Pipeline welding pool feature recognition method and device and storage medium
By using an image acquisition device and a target detection model during pipeline welding, combined with preprocessing and spatial attention modules, the problem of insufficient accuracy in molten pool feature recognition was solved, enabling real-time monitoring of molten pool features and accurate adjustment of welding parameters, thus ensuring welding quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-03-13
AI Technical Summary
In existing technologies, the shape and position of the molten pool during pipeline welding are affected by the welding position, arc light interference, and dynamic changes, resulting in insufficient accuracy in molten pool feature recognition. This makes it impossible to effectively replace manual recognition and affects welding quality.
The image acquisition device (including interference filter, polarizer and analyzer) is used to acquire the image of the molten pool. Combined with preprocessing and target detection model, the property characteristics of the molten pool are identified through median filtering and spatial attention module, and the welding parameters are adjusted in real time.
It effectively suppresses arc light interference, improves the accuracy of molten pool feature recognition, enables real-time monitoring and parameter adjustment of molten pool features, and ensures welding quality.
Smart Images
Figure CN121661431A_ABST
Abstract
Description
Technical Field
[0001] This application relates to a method for identifying the characteristics of a weld pool in pipeline welding, a device for identifying the characteristics of a weld pool in pipeline welding, an electronic device, and a computer-readable storage medium, belonging to the field of pipeline welding. Background Technology
[0002] In current related technologies, the state of the molten pool during pipeline welding contains a wealth of information about welding quality, and accurate identification of the molten pool's features across all positions is crucial for ensuring welding quality. However, current molten pool feature recognition technologies suffer from limitations because the shape and position of the molten pool change with the welding location during pipeline welding. Furthermore, the image of the molten pool is significantly affected by factors such as arc light interference and dynamic changes in the molten pool, resulting in limited accuracy in identifying the current features of the molten pool. This makes it impossible to effectively replace manual molten pool feature recognition, thus impacting welding quality. Summary of the Invention
[0003] This application provides a method for identifying the characteristics of a weld pool in a pipeline, a device for identifying the characteristics of a weld pool in a pipeline, an electronic device, and a computer-readable storage medium.
[0004] The pipeline welding molten pool feature identification method in this application includes the following steps: Based on the image acquisition device, images of the molten pool are acquired during the pipeline welding process; Preprocessing is performed on the molten pool image to eliminate noise points in the molten pool image; Based on a pre-trained target detection model, the attribute features of the molten pool image are obtained according to the pre-processed molten pool image. The current state of the molten pool is identified based on preset baseline features and the attribute features.
[0005] In some embodiments, the image acquisition device acquires a molten pool image, including: The image acquisition device is controlled to acquire multiple images of the molten pool at different locations during the pipe welding process; The image acquisition device includes a molten pool camera, and an interference filter, a polarizer, and an analyzer are sequentially arranged at the lens of the molten pool camera.
[0006] In some embodiments, controlling the image acquisition device to acquire multiple images of the molten pool at different locations during pipe welding includes: Control the image acquisition device to acquire one or more background images; Based on the background image, the average gray level and peak gray level of the corresponding background region are determined to determine the reference contrast of the molten pool image; Based on the reference contrast, the polarization axis angles of the polarizer and analyzer are adjusted so that the grayscale contrast of the molten pool image satisfies a preset relationship with the reference contrast; Under the given grayscale contrast, multiple images of the molten pool at different locations during the pipe welding process are acquired.
[0007] In some embodiments, the preprocessing of the molten pool image includes: Based on the median filtering algorithm, noise reduction processing is performed on the molten pool image to eliminate noise points in the molten pool image.
[0008] In some embodiments, performing preprocessing on the molten pool image further includes: The molten pool image is converted to grayscale to adapt it to the target detection model.
[0009] In some implementations, the target detection model incorporates a spatial attention module that integrates physical features of the molten pool; The pre-trained target detection model obtains attribute features of the molten pool image based on the preprocessed molten pool image, including: Based on the processed molten pool image, obtain the semantic features corresponding to the molten pool image; Based on the spatial attention module, spatial attention weighting processing is performed on the semantic features to determine the attribute features of the melt pool image, thereby improving the accuracy of the attribute features.
