under various working conditions.

[0128] Preferably, this embodiment adopts the first fixed ROI region extraction method, which extracts the fixed ROI region based on the prior position information of structured light in the image, and limits the processing area to a rectangular window with a width of 800 pixels centered on the laser stripe, so as to ensure the real-time requirements of the welding process while ensuring the processing accuracy.

[0129] S12. Perform binarization and denoising on the target region image to obtain the denoised image;

[0130] The denoised image refers to the black and white image obtained after adaptive threshold binarization of the target area image. The pixel grayscale values ​​only include two values: 0 (black) and 255 (white). By separating the structured light stripes at the bevel from the background, the contrast of the bevel features is enhanced.

[0131] Adaptive threshold binarization is applied to the target region image, which dynamically calculates the segmentation threshold for each pixel by analyzing the gray-level distribution characteristics of the local image region. The threshold calculation formula can be expressed as:

[0132]

[0133] In the formula, For local adaptive threshold, The grayscale mean of the local neighborhood. is the standard deviation of this neighborhood, and k is an adjustment factor set according to the on-site reflectivity conditions (usually between 1.5 and 2.0).

[0134] It should be noted that the range of values ​​for the coefficient k is an empirical value obtained through statistical analysis of a certain number of bevel images under different reflective conditions, which can achieve the optimal segmentation effect in this type of welding scenario.

[0135] It is worth mentioning that by binarizing the image, the problem of local over-brightness or under-darkness caused by uneven reflection on the welding bevel surface can be overcome. This allows the structured light stripes representing the bevel to be distinguished from the background to the greatest extent in the final image, thus enhancing the contrast between the two.

[0136] Furthermore, an optimization algorithm combining morphological erosion and dilation is used to denoise the binary image. Specifically, the original binary image is eroded to effectively eliminate isolated noise points and small burrs with areas smaller than the structuring element. Then, the eroded image is dilated to restore the effective contour area that has shrunk due to erosion and fill in the breaks and holes in the contour, finally obtaining a denoised image with good connectivity and clear boundaries.

[0137] Furthermore, the denoised image is processed using edge detection algorithms, including but not limited to Sobel, Canny, Prewitt, and Laplacian operators, which can be selected according to actual needs.

[0138] Preferably, this embodiment uses the Sobel operator to process the denoised image. By calculating the approximate gradient values ​​of each pixel in the horizontal and vertical directions, it effectively captures the gray-level change features of the bevel contour in the X and Y directions. Specifically:

[0139] 3×3 Sobel convolution kernels were used respectively and Perform convolution operations on the denoised image and calculate the gradient magnitude of each pixel. and direction By setting an appropriate gradient threshold, edge pixels with amplitudes greater than the threshold are retained to form a continuous bevel boundary profile.

[0140] It should be noted that an initial gradient threshold can be set according to the actual detection requirements of the weld bevel. In actual application, this threshold can be adaptively adjusted. That is, when the system detects obvious breakage in the boundary contour, the gradient threshold is automatically and appropriately reduced to enhance edge continuity, or when the boundary contour contains too many noise points, the gradient threshold is increased accordingly to improve edge quality. The dynamic adjustment strategy ensures that the optimal edge detection effect can be obtained under different working conditions.

[0141] S13. Perform edge detection and skeleton extraction on the denoised image to obtain the center contour line of the bevel.

[0142] The bevel centerline refers to the single-pixel-width bevel centerline obtained by edge detection and skeleton extraction of a binary image. This outline accurately describes the geometric features of the bevel cross-section.

[0143] It should be noted that in this embodiment, the Zhang-Suen parallel thinning algorithm is used to extract the skeleton of the bevel boundary contour, the two-pixel width edge is thinned into a single-pixel width center contour line, and the contour line is smoothed by cubic spline interpolation to obtain the center contour line of the bevel.

[0144] It is worth mentioning that in weld recognition, skeleton extraction can highlight the centerline or main features of the weld, making the shape of the weld easier to describe and analyze. Through skeleton extraction, the complexity of image data can be effectively reduced, noise interference can be reduced, and the accuracy and stability of weld recognition can be improved. In addition, skeleton extraction can also help detect the connectivity and integrity of the weld, further optimizing the quality control and process analysis of the weld.

[0145] S14. Based on the center contour line of the bevel, calculate the bevel morphology characteristic parameters of the pipe to be welded through coordinate transformation.

[0146] Equally spaced sampling is performed on the center contour line of the bevel to obtain a series of discrete two-dimensional pixel coordinate points as feature points representing the geometry of the bevel. The extracted bevel feature points are then transformed from two-dimensional pixel coordinates to three-dimensional world coordinates using a preset coordinate transformation model.

