Computer vision-based weld point insulation wrap detection method, medium, and product

Abstract

The application discloses a kind of based on computer vision's welding point heat preservation package detection method, comprising the following steps: S1, video acquisition and pretreatment;S2, welding point and heat preservation material identification: based on pre-trained deep learning model is analyzed to video frame, while identifying steel pipe welding point and surrounding heat preservation material;S3, object tracking and counting: the welding point detected is accurately tracked across frame, and each welding point is assigned unique identifier, while realizing the automatic counting of steel pipe welding point;S4, package state analysis: extract welding point key vertex, analyze the space relationship of welding point and heat preservation material;S5, package duration calculation: record the start time and duration of welding point being wrapped, ensure to reach the heat preservation duration of process requirement;S6, result statistics and storage;S7, alarm push and record.The application can replace traditional manual monitoring and infrared temperature measurement scheme by computer vision technology, improve detection efficiency and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent industrial vision inspection, and particularly relates to a welding point heat preservation wrapping detection method, medium and product based on computer vision. Background Art

[0002] In the process of steel pipe welding production, the welding point is a key part of the steel pipe structure, and its quality directly affects the overall strength and service life of the steel pipe. After welding, it is necessary to use special heat preservation materials to wrap and insulate the welding point to ensure that the welding point cools slowly, avoiding the generation of welding stress and cracks. And it is necessary to ensure that the heat preservation wrapping time of the welding point meets the process requirements before it can be considered qualified.

[0003] To ensure the quality and timeliness of the heat preservation wrapping of the welding point, a variety of detection methods have been adopted in the industry, mainly including the following two categories: 1. Manual inspection method In the prior art, this process mainly relies on manual inspection, which has the following limitations: - Low inspection efficiency: Special personnel are required to monitor the heat preservation state of the welding point in real time, wasting human resources; - Strong subjectivity of results: Different inspectors may have inconsistent judgment criteria for the heat preservation wrapping of the welding point; - Difficult to accurately record the wrapping time: Manual timing is prone to errors, affecting the quality of the welding point; - High missed inspection rate: Manual inspection is prone to fatigue, resulting in the situation of incomplete heat preservation wrapping of the welding point being ignored; - Difficult to trace data: The inspection results lack systematic records and are difficult to conduct subsequent quality analysis and traceability.

[0004] 2. Infrared temperature measurement-based scheme This scheme indirectly judges the heat preservation wrapping state by detecting the temperature change of the welding point. However, this scheme has the following significant limitations: - High deployment complexity and low resource utilization rate: Infrared temperature measurement equipment needs to be installed separately at each welding station, with a high deployment cost, and the existing monitoring camera resources cannot be utilized; - Insufficient target recognition ability: When steel pipes are densely stacked, infrared temperature measurement technology cannot effectively distinguish the welding points of different steel pipes, resulting in a decrease in the accuracy of the detection results; - High false alarm rate: It is easy to identify high-temperature equipment other than steel pipes (such as welding torches, hot air guns, etc.) as targets, generating false alarms; - Indirect detection method: It can only indirectly infer the wrapping state through temperature changes and cannot directly identify the actual coverage of the heat preservation material; - Poor environmental adaptability: External factors such as environmental temperature and air flow will have a significant impact on the temperature measurement accuracy; - Insufficient timeliness: It can only be detected and alarmed after the temperature has dropped. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a computer vision-based method for detecting insulation wrapping at welding points. For the quality control of insulation wrapping at welding points, computer vision technology can replace traditional manual monitoring and infrared temperature measurement methods, thereby improving detection efficiency and accuracy.

[0006] To address the aforementioned technical problems, this invention provides a computer vision-based method for detecting insulation wrapping at weld points, comprising the following steps: S1, Video Acquisition and Preprocessing: Managing multiple video sources, supporting switching between different welding station videos, acquiring video streams of the insulation process at steel pipe weld points through a camera, and preprocessing the videos; S2, Weld Point and Insulation Material Identification: Analyzing video frames based on a pre-trained deep learning model, simultaneously identifying the steel pipe weld points and surrounding insulation materials; S3, Object Tracking and Counting: Performing precise cross-frame tracking of detected weld points, assigning a unique identifier to each weld point, and simultaneously achieving automatic counting of steel pipe weld points; S4, ... Wrapping Status Analysis: Extract key vertices of the welding points, analyze the spatial relationship between the welding points and the insulation material, and use a vertex wrapping coverage algorithm to calculate the wrapping coverage rate to determine whether the welding points are completely wrapped by the insulation material; S5, Wrapping Duration Calculation: Record the start time and duration of the welding point wrapping to ensure that the insulation duration required by the process is met; S6, Result Statistics and Storage: Statistics on the number of welding points, wrapping status, and wrapping duration, store historical data, and support historical data query, traceability by welding point ID, and quality analysis; S7, Alarm Push and Recording: When incomplete wrapping of welding points, insufficient wrapping duration, or abnormal video source is detected, alarm information is pushed and alarm details are recorded.