[0010] In some embodiments, identifying the current state of the molten pool based on preset reference features and the attribute features to control the parameters of the pipe welding process includes: Based on the baseline features and the attribute features, the feature differences are determined; If the feature difference exceeds a preset threshold range, the current state of the molten pool is identified as the first state in order to control and adjust the parameters of the pipe welding process; If none of the feature differences exceed a preset threshold range, the current state of the molten pool is identified as the second state, wherein the second state indicates that the molten pool is currently in a normal state.
[0011] The pipeline welding molten pool feature recognition device in this application includes: The image acquisition module is used to acquire images of the molten pool during the pipeline welding process based on the image acquisition device. An image processing module is used to perform preprocessing on the molten pool image to eliminate noise points in the molten pool image; The feature acquisition module is used to acquire the attribute features of the molten pool image based on the pre-trained target detection model and the pre-processed molten pool image. The feature recognition module is used to identify the current state of the molten pool based on preset reference features and the attribute features, so as to control the parameters of the pipeline welding process.
[0012] The electronic device in this application includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the pipe welding molten pool feature recognition method in the above-described embodiments is implemented.
[0013] The computer-readable storage medium in this application embodiment stores a computer program that, when executed by one or more processors, implements the pipe welding molten pool feature recognition method in the above embodiment.
[0014] The beneficial effects of this application are: This application can suppress arc light interference in the molten pool image through a specially set filter device and target detection model on the image acquisition device, improve the accuracy of feature extraction, and further meet the requirements for real-time monitoring of molten pool features during pipeline welding by performing real-time processing on the acquired molten pool image, which helps to adjust the welding process parameters in a timely manner and thus ensure welding quality. Attached Figure Description
[0015] Figure 1 This is one of the flowcharts illustrating the pipeline welding molten pool feature identification method in the embodiments of this application; Figure 2 This is the second flowchart illustrating the pipeline welding molten pool feature identification method in the embodiments of this application; Figure 3 This is the third flowchart illustrating the pipeline welding molten pool feature identification method in the embodiments of this application; Figure 4 This is the fourth flowchart illustrating the pipeline welding molten pool feature identification method in the embodiments of this application; Figure 5 This is the fifth flowchart illustrating the pipeline welding molten pool feature identification method in the embodiments of this application; Figure 6 This is the sixth flowchart illustrating the pipeline welding molten pool feature identification method in the embodiments of this application. Detailed Implementation
[0016] Please see Figure 1 The pipeline welding molten pool feature identification method in this application includes the following steps: Step 01: Based on the image acquisition device, acquire the image of the molten pool during the pipeline welding process.
[0017] Please see Figure 2 Furthermore, step 01 includes: Step 011: Control the image acquisition device to acquire multiple molten pool images at different locations during the pipe welding process; The image acquisition device includes a molten pool camera, and an interference filter, a polarizer, and an analyzer are sequentially arranged at the lens of the molten pool camera.
[0018] Specifically, the pipeline welding molten pool feature recognition method in this application mainly targets the feature recognition of the molten pool during the pipeline welding process. Based on the recognized features, it compares them with reference features to determine whether the current state of the molten pool is normal, and then determines whether parameter adjustments are needed for the pipeline welding process to ensure the quality of the pipeline process. To accurately identify molten pool features, it is necessary to first acquire an image of the molten pool, and then further identify its size, shape, position, and other attributes based on the image to determine the current state of the molten pool.
[0019] Therefore, by way of example, based on an image acquisition device such as a molten pool camera for acquiring images of the current shape of the molten pool, images of the molten pool at multiple locations in different positions can be acquired during the pipeline welding process. Each molten pool can acquire one or more images, and finally form a set of molten pool images corresponding to the pipeline welding process.
