Pipe full-position automatic welding molten pool monitoring method and system based on deep learning

Abstract

The application discloses a kind of pipeline full-position automatic welding molten pool monitoring method and system based on deep learning, it is related to welding and intelligent monitoring technical field, comprising: selecting pipeline, using set welding parameter to carry out full-position automatic continuous welding to pipeline, obtain the molten pool image at different welding positions in welding process, the molten pool area in image is identified using the target detection model based on improved YOLO v8, and the molten pool size is extracted;With the molten pool size at initial welding position as reference molten pool size, determine the change of molten pool size at other different welding positions, the correlation between molten pool size and welding parameter is combined, and the set welding parameter corresponding to different welding positions is adjusted, so that the molten pool size in welding process remains consistent;Based on adjusted welding parameter, carry out next pipeline full-position automatic continuous welding to selected pipeline.The application can realize pipeline full-position high-quality welding, effectively improve the quality and efficiency of pipeline full-position welding.

Description

Technical Field

[0001] This invention relates to the field of welding and intelligent monitoring technology, and in particular to a method and system for automatic monitoring of weld pools in all positions of pipelines based on deep learning. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] All-position automated welding of pipelines is a rapidly developing pipeline welding technology in recent years, offering advantages such as reliable weld quality, high welding efficiency, and low manual labor load. The molten pool refers to the pool-like portion of the base material melted by the heat of the welding arc. During fusion welding, the liquid metal portion with a specific geometric shape formed on the workpiece is called the molten pool. The molten pool contains a wealth of information about the physical state changes during welding, directly reflecting the weld's appearance and internal formation, thus judging the weld quality. The combination of information sensing and processing technologies with traditional welding manufacturing is becoming a trend in welding industry innovation. Applying intelligent control algorithms to welding visual inspection can simulate welder behavior by tracking the weld, monitoring the molten pool's changing state during welding, and adjusting welding parameters to achieve efficient welding and liberate human labor. Currently, in all-position pipeline welding, to ensure welding quality, it is necessary to maintain stable molten pool information (such as molten pool size). Existing methods typically use image processing and machine learning to extract the molten pool size during pipeline welding and adjust subsequent welding parameters based on the molten pool size to achieve higher-quality all-position pipeline welding.

[0004] However, existing methods still have certain problems: On the one hand, adjusting the next welding parameters by monitoring changes in the molten pool size is problematic because the accuracy of monitoring the molten pool size affects the subsequent adjustment of welding parameters, and the adjustment of welding parameters has a certain degree of uncertainty, making it difficult to guarantee good welding quality after adjustment. On the other hand, during the welding process, under the interference of complex environments such as strong arc light reflection, it is difficult to obtain molten pool images with clearer edges and richer details. Traditional image processing methods, such as edge detection and threshold segmentation, can extract the molten pool size, but they have high requirements for image quality and are easily interfered with in complex backgrounds, resulting in inaccurate extracted molten pool size, which is difficult to meet the needs of high frame rate molten pool monitoring. While machine learning methods, especially those based on convolutional neural networks, can more accurately identify and extract the molten pool size by learning from a large amount of molten pool image data, they have high computational complexity, low efficiency, rely on GPU acceleration, have high hardware costs, and lack real-time performance when processing high frame rate videos. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a deep learning-based method and system for monitoring the molten pool in all-position automatic welding of pipelines. The improved target detection algorithm rapidly detects the molten pool size, enhancing detection efficiency and accuracy. Simultaneously, it considers the influence of welding position on changes in molten pool size, analyzing the dynamic correlation between welding position, welding parameters, and molten pool size. This adds a dynamic compensation mechanism for the welding position, enabling high-quality automatic welding of pipelines in all positions and effectively improving the quality and efficiency of all-position welding of pipelines.

[0006] In a first aspect, the present invention provides a method for automatic monitoring of weld pool in all positions of pipelines based on deep learning.

[0007] A deep learning-based method for automated all-position welding pool monitoring in pipelines includes: Select a pipe and perform automatic continuous welding of the pipe in all positions using set welding parameters. Obtain images of the molten pool at different welding positions during the welding process. Use a target detection model based on improved YOLO v8 to identify the molten pool region in the image and extract the molten pool size. Using the molten pool size at the initial welding position as the reference molten pool size, the changes in the molten pool size at other different welding positions relative to the reference molten pool size are determined. By combining the correlation between the molten pool size and welding parameters, the welding parameters corresponding to different welding positions are adjusted to keep the molten pool size consistent during the welding process. Based on the welding parameters after the pipeline is adjusted in all positions, the next pipeline is automatically and continuously welded in all positions.