[0007] Furthermore, in step S1, the video is preprocessed, including frame skipping, noise reduction, and dynamic compensation to improve analysis efficiency.

[0008] Furthermore, step S1 performs frame skipping processing by processing one frame every 1-50 frames.

[0009] Further, the training process of the deep learning model in step S2 is as follows: acquire image data of different types of steel pipe welding points and insulation materials collected in step S1; annotate the collected images to mark the location and category of welding points and insulation materials; use the SAM3 network to perform semantic target segmentation on the annotated data to extract the precise contours of welding points and insulation materials; expand the original dataset through the AI ​​large model to generate more training samples and improve the model's generalization ability; train the YOLO segmentation model using the expanded dataset to optimize the recognition accuracy of welding points and insulation materials; optimize model performance through data augmentation and parameter tuning to improve adaptability in different scenarios.

[0010] Further, step S4 includes: vertex extraction, extracting the key vertex coordinate set of the welding point. Where n is the number of vertices at the welding points; extraction of the insulation material coverage area; calculation of vertex coverage rate for each vertex. Calculate whether it is located within the area covered by the insulation material using the following formula: C represents the pixel area of ​​the thermal insulation material; When the coverage reaches the threshold, step S5 records the start time and calculates the duration.

[0011] Furthermore, the vertices extracted in step S4 are located within a user-preset region of interest.

[0012] Furthermore, the alarm details in step S7 include alarm type, time, location, and processing status.

[0013] The present invention also provides a computer-readable storage medium having computer instructions stored thereon, which are executed by a processor to implement the above-described method for detecting the insulation wrapping of weld points.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for detecting the insulation wrapping of weld points.

[0015] Compared with existing technologies, the present invention has the following advantages: The computer vision-based welding point insulation wrapping detection method provided by the present invention combines deep learning, target segmentation, object tracking, and mathematical algorithms for the quality control of welding point insulation wrapping, realizing real-time monitoring and intelligent analysis of the steel pipe welding point insulation wrapping process; it improves detection efficiency and accuracy by replacing traditional manual monitoring and infrared temperature measurement schemes with computer vision technology; and it is applicable to industrial fields that require post-weld insulation treatment, such as steel pipe manufacturing, petrochemical pipelines, power pipelines, and building steel structures. Attached Figure Description

[0016] Figure 1 This is a flowchart of the computer vision-based detection process for insulation wrapping of weld points according to the present invention. Figure 2 This is a diagram illustrating the actual application effect of the invention on a steel pipe welding production line. Detailed Implementation

[0017] The present invention will now be further described with reference to the accompanying drawings and embodiments.

[0018] Figure 1 This is a flowchart of the computer vision-based detection process for insulation wrapping of weld points according to the present invention.

[0019] Please see Figure 1 The computer vision-based detection method for insulation wrapping of weld points provided by this invention includes the following steps: S1. Video Acquisition and Preprocessing: Manages multiple video sources, supports switching between different welding station videos, acquires video streams of the insulation process at steel pipe welding points through cameras, and preprocesses the videos, including frame skipping, noise reduction, and dynamic compensation to improve analysis efficiency; frame skipping is performed in a manner of processing 1 frame every 1-50 frames, with a default of processing 1 frame every 25 frames. S2. Identification of Welding Points and Insulation Materials: Based on a pre-trained deep learning model (through SAM3 semantic segmentation, dataset expansion, and model training optimization), video frames are analyzed to identify the steel pipe welding points and the surrounding insulation materials. S3. Object Tracking and Counting: Accurately track detected welding points across frames, assign a unique identifier to each welding point, and automatically count the welding points of the steel pipe. S4. Wrapping Status Analysis: Extract key vertices of the welding point, analyze the spatial relationship between the welding point and the insulation material, use the vertex wrapping coverage algorithm to calculate the wrapping coverage rate, and determine whether the welding point is completely wrapped by the insulation material. S5. Wrapping time calculation: Record the start time and duration of wrapping the welding point to ensure that the required insulation time is met; S6. Results Statistics and Storage: Statistics on the number of welding points, package status and package duration; storage of historical data; support for historical data query, traceability by welding point ID and quality analysis. S7. Alarm Push and Recording: When incomplete welding point wrapping, insufficient wrapping time, abnormal video source, etc. are detected, alarm information is pushed and alarm details are recorded, including alarm type, time, location and processing status, so as to promptly remind staff to handle the situation.