[0147] It should be noted that this transformation model is based on camera calibration parameters. Its transformation matrix integrates the camera's intrinsic and extrinsic parameter matrices. By solving the geometric relationship between the structured light plane equation and the camera's optical center, the model uses the principle of triangulation to calculate the precise coordinates of the bevel profile feature points in three-dimensional space point by point, thereby constructing three-dimensional point cloud data of the bevel cross-sectional shape.

[0148] Therefore, the system processes and analyzes the three-dimensional feature point cloud through geometric algorithms to calculate the key bevel morphology parameters: by fitting the inclined surfaces on both sides of the bevel and calculating their included angle, the bevel angle is obtained; by locating and measuring the vertical distance between the plane where the bottom blunt edge of the bevel is located and the reference plane at the top of the bevel, the blunt edge thickness and bevel depth are obtained; and by calculating the horizontal distance between the feature points on both sides of the opening at the top of the bevel, the bevel width is obtained.

[0149] In this embodiment of the invention, the five-step processing flow of fixed ROI region extraction, adaptive threshold binarization, erosion and dilation denoising, Sobel operator edge detection, and skeleton extraction effectively improves the recognition accuracy of bevel size by active vision.

[0150] Step 203: Based on the bevel morphology feature parameters, determine the corresponding basic layer layout rules and benchmark process parameters from the preset layer layout database, and generate a welding process scheme that matches the current bevel by combining the bevel morphology feature parameters.

[0151] Preferably, step 203 may include the following sub-steps:

[0152] S21. Match the bevel morphology feature parameters with the bevel types in the preset layer layout database to determine the successfully matched bevel types.

[0153] Bevel type refers to the classification based on the macroscopic geometry of the bevel, and is the primary basis for selecting the basic layer layout rules and benchmark process parameters. In this embodiment, it specifically refers to the standard bevel form determined by analyzing bevel morphological characteristic parameters (such as bevel angle, bottom shape, transition arc radius, etc.).

[0154] Understandably, after receiving the bevel morphology feature parameters, the system first calls the bevel type discrimination logic built into the database, which makes a decision based on the threshold of the key parameters.

[0155] For example, when the radius of the transition arc at the bottom of the bevel is detected to be R ≈ 0 mm and the bevel angle θ ≥ 37.5°, it is determined to be a V-shaped bevel; when a significant transition arc is detected at the bottom of the bevel (e.g., R ≥ 5 mm) and the bevel angle θ ≤ 20°, it is determined to be a U-shaped bevel. This matching method based on rules and geometric features can effectively overcome the identification interference caused by bevel processing tolerances and ensure the accuracy of classification.

[0156] S22. Based on the successfully matched bevel type, retrieve the corresponding basic layer layout rules and benchmark process parameters from the preset layer layout database.

[0157] The basic layer layout rules refer to a set of mathematical calculation models or empirical logic encapsulated in a database. The input is the geometric dimensions of the bevel, and the output is the spatial layout strategy of the weld beads, such as the number of weld beads per layer, the weld bead spacing, and the layout order.

[0158] The reference process parameters refer to a set of standard welding parameters that have been verified through a large number of basic process tests for a certain type of bevel and can be used as a calculation reference. These parameters include, but are not limited to, welding current, arc voltage, welding speed, and wire feed speed.

[0159] Understandably, after successfully matching the bevel type, the system uses the database query interface to quickly retrieve and call the basic layer layout rules bound to it, using the bevel type as an index, and at the same time obtains a set of optimized baseline process parameters.

[0160] It should be noted that the reference process parameters here are usually reference values ​​set for the root weld or standard filler layer of this type of bevel.

[0161] S23. Based on the basic layer layout rules, bevel morphology parameters and benchmark process parameters, determine the number of weld beads required for each layer of the current bevel, the weld bead layout position and the corresponding welding process parameters.

[0162] The number of weld passes required per layer refers to the total number of single weld passes arranged in parallel to completely fill the bevel cross-sectional space of a specific weld layer; it is an integer greater than or equal to 1.

[0163] The weld bead layout position refers to the precisely planned geometric center position for each weld bead within the cross-sectional coordinate system of the bevel.

[0164] Welding process parameters refer to a set of equipment operation command values ​​set to perform a specific weld operation. They directly control the energy input and molten pool behavior during the welding process and are the core commands that ultimately drive the welding equipment to execute.

