A machine vision-based cable insulation layer defect detection device

Abstract

The application relates to the technical field of cable defect detection, and particularly relates to a cable insulation layer flaw detection device based on machine vision, which comprises a bending conveying module, an image module and an analysis control module. The bending conveying module is used for setting the detection path of the cable insulation layer to be detected as a serpentine bending path, and makes the cable insulation layer move along the serpentine bending path. The image module is used for acquiring optical images of each arc-shaped convex side in the serpentine bending path of the cable insulation layer to be detected. The analysis control module is used for analyzing flaw features on the surface and near-surface of the cable insulation layer to be detected based on the optical images, and outputting analysis results. The cable is bent along the serpentine bending path to generate stress, so that the hidden flaws such as internal delamination and micro cracks are exposed on the arc-shaped convex side and form identifiable surface responses. In combination with accurate image acquisition and multi-dimensional feature analysis, the flaws and normal deformation can be effectively distinguished, the flaw types on the surface and near-surface are comprehensively covered, and the accuracy and comprehensiveness of detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cable defect detection, and particularly relates to a cable insulation layer flaw detection device based on machine vision. BACKGROUND

[0002] As the core carrier of power transmission and signal transmission, cables are widely used in industrial production, infrastructure construction and other fields. The cable is mainly composed of a conductor, an insulation layer, a shielding layer and a sheath. The insulation layer is a key component of the cable, and its core function is to isolate the conductor from the external environment, prevent current leakage, avoid short circuit accidents, and protect the conductor from mechanical damage and corrosion, directly determining the safety and service life of the cable. In the cable production process, due to factors such as raw material purity, processing precision, equipment running state, etc., the insulation layer is prone to various defects, including air bubbles, micro-cracks, impurity inclusions, uneven thickness, etc. These defects can seriously weaken the insulation performance of the insulation layer, and may cause insulation breakdown during long-term use, leading to power interruption, equipment damage, even fire and other safety hazards, causing significant losses to production and life.

[0003] In order to ensure the quality of the cable, the insulation layer is usually systematically detected to eliminate unqualified products and avoid safety risks. The current mainstream detection methods include manual visual inspection, pressure test, ultrasonic detection, eddy current detection, etc. Manual visual inspection relies on the experience of the operator, and is low in efficiency and easy to miss small defects. The pressure test focuses on the verification of insulation performance, and it is difficult to locate the specific defect position. Although ultrasonic and eddy current detection can penetrate to detect internal defects, the equipment cost is high and the detection speed is limited in adaptability. Therefore, the surface and near-surface (0.1-1.0mm) detection of some defects can also be carried out by machine vision, which has the advantages of high efficiency, objectivity and quantifiability, and gradually becomes an important supplement to the detection of cable insulation layer.

[0004] But in practical application, due to the fact that most insulation layer defects are small in size and irregular in shape, and are easily confused with surface texture and light changes of the insulation layer, the recognition degree of the defect features in the image is low, and the conventional machine vision cannot accurately distinguish them, resulting in frequent missed detection and false detection. Secondly, when detecting through machine vision, the detection path of part of the cable is usually a straight line (for example, a kind of cable insulation layer detection device and method disclosed in patent publication No. CN119936062A, which marks the surface defects through paint to improve the image recognition effect), although it can effectively capture the visible defects (such as scratches and stains) on the surface of the cable, but for the internal invisible problems (such as internal delamination, micro cracks and local material unevenness) near the surface of the insulation layer, such problems are difficult to respond on the surface under the condition of the existing straight detection path, and thus are difficult to be recognized by 2D or 3D imaging, resulting in high risk of missed detection. Therefore, the present application provides a kind of cable insulation layer defect detection device based on machine vision to solve the above problems. SUMMARY

[0005] To solve the above problems, the present application provides a kind of cable insulation layer defect detection device based on machine vision, which is used for accurately detecting the defects such as bubbles, micro cracks and internal delamination on the surface and near the surface of the cable insulation layer, capturing the hidden defects difficult to expose under the straight detection path, improving the detection accuracy and comprehensiveness, and ensuring the safety and service life of the cable.

