A 3D model fitting generation device and a defect detection method for cable defects

Abstract

The application discloses a kind of 3D model fitting generation device and defect detection method of cable defect;Through the combination of two groups of ray machines and DR imaging plates, the 360° internal structure profile of each layer of cable is obtained, and the cable surface modeling is obtained by two laser camera modules;Two modeling superimposes to form the model to be evaluated;Through artificial synthesis and difficult-to-obtain defect pattern merging conventional defect pattern, a more complete defect model training library is constructed, a cable defect deep learning model is formed using the training library deep learning neural network training, and a more complete and accurate defect evaluation report is obtained by comparing the model to be evaluated with the defect deep learning model;The structure of the device is simple, and compared with the traditional CT machine rotating structure, it is lighter and has lower cost;And the defect model library is more perfect than the conventional model library, so the defect evaluation is more accurate.

Description

Technical Field

[0001] This invention relates to the field of 3D imaging, and more specifically, to a device for generating a 3D model of cable defects and a method for detecting defects. Background Technology

[0002] Power cables are valuable production materials. Their internal and external structural features may contain defects, necessitating accurate and rapid inspection techniques during production, warehousing, and quality control. In recent years, with industrialization and urbanization, high-voltage transmission cables have been widely used in my country's urban power grids, becoming crucial power equipment in the power transmission and transformation process. Buried underground, cables suffer numerous defects in their outer sheath due to operating environment, design and construction quality, and external forces. Current technology often uses the rotating mechanism of a CT scanner for defect detection, but this method has too few acquisition angles compared to conventional CT principles, resulting in very coarse estimations of other angular contour values ​​and surface reconstruction. Laser cameras, however, can accurately model surfaces that may contain defects from a 360-degree perspective. The advantage of this sensor is its high accuracy in surface modeling, but the disadvantage is that it provides no information about internal structural features.

[0003] For example, a method for judging differentiated insulation defects in the outer sheath of high-voltage cables, disclosed in Chinese patent literature (publication number CN104865500A), includes the following steps: familiarizing oneself with the high-voltage cable structure according to the needs of judging the insulation status of the outer sheath; classifying the line sections with decreased outer sheath insulation into two categories according to the laying method based on the operational impact and external manifestations: direct burial or conduit laying and tunnel or cable bridge laying; if it is a first-category defect, determining whether it is a decrease in outer sheath insulation or damage to the outer sheath, and classifying the damage as localized or regional; if it is a second-category defect, judging the degree of decrease in outer sheath resistance and the increase in circulating current; under different laying methods, formulating corresponding inspection and judgment methods based on the operational impact, development speed, and economic efficiency of repair of the outer sheath defects; the judgment method proposed in this invention can quickly identify the specific causes of insulation defects in the outer sheath of high-voltage cables, providing technical support for subsequent specific maintenance. However, this invention can only assess the defect status of the cable's outer sheath, while it lacks evaluation standards for internal cable damage, especially for some internal damage that cannot be described by actual images. Therefore, it cannot accurately reconstruct the cable's defect status. Summary of the Invention

[0004] To address the problem of defect identification in power cables, this invention provides a 3D model fitting and generation device and a defect detection method for cable defects. The device uses a combination of two X-ray machines and a DR imaging plate to obtain 360° contours of the internal structure of each layer of the cable, and two laser camera modules to acquire surface models of the cable. The two models are superimposed to form the model to be evaluated. A relatively complete defect model training library is constructed by merging artificially synthesized and difficult-to-obtain defect graphics with conventional defect graphics. A deep learning neural network from the training library is used to train a deep learning model of the cable defects. By comparing the model to be evaluated with the deep learning model of defects, a more complete and accurate defect assessment report is obtained. The device has a simple structure and is lighter and less expensive than the rotating structure of a traditional CT scanner. Furthermore, the defect model library is more comprehensive than conventional model libraries, resulting in more accurate defect assessment.

[0005] The technical solution adopted by this invention to solve its technical problem is: A 3D model fitting and generation device for cable defects includes two sets of dynamic X-ray imaging devices. Each dynamic X-ray imaging device includes an X-ray machine and a DR imaging plate, which are respectively positioned on both sides of the cable. The two sets of dynamic imaging devices are arranged intersectingly around the cable. A laser camera module is positioned between the two sets of dynamic X-ray imaging devices. The device obtains the 360° structural contours of each layer inside the cable through the combination of the two sets of X-ray machines and the DR imaging plate, and acquires a surface model of the cable through the two laser camera modules. The two models are superimposed to form the model to be evaluated. This device has a simple structure and is lighter and less expensive than the traditional rotating structure of a CT scanner.

[0006] Preferably, the X-ray machine is positioned at the top of the cable; the DR imaging plate is positioned at the bottom of the cable; and a laser camera module is positioned between the two sets of X-ray machines and the two sets of DR imaging plates. The left-right crossover design ensures that the image can be captured and imaged from all angles, while the vertical arrangement of the laser cameras ensures that the cameras can capture a 360-degree image view of the cable surface, and the images captured by the laser cameras have better clarity, facilitating more accurate subsequent modeling.