[0008] Secondly, the present invention provides a deep learning-based automatic welding pool monitoring system for pipelines in all positions.

[0009] A deep learning-based automated all-position welding pool monitoring system for pipelines includes: The molten pool size extraction module is used to select the pipeline, perform automatic continuous welding of the pipeline in all positions using set welding parameters, acquire molten pool images at different welding positions during the welding process, and use an object detection model based on improved YOLO v8 to identify the molten pool region in the image and extract the molten pool size. The welding parameter adjustment module is used to determine the changes in the size of the molten pool at other welding positions relative to the reference molten pool size, based on the molten pool size at the initial welding position. It then combines the relationship between the molten pool size and the welding parameters to adjust the set welding parameters for different welding positions, so that the molten pool size remains consistent during the welding process. The pipeline all-position welding module is used to perform automatic continuous all-position welding on the selected pipeline based on the welding parameters adjusted for pipeline all-position.

[0010] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-described deep learning-based automatic welding pool monitoring method for pipelines in all positions when executing the executable instructions stored in the memory.

[0011] Fourthly, the present invention also provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-described deep learning-based method for monitoring the molten pool of welded pipes in all positions.

[0012] Fifthly, the present invention also provides a computer program product comprising executable instructions stored in a computer-readable storage medium; wherein, when the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the above-mentioned deep learning-based method for monitoring the molten pool of welded pipes in all positions is realized.

[0013] The above one or more technical solutions have the following beneficial effects: 1. This invention proposes a deep learning-based method and system for monitoring the molten pool in all positions of pipeline welding. By improving the target detection algorithm, the molten pool size can be quickly detected, thereby improving detection efficiency and accuracy. At the same time, the influence of welding position on the change of molten pool size is considered, and the dynamic correlation between welding position, welding parameters and molten pool size is analyzed to increase the dynamic compensation mechanism of welding position, realize high-quality automatic welding of pipelines in all positions, and effectively improve the quality and efficiency of pipeline welding in all positions.

[0014] 2. Based on the characteristics of actual molten pool images, this invention designs a molten pool image processing algorithm. During model training, the influence of different welding positions and parameters on the molten pool size during pipeline welding is mainly considered. An improved YOLO v8-based target detection model is introduced for molten pool detection. By introducing a spatial attention module on top of the existing YOLO v8 model, the network can focus more on regions with molten pool characteristics, suppressing non-critical backgrounds and significantly improving the accuracy of welding defect detection. Simultaneously, it maintains the advantage of zero deployment cost, achieving rapid and accurate molten pool detection and real-time output of molten pool size changes. The application of this model not only improves detection accuracy and speed but also possesses strong generalization ability, maintaining stable performance under different welding scenarios and conditions.

[0015] 3. The key point of this invention is to establish a correlation model by considering the influence of gravity on the molten pool at different welding positions on the pipeline, analyze the correlation between the molten pool size and the welding position, and analyze the variation law of the molten pool corresponding to different welding parameters within the same welding position range. Based on this, taking the molten pool size at a set position as a benchmark, and according to the variation law obtained from the analysis, suggestions for modifying the welding parameters at other positions are given. Then, the modified welding parameters are used to perform automatic welding of the pipeline at all positions, which can make the molten pool size basically consistent throughout the entire welding position range, effectively maintain the stability of the weld performance, and improve the welding quality.