[0165] Understandably, this step is achieved through a layer layout decision model trained on a large amount of welding process data. That is, by inputting the basic layer layout rules, bevel morphology characteristic parameters and benchmark process parameters into the model, the required number of weld beads for each layer of the current bevel, the weld bevel layout position and the corresponding welding process parameters can be directly output.

[0166] It should be noted that the layer layout decision model is a data-driven model trained with a large amount of welding process test data. The model takes groove morphology characteristic parameters, basic layer layout rules and benchmark process parameters as inputs, and through the complex mapping relationship learned internally, it directly outputs the number of weld beads per layer required to completely fill the current groove, the precise layout position of each weld bead in the groove section and the matching refined welding process parameters.

[0167] Understandably, the model construction involves: collecting a large amount of sample data containing different bevel morphologies (such as V-shaped, U-shaped and their specific dimensions), welding parameters (current, voltage, speed) and corresponding weld formation quality (such as weld width, reinforcement height, number of passes). Then, based on this sample dataset, a random forest regression algorithm is used for supervised learning training. The bevel morphology feature parameters, basic pass arrangement rules and benchmark process parameters are used as input features, and the validated optimal number of weld passes, arrangement position and process parameters are used as output labels, thereby establishing a nonlinear mapping relationship from input features to output results.

[0168] S24. Based on the number of weld beads required for each layer, the weld bead layout position, and the welding process parameters, generate a welding process scheme that matches the current bevel.

[0169] Understandably, the system integrates all the structured data calculated by S23, namely the number of welding layers, the number of weld beads in each layer, the precise arrangement position of each weld bead, and the refined welding process parameters corresponding to each weld bead, into a complete instruction sequence that can be read and executed by the welding execution system (such as a welding robot or intelligent welding machine).

[0170] To more clearly illustrate how this embodiment adapts to different bevels, two specific examples are provided below:

[0171] Example 1 (V-shaped bevel): The system acquires an image of a carbon steel pipe bevel through visual perception. After calculation, the bevel angle is found to be 38°, the blunt edge is 1.2mm, and the bevel width is 12mm. It is matched as a "V-shaped bevel" type. The corresponding rules and initial parameters are retrieved from the database, and the process plan is finally generated through the mathematical model of weld bead arrangement: a total of 5 layers of welding are performed. The first layer (root pass) is 1 pass with a current of 150A; the second layer is 2 passes with a current of 180A and a center-to-center distance of 4.5mm between the two passes; the third to fifth layers are 2 passes, 3 passes, and 1 pass (capping pass) respectively. The welding path and parameters for each layer are given.

[0172] Example 2 (U-shaped bevel): The system processes a stainless steel pipe bevel, whose characteristic parameters show it to be U-shaped with a bottom arc radius R=6mm and a bevel depth of 20mm. After successful system matching, the model generates the following process scheme: a total of 8 welding layers, with the layer arrangement adopting a "bottom-concentrated, upward-divergent" strategy. The overlap rate of each weld pass in the middle filler layer is controlled at 40%~50%, and the wire feed speed is fine-tuned through the model to adapt to good fusion of the U-shaped bevel sidewalls.

[0173] In this embodiment of the invention, a welding process scheme matching the current bevel is generated based on the number of weld beads required for each layer, the weld bead arrangement position, and the welding process parameters.

[0174] Step 204: Based on the welding process plan, weld the current weld bead inside the bevel of the pipe to be welded.

[0175] Welding the current weld bead refers to the process of controlling the welding torch at a designated position on the pipe bevel according to the generated welding process plan, and performing a continuous weld deposition operation according to the predetermined path and parameters to form a single weld bead with specific geometric dimensions and metallurgical quality.

[0176] Preferably, step 204 may include the following sub-steps:

[0177] S31. Based on the weld bead arrangement position, calculate the spatial position coordinates of each weld bead in the current layer.

[0178] Spatial position coordinates refer to the six-dimensional pose data of the path points that the welding torch tip needs to follow in the pipeline world coordinate system, which is obtained by three-dimensional spatial mapping calculation. It includes three-dimensional position (X, Y, Z) and three-dimensional orientation (Rx, Ry, Rz).

[0179] In practice, the center point of the weld bead in the two-dimensional groove section coordinate system is mapped to three-dimensional space to obtain the spatial position coordinates of each weld bead. The specific conversion method is a conventional technique in this field and will not be elaborated here.

[0180] S32. Generate the welding path of the welding gun based on the spatial coordinates of each weld bead.

[0181] Welding path refers to the continuous trajectory of the welding torch traversing all spatial coordinate points according to a predetermined sequence and motion law.