[0006] In order to achieve the above purpose, the technical scheme of the present application is as follows: a kind of cable insulation layer defect detection device based on machine vision, comprising:

[0007] The curved conveying module is used for conveying the cable in a non-linear form;

[0008] The image module is used for acquiring the optical image of each arc-shaped convex side of the cable in the non-linear form; wherein the coverage angle of the optical image is greater than 180°;

[0009] The analysis control module is used for identifying the defect features on the surface and near the surface of the cable insulation layer based on the optical image, and outputting the identification result.

[0010] Further, the curved conveying module includes a traction subunit and a plurality of arc-shaped path subunits;

[0011] The arc-shaped path subunits are connected in sequence and end to end, and are used for building the detection path for conveying the cable in a non-linear form;

[0012] The traction subunit is used for pulling the cable in the arc-shaped path subunit, so that the cable is conveyed along the detection path.

[0013] Further, the image module is configured to acquire a two-dimensional image of the texture, color and planar defects of the cable convex side surface based on a 2D image, and obtain three-dimensional topographic data of the cable convex side surface based on a 3D image.

[0014] Further, the analysis control module is configured to receive the two-dimensional image and the three-dimensional topographic data, perform image preprocessing, extract defect feature parameters of the cable arc convex side region based on the preprocessed image, compare the extracted feature parameters with a preset defect feature library, distinguish the defect type and normal deformation by combining a machine learning model, quantitatively analyze the severity of each defect, determine whether the cable insulation layer to be detected is qualified according to a preset quality standard, and output an analysis report containing the defect type, location, size parameter and qualification determination result.

[0015] The defect feature parameters include 2D dimensional shape, area, gray distribution, and 3D dimensional depth, height and slope change.

[0016] The above scheme has the following beneficial effects: the scheme combines the serpentine curved path design with multi-dimensional detection technology, realizes the comprehensiveness and precision improvement of cable insulation layer defect detection, and has the following core beneficial effects:

[0017] Firstly, the bending structure formed by the arc path subunit of the bending conveying module applies controllable stress to the cable insulation layer, so that the near-surface (0.1-1.0 mm) hidden defects such as internal delamination and micro-cracks that are difficult to expose under a conventional straight path produce obvious surface responses at the arc convex side (stress concentration area), effectively making up for the missed defects of traditional straight-line conveying detection, and simulating the bending working condition of the cable in actual use, thereby reducing the quality risk of “static qualification, dynamic failure”;

[0018] Secondly, the image module adopts a 2D+3D joint imaging method with a coverage angle of more than 180°, which can accurately capture surface texture, color-related defects (such as impurities and scratches) through 2D images, and quantitatively measure three-dimensional features such as bump height and depression depth by means of 3D topographic data, and with a coverage range of more than 180°, the detection blind area of the arc surface is eliminated, and the recognition degree of defect features is improved.

[0019] Thirdly, the analysis control module can accurately distinguish defects and normal deformation caused by bending by means of multi-dimensional parameter extraction (2D+3D) and machine learning model application, and reduce false positives; at the same time, the severity of the defect is quantified and a detailed analysis report is output, which provides data support for quality control, retains the advantages of machine vision in efficiency and objectivity, and solves the problem of “difficulty in quantification and distinction” in traditional detection.

[0020] Further, the arc-shaped path sub-units each include a mounting plate, the surface of the mounting plate is symmetrically provided with an inlet and an outlet, the surface of the mounting plate is fixedly connected with an arc-shaped guide frame, the side surface of the guide frame away from the mounting plate is provided with a guide sliding groove along the length direction of the guide frame, and the two ends of the guide sliding groove correspond to the inlet and the outlet respectively; the surface of the mounting plate close to the guide frame is symmetrically provided with an arc-shaped sealing plate in up and down directions.

[0021] The traction sub-unit includes a traction roller shaft for traction of the cable, and the driving part of the traction roller shaft is signal connected with the analysis control module.