[0016] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0018] Figure 1 This is an overall flowchart of the deep learning-based automatic welding pool monitoring method for pipelines in all positions, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the molten pool monitoring system in an embodiment of the present invention; Figure 3 This is a schematic diagram of the pipe welding position in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the training process of the improved target detection model in this embodiment of the invention. Figure 5 This is a schematic diagram of the camera lens calibration plate selected in an embodiment of the present invention; Figure 6 This refers to the recognition result during the training process of the target detection model in this embodiment of the invention; Figure 7 This is a schematic diagram of the loss during the training process of the target detection model in an embodiment of the present invention; where (a) is the result of training the training set, and (b) is the result of training the test set; Figure 8 This is a schematic diagram showing the changes in the width and length of the molten pool when the wire feeding speed is 8 cm / min in an embodiment of the present invention; where (a) represents the change in the width of the molten pool and (b) represents the change in the length of the molten pool. Figure 9 This is a schematic diagram showing the changes in the width and length of the molten pool when the wire feeding speed is 8.6 cm / min in an embodiment of the present invention; where (a) represents the change in the width of the molten pool and (b) represents the change in the length of the molten pool. Figure 10Suggestions for changes in molten pool size and process adjustments under a wire feeding speed of 8.6 cm / min in this embodiment of the invention; Figure 11 The accuracy of the target detection model before and after improvement in this embodiment of the invention is shown in (a) and (b) is the accuracy after model improvement. Figure 12 The following are performance comparison results of the target detection model before and after improvement in the embodiments of the present invention; where (a) is the comparison of bounding box regression loss, (b) is the comparison of classification loss, (c) is the comparison of accuracy, (d) is the comparison of average accuracy with IoU 0.5, and (e) is the comparison of average accuracy with IoU 0.95. Detailed Implementation

[0019] It should be noted that the following detailed descriptions are exemplary and are intended only to describe specific embodiments and to provide further explanation of the invention, and are not intended to limit the scope of exemplary embodiments of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0020] Existing automated all-position welding solutions for pipelines suffer from poor real-time performance in traditional molten pool identification, making it difficult to meet the monitoring requirements of high frame rate molten pools. Furthermore, processing high frame rate video involves high computational complexity, reliance on GPU acceleration, and high hardware costs. Additionally, existing automated all-position welding methods for pipelines often result in poor welding quality. Therefore, this invention considers YOLO v8 as an advanced target detection algorithm with high real-time performance and high detection accuracy. Based on this, and combined with the specific physical characteristics of the molten pool during the welding process, an improved YOLO-based method is proposed. The V8 target detection method, combined with nonlocal mean denoising to improve image quality, achieves rapid and accurate detection of the molten pool size. Simultaneously, regarding the issue of how to improve welding quality, considering that the molten pool position in all-position pipeline welding is a gradual process, and that the stress on the molten pool is constantly changing due to this change, the welding position is one of the important influencing factors of molten pool change during all-position pipeline welding. Based on this, this invention analyzes the dynamic correlation between welding position, welding parameters, and molten pool size, providing an accurate reference for adjusting welding parameters. Based on these adjusted welding parameters, continuous and uninterrupted welding of the next pipeline is performed, achieving high-quality all-position pipeline welding.

[0021] Example 1 This embodiment provides a deep learning-based method for automatic all-position welding pool monitoring in pipelines, such as... Figure 1As shown, the specific steps include: Step S1: Select the pipe, perform automatic continuous welding of the pipe in all positions using the set welding parameters, obtain images of the molten pool at different welding positions during the welding process, use the target detection model based on improved YOLO v8 to identify the molten pool area in the image, and extract the molten pool size.

[0022] Specifically, before pipe welding, a fully automated all-position welding molten pool monitoring system based on a CMOS camera is installed on the pipe. This system includes welding equipment, pipe materials, and image acquisition equipment. The image acquisition equipment includes a high-resolution, high-frame-rate CMOS camera with a suitable optical lens to capture real-time images of the molten pool during the welding process. The camera's installation position and angle are adjusted in advance to ensure image clarity and integrity. This well-designed monitoring system provides a foundation for subsequent image processing and target detection.

[0023] like Figure 2 As shown, X70 pipeline steel was selected as the welding material. A CMOS camera-based automated all-position welding molten pool monitoring system was built on the pipeline. This system includes a welding carriage 3 that moves circumferentially along the pipeline, a welding torch 3 and a CMOS camera 5 connected to the welding carriage, a data processing system 2 equipped with a target detection algorithm (this data processing system is used to acquire and process the welding molten pool images captured by the CMOS camera), and a welding power supply 1 to power each device. Based on this monitoring system, molten pool images during the welding process can be acquired via the CMOS camera.

[0024] Step S1.1: Select the pipe and perform automatic continuous welding of the pipe in all positions using the set welding parameters.