[0182] Understandably, the generation of this path first follows a preset optimization principle, automatically planning the welding sequence of all weld passes. For example, for a layer containing multiple weld passes, the system will adopt a welding sequence of "center first, then both sides" or "alternating left and right jumps" to balance heat input and control welding stress and deformation.

[0183] After determining the welding sequence, the continuous welding torch movement trajectory is constructed: for the selected first weld pass, its spatial coordinate sequence is directly defined as the main welding path of that pass. Once the first weld pass is completed, the system does not immediately move to the starting point of the next pass. Instead, it first generates an idle movement path: the welding torch is first vertically raised from the end point of the current weld pass to a preset safe height, then moves rapidly in a straight line on that height plane to directly above the starting point of the next weld pass, and finally descends vertically to that starting point. This "lift-translation-descent" idle movement pattern is repeatedly applied to the connection between all subsequent weld passes to ensure that there is no risk of collision between the welding torch and the workpiece or fixture during movement.

[0184] Finally, the main welding paths of all weld beads arranged in the optimized order are connected end to end with the empty movement paths that connect them, forming a complete welding path that starts from a safe position and traverses all weld beads to be welded.

[0185] S33. Based on the welding path and welding process scheme, welding materials are deposited in the bevel of the pipe to be welded to form a weld bead.

[0186] It is understandable that, based on the welding path and welding process parameters in the welding process plan obtained in S22, the welding materials are deposited collaboratively within the groove to form a weld bead.

[0187] In this embodiment of the invention, based on the welding process scheme, the current weld bead is welded within the bevel of the pipe to be welded.

[0188] Step 205: Acquire an image of the formed weld bead using a passive vision sensor.

[0189] Passive vision sensors refer to visual inspection devices that do not actively project structured light sources, but instead rely on natural ambient light or the arc light generated during the welding process as illumination sources to directly acquire images of the target area. In this embodiment, it specifically refers to a high dynamic range industrial camera installed behind the welding torch at a fixed offset angle to the direction of the welding torch's movement. This camera integrates an optical bandpass filter and is equipped with an active cooling device to cope with the high-temperature environment of the welding zone.

[0190] An image of a formed weld bead refers to a two-dimensional digital image captured by a passive vision sensor after the welding arc has been extinguished but before the molten pool has completely solidified or cooled to room temperature. The image contains the current weld bead and the heat-affected zones on both sides. The image clearly records the weld bead's reinforcement height, weld width, fusion with the bevel sidewall, and surface forming features (such as weld bead, undercut, etc.).

[0191] It should be noted that when the system detects that the current weld has ended (welding current and voltage return to zero), it will wait for a preset delay time (usually 100-500 milliseconds). Once the arc has extinguished, the strong light interference has disappeared, and the surface of the molten pool has initially solidified, the passive vision sensor will be immediately triggered to acquire an image. The sensor's exposure parameters have been pre-optimized to ensure that a high-contrast weld contour image can be obtained under ambient light and residual heat radiation conditions.

[0192] It is worth mentioning that the optical bandpass filter used in this embodiment typically has a center wavelength selected in the near-infrared band above 800nm, or corresponds to a specific dark line region in the welding arc spectrum, thereby effectively filtering out most of the interference from welding residual radiation and ambient stray light, highlighting the grayscale contrast between the weld bead and the base material due to differences in temperature and surface condition, and improving the image signal-to-noise ratio.

[0193] In practice, the passive vision sensor establishes a high-speed connection with the industrial computer via a GigE Vision or Camera Link interface. The sensor is mounted on a finely adjustable attitude mechanism, with its optical axis and the welding torch axis tilted backward in the welding direction typically set between 15° and 25°, ensuring its field of view completely covers the current weld bead and its adjacent areas before and after it. During image acquisition, the system simultaneously records the layer number and welding position information corresponding to the weld bead for precise correlation with the process plan.

[0194] Preferably, the backward tilt angle is in the range of 20°±2°. This angle can achieve the best side view effect of the weld bead profile while avoiding the welding torch body from obstructing the field of view.

[0195] It should be noted that before the actual data acquisition, the system will activate its built-in background light compensation algorithm to dynamically adjust the camera gain based on the real-time monitored ambient light level. If obvious smoke or spatter is detected on the weld surface, the system can automatically trigger the air blowing device installed next to the sensor to clean it, ensuring that the acquired weld image is clear and unobstructed.

[0196] The acquired raw weld images will be temporarily stored in the circular image buffer of the industrial computer in 8-bit or 12-bit grayscale format. The image resolution is usually set to 1280×1024 pixels to meet the dual requirements of real-time performance and accuracy of subsequent image processing algorithms.