[0020] Image acquisition devices such as molten pool cameras are typically mounted on welding robots. During the welding process, the molten pool camera can acquire real-time images of the molten pool. Furthermore, the lens of the molten pool camera is equipped with an interference filter, a polarizer, and an analyzer. Light entering the lens passes through these components sequentially. The interference filter's main function is to filter out specific wavelengths of light during pipe welding, thus achieving initial light interference reduction in the wavelength dimension. The polarizer's core function is polarization state conversion; it converts the light filtered by the interference filter into linearly polarized light, straightening the polarization direction of the light and preventing the disordered superposition of light with different polarization states, further regulating the light entering the lens from the perspective of polarization direction. The core function of the analyzer is polarization state screening. By adjusting its own polarization axis angle, it can selectively allow linearly polarized light converted by the polarizer to pass through, while blocking reflected and scattered light that does not conform to the polarization direction. This allows for precise reduction of the remaining interference light from the dimension of polarization direction screening, thereby minimizing the influence of welding arc light on the light used for imaging.
[0021] In some implementations, please refer to Figure 3 Step 001 further includes: 0111: Control the image acquisition device to acquire one or more background images; 0112: Based on the background image, determine the average gray level and peak gray level of the corresponding background area to determine the baseline contrast of the molten pool image; 0113: Based on the reference contrast, adjust the polarization axis angles of the polarizer and analyzer so that the grayscale contrast of the molten pool image meets the preset relationship with the reference contrast; 0114: Under grayscale contrast, acquire multiple molten pool images at different locations during the pipe welding process.
[0022] Specifically, based on the above embodiments, the specific method for acquiring molten pool images using the image acquisition device in the above embodiments can be implemented with reference to the following example.
[0023] In some examples, before acquiring the molten pool image during the pipe welding process, the background lighting conditions of the pipe welding process need to be pre-set, considering the overall contrast and grayscale of the image. Therefore, one or more background images of the welding environment are first acquired using an image acquisition device such as a molten pool camera. Based on the aforementioned background images, the average grayscale and peak grayscale of the background area corresponding to the welding process can be calculated using calculation methods currently available in related technologies. Furthermore, the reference contrast of the molten pool image to be acquired subsequently is calculated based on the average grayscale and peak grayscale.
[0024] The aforementioned reference contrast is the data benchmark for the grayscale contrast of the molten pool image. Generally, the grayscale contrast threshold of the molten pool image is set to 1.2 to 1.5 times the reference contrast (corresponding to the grayscale contrast of the molten pool image satisfying the preset relationship with the reference contrast). The specific multiple can be adjusted according to the actual situation. The contrast of the molten pool image is generally affected by the background light and the arc light during the welding process. Therefore, the polarization axis angle of the polarizer and analyzer can be used to adjust the normalization parameters of the light entering the lens of the molten pool camera, thereby adjusting the image contrast of the molten pool image affected by the background light and the arc light.
[0025] Having determined the grayscale contrast of the molten pool image by adjusting the polarization axis angles of the polarizer and analyzer, it is further possible to use a molten pool camera to acquire molten pool images corresponding to multiple molten pools at different locations during the pipeline welding process, thereby providing an image data basis for subsequent molten pool feature recognition.
[0026] In some implementations, please refer to [the relevant documentation]. Figure 1 The pipeline welding molten pool feature identification method in this application embodiment further includes: Step 02: Perform preprocessing on the molten pool image to eliminate noise points in the molten pool image.
[0027] Specifically, based on the above implementation method, for example, when multiple molten pool images are acquired, considering the impact of image quality on feature recognition, it is necessary to preprocess the molten pool images before performing feature recognition. By eliminating noise points present in the molten pool images, the image quality of the molten pool images can be improved, thereby increasing the accuracy of feature recognition of the molten pool images.
[0028] Further, please refer to Figure 4 Step 02 specifically includes: Step 021: Based on the median filtering algorithm, perform noise reduction processing on the molten pool image to eliminate noise points in the molten pool image.
[0029] In particular, in some embodiments, step 02 further includes: The molten pool image is converted to grayscale to adapt it to the median filtering algorithm.