[0020] The present invention implements the above process using the following technical solution: 1. Video Processing and ROI (Region of Interest) Management Module: Responsible for video loading, playback, and preprocessing. It supports frame skipping to improve analysis efficiency and allows users to define regions of interest, analyzing only the welding points within the ROI. 2. Target Recognition Module: Uses a pre-trained deep learning model to simultaneously identify the steel pipe welding points and the surrounding insulation material in the video; 3. Object Tracking and Counting Module: Enables accurate cross-frame tracking of welding points, assigns a unique ID to each welding point, and also has the function of counting steel pipe welding points; 4. Wrapping Analysis Module: Calculates and determines the wrapping status based on the spatial relationship between the welding point and the insulation material, and also calculates the duration for which the welding point is wrapped by the insulation material; 5. Data Management Module: Records and stores analysis results, supports historical data query, traceability by weld point ID, and quality analysis; 6. Alarm Module: When incomplete welding point wrapping, insufficient wrapping time, abnormal video source, etc. are detected, alarm information is pushed and alarm details are recorded, including alarm type, time, location and processing status.

[0021] The main innovative points of this invention include: 1. Precise identification of welding points: It realizes the automatic identification and positioning of welding points of steel pipes, ensuring the accuracy of detection and solving the problem that infrared temperature measurement schemes cannot distinguish stacked steel pipes; 2. Automated inspection process: It realizes automated inspection of the insulation wrapping status and duration of the welding points, replacing manual inspection and improving inspection efficiency and consistency; 3. Deep learning applications: SAM3 is used for semantic object segmentation, and YOLO-segment is used for object recognition to improve detection accuracy and speed; 4. Steel pipe target tracking and counting: Achieve accurate cross-frame tracking of welding points in single-camera scenarios, while also having the function of counting steel pipe welding points to ensure the continuity of package status analysis and avoid missed detections; 5. Intelligent Wrapping Relationship Analysis: Based on vertex wrapping coverage algorithm and mathematical formula calculation, the wrapping status of welding points is accurately determined, solving the problem of indirect inference in infrared temperature measurement scheme; 6. Data Management and Alarms: Enables storage of analysis results and alarms for abnormal situations, improving production management efficiency and supporting subsequent quality traceability; 7. Deployment flexibility: Utilizing existing camera resources, no special equipment needs to be installed, reducing deployment costs and solving the problem of cumbersome deployment in traditional solutions; 8. Environmental adaptability: Unaffected by ambient temperature, light, and air flow, it is stable and reliable, solving the problem of infrared temperature measurement solutions being greatly affected by the environment; 9. Multi-station parallel monitoring: A single system can monitor multiple welding stations simultaneously, improving system utilization; 10. Automatic generation of analysis reports: The system can automatically generate analysis reports, reducing manual workload.

[0022] As a preferred embodiment, the training phase of the deep learning model of the present invention is as follows: 1. Data Acquisition: Acquire image data of welding points of different types of steel pipes and insulation materials; 2. Data annotation: Annotate the acquired images, marking the location and type of welding points and insulation materials; 3. SAM3 Semantic Segmentation: The SAM3 (Segment Anything Model 3) network is used to perform semantic target segmentation on the labeled data, extracting the precise contours of the welding points and insulation materials; 4. Dataset Expansion: Expand the original dataset using a large AI model to generate more training samples and improve the model's generalization ability; 5. Model Training: The YOLO segmentation model was trained using the expanded dataset to optimize the recognition accuracy of weld points and insulation materials; 6. Model Optimization: Optimize model performance through data augmentation and parameter tuning to improve adaptability in different scenarios.