[0197] In this embodiment of the invention, an image of the formed weld bead is acquired using a passive vision sensor.

[0198] Step 206: Input the image of the formed weld bead into the preset image segmentation model to obtain the contour information of the formed weld bead.

[0199] The preset image segmentation model refers to a machine learning model that has been trained on a large number of weld bead image samples and can accurately identify and segment the pixels of the weld bead region from the input weld bead image. In this embodiment, it specifically refers to a deep learning model based on the U-Net architecture. This model, through an encoder-decoder structure combined with skip connections, can achieve end-to-end pixel-level prediction of the weld bead contour while ensuring segmentation accuracy.

[0200] Contour information refers to the digital information output by a pre-defined image segmentation model that describes the geometry of the weld bead. Specifically, it is represented by a binary mask of the same size as the input image, where the pixel value of the weld bead region is 1 (or 255) and the pixel value of the background region is 0. This information can also be represented as a sequence of sub-pixel precision coordinate points of the weld bead contour edge.

[0201] It should be noted that a series of image preprocessing operations are required before inputting the weld image into the model. These include: firstly, performing flat-field correction on the original image to eliminate the effects of lens vignetting and uneven lighting; then, applying a contrast-limited adaptive histogram equalization algorithm to enhance the local contrast between the weld and the background; and finally, normalizing the image pixel values ​​to the [0,1] range and scaling them to the fixed input size specified by the model (e.g., 512×512 pixels).

[0202] It is worth mentioning that the U-Net model used in this embodiment uses a ResNet-34 network pre-trained on the ImageNet dataset for weight initialization in its encoder part, thereby improving the model's feature extraction capability and training convergence speed. The model's training data comes from thousands of weld images collected under different welding processes, lighting conditions, and interference backgrounds, and has been precisely annotated at the pixel level by professionals to ensure the model's robustness and generalization ability in practical applications.

[0203] In practice, the trained model is deployed on an industrial control computer using an inference engine such as ONNX or TensorRT. When the system calls the model, the preprocessed weld bead image is used as input, and the model outputs the corresponding probability map after forward inference. Then, a fixed threshold (usually set to 0.5) is applied to the probability map for binarization, followed by morphological closing operations to fill in small holes caused by noise in the binary mask and smooth the contour boundaries. Finally, an edge tracking algorithm is used to extract the closed contour line of the weld bead region with a single pixel width from the binary mask, and the coordinates of this contour line are the required contour information.

[0204] Preferably, the fixed threshold can be finely adjusted within the range of 0.3 to 0.7 according to the actual welding conditions. For welds made of highly reflective materials, the threshold can be appropriately increased to suppress false contours, while for welds with low contrast, the threshold can be appropriately decreased to ensure the integrity of the contour.

[0205] It should be noted that the system will periodically use newly acquired, labeled weld images to perform online evaluation of the model. When the model's segmentation accuracy (such as the IoU index) continues to be lower than the preset threshold, the system will prompt the operator to start the incremental learning process of the model, using new data to fine-tune the model so that it can continuously adapt to the slow changes in production line conditions.

[0206] In this embodiment of the invention, the image of the formed weld bead is input into a preset image segmentation model to obtain the contour information of the formed weld bead.

[0207] Step 207: Perform three-dimensional reconstruction based on the contour information to obtain the three-dimensional model of the current weld bead.

[0208] 3D reconstruction refers to the process of using weld contour information extracted from 2D images and combining it with the spatial pose parameters of a vision sensor to calculate and recover the geometric shape of the weld surface in 3D space.

[0209] The current 3D model of the weld bead refers to a 3D digital representation obtained through reconstruction that can fully express the geometric features of the weld bead surface. In this embodiment, it specifically refers to a curved surface model composed of triangular meshes, which accurately reflects key dimensions such as the width, height, and cross-sectional shape of the weld bead.

[0210] Preferably, step 207 may include the following sub-steps:

[0211] S41. Obtain the acquisition time series of the passive vision sensor.

[0212] Acquisition time series refers to the high-precision timestamp sequence recorded by the system for each frame of an image during the acquisition of weld seam images by a passive vision sensor.

[0213] It should be noted that the time series was acquired using a high-precision clock source on the industrial control computer, ensuring that the timestamp accuracy is at the millisecond level. This series records the absolute time of each image acquisition or the relative time relative to the welding start point.

[0214] S42. Based on the collected time series and contour information, determine the temporal position data of the weld contour.