[0030] Specifically, for the preprocessing of the molten pool image, median filtering can be used as an example, as described in related technologies. It is important to note that since the median filtering algorithm processes a single-color-channel grayscale image, the image acquired by the molten pool camera may be a color image similar to RGB three-channel or other channel types. Therefore, before executing the median filtering algorithm, the color type of the molten pool image needs to be confirmed. If it is a single-channel grayscale image, the median filtering algorithm can be executed directly. However, if it is an RGB three-channel or other channel type color image, a grayscale conversion process needs to be performed first to convert the molten pool image into a grayscale image suitable for the median filtering algorithm.
[0031] Next, a median filtering algorithm is performed on the grayscale image or the melt pool image that has been adjusted to grayscale. First, a square window of a defined size is placed at the starting position in a corner of the grayscale image. The entire grayscale image is traversed pixel by pixel, moving one pixel at a time to ensure the window covers all pixels in the image. Specifically, edge pixels can be covered by padding with zeros or copying edge pixels. The window size can generally be 3×3 or 5×5, but can be adjusted according to the actual size of the image; this application does not impose specific limitations.
[0032] Then, for each slid-out window, the grayscale values of all pixels within the window are extracted, and the obtained grayscale values are arranged in ascending or descending order to obtain an ordered grayscale value sequence. Finally, the median of the grayscale value sequence is selected as the median of the window, and this median is used to replace the original grayscale value of the current window's center pixel. In this way, noise points within the window can be replaced with the median grayscale value of surrounding normal pixels, thereby removing noise points. At the same time, because the median value can reflect the overall brightness trend of surrounding pixels, the edge contour and detail information of the melt pool can be preserved.
[0033] In this way, random noise points in the grayscale image of the molten pool can be effectively removed without damaging the edge areas and image details of the molten pool image. This provides a high-quality image data source for subsequent feature recognition and extraction processes, thereby ensuring the accurate extraction of the size, shape, position and other attribute features of the molten pool.
[0034] In some implementations, please refer to [the relevant documentation]. Figure 1 The pipeline welding molten pool feature identification method in this application embodiment further includes: Step 03: Based on the pre-trained target detection model, obtain the attribute features of the molten pool image according to the pre-processed molten pool image.
[0035] Specifically, based on the above implementation method, by way of example, after preprocessing the molten pool image to remove noise points, the preprocessed molten pool image can be input into a pre-trained target detection model for feature recognition. The target detection model takes the preprocessed molten pool image as input and outputs the attribute features such as the size, shape, and position of the molten pool corresponding to each molten pool image, that is, it obtains the molten pool features through the target detection model.
[0036] Furthermore, the object detection model introduces a spatial attention module that integrates physical features of the molten pool. Based on this, please refer to [link / reference needed]. Figure 5 Step 03 specifically includes: Step 031: Obtain the semantic features corresponding to the processed molten pool image; Step 032: Based on the spatial attention module, perform spatial attention weighting processing on the semantic features to determine the attribute features of the melt pool image, so as to improve the accuracy of the attribute features.
[0037] Specifically, for example, the target detection model in the above embodiments can be based on the YOLO v8 target detection model in the current related technologies, and introduce a spatial attention mechanism that integrates the physical features of the molten pool. The spatial attention module corresponding to the spatial attention mechanism 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 processing 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 molten pool region, thereby accurately extracting features such as the size, shape, and position of the molten pool.
[0038] Therefore, under the conditions of the target detection model in the above example, the specific method for obtaining the attribute features of the molten pool corresponding to the molten pool image is as follows: The preprocessed molten pool image is input into the target detection model. The target detection model uses its backbone network to extract semantic features based on the input data, and uses a spatial attention module to perform spatial attention weighting processing on the semantic features to enhance the model's ability to represent molten pool features. Finally, the target detection model outputs the feature information of the molten pool corresponding to the molten pool image, which generally includes at least the bounding box coordinates, size, and shape of the molten pool.
[0039] In some implementations, please refer to [the relevant documentation]. Figure 1 The pipeline welding molten pool feature identification method in this application embodiment further includes: Step 04: Identify the current state of the molten pool based on preset baseline features and attribute features.