[0007] A cable defect detection method, comprising: Step S1: The X-ray dynamic imaging device captures the layered structure and metal contour of the cable insulation layer; the contour model is then derived from the contour image. Step S2: The laser camera models the surface of the cable; Step S3: Overlay the obtained surface modeling and contour modeling, fit the 3D model data, and the generated data model is the model to be identified; Step S4: Compare the cable defect deep learning model to generate cable defect information.

[0008] The development packages provided by OpenGL and VTK offer sufficient tools for 3D model construction. A relatively complete defect model training library is built by merging artificially synthesized and hard-to-obtain defect images with conventional defect images. A deep learning neural network from this library is then used to train a deep learning model of cable defects. By comparing the model to be evaluated with the deep learning defect model, a more complete and accurate defect assessment report is obtained. Furthermore, the defect model library is more comprehensive than conventional model libraries, resulting in more accurate defect assessments.

[0009] Preferably, before establishing the contour modeling and surface modeling, the acquired cable image is first subjected to noise reduction, grayscale conversion, and filtering preprocessing to form image data to be identified. Processing the image data can speed up data processing and facilitate subsequent model construction and evaluation.

[0010] Preferably, the construction of the cable defect deep learning model includes the following steps: S51. Set up and build a deep learning neural network for cable defect identification; S52. Extract various defect features from the typical cable defect library, train a deep learning neural network using the defect features, and establish a deep learning model for cable defects. S53. After pre-evaluating the model to be identified, input the model to be identified into the cable defect deep learning model and compare the defect features. S54. Output the defect judgment results and generate a defect identification report.

[0011] By extracting typical defect features to form a defect training library, and using the training library to train a deep learning neural network to form a deep learning model of cable defects, a more complete and accurate defect assessment report is obtained by comparing the model to be evaluated with the defect deep learning model, which facilitates quick determination of the defect type of the tested cable.

[0012] Preferably, in step S53, the feature map of the model to be identified is pre-assessed based on the various defect characteristics of the typical cable defect library. Pre-assessment helps to speed up the subsequent assessment and judgment process, and the prediction and pre-classification also ensure more accurate defect judgment.

[0013] Preferably, the cable defect deep learning model uses the differentiated characteristics of cable grounding method, laying method, and degree of impact as a judgment strategy to pre-evaluate the model to be identified. Cable defects are classified according to the differences in defect states; labeling defect states helps to subsequently train the neural network using different defects, and accurate classification helps to accelerate the defect assessment and judgment speed.

[0014] Preferably, the deep learning neural network for cable defect identification is constructed as follows: Step S81: Establish a cable defect image database, expand the images through data augmentation and label the defect categories, and divide the training set and test set according to a preset ratio; Step S82: Construct the YOLOv3 defect detection model; Step S83: Train the defect model using the training set.

[0015] The YOLOv3 algorithm is commonly used for defect detection. As a mature defect detection model, it can be used to train deep learning neural networks.

[0016] Preferably, if there are defect categories that cannot be represented by actual cable defect images, the defect information is fitted using artificial image synthesis to expand the training and testing sets. Increasing the number of defect types in the training library makes the obtained defect detection model more complete.

[0017] The beneficial effects of this invention are mainly reflected in: 1. By capturing and constructing cable models using two sets of dynamic X-ray imaging devices and two sets of laser cameras, this invention achieves higher acquisition accuracy and a larger acquisition angle compared to conventional CT acquisition, resulting in a more realistic model. Furthermore, the size and weight of the device are significantly reduced, leading to an order-of-magnitude increase in processing speed for production processes.

[0018] 2. By combining artificially synthesized and hard-to-obtain defect images with regular defect images, a relatively complete defect model training library is constructed, making defect assessment more accurate. Attached Figure Description

[0019] Figure 1 This is a structural diagram of the device of the present invention; Figure 2 This is a flowchart of the method of the present invention; In the picture: 1-X-ray machine, 3-X-ray machine; 5-DR imaging plate, 7-DR imaging plate; 2-laser camera module, 6-laser camera module; 4-cable. Detailed Implementation

[0020] It should be understood that the embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0021] The technical solution of the present invention will be further described in detail below through embodiments.

[0022] A 3D model fitting and generation device for cable defects, such as Figure 1As shown, the device includes two sets of cross-arranged dynamic X-ray imaging devices. Each device comprises an X-ray machine and a DR imaging panel, with the X-ray machine positioned at the top of the cable and the DR imaging panel at the bottom. Laser camera modules are positioned between the two sets of X-ray machines and the two sets of DR imaging panels. This device can acquire 360° contours of the cable's surface and internal structural layers, and the two models are superimposed to form the model to be evaluated.