[0215] Temporal location data refers to a sequence of contour data with spatial positional relationships obtained by associating two-dimensional contour information with acquisition time and welding speed.

[0216] It should be noted that the system calculates the longitudinal position coordinate Y corresponding to each contour based on the acquired time series and the known welding speed (obtained from the welding process parameters). The specific calculation formula is: Yᵢ = V × (tᵢ - t0), where V is the welding speed, tᵢ is the acquisition time of the i-th frame image, and t0 is the welding start time. In this way, each two-dimensional contour point cloud is assigned a precise position in the welding direction.

[0217] S43. Based on the preset camera calibration parameters and time-series position data, obtain the three-dimensional point cloud data of the weld contour.

[0218] Preset camera calibration parameters refer to the internal and external parameters of the passive vision sensor obtained in advance through camera calibration experiments.

[0219] 3D point cloud data refers to the collection of 3D spatial points obtained by transforming 2D contour points to the world coordinate system through coordinate transformation.

[0220] It should be noted that the system uses the camera's intrinsic parameter matrix and distortion coefficients to perform distortion correction on the extracted 2D contour pixel coordinates to obtain normalized camera coordinates. Then, combined with the extrinsic parameter matrix obtained through hand-eye calibration, the normalized camera coordinates are fused with the longitudinal coordinate Yᵢ obtained in S42. The coordinates (X, Y, Z) of each contour point in the 3D world coordinate system are calculated using the coordinate transformation formula, thereby generating dense 3D point cloud data.

[0221] S44. Filter and fit the three-dimensional point cloud data to obtain the three-dimensional model of the current weld bead.

[0222] Filtering refers to using algorithms such as statistical filtering or radius filtering to remove outliers and noise points in a 3D point cloud caused by measurement noise, splashes, or dust interference.

[0223] Surface fitting refers to constructing a continuous and smooth weld bead surface model based on filtered clean point clouds using triangulation algorithms or B-spline surface fitting algorithms based on sliding windows.

[0224] It should be noted that the system first uses statistical outlier removal filtering to calculate the average distance from each point to its K nearest neighbors and remove all points whose distance exceeds the mean plus n times the standard deviation. Then, the filtered point cloud is converted into a triangular mesh model.

[0225] In this embodiment of the invention, a three-dimensional reconstruction is performed based on the contour information to obtain a three-dimensional model of the current weld bead. This model realizes the accurate and digital restoration of the weld bead geometry from two dimensions to three dimensions.

[0226] Step 208: Compare the 3D model with the preset weld bead model to determine the real-time deviation.

[0227] A pre-defined weld bead model refers to a three-dimensional digital model that is pre-generated based on the welding process scheme and groove morphology parameters, through theoretical calculation or numerical simulation, and represents the ideal geometric shape that the weld bead should have at the current layer and position.

[0228] Real-time deviation refers to the difference in geometric dimensions in a specific direction, which is quantitatively calculated by comparing the actual weld bead 3D model with the preset ideal model.

[0229] Preferably, step 208 may include the following sub-steps:

[0230] S51. Perform geometric registration between the current weld bead's 3D model and the preset weld bead model to obtain the registered model.

[0231] Geometric registration refers to the process of precisely aligning the actual 3D model of the weld bead with the preset weld bead model in the same coordinate system through spatial transformation, so as to ensure that the comparison is carried out under the same spatial reference.

[0232] It should be noted that the system uses an iterative nearest-point algorithm for registration. This algorithm first selects a set of key points on a pre-defined weld bead model and then finds the corresponding nearest points on the actual 3D weld bead model. Next, it calculates an optimal rigid body transformation matrix (including rotation and translation) using mathematical methods such as singular value decomposition, minimizing the average distance between the two sets of corresponding points. This process is iterative until the transformation converges, ultimately achieving spatial alignment of the two models and obtaining the registered model.

[0233] It is worth mentioning that in this embodiment, the registration process prioritizes the selection of feature points at the junction of the weld bead and the base material on both sides, as well as the center line of the weld bead, for preliminary matching. This effectively improves the accuracy and speed of registration and avoids mismatch caused by poor forming of the top of the weld bead.

[0234] S52. Calculate the Euclidean distance between corresponding points in the horizontal and depth directions of the registered model and the preset weld bead model, and use it as the real-time deviation.

[0235] It should be noted that after completing the geometric registration, the system takes multiple cross-sections along the length of the weld bead at fixed intervals (e.g., 1 mm). On each cross-section, a series of corresponding points are sampled at equal intervals along the contour line of the preset weld bead model. For each corresponding point, the system calculates its projection point on the actual model surface after registration, and then calculates the three-dimensional Euclidean distance between the projection point and the corresponding point on the preset model.