[0040] Specifically, based on the above implementation method, and assuming that the attribute features of each melt pool feature have been obtained by using the target detection model, for example, the obtained attribute features can be compared with the preset benchmark features, and the difference between the two can be used to identify whether the current state of the corresponding melt pool is normal, thereby realizing the identification of whether the current state of the corresponding melt pool is normal based on each melt pool image.
[0041] Further, please refer to Figure 6 Step 04 specifically includes: Step 041: Determine the feature differences based on the baseline features and attribute features; Step 042: If the feature difference exceeds the preset threshold range, identify the current state of the molten pool as the first state in order to control and adjust the parameters of the pipeline welding process; Step 043: If the feature differences do not exceed the preset threshold range, identify the current state of the melt pool as the second state. The second state indicates that the molten pool is currently in a normal state.
[0042] Specifically, to determine whether the current state of the molten pool is normal, in some examples, it is first necessary to compare the baseline features and the acquired attribute features to calculate and determine the feature differences between the two. For example, based on the above implementation method, attribute features generally include at least bounding box coordinates, size, shape, and other feature data. Correspondingly, the baseline features should also include at least bounding box coordinates, size, shape, and other data, which correspond one-to-one with the data types included in the attribute features.
[0043] When comparing baseline features and attribute features, the data included in both should be compared according to their types. For example, the bounding box coordinates in the attribute features should be compared with the bounding box coordinates included in the baseline features, the size in the attribute features should be compared with the size included in the baseline features, and the shape in the attribute features should be compared with the shape included in the baseline features.
[0044] Furthermore, the identification and judgment of the current state of the molten pool is mainly based on calculating the corresponding difference data through the aforementioned comparison method, and then determining the state based on the relationship between the difference data and the corresponding threshold range. For example, for the bounding box coordinates, a coordinate difference boundary threshold can be set for both the horizontal and vertical coordinates. The feature difference calculated based on the bounding box coordinates in the attribute features and the bounding box coordinates in the baseline features generally includes the difference in the horizontal coordinate and the difference in the vertical coordinate. Then, the obtained horizontal coordinate difference is compared with the corresponding coordinate difference boundary threshold, and the obtained vertical coordinate difference is compared with the corresponding coordinate difference boundary threshold. If either of these two conditions is greater than the corresponding coordinate difference boundary threshold, it can be directly determined that the current state of the molten pool is abnormal (corresponding to the first state). This indicates that there is an abnormality or hidden danger in the pipeline welding process, and the relevant parameters used in the pipeline welding process need to be adjusted to eliminate the abnormality or hidden danger. Only when both of these conditions are less than or equal to the corresponding coordinate difference, and the difference data between other types of attribute features and the baseline features does not exceed the corresponding threshold range, can it be determined that the current state of the molten pool is normal (corresponding to the second state). Therefore, only when the attribute characteristics of the molten pools corresponding to all molten pool images meet the above conditions can all molten pools be considered to be in a normal state, and only then can it be determined that the parameters of the entire pipeline welding process are normal and do not require adjustment. The above comparison and judgment process can be implemented using computer programs in current related technologies, and can be written or selected according to actual conditions; this application does not impose specific limitations.
[0045] Thus, this application can suppress arc interference in the molten pool image through a specially designed filter device and target detection model on the image acquisition device, thereby improving the accuracy of feature extraction. Furthermore, by performing real-time processing on the acquired molten pool image, it meets the requirements for real-time monitoring of molten pool features during pipeline welding, which helps to adjust welding process parameters in a timely manner, thereby ensuring welding quality.
[0046] The pipeline welding molten pool feature recognition device in this application includes: The image acquisition module is used to acquire images of the molten pool during the pipeline welding process based on the image acquisition device. The image processing module is used to perform preprocessing on the molten pool image to eliminate noise points in the molten pool image; The feature acquisition module is used to acquire the attribute features of the molten pool image based on the pre-trained target detection model and the pre-processed molten pool image. The feature recognition module is used to identify the current state of the molten pool based on preset baseline features and attribute features.