[0023] As a preferred embodiment, the detection phase process of the deep learning model of the present invention is as follows: 1. Video Loading: Load the video file of the insulation process of the welding point to be analyzed; 2. Model Loading: Load the pre-trained YOLO segmentation model; 3. Frame processing: Process video frames, skipping one frame every 25 frames; 4. Welding point and insulation material identification: The YOLO-segment network is used to analyze video frames and identify the steel pipe welding points and the surrounding insulation material. 5. Object tracking: Track detected weld points across frames and assign a unique identifier to each weld point; 6. Vertex Extraction: Extract the coordinates of key vertices at the welding points; 7. Wrapping State Analysis: The spatial relationship between the welding points and the insulation material is analyzed using the vertex wrapping coverage algorithm; 8. Wrapping time calculation: Record the start time and duration of wrapping the weld joint; 9. Results Output: Displays analysis results and stores historical data.

[0024] In a preferred embodiment, the vertex wrapping coverage algorithm of the present invention calculates the insulation material wrapping coverage of the welding point through the following steps: , 1. Vertex Extraction: Extract the coordinate set of key vertices of the welding point. , where n is the number of vertices at the welding point; 2. Extraction of the insulation material coverage area: Extract the coverage area C of the insulation material as the pixel area of ​​the insulation material; 3. Vertex cover calculation: For each vertex Calculate whether it is located within the area covered by the insulation material; 4. Coverage Calculation: Calculate the percentage of vertex coverage; 5. Package duration calculation: When the coverage reaches the threshold, record the start time and calculate the duration.

[0025] The advantages of the algorithm described above in this invention are as follows: • High accuracy: The vertex-coverage-based calculation method can accurately assess the wrapping state of the weld point. • Good real-time performance: Low computational complexity, suitable for real-time video analysis • High adaptability: Unaffected by the shape and size of the weld joint • High interpretability: The computation process is transparent and the results are easy to understand. As a preferred embodiment, the management phase process of the present invention is as follows: 1. ROI Settings: Users set the region of interest, and only the weld points within the ROI are analyzed; 2. Parameter Configuration: Configure analysis parameters, such as confidence threshold, IOU (Intersection over UnionThreshold) threshold, etc. 3. Video source management: Manage multiple video sources and support switching between different welding station videos; 4. Results Query: Query historical analysis results and statistical data, and support tracing by weld point ID.

[0026] The computer vision-based detection method for insulation wrapping of weld points provided by this invention has the following technical advantages: 1. Improve inspection efficiency: Automated inspection replaces manual inspection, improving the efficiency and accuracy of inspecting insulation wrapping at weld points; 2. Ensure welding quality: Ensure that the insulation wrapping of the steel pipe welding points meets the process requirements to improve welding quality; 3. Reduced manpower: No need for dedicated personnel to monitor the insulation status of welding points in real time, saving manpower; 4. Data traceability: The system records and analyzes the results, facilitating subsequent quality inquiries and traceability; 5. Real-time alarms: Promptly detect abnormalities in the insulation wrapping of welding points, improving production management efficiency; 6. High adaptability: Supports the inspection of different types of steel pipe welding points and insulation wrapping in different scenarios.

[0027] This invention is applicable to the following application scenarios: 1. Steel pipe welding production line: Real-time monitoring of the execution of the insulation process at the welding points; 2. Industrial production quality control: Ensure that the wrapping time of the insulation material at the welding points meets the process requirements; 3. Production process optimization: Optimize the insulation process at welding points based on data analysis to improve production efficiency; 4. Quality Traceability: Record the execution of the insulation process at the welding points to facilitate traceability of quality issues.

[0028] Comparative analysis with existing solutions 1. Comparison with manual video surveillance

[0029] 2. Economic Benefit Comparison (1 region, 16 testing sites)

[0030] 3. Comprehensive Assessment Compared with traditional manual video surveillance methods and infrared temperature measurement solutions, this invention offers significant improvements in detection performance, economic benefits, and technological advantages. • Compared to manual monitoring: – Performance improvement: Detection accuracy and completeness are significantly higher than manual monitoring, and the false negative rate is greatly reduced; – Cost savings: Although there is an initial investment, the long-term operating costs are significantly lower than those of manual monitoring, and the investment payback period is short; – Technical advantages: Possesses multiple technical advantages such as real-time alarms, data storage, and consistency detection; – Adaptability: Not limited by working hours, can work continuously for 24 hours, suitable for various industrial environments.