[0236] Specifically, the three-dimensional Euclidean distance vector is decomposed into horizontal and depth components, and the distance components in these two directions are recorded as the real-time deviation at that point. Finally, a series of horizontal and depth deviation data along the entire length of the weld are output.

[0237] In this embodiment of the invention, the three-dimensional model is compared with the preset weld bead model to determine the real-time deviation, thereby realizing online and quantitative evaluation of the welding forming quality. This transforms the macroscopic weld bead morphology problem into specific and operable numerical information, effectively ensuring the final forming accuracy and internal quality of multi-layer and multi-pass welding.

[0238] Step 209: Based on the real-time deviation, update the welding process plan until the deposition of each layer in the welding process plan is completed within the bevel of the pipe to be welded, forming a weld.

[0239] It is worth mentioning that after receiving the real-time deviation, the system first determines whether the deviation exceeds the preset process tolerance window. If it does not exceed the window, the original welding process plan is maintained and continues to be executed; if it does exceed the window, the dynamic update of the welding process plan is immediately triggered.

[0240] Specifically, for horizontal offsets, the spatial coordinates of the welding torch for the next weld bead are corrected in real time through coordinate transformation to ensure accurate weld bead filling position. For depth offsets, key process parameters such as welding current and wire feed speed are dynamically adjusted based on the magnitude of the deviation and the remaining filling amount of the current layer.

[0241] It is worth mentioning that this embodiment not only compensates for the current deviation, but also predicts the thermal deformation trend in the subsequent welding process based on the welding heat accumulation model, and makes preventive adjustments to the process parameters of the subsequent 2 to 3 weld passes in advance, including the pre-correction of the welding torch posture and the optimized allocation of welding heat input.

[0242] After each update to the welding process plan, welding operations continue, and after each weld pass is completed, 3D reconstruction and deviation calculations are performed again, forming a closed-loop control cycle of welding-inspection-comparison-update. This process continues iteratively until all weld layers are deposited according to the updated process plan, ultimately forming a high-quality weld that meets nuclear power standards.

[0243] In this embodiment of the invention, the welding process plan is updated based on the real-time deviation until the deposition of each layer in the welding process plan is completed within the bevel of the pipe to be welded, forming a weld, thus ensuring the stability and consistency of the welding quality.

[0244] Example 3:

[0245] Please see Figure 11 The present invention provides a data-driven intelligent pipeline welding system, comprising:

[0246] The vision perception module 101 is used to respond to welding requests and acquire bevel images of the pipe to be welded through an active vision sensor.

[0247] The feature processing module 102 is used to determine the bevel morphology feature parameters of the pipe to be welded based on the bevel image.

[0248] The intelligent decision-making module 103 is used to determine the corresponding basic layer layout rules and benchmark process parameters from the preset layer layout database based on the bevel morphology feature parameters, and generate a welding process scheme that matches the current bevel by combining the bevel morphology feature parameters.

[0249] The welding execution module 104 is used to perform welding operations within the bevel of the pipe to be welded based on the welding process plan.

[0250] Since the above is a system corresponding to a data-driven intelligent pipeline welding method, and its implementation principle is the same as that of a data-driven intelligent pipeline welding method, for the sake of convenience and brevity, those skilled in the art can clearly understand that the specific working process of the system and modules described above can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here.

[0251] Example 4:

[0252] An electronic device according to an embodiment of the present invention includes: a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs a data-based intelligent pipe welding method as described in any of the above embodiments.

[0253] The memory can be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. The memory has storage space for program code used to perform any of the method steps described above. For example, the storage space for program code may include individual program codes for implementing the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, compact discs (CDs), memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above.

[0254] Example 5:

[0255] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed, implements the data-based intelligent pipe welding method of any of the above embodiments.