[0022] Beneficial effects: the arc-shaped guide sliding groove of the guide frame and the inlet and outlet of the mounting plate are precisely connected, cooperating with the traction of the traction roller shaft, so that the cable can be forced to move along the preset serpentine bending track stably, reducing the deviation and shaking in the detection process, ensuring that the arc-shaped protruding side is always within the collection range of the image module, and ensuring the imaging stability. The arc-shaped sealing plates symmetrically arranged in up and down directions can form a relatively closed detection space, effectively isolating the external interference such as dust and stray light in the workshop.

[0023] Further, the image module includes a plurality of protective covers, each of which is hingedly connected to a mounting plate and corresponds to a sealing plate; the side of each protective cover close to the guide frame is fixedly connected with a plurality of image sensors signal connected with the analysis control module, and the collection end of the image sensor faces the arc peak area of the guide frame.

[0024] Beneficial effects: the protective cover is hingedly connected to the mounting plate and corresponds to the sealing plate, which can not only improve the closed detection space together with the sealing plate to effectively protect the image sensor from dust and impact, but also can be easily opened and closed for maintenance through the hinged design; the image sensor is precisely aligned with the arc peak position of the guide frame, and the arc peak is the stress concentration core area when the cable is bent, and the hidden defects are most fully exposed, ensuring that the collected image focuses on the key detection area, providing direct protection for the subsequent 180° or more coverage imaging and accurate identification of the analysis control module, and improving the pertinence and effectiveness of defect capture.

[0025] Further, a pump assembly for providing gas is arranged in each mounting plate, the pump assembly is signal connected with the analysis control module, and the inner wall of the inlet and outlet is provided with a gas bag ring in communication with the pump assembly.

[0026] Beneficial effects: after the pump assembly is inflated, the air bag ring can be self-adapted to inflate and fit according to the cable diameter, without the need to replace the adapter to be compatible with cables of different diameters, and the adaptability is flexible and efficient; at the same time, the air bag ring further fixes the cable through a flexible clamping mode, which can effectively limit the deviation and shaking of the cable at the inlet and outlet, ensure the stable movement of the cable along the guide groove, and the flexible contact will not damage the surface of the insulation layer, providing reliable positioning protection for the accurate alignment of the image sensor to the arc peak detection area and the collection of clear and effective images. In addition, during the sliding process of the cable, the air bag ring forms friction with the surface of the cable, which can automatically clean the surface dust, oil stains and other dirt, reduce the pollution of the insulation layer, and improve the cleanliness and clarity of the image collected by the image sensor.

[0027] Further, the inner wall of the guide groove is linearly arrayed with a plurality of rotating columns connected with the guide frame in the length direction of the guide frame.

[0028] Beneficial effects: the scheme converts the sliding friction between the cable and the guide groove into rolling friction, reduces the resistance of the cable when moving along the serpentine bending path, effectively reduces the problem of jamming and sticking caused by excessive friction, and ensures the continuous and smooth detection process. At the same time, rolling contact can reduce the friction damage to the surface of the cable insulation layer, and reduce the influence of secondary defects on the detection results.

[0029] Further, the surface of the air bag ring is annularly arrayed with a plurality of cleaning pipes; the guide frame is internally provided with a plurality of flow guide cavities corresponding to the rotating columns one by one, the flow guide cavities are internally provided with a main gear and a secondary gear meshing with the main gear, the side surfaces of the main gear and the secondary gear are always in contact with the side wall of the flow guide cavity, the secondary gear is rotatably connected to the inner wall of the flow guide cavity, one end of the rotating column extends into the corresponding flow guide cavity and is fixedly connected with the main gear in a coaxial manner; the flow guide cavities are respectively provided with an air inlet and an air outlet on the two sides, which are communicated with the outside and the cleaning pipes respectively; the mutually close sides of the sealing plates are provided with arc-shaped diffuse strip light sources.