[0031] • Compared to infrared temperature measurement solutions: – Deployment advantages: Utilize existing cameras, eliminating the need for dedicated equipment installation, thus reducing deployment costs; – Detection accuracy: It can accurately identify individual steel pipes and welding points, avoiding false detection of high-temperature equipment; – Package status detection: Directly detect the status of the insulation material package, rather than inferring it indirectly; – Environmental adaptability: Unaffected by ambient temperature and light, stable and reliable; – Cost-effectiveness: Low long-term operating costs and fast return on investment.

[0032] The actual application effect diagram on a certain steel pipe welding production line is shown below. Figure 2 As shown, the present invention achieves the following effects: • The pass rate for insulation wrapping at weld points increased from 87% to 99.5%; • The rate of missed detections decreased from 5.3% to 0.2%, reducing scrap and rework by 126 times due to welding overheating; • Each production shift saves approximately 3,000 yuan in labor costs; • The time to trace quality issues has been reduced from an average of 2 hours to 1 minute; These data fully demonstrate the effectiveness and advantages of this system in actual industrial production.

[0033] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be defined by the claims.

Claims

1. A computer vision-based method for detecting insulation wrapping at weld points, characterized in that, Includes the following steps: S1. Video Acquisition and Preprocessing: Manage multiple video sources, support switching between different welding station videos, acquire video streams of the insulation process at steel pipe welding points through cameras, and preprocess the videos. S2. Welding point and insulation material identification: Based on a pre-trained deep learning model, video frames are analyzed to identify the steel pipe welding points and the surrounding insulation material. S3, Object Tracking and Counting: Accurately track detected welding points across frames, assign a unique identifier to each welding point, and automatically count the welding points of the steel pipe. S4. Wrapping Status Analysis: Extract key vertices of the welding point, analyze the spatial relationship between the welding point and the insulation material, use the vertex wrapping coverage algorithm to calculate the wrapping coverage rate, and determine whether the welding point is completely wrapped by the insulation material. S5. Wrapping time calculation: Record the start time and duration of wrapping the welding point to ensure that the required insulation time is met; S6. Results Statistics and Storage: Statistics on the number of welding points, package status and package duration; storage of historical data; support for historical data query, traceability by welding point ID and quality analysis. S7. Alarm Push and Recording: When incomplete welding point wrapping, insufficient wrapping time, or abnormal video source is detected, alarm information is pushed and alarm details are recorded.

2. The computer vision-based detection method for insulation wrapping of weld points as described in claim 1, characterized in that, In step S1, the video is preprocessed, including frame skipping, noise reduction, and dynamic compensation to improve analysis efficiency.

3. The computer vision-based detection method for insulation wrapping of weld points as described in claim 2, characterized in that, Step S1 involves frame skipping, processing one frame every 1-50 frames.

4. The computer vision-based detection method for insulation wrapping of weld points as described in claim 1, characterized in that, The training process of the deep learning model in step S2 is as follows: Acquire image data of different types of steel pipe welding points and insulation materials collected in step S1; The acquired images are labeled, marking the location and type of welding points and insulation materials; The SAM3 network was used to perform semantic target segmentation on the labeled data to extract the precise contours of the welding points and insulation materials. By expanding the original dataset with large AI models, more training samples are generated, thereby improving the model's generalization ability. The YOLO segmentation model was trained using the expanded dataset to optimize the recognition accuracy of weld points and insulation materials. Optimize model performance through data augmentation and parameter tuning to improve adaptability in different scenarios.

5. The computer vision-based detection method for insulation wrapping of weld points as described in claim 1, characterized in that, Step S4 includes: Vertex extraction: Extracting the coordinate set of key vertices at the welding points. , where n is the number of vertices at the welding point; Extraction of the area covered by the thermal insulation material; Vertex coverage calculation, for each vertex Calculate whether it is located within the area covered by the insulation material using the following formula: C represents the pixel area of ​​the thermal insulation material; When the coverage reaches the threshold, step S5 records the start time and calculates the duration.

6. The computer vision-based detection method for insulation wrapping of weld points as described in claim 5, characterized in that, The vertices extracted in step S4 are located within the user-preset region of interest.

7. The computer vision-based detection method for insulation wrapping of weld points as described in claim 1, characterized in that, The alarm details in step S7 include alarm type, time, location, and processing status.

8. A computer-readable storage medium storing computer instructions thereon, characterized in that, The computer instructions are executed by the processor to implement the welding point insulation wrapping detection method as described in any one of claims 1-7.

9. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the welding point insulation wrapping detection method as described in any one of claims 1-7.

Images