[0025] Specifically, in this embodiment, X70 pipeline steel is selected, and gas metal arc welding (GMAW) is used to continuously weld on the pipeline within a set welding angle range of 0-180° (i.e., all positions). Figure 3 As shown, firstly, the initial welding position for automatic all-position welding of the pipeline is selected as the reference position. This position can be selected from different positions such as flat welding, vertical welding, and overhead welding. Welding parameters are then determined, including welding speed, welding current, welding voltage, wire feed speed, oscillation width, and slot size (i.e., the distance between the two pipe sections to be welded). Continuous welding is then initiated at the reference position, and the aforementioned monitoring system is used to acquire images of the molten pool at each welding position during the welding process. It should be noted that after starting continuous welding, the welding robot will continuously and uninterruptedly weld the entire pipeline according to the set welding parameters, without any adjustments during the welding process, until the welding is completed.

[0026] Step S1.2: Obtain images of the molten pool at different welding positions during the welding process, identify the molten pool region in the image using an improved YOLO v8 target detection model, and extract the molten pool size.

[0027] To avoid the problems of low detection accuracy and low efficiency in traditional image detection and machine learning methods, this embodiment proposes a deep learning-based method for dynamic monitoring of weld pools in all positions of pipelines. An improved lightweight target detection model is designed, and nonlocal mean denoising is combined to improve image quality, thereby achieving rapid detection of weld pool targets.

[0028] Specifically, in this embodiment, an improved target detection model was built, namely, an improved YOLO v8 target detection model. This model adopts the YOLO v8 target detection model. Based on this, considering the specific physical characteristics of the molten pool during the welding process, namely that the molten pool is mostly Gaussian distributed (bright in the center and slightly darker at the edges), mostly elongated elliptical in shape, with certain regular gradient directions at the edges, and usually located in a specific region of the image (near the center of the arc), the YOLOv8 model was structurally optimized according to the above characteristics. A spatial attention mechanism (Spatial PriorAttention) that integrates the physical features of the molten pool was introduced. This spatial attention module was embedded at the end of the backbone network of the YOLO v8 target detection model (i.e., after the SPPF layer). This module was used to perform spatial attention weighting on the high-level semantic features output by the Feature Pyramid Pooling Layer (SPPF) in the backbone network, thereby enhancing the representation ability of the molten pool region through feature reweighting.

[0029] Furthermore, the process of identifying and detecting the melt pool using the improved target detection model described above is as follows: (1) The melt pool image The input is fed into the model, which extracts high-level semantic feature maps of the image based on the YOLO v8 algorithm; where R represents the set of real numbers, B represents the batch size, C represents the number of channels, H represents the height, and W represents the width.

[0030] (2) Using the high-level semantic feature map as the original feature map, input it into the spatial attention module, calculate the channel mean and maximum value of the feature respectively, and extract the channel mean feature and peak feature.

[0031] In this process, the mean value is calculated along the channel dimension of the original features, and the channel mean feature avg_out is extracted. This feature reflects the overall energy distribution of the melt pool region and can be expressed as: ; Simultaneously, the maximum value from the original features is taken to extract the peak feature max_out, which reflects the high-temperature region of the molten pool core and can be represented as: ; In the above formula, Representing dimension, This is a function in PyTorch used to calculate the mean of a tensor. This is a function in PyTorch used to return the maximum value in an input tensor.

[0032] Through the above operations, the channel dimension is compressed to obtain a feature with dimension [B,1,H,W].

[0033] (3) The mean feature and peak feature are fused with physical features to generate a feature map guided by the physical features of the melt pool. Then, attention weights are obtained by Sigmoid normalization. The original feature map is weighted based on the attention weights to obtain an enhanced feature map.

[0034] Specifically, the mean feature and the peak feature are added element-wise and then divided by 2 to obtain the arithmetic mean of the two features, thus yielding the fused guide map, which can be represented as: ; (4) The original feature map and the enhanced feature map are merged by residual connection to obtain the final fused feature map.

[0035] (5) Based on the fused feature map, the final detection result and detection area are output through the fully connected layer.

[0036] By introducing the spatial attention module, the network can focus more on regions with molten pool characteristics, suppressing non-critical backgrounds (such as strong arc sputtering points and workpiece textures), significantly improving the accuracy of welding defect detection while maintaining the advantage of zero deployment cost. Furthermore, this improved module has only two trainable parameters, significantly enhancing the visual feature representation of the molten pool region while maintaining model computational efficiency, with a computational complexity increase of only 0.01 GFLOPs, achieving a leap in accuracy while maintaining real-time performance (>30 FPS).