[0023] A cable defect detection method, comprising: I. A dynamic X-ray imaging device captures the layered structure and metal contour of the cable insulation layer; contour modeling is then derived from the contour images; a laser camera models the cable surface; before modeling, the images acquired by both methods need to be denoised, converted to grayscale, and filtered preprocessed to form the image data to be identified. The image processing process can accelerate the subsequent model construction, and the superposition of the two methods yields the model to be identified. This 3D model is completed using the development packages provided by OpenGL and VTK.

[0024] II. Constructing a cable defect model.

[0025] Configure and build a deep learning neural network for cable defect identification; Extract various defect features from the typical cable defect library, train a deep learning neural network using the defect features, and establish a deep learning model for cable defects. The cable defect deep learning model uses the differentiated characteristics of cable grounding method, laying method and degree of impact as the judgment strategy for cable defects to pre-evaluate the model to be identified; during the pre-evaluation process, the cable condition is compared with various defect features in the typical cable defect library; after the pre-evaluation of the model to be identified, the model to be identified is input into the cable defect deep learning model for defect feature comparison.

[0026] 3. Output the defect judgment results and generate a defect identification report.

[0027] The steps for constructing a deep learning neural network for cable defect identification are as follows: (1) Establish a cable defect image database, expand the images through data augmentation and label the defect categories. If there are defect categories that cannot be represented by actual cable defect images, fit the defect information through manual image synthesis and expand the training set and test set. Increase the types of defects in the training library to make the obtained defect detection model more complete; divide the training set and test set according to a preset ratio; this ratio can be set manually.

[0028] (2) Construct a YOLOv3 defect detection model. The YOLOv3 algorithm is often used for defect detection. As a mature defect detection model, it can be used to train deep learning neural networks.

[0029] (3) Use the training set to train the defect model.

Claims

1. A 3D model fitting and generation device for cable defects, characterized in that, The system includes two sets of dynamic X-ray imaging devices. Each device comprises an X-ray machine and a DR imaging plate, positioned on opposite sides of the cable. The two sets of dynamic imaging devices are arranged crosswise around the cable. A laser camera module is positioned between the two sets of dynamic X-ray imaging devices. The X-ray machine is positioned at the top of the cable, and the DR imaging plate is positioned at the bottom. Laser camera modules are positioned between the two sets of X-ray machines and the two sets of DR imaging plates. The system acquires a 360° view of the internal structure of each layer of the cable using the combination of the two sets of X-ray machines and the DR imaging plate. The two laser camera modules acquire surface models of the cable. These two models are superimposed to form the model to be evaluated. A defect model training library is constructed by merging artificially synthesized and difficult-to-obtain defect graphics with conventional defect graphics. A deep learning neural network from the training library is used to train a deep learning model of cable defects, which is then used to evaluate the cable defects.

2. A method for using a 3D model fitting and generation device for cable defects, applied to the 3D model fitting and generation device for cable defects as described in claim 1, characterized in that, include: Step S1: The X-ray dynamic imaging device captures the layered structure and metal contours of the cable insulation layer; Contour modeling is derived by reverse engineering from contour images; Step S2: The laser camera models the surface of the cable; Step S3: Overlay the obtained surface modeling and contour modeling, fit the 3D model data, and the generated data model is the model to be identified; Step S4: Compare the cable defect deep learning model to generate cable defect information.

3. The method of using the 3D model fitting and generation device for cable defects according to claim 2, characterized in that, Before establishing contour modeling and surface modeling, the acquired cable images are first subjected to noise reduction, grayscale conversion, and filtering preprocessing to form image data to be identified.

4. The method of using the 3D model fitting and generation device for cable defects according to claim 2, characterized in that, The construction of a deep learning model for cable defects includes the following steps: S51. Set up and build a deep learning neural network for cable defect identification; S52. Extract various defect features from the typical cable defect library, use the defect features to train a deep learning neural network, and establish a deep learning model for cable defects. S53. After pre-evaluating the model to be identified, input the model to be identified into the cable defect deep learning model and compare the defect features. S54. Output the defect judgment results and generate a defect identification report.

5. The method of using the 3D model fitting and generation device for cable defects according to claim 4, characterized in that, In step S53, the feature map of the model to be identified is pre-evaluated based on the various defect characteristics of the typical cable defect library.

6. The method of using the 3D model fitting and generation device for cable defects according to claim 5, characterized in that, The cable defect deep learning model uses the differentiated characteristics of cable grounding method, laying method and degree of impact as the judgment strategy for the identification model to pre-evaluate the cable defect.

7. The method of using the 3D model fitting and generation device for cable defects according to claim 2, characterized in that, The deep learning neural network for cable defect identification is constructed as follows: Step S81: Establish a cable defect image database, expand the images through data augmentation and label the defect categories, and divide the training set and test set according to a preset ratio; Step S82: Construct the YOLOv3 defect detection model; Step S83: Train the defect model using the training set.

8. The method of using the 3D model fitting and generation device for cable defects according to claim 7, characterized in that, If there are defect categories that cannot be represented by actual cable defect images, the defect information is fitted by artificial image synthesis to expand the training and test sets.

Images