[0030] Beneficial effects: the cable rotates the rotating column when moving, which in turn drives the main gear and the secondary gear to mesh and extrude the flow guide cavity, forming a high-speed airflow and spraying out through the cleaning pipe, accurately flushing the stubborn dirt embedded in the flaw gap, further ensuring the cleanliness of the image collected by the image sensor, reducing the dirt blocking or confusing the flaw features, laying a more reliable foundation for the analysis control module to accurately identify the surface and near-surface flaws, and reducing the risk of missed detection and false detection caused by dirt interference. The arc-shaped diffuse strip light source fits the shape of the cable, can form uniform and soft diffuse lighting on the surface of the insulation layer, effectively eliminate local glare and shadow interference, enhance the contrast of small flaws (such as micro-cracks and bubbles) and background, make it easier for the machine vision system to capture flaw features, thereby improving detection accuracy and reducing missed detection and false detection.

[0031] Further, the included angle of the two adjacent mounting plates in the same horizontal plane is not 180°.

[0032] Beneficial effect: the insulation layer of the cable circumferential (front and back, up and down) all areas can bear stress evenly, the upper and lower two sides of the hidden defects under the original 180° angle are fully exposed under the stress, and the detection blind area is further eliminated.

[0033] Additional aspects and advantages of the application will be set forth in part in the following description, will become apparent from the following description, or will be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 It is the overall axonometric view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0035] Figure 2 It is the overall sectional view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0036] Figure 3 It is the cable, mounting plate and guide frame axonometric view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0037] Figure 4 It is the mounting plate and guide frame axonometric view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0038] Figure 5 It is the mounting plate lateral axonometric view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0039] Figure 6 It is the Figure 5 Enlarged view of part A of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0040] Figure 7 It is the protective cover axonometric view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0041] Figure 8 It is the internal structure schematic view of the flow guide cavity of the embodiment of the cable insulation layer defect detection device based on machine vision of the application;

[0042] Figure 9 It is the flow process schematic view of the embodiment of the cable insulation layer defect detection device based on machine vision of the application.

[0043] The reference signs in the drawings of the specification include: 1, cable; 2, protective cover; 201, image sensor; 3, traction roller; 4, mounting plate; 5, guide frame; 501, rotating column; 502, inlet and outlet; 503, air bag ring; 504, cleaning pipe; 6, flow guide cavity; 601, main gear; 602, auxiliary gear. DETAILED DESCRIPTION

[0044] The technical solutions of the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", "third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0046] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0047] The specific embodiments will be described in detail below:

[0048] Embodiment 1, as shown in the accompanying drawings: a cable insulation layer defect detection device based on machine vision, comprising a bending conveying module, an image module and an analysis control module; Figure 9 Specifically, the bending conveying module comprises a traction subunit and a plurality of arc path subunits:

[0049] Specifically, as

[0050] , Figure 1 , Figure 2 , Figure 3 and Figure 4As shown, the arc-shaped path sub-units each include a mounting plate 4, the surface of the mounting plate 4 is symmetrically provided with an inlet and an outlet 502, and the surface of the mounting plate 4 is welded with an arc-shaped guide frame 5, the side surface of the guide frame 5 away from the mounting plate 4 is provided with a guide sliding groove along the length direction of the guide frame 5, and the two ends of the guide sliding groove correspond to the two inlets and outlets 502 respectively. The surface of the mounting plate 4 close to the guide frame 5 is symmetrically provided with an arc-shaped sealing plate.

[0051] In combination Figure 2 As shown, taking two mounting plates 4 as an example, the splicing and forming mode of adjacent arc-shaped path sub-units is described in detail: the planar ends of the left mounting plate 4 and the right mounting plate 4 are close to each other, any one inlet and outlet 502 of the left mounting plate 4 is accurately aligned and coaxial with one inlet and outlet 502 of the right mounting plate 4, and the coaxial two inlets and outlets 502 ensure that the cable 1 can smoothly transition from the guide sliding groove of the left mounting plate 4 to the guide sliding groove of the right mounting plate 4. At the same time, the other inlet and outlet 502 of the left mounting plate 4 and the other inlet and outlet 502 of the right mounting plate 4 are distributed in a staggered manner (there is no overlapping area between the two inlets and outlets 502), and do not form a straight line through. Preferably, the two coaxial inlets and outlets 502 and the two inlets and outlets 502 without overlapping area are in the same horizontal plane / horizontal height.