[0037] As one implementation method, such as Figure 4 As shown, the training process of the improved object detection model described above is as follows: Step 1: Select X70 pipeline steel, determine welding parameters, and employ gas metal arc welding (GMAW). Conduct welding experiments with welding position and parameters as variables. Simultaneously, construct a suitable monitoring system using a high-resolution, high-frame-rate CMOS camera and appropriate optical lenses to capture real-time images of the molten pool during the welding process. Preferably, such as... Figure 5 As shown, camera calibration is performed on the camera used. That is, the camera is calibrated using a standard-sized calibration plate, and the actual size of the camera's field of view is output to ensure that the acquired molten pool image is consistent.

[0038] Step 2: Preprocess the acquired molten pool image and perform molten pool feature annotation on the preprocessed image. Specifically, image preprocessing includes: comparing the effects of different filtering algorithms, selecting the nonlocal average denoising algorithm from the OpenCV library to remove noise from the image, thereby improving image quality and ensuring the efficiency and accuracy of image processing; then using the Labelme tool to perform feature annotation on the processed image, extracting the boundary and morphological features of the molten pool, and generating a JSON file containing data such as image type, marker coordinates, and molten pool width and height.

[0039] Step 3: Divide the molten pool images collected in each welding position interval into two parts: a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%.

[0040] Step 4: Train the improved YOLO v8-based object detection model using the training set. This involves inputting labeled image data into the model and optimizing its performance by adjusting the model's parameters and hyperparameters. During training, metrics such as loss function, precision, and recall are used to evaluate the model's performance. Through continuous training and optimization, the model is ensured to accurately identify and locate melt pools. The recognition results during model training are shown below. Figure 6 As shown in the figure, the red area in the upper left corner indicates the location of the molten pool, and the blue area indicates the range of its location. Figure 6 The image shows the different locations of the molten pool under different welding conditions.

[0041] Step 5: Test the trained optimal object detection model using the test set to ensure model performance. The results of model training using the training and test sets are as follows: Figure 7 As shown.

[0042] Preferably, the trained model is subjected to interpretability analysis to evaluate its performance under different welding positions and parameters, and the reliability and accuracy of the model are verified by indicators such as precision curves, recall curves and confusion matrices.

[0043] As a real-time approach, YOLO v5 can be used instead of YOLO v8, by sacrificing some accuracy to reduce hardware requirements.

[0044] The above method is used to train and obtain the target detection model with the best performance. Based on this, the molten pool video (i.e., sequential molten pool video frame images) captured during the welding process is input into the target detection model. The model detects the molten pool image, quickly and accurately identifies the molten pool area, and extracts the molten pool size (including molten pool length and molten pool width).

[0045] Step S2: Using the molten pool size at the initial welding position as the reference molten pool size, determine the changes in the molten pool size at other different welding positions relative to the reference molten pool size. Combine the relationship between the molten pool size and welding parameters, and adjust the set welding parameters corresponding to different welding positions to keep the molten pool size consistent during the welding process.

[0046] The key point of this embodiment is to establish a model considering the influence of different welding positions (including flat welding, vertical welding, overhead welding, etc.) on the gravity of the molten pool, analyze the correlation between the size of the molten pool and the welding position in real time, and analyze and explore the correlation between welding parameters and molten pool changes through multiple welding experiments with different welding parameters within the same welding position range, so as to clarify the molten pool change law.

[0047] Specifically, welding experiments at different welding positions and with different welding parameters were conducted to monitor the dynamic behavior of the molten pool at different welding positions and with different welding parameters in the automated welding of pipelines in all positions. The molten pool image was processed using the improved YOLO v8 target detection model to extract the molten pool size, and the correlation between the actual molten pool size and the welding position and welding parameters was analyzed. Specifically, continuous welding was performed on the pipeline in all positions using the same welding parameters to observe the changes in the molten pool size and analyze the correlation between the welding position and the molten pool size. Then, continuous welding was performed again on the pipeline in all positions using different welding parameters, and the changes in the molten pool size were observed. This process was repeated multiple times to obtain the dynamic changes in the molten pool at different welding positions and with different welding parameters, thus clarifying the dynamic correlation between the welding position, welding parameters, and molten pool size, providing a reference for adjusting the welding parameters. Finally, the adjusted parameters were used to perform automated welding on the pipeline in all positions, ensuring that the molten pool size remained basically consistent throughout the welding process, thereby obtaining circumferential welds with similar performance and improving welding quality.