[0047] The embodiments of this application include a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the pipe welding molten pool feature recognition method in the above embodiments is implemented.
[0048] The computer-readable storage medium in the embodiments of this application stores a computer program, which, when executed by one or more processors, implements the pipe welding molten pool feature recognition method in the above embodiments.
[0049] The above description is merely a preferred embodiment of this application and is not intended to limit this application in any way. Although this application has disclosed the preferred embodiment as above, it is not intended to limit this application. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the technical solution of this application. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the technical solution of this application, based on the technical essence of this application and within the spirit and principles of this application, shall still fall within the protection scope of the technical solution of this application.
Claims
1. A method for identifying the characteristics of a weld pool in pipe welding, characterized in that, The method includes: Based on the image acquisition device, images of the molten pool are acquired during the pipeline welding process; Preprocessing is performed on the molten pool image to eliminate noise points in the molten pool image; Based on a pre-trained target detection model, the attribute features of the molten pool image are obtained according to the pre-processed molten pool image. The current state of the molten pool is identified based on preset baseline features and the attribute features.
2. The method according to claim 1, characterized in that, The image acquisition device acquires images of the molten pool, including: The image acquisition device is controlled to acquire multiple images of the molten pool at different locations during the pipe welding process; The image acquisition device includes a molten pool camera, and an interference filter, a polarizer, and an analyzer are sequentially arranged at the lens of the molten pool camera.
3. The method according to claim 2, characterized in that, The control of the image acquisition device to acquire multiple images of the molten pool at different locations during pipe welding includes: Control the image acquisition device to acquire one or more background images; Based on the background image, the average gray level and peak gray level of the corresponding background region are determined to determine the reference contrast of the molten pool image; Based on the reference contrast, the polarization axis angles of the polarizer and analyzer are adjusted so that the grayscale contrast of the molten pool image satisfies a preset relationship with the reference contrast; Under the given grayscale contrast, multiple images of the molten pool at different locations during the pipe welding process are acquired.
4. The method according to claim 1, characterized in that, The preprocessing of the molten pool image includes: Based on the median filtering algorithm, noise reduction processing is performed on the molten pool image to eliminate noise points in the molten pool image.
5. The method according to claim 4, characterized in that, The preprocessing of the molten pool image further includes: The molten pool image is converted to grayscale to adapt it to the median filtering algorithm.
6. The method according to claim 1, characterized in that, The target detection model incorporates a spatial attention module that integrates physical features of the molten pool; The pre-trained target detection model obtains attribute features of the molten pool image based on the preprocessed molten pool image, including: Based on the processed molten pool image, obtain the semantic features corresponding to the molten pool image; Based on the spatial attention module, spatial attention weighting processing is performed on the semantic features to determine the attribute features of the melt pool image, thereby improving the accuracy of the attribute features.
7. The method according to claim 1, characterized in that, The step of identifying the current state of the molten pool based on preset reference features and attribute features, in order to control the parameters of the pipe welding process, includes: Based on the baseline features and the attribute features, the feature differences are determined; If the feature difference exceeds a preset threshold range, the current state of the molten pool is identified as the first state in order to control and adjust the parameters of the pipe welding process; If none of the feature differences exceed a preset threshold range, the current state of the molten pool is identified as the second state, wherein the second state indicates that the molten pool is currently in a normal state.
8. A device for identifying the characteristics of weld pools in pipe welding, characterized in that, The device includes: The image acquisition module is used to acquire images of the molten pool during the pipeline welding process based on the image acquisition device. An image processing module is used to perform preprocessing on the molten pool image to eliminate noise points in the molten pool image; The feature acquisition module is used to acquire the attribute features of the molten pool image based on the pre-trained target detection model and the pre-processed molten pool image. The feature recognition module is used to identify the current state of the molten pool based on preset benchmark features and the attribute features.
9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program that, when executed by the processor, implements the pipe welding molten pool feature identification method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by one or more processors, implements the pipe welding molten pool feature identification method as described in any one of claims 1-7.