[0052] Then, the abutting planes of the two mounting plates 4 are locked and fixed by circumferentially uniformly arranged screws, so that the arc-shaped guide frames 5 of the two mounting plates 4 form a continuous reverse arc-shaped track, and constitute a serpentine curved path. When the cable 1 is conveyed along the serpentine curved path, it will produce corresponding bending deformation under the constraint of each arc-shaped guide frame 5, so that uniform stress distribution is formed inside the insulation layer, and the arc peak position becomes the stress concentration core area. The near-surface (0.1-1.0mm) hidden defects (such as internal delamination, micro cracks, local material unevenness) that are difficult to expose in conventional straight path detection produce obvious surface response (such as crack opening increase, delamination slight bulge) under the action of stress, thereby being converted into visual features that can be captured by the image module. At the same time, the serpentine curved path simulates the bending working condition of the cable 1 in actual use, reduces the quality hidden danger of static detection qualification and dynamic use failure, and can more comprehensively cover various defects on the surface and near-surface, thereby providing a key premise for subsequent image acquisition and accurate analysis.

[0053] In addition, a flexible buffer layer (such as a silicone rubber gasket) can be additionally arranged at the abutting connection of the two mounting plates 4, which can not only absorb the slight vibration generated when the cable 1 moves, reduce the wear caused by rigid collision of the structure, but also enhance the sealing performance of the connection, prevent external dust and stray light from invading the detection area, and ensure the stability of the detection environment.

[0054] The traction sub-unit includes a traction roller shaft 3 for pulling the cable 1, and the driving member (such as a three-phase motor) of the traction roller shaft 3 is signal connected with the analysis control module. In combination Figure 1As shown, for the traction work of the cable 1, the traction roller shaft 3 in the conventional technology is adopted, for example, a double roller pair pressing type traction roller shaft 3 (which is composed of a driving roller, a driven roller and a driving motor, the driving motor is drivingly connected with the driving roller through a speed reducer to output stable traction force, the driven roller is arranged in upper and lower correspondence with the driving roller through an elastic pressing assembly, and the pressing gap and the pressing force can be self-adaptively adjusted according to the diameter of the cable 1), so as to drive the cable 1 to smoothly pass through the preset serpentine bending path, lay a foundation for the accurate synchronous collection of the arc-shaped protrusion side image by the image module and the stable identification of the defect features by the analysis and control module.

[0055] In addition, in the present scheme, the pump assembly for providing gas is arranged in the mounting plate 4, the pump assembly is signal connected with the analysis and control module, and the pump assembly is preferably a micro direct current air pump. The inner wall of the inlet and outlet 502 is provided with an air bag ring 503 in communication with the pump assembly, so as to realize the flexible fixing and multi-specification adaptation of the cable 1. When the air pump is started to inflate the air bag ring 503, the air bag ring 503 can be self-adaptively inflated according to the actual diameter of the cable 1 to be detected, and the cable 1 is clamped from the circumferential direction in a flexible manner, which not only reduces the damage to the surface of the insulation layer caused by rigid fixing, but also can adapt to the detection requirements of cables 1 of different diameters. In addition, when the cable 1 slides through the air bag ring 503, the air bag ring 503 forms friction with the surface of the cable 1, which can automatically clean the surface dust, oil stains and other dirt, reduce the pollution of the insulation layer defects or misjudgment as defects, and improve the cleanliness and clarity of the image collected by the image sensor 201.