[0048] In step S2, the pipeline is welded in all positions using the set welding parameters. The molten pool size during the welding process is obtained through step S1 above. The initial welding position is the 0-15° flat welding position, and the molten pool size corresponding to this initial welding position is the reference molten pool size. At this time, the changes in the molten pool size relative to the reference molten pool size at other welding positions can be determined. Based on these changes, and in conjunction with the correlation between the molten pool size and the welding parameters, suggestions for modifying the welding parameters at other positions are given. The set welding parameters corresponding to different welding positions are adjusted so that the molten pool size is basically consistent within the 0-180° welding position range, thereby maintaining the stability of the weld performance.

[0049] The relationship between the molten pool size and the welding position and parameters was determined through multiple welding experiments. Specifically, X70 pipeline steel was selected for welding experiments at different positions and with different parameters. Images of the molten pool were obtained, and the molten pool size was extracted from the images using a trained target detection model. The variation law of the molten pool size at different welding parameters and positions was analyzed. Preferably, the relationship between the molten pool size and welding parameters and position can be revealed through experimental data and theoretical analysis, and a dynamic correlation model of welding position, welding parameters, and molten pool size can be constructed.

[0050] First, regarding the relationship between the molten pool size (including the length and width of the molten pool) and the welding position, based on the data obtained from the above experiments, a Gaussian function can be used for function fitting, expressed as: ; in, , , , Represents the fitted parameters, Indicates the welding position. This indicates the length or width of the molten pool.

[0051] Based on this, according to Figure 8 By fitting the changes in the weld pool width and length at a wire feed speed of 8 cm / min, a fitting function for the weld pool width and length under these welding parameters can be obtained; similarly, as shown... Figure 9 The changes in the width and length of the molten pool when the wire feed speed is 8.6 cm / min can also be used to obtain the fitting function of the width and length of the molten pool under this welding parameter.

[0052] Secondly, based on the variation pattern of the molten pool size at the same welding position under different welding parameters, the correlation between the molten pool size and welding parameters is analyzed. In this embodiment, the different gravity and heat conduction conditions caused by positional changes during pipeline welding are also considered. For example, the molten pool experiences uniform gravity during flat welding, the molten pool tends to flow during downhill welding, the heat accumulation is different at vertical welding positions, and the molten pool tends to sag during overhead welding. The relationship between the molten pool size variation at different welding positions and welding parameter adjustments is analyzed, along with corresponding welding parameter modification suggestions, including: (1) The welding parameters applied to the flat weld zone (PA) with an angle range of 0°-30° are used as standard parameters, and the welding parameters of the flat weld zone are maintained. (2) For the downslope welding transition zone (PG) with an angle range of 30°-90°, after comparison and judgment, when the molten pool is too long, the welding speed can be increased, the current can be reduced, and the voltage can be reduced slightly (e.g., reduced by 0.5-1V) in order of priority; when the molten pool is too short, the welding speed can be reduced slightly; when the molten pool is too wide, the welding speed can be increased and the voltage can be reduced in order of priority; when the molten pool is too narrow, the voltage can be increased slightly and narrow-amplitude high-frequency oscillation can be performed in order of priority.

[0053] (3) For vertical and overhead welding areas (PE) with an angle range of 90°-180°, after comparison and judgment, when the molten pool is too long, the current can be reduced, the welding speed can be increased, and the voltage can be reduced slightly in order of priority; when the molten pool is too short, the current can be increased, the welding speed can be reduced slightly, and the voltage can be increased slightly in order of priority; when the molten pool is too wide, the voltage can be reduced and the welding speed can be increased in order of priority; when the molten pool is too narrow, the voltage can be increased slightly in order of priority, narrow-amplitude high-frequency oscillation can be performed, and the wire alignment / extension can be checked.

[0054] The above adjustments mainly consider the welding speed, welding current, welding voltage, wire feed speed, and oscillation width. The slot size can ensure the stability of the weld pool width, and is usually a fixed value without adjustment.

[0055] As another implementation method, the relationship between the size of the weld pool and the welding parameters and position can be revealed through experimental data and theoretical analysis, and a dynamic correlation model between the welding position, welding parameters and the size of the weld pool can be obtained by further fitting.

[0056] As one implementation method, an infrared thermal imager can be added to the molten pool monitoring system. This infrared thermal imager assists in monitoring the temperature field of the molten pool, and an alarm is triggered when an abnormal temperature is detected, thereby enhancing the defect early warning capability and ensuring the normal progress of welding.