[0256] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0257] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0258] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0259] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0260] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0261] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data-driven intelligent pipeline welding method, characterized in that, include: In response to a welding request, an active vision sensor acquires an image of the bevel of the pipe to be welded. Based on the bevel image, the bevel morphology characteristic parameters of the pipe to be welded are determined; Based on the bevel morphology feature parameters, the corresponding basic layer layout rules and benchmark process parameters are determined from the preset layer layout database, and a welding process scheme matching the current bevel is generated by combining the bevel morphology feature parameters. Based on the aforementioned welding process scheme, welding operations are performed within the bevel of the pipe to be welded. The step of determining the corresponding basic layer layout rules and benchmark process parameters from a preset layer layout database based on the bevel morphology feature parameters, and generating a welding process scheme matching the current bevel by combining the bevel morphology feature parameters, includes: The bevel morphology feature parameters are matched with the bevel types in the preset layer layout database to determine the successfully matched bevel types. Based on the successfully matched bevel type, the corresponding basic layer layout rules and benchmark process parameters are retrieved from the preset layer layout database. Based on the basic layer layout rules, the bevel morphology characteristic parameters and the benchmark process parameters, the number of weld beads required for each layer of the current bevel, the weld bevel layout position and the corresponding welding process parameters are determined. Based on the number of weld beads required for each layer, the weld bead arrangement position, and the welding process parameters, a welding process plan matching the current bevel is generated, wherein the welding process plan is a complete set of work instruction data generated to complete the welding of a specific bevel; The welding operation performed within the bevel of the pipe to be welded, based on the aforementioned welding process scheme, includes: Based on the aforementioned welding process scheme, the current weld bead is welded within the bevel of the pipe to be welded; An image of the current weld bead is acquired using a passive vision sensor; The image of the current weld bead is input into a preset image segmentation model to obtain contour information; Based on the contour information, a three-dimensional reconstruction is performed to obtain a three-dimensional model of the current weld bead, wherein the three-dimensional model of the current weld bead is used to reflect the width, reinforcement height, and cross-sectional shape of the weld bead; The three-dimensional model is compared with the preset weld bead model to determine the real-time deviation. Based on the real-time deviation, the welding process plan is updated until the deposition of each layer in the welding process plan is completed within the bevel of the pipe to be welded, forming a weld.

2. The intelligent pipeline welding method based on data decision-making according to claim 1, characterized in that, The step of determining the bevel morphology characteristic parameters of the pipe to be welded based on the bevel image includes: The bevel image is subjected to region interest extraction to obtain the target region image; The target region image is binarized and denoised to obtain a denoised image; Edge detection and skeleton extraction are performed on the denoised image to obtain the center contour line of the bevel; Based on the center contour line of the bevel, the bevel morphology characteristic parameters of the pipe to be welded are calculated through coordinate transformation.

3. The intelligent pipeline welding method based on data decision-making according to claim 1, characterized in that, The step of welding the current weld pass within the bevel of the pipe to be welded, based on the aforementioned welding process scheme, includes: Based on the weld bead arrangement position, calculate the spatial position coordinates of each weld bead in the current layer; Based on the spatial coordinates of each weld bead, the welding path of the welding gun is generated; Based on the welding path and the welding process parameters, welding materials are deposited in the bevel of the pipe to be welded to form a weld bead.

4. The intelligent pipeline welding method based on data decision-making according to claim 1, characterized in that, The step of performing three-dimensional reconstruction based on the contour information to obtain a three-dimensional model of the current weld bead includes: Obtain the acquisition time series of the passive vision sensor; Based on the acquired time series and the contour information, the temporal position data of the weld contour is determined; Based on the preset camera calibration parameters and the time-series position data, the three-dimensional point cloud data of the weld contour is obtained; The three-dimensional point cloud data is filtered and fitted with a surface to obtain the three-dimensional model of the current weld bead.

5. The intelligent pipeline welding method based on data decision-making according to claim 1, characterized in that, The step of comparing the three-dimensional model with the preset weld bead model to determine the real-time deviation includes: Geometric registration is performed between the current weld bead's 3D model and the preset weld bead model to obtain the registered model; The Euclidean distance between corresponding points in the horizontal and depth directions of the registered model and the preset weld bead model is calculated and used as the real-time deviation.

6. A data-driven intelligent pipeline welding system, characterized in that, The system for implementing the data-driven intelligent pipe welding method as described in claim 1 includes: The vision perception module is used to respond to welding requests and acquire bevel images of the pipe to be welded through an active vision sensor; The feature processing module is used to determine the bevel morphology feature parameters of the pipe to be welded based on the bevel image. The intelligent decision-making module is used to determine the corresponding basic layer layout rules and benchmark process parameters from the preset layer layout database based on the bevel morphology feature parameters, and generate a welding process scheme that matches the current bevel by combining the bevel morphology feature parameters. The welding execution module is used to perform welding operations within the bevel of the pipe to be welded, based on the welding process scheme.

7. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program capable of being loaded by the processor and executed as described in any one of claims 1 to 5, which is a data-driven intelligent pipe welding method.

8. A computer-readable storage medium, characterized in that, The computer program is stored and can be loaded by a processor and executed as described in any one of claims 1 to 5, which is a data-driven intelligent pipe welding method.

Images