[0056] Secondly, the inner wall of the guide chute is linearly arrayed along the length direction of the guide frame 5, and a plurality of rotating columns 501 are rotationally connected with the guide frame 5, which reduces the friction resistance and reduces the jamming when the cable 1 moves along the guide chute. The rotating column 501 can be preferably made of stainless steel and polished. The two ends of the rotating column 501 are rotationally connected with the guide frame 5 through bearings. When the cable 1 passes through the guide chute under the driving of the traction roller shaft 3, the surface of the cable 1 contacts with the rotating column 501 to drive the rotating column 501 to roll, so as to convert the sliding friction into rolling friction, greatly reduce the movement resistance, and reduce the scratches or secondary defects on the surface of the cable 1 due to friction, so as to ensure the smooth movement of the cable 1 along the serpentine bending path.

[0057] In the aspect of machine vision, the present scheme is specifically as follows: Figure 7As shown, the image module includes a plurality of protective covers 2 corresponding to the mounting plates 4, one end of each protective cover 2 is hinged to the corresponding mounting plate 4 and corresponds to the corresponding sealing plate, and the free end edge of the protective cover 2 is embedded with a magnetic sealing strip. When the protective cover 2 is closed, it can be quickly adsorbed and fixed with the sealing plate, which not only ensures the sealing of the detection area, but also facilitates quick opening and closing for maintenance or cable 1 threading. The protective cover 2 is screw-connected to a plurality of image sensors 201 connected to the analysis control module on the side close to the guide frame 5, and the image sensors 201 correspond to the arc peak positions of the guide frame 5, and the detection coverage angle of the image sensors 201 is greater than 180°. The image sensor 201 specifically includes a 2D area array camera and a 3D structured light sensor. The 2D area array camera is used to collect high-definition two-dimensional images of the texture, color and plane defects (such as scratches and impurities) on the convex side surface of the cable 1, and the 3D structured light sensor is used to obtain surface three-dimensional topographic data (such as bump height, recess depth, and crack opening). The two work together to realize multi-dimensional imaging coverage.

[0058] Secondly, the sealing plates are provided with arc-shaped diffuse strip-shaped light sources on the side close to each other. The arc-shaped diffuse strip-shaped light sources are adapted to the arc-shaped profile of the sealing plates and are arranged along the length of the sealing plates to achieve circumferential coverage of the cable 1 without dead angles. The arc-shaped diffuse strip-shaped light sources are internally provided with high color rendering index (CRI≥90) patch type LED lamp beads, and the light emitting surface is covered with a frosted diffuser cover to convert the light into uniform and soft diffuse light, reducing the generation of dazzling glare or local strong light reflection. At the same time, the brightness of the arc-shaped diffuse strip-shaped light source supports stepless adjustment within the range of 0-1000 lux, which is synchronized and adapted with the collection frequency of the analysis control module and the image sensor 201, and dynamically adjusts the light intensity according to the reflection characteristics of different materials (such as PVC and cross-linked polyethylene) and the types of defects (such as micro-cracks and near-surface bubbles).

[0059] The analysis control module is used to receive two-dimensional images and three-dimensional topographic data, and then perform image preprocessing. Based on the preprocessed images, the characteristic parameters of each cable 1 convex side region are extracted. Then the extracted characteristic parameters are compared with the pre-set defect feature library, and the machine learning model is combined to distinguish the defect type and normal deformation. Then the severity of each defect is quantitatively analyzed, and whether the insulation layer of the cable 1 to be detected is qualified is determined according to the pre-set quality standard, and an analysis report containing the defect type, location, size parameter and qualification determination result is output. The defect characteristic parameters include 2D dimensional shape, area, gray distribution and 3D dimensional depth, height and slope change.

[0060] For image data processing, the data preprocessing process is as follows:

[0061] For the high-definition two-dimensional image collected by the 2D area array camera, the weighted average method is used to convert it into a single-channel 2D gray image, then the random noise is removed through 3*3 window median filtering, the image is smoothed through Gaussian filtering with σ=0.8, and finally the contrast is improved through histogram equalization algorithm; for the three-dimensional point cloud data obtained by the 3D structured light sensor, the statistical filtering method is used to remove isolated noise points, and the cable axis is taken as the reference to map to the cylindrical coordinate system, and the continuous three-dimensional grid model is generated through Poisson surface reconstruction, and then the least square method is used to fit the normal area surface as the reference surface, and the height difference (ΔZ) between the target area and the reference surface is calculated, that is, the protrusion / recession depth data.