[0057] Step S3: Based on the welding parameters after the pipeline is adjusted in all positions, the next pipeline is automatically and continuously welded in all positions. This ensures that the size of the molten pool is basically the same throughout the welding process, guarantees similar welding performance, and improves welding quality.

[0058] The superiority of the method proposed in this embodiment is further verified through the following examples.

[0059] Select X70 pipeline steel and perform continuous welding at angles of 0-180° on the pipeline. For example... Figure 11 and Figure 12As shown, compared to the unmodified YOLO v8 model, the improved model achieves better detection results: Precision breakthrough: mAP@0.5 increased to 0.995 (an increase of 1.7%); the stringent localization standard mAP@0.95 jumped 31.8% to 0.874, achieving 100% precision at 98% confidence; recall remained stable at 100%, achieving zero false negatives, while reducing the high-confidence false positive rate by 98%; training efficiency improved, with a 9-fold increase in convergence speed, reaching optimal precision in only 3 epochs (the original model required 27 epochs), and a 56.8% reduction in training loss.

[0060] Based on the improved model, a process window is created during continuous welding. Using the molten pool size at the 0-15° flat welding position as a benchmark, the changes in the molten pool size at other positions during welding are detected, and suggestions for adjusting welding parameters are provided, such as... Figure 10 As shown, the changes in molten pool size and process adjustment suggestions are presented at a wire feed speed of 8.6 cm / min, so as to make the molten pool size approximately the same at all positions from 0 to 180° and maintain stable weld performance.

[0061] Example 2 This embodiment provides a deep learning-based automated welding pool monitoring system for pipelines in all positions, including: The molten pool size extraction module is used to select the pipeline, perform automatic continuous welding of the pipeline in all positions using set welding parameters, acquire molten pool images at different welding positions during the welding process, and use an object detection model based on improved YOLO v8 to identify the molten pool region in the image and extract the molten pool size. The welding parameter adjustment module is used to determine the changes in the size of the molten pool at other welding positions relative to the reference molten pool size, based on the molten pool size at the initial welding position. It then combines the relationship between the molten pool size and the welding parameters to adjust the set welding parameters for different welding positions, so that the molten pool size remains consistent during the welding process. The pipeline all-position welding module is used to perform automatic continuous all-position welding on the selected pipeline based on the welding parameters adjusted for pipeline all-position.

[0062] Example 3 This embodiment provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided in this embodiment.

[0063] Example 4 This embodiment also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, will cause the processor to execute the method described above in this embodiment.

[0064] Example 5 This embodiment provides a computer program product including executable instructions, which are computer instructions; the executable instructions are stored in a computer-readable storage medium. When the processor of an electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the electronic device performs the method described in this embodiment.

[0065] The steps and methods involved in Embodiments 2 to 5 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0066] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0067] The above description is only a preferred embodiment of the present invention. Although the specific implementation of the present invention has been described in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the present invention.

Claims

1. A method for automatic monitoring of weld pool in all positions of pipelines based on deep learning, characterized in that, include: Select a pipe and perform automatic continuous welding of the pipe in all positions using set welding parameters. Obtain images of the molten pool at different welding positions during the welding process. Use a target detection model based on improved YOLO v8 to identify the molten pool region in the image and extract the molten pool size. Melt pool identification and detection are performed using an improved YOLO v8-based target detection model, including: The molten pool image is input into the target detection model, and high-level semantic feature maps of the image are extracted based on the YOLO v8 algorithm; Using the high-level semantic feature map as the original feature map, it is input into the spatial attention module to calculate the channel mean and maximum value of the feature, and extract the channel mean feature and peak feature; The mean and peak features are fused with physical features to generate a feature map guided by the physical features of the melt pool. Then, attention weights are obtained by Sigmoid normalization. The original feature map is weighted based on the attention weights to obtain an enhanced feature map. The original feature map and the enhanced feature map are merged by residual connection to obtain the final fused feature map; Based on the fused feature map, the final detection result and detection region are output through a fully connected layer; Using the molten pool size at the initial welding position as the reference molten pool size, the changes in the molten pool size at other different welding positions relative to the reference molten pool size are determined. By combining the correlation between the molten pool size and welding parameters, the set welding parameters corresponding to different welding positions are adjusted to keep the molten pool size consistent during the welding process. Based on the welding parameters after the pipeline is adjusted in all positions, the next pipeline is automatically and continuously welded in all positions.

2. The method for automatic monitoring of weld pool in all positions of a pipeline based on deep learning as described in claim 1, characterized in that, The improved YOLO v8-based target detection model is based on the YOLO v8 target detection model and introduces a spatial attention mechanism that integrates the physical features of the melt pool. The spatial attention module is embedded at the end of the backbone network of the YOLO v8 target detection model. The spatial attention module is used to perform spatial attention weighting on the high-level semantic features output by the feature pyramid pooling layer in the backbone network to enhance the ability of features to represent the melt pool region.

3. The method for automatic all-position welding pool monitoring of pipelines based on deep learning as described in claim 1, characterized in that, The training process of the object detection model based on the improved YOLO v8 includes: Select the pipe type, determine different welding parameters, and conduct welding experiments with welding position and welding parameters as variables, and collect images of the molten pool during the welding experiment; The acquired molten pool image is preprocessed, and the molten pool feature is annotated on the preprocessed image. Among them, the nonlocal average denoising algorithm is used to remove noise in the image, and the preprocessed image is annotated to extract the boundary and morphological features of the molten pool, and to annotate the image type, the coordinates of the marker points, and the width and height of the molten pool. The molten pool images collected in each welding location range are divided into training and testing sets. The training set is used to train the target detection model based on the improved YOLO v8, and the testing set is used to test the best target detection model obtained by training, thus completing the model training.

4. The method for automatic monitoring of weld pool in all positions of pipelines based on deep learning as described in claim 1, characterized in that, The welding positions include flat welding, vertical welding, and overhead welding positions, and the welding parameters include welding speed, welding current, welding voltage, wire feed speed, and oscillation width.

5. The method for automatic monitoring of weld pool in all positions of pipelines based on deep learning as described in claim 1, characterized in that, Determining the relationship between weld pool size and welding parameters includes: Select a pipeline and conduct multiple automatic all-position welding experiments with different welding parameters. Monitor the dynamic behavior of the molten pool during the welding process, acquire molten pool images, and then use an improved YOLO v8-based target detection model to process the molten pool images and extract the molten pool size. Based on the variation pattern of the molten pool size at the same welding position under different welding parameters, and combined with the different gravity and heat conduction conditions caused by position changes during pipeline welding, the relationship between the variation of the molten pool size at different welding positions and the adjustment of welding parameters is analyzed.

6. A deep learning-based all-position automatic welding pool monitoring system for pipelines, characterized in that, include: The molten pool size extraction module is used to select the pipeline, perform automatic continuous welding of the pipeline in all positions using set welding parameters, acquire molten pool images at different welding positions during the welding process, and use an object detection model based on improved YOLO v8 to identify the molten pool region in the image and extract the molten pool size. Melt pool identification and detection are performed using an improved YOLO v8-based target detection model, including: The molten pool image is input into the target detection model, and high-level semantic feature maps of the image are extracted based on the YOLO v8 algorithm; Using the high-level semantic feature map as the original feature map, it is input into the spatial attention module to calculate the channel mean and maximum value of the feature, and extract the channel mean feature and peak feature; The mean and peak features are fused with physical features to generate a feature map guided by the physical features of the melt pool. Then, attention weights are obtained by Sigmoid normalization. The original feature map is weighted based on the attention weights to obtain an enhanced feature map. The original feature map and the enhanced feature map are merged by residual connection to obtain the final fused feature map; Based on the fused feature map, the final detection result and detection region are output through a fully connected layer; The welding parameter adjustment module is used to determine the changes in the size of the molten pool at other welding positions relative to the reference molten pool size, based on the molten pool size at the initial welding position. It then combines the relationship between the molten pool size and the welding parameters to adjust the set welding parameters for different welding positions, so that the molten pool size remains consistent during the welding process. The pipeline all-position welding module is used to perform automatic continuous all-position welding on the selected pipeline based on the welding parameters adjusted for pipeline all-position.

7. An electronic device, characterized in that, include: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the deep learning-based automatic welding pool monitoring method for pipelines in all positions as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The device stores executable instructions that, when executed by a processor, implement the deep learning-based automatic welding pool monitoring method for pipelines in all positions as described in any one of claims 1-5.

9. A computer program product, characterized in that, The computer program product includes executable instructions stored in a computer-readable storage medium; When the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, it implements the deep learning-based automatic welding pool monitoring method for pipelines in all positions as described in any one of claims 1-5.

Images