[0062] After the pretreatment is completed, the cooperative fusion process is entered: based on the 2D gray image, the Otsu adaptive threshold segmentation algorithm is used to preliminarily screen the gray abnormal area, and the 2D characteristic parameters such as shape, area and gray distribution standard deviation are extracted, if the 2D characteristic parameters exceed the normal range (for example: area>0.01mm², gray distribution standard deviation>30), it is marked as "2D suspected defect area" and the coordinates are recorded; the corresponding 3D topographic data is extracted for the "2D suspected defect area", if the height difference ΔZ≤±0.02mm (normal insulating layer surface roughness allowable range), it is determined as light and shadow interference or surface texture, and the defect is excluded, if ΔZ exceeds the normal insulating layer surface roughness allowable range, it is determined as a real defect related area, and further 3D characteristic parameters such as depth / height, slope change rate and volume are extracted, at the same time, for the area which is not screened out in the 2D gray image but the 3D data shows that there is continuous height mutation, it is marked as "3D suspected defect area" in reverse and the corresponding 2D image features are traced back to avoid missing detection; finally, the 2D characteristic parameters and 3D characteristic parameters of the verified defect area are fused to form a "2D+3D joint characteristic vector".

[0063] In addition, in actual application, the cooperative logic for different defect types has different focuses: microcracks are located by 2D gray image to determine the direction and length, and 3D data is used to supplement the quantitative crack depth and side wall steepness; internal delamination has no obvious gray difference in 2D gray image, and the suspected area is marked by capturing the slight protrusion (ΔZ=0.03-0.1mm) at the delamination through 3D data, and then the defect is confirmed in combination with the texture disorder characteristics of the area in the 2D gray image; bubbles are screened by 2D gray image to determine the circular / elliptical shape and gray uniformity characteristics, and 3D data is used to quantify the bubble height and volume to distinguish between slight and serious bubbles.

[0064] The specific establishment process of the machine learning model is as follows: first, a combination model of CNN (Convolutional Neural Network) + RF (Random Forest) is adopted, and a lightweight MobileNetV3 architecture is adopted for feature extraction network, taking "2D+3D joint feature vector" as input, automatically extracting deep semantic features through 12 convolutional layers (including depth separable convolution), 4 pooling layers and 2 fully connected layers, and outputting a 256-dimensional feature vector; the classification decision network splices the 256-dimensional feature vector with the 12-dimensional "2D+3D joint feature vector" (including shape, area, gray distribution, depth and slope change rate, etc.) extracted manually, inputs into a random forest classifier containing 100 decision trees, each decision tree selects a split feature by random sampling, and finally outputs the classification result (normal deformation and defect type) through a voting mechanism.

[0065] At the same time, in the data set construction and labeling link of the machine learning model, samples of different specifications (diameter 5-50 mm) and different materials (PVC, cross-linked polyethylene, rubber) of cable insulation layer are collected, covering 6 types of target defects (bubbles, micro-cracks, internal delamination, impurity inclusions, uneven thickness) and normal deformation, and the number of samples of each type is not less than 500, and the total number of samples is ≥3500; the position and type of the defect in the 2D image are labeled by using LabelImg tool, and the three-dimensional boundary and feature parameters of the defect in the 3D point cloud data are labeled by using CloudCompare tool, and a number of technical personnel independently label, the sample with a consistent rate of ≥95% is included in the training set, the inconsistent sample is determined by collective review, and finally the training set, the validation set and the test set are divided according to the ratio of 7:2:1.

[0066] The model training is divided into two stages: in the first stage, the random forest parameters are fixed, only the CNN network is trained, the Adam optimizer is adopted, the cross-entropy loss function is adopted, and the training is performed for 20 epochs until the validation set accuracy is stable at more than 85%; in the second stage, the CNN network and the random forest classifier are trained simultaneously, the Adam optimizer is adjusted to the SGD optimizer, the weighted cross-entropy loss is adopted, and the training is performed for 30 epochs until the validation set accuracy is ≥98% and the test set accuracy is ≥97%, the early stopping strategy and the dropout layer are adopted to inhibit overfitting during the training process, and finally the optimal model parameters are saved.

[0067] During the inference of the machine learning model, the pre-processed "2D+3D joint feature vector" is input first, the deep semantic features are extracted by the CNN network, and the features extracted manually are spliced, then the random forest algorithm is used to output the probability values of each category, the category with the highest probability is selected as the final recognition result, and if the probability of all categories is less than 85%, it is marked as "suspected defect" and the manual review process is triggered.

[0068] The flaw severity is based on the "2D+3D combined feature vector" to make quantitative index and grading standard:

[0069] The bubble takes diameter (D) and convex height (AH) as core indexes, D≤0.5mm and AH≤0.05mm as slight flaw (grade I), 0.5mm<D≤1.0mm or 0.05mm<AH≤0.1mm as moderate flaw (grade II), and D>1.0mm or AH>0.1mm as serious flaw (grade III);

[0070] The micro crack takes length (L) and depth (d) as core indexes, L≤1.0mm and d≤0.1mm as grade I, 1.0mm<L≤3.0mm or 0.1mm<d≤0.3mm as grade II, and L>3.0mm or d>0.3mm as grade III;

[0071] The internal layering takes area (S) and convex height (AH) as core indexes, S≤1.0mm² and AH≤0.05mm as grade I, 1.0mm²<S≤3.0mm² or 0.05mm<AH≤0.1mm as grade II, and S>3.0mm² or AH>0.1mm as grade III;

[0072] The impurity inclusion takes area (S) and embedding depth (d) as core indexes, S≤0.3mm² and d≤0.05mm as grade I, 0.3mm²<S≤1.0mm² or 0.05mm<d≤0.1mm as grade II, and S>1.0mm² or d>0.1mm as grade III; the uneven thickness takes thickness deviation rate (δ=|actual thickness-standard thickness| / standard thickness) as core index, δ≤5% as grade I, 5%<δ≤10% as grade II, and δ>10% as grade III, wherein the standard thickness is the design thickness of the cable insulation layer (for example, the standard thickness of the 10kV cable insulation layer is 3.4mm).

[0073] The quality standard judgment rule is clear:

[0074] 1. No grade III serious flaw, the number of grade II moderate flaw≤2 / m and the distance between any two is≥50mm, the number of grade I slight flaw≤5 / m and≤2 in the same 100mm length, which is determined as qualified and can be put into use;

[0075] 2. There is 1 grade III serious flaw, the number of grade II moderate flaw>2 / m or the distance between any two is<50mm, the number of grade I slight flaw>5 / m or>2 in the same 100mm length, which is determined as needing rework and rectification;

[0076] 3. There are 2 and more grade III serious flaws, grade III and grade II flaws are adjacent (the distance is<30mm) or the maximum deviation rate δ of uneven thickness is>15%, which is determined as unqualified and prohibited from being put into use.

[0077] The quantitative analysis and judgment process is specifically as follows: the analysis control module receives the machine learning model identification result, automatically matches the core quantitative indicators of the flaw type, calculates the specific value, determines the severity level, counts the number, interval and distribution of each type of flaw in 1 meter, outputs the "qualified / rectification / unqualified" judgment result according to the above rules, and finally generates an analysis report containing the type, severity level, position coordinates, core quantitative parameters and eligibility judgment conclusion of each flaw.

[0078] In the actual industrial detection scene, the flaw gap of the cable 1 insulation layer surface, such as micro-cracks and bubble depressions, often has more stubborn stains. These stains are mostly oil stains, long-term accumulated deep dust, or impurities slightly adhered to the insulation layer material. They have strong adhesion and hide in the gap. It is difficult to reach the inside of the gap for thorough cleaning by relying on the surface friction of the air bag ring 503. These residual stubborn stains not only directly block the real form of the flaw, causing the flaw features in the image collected by the image sensor 201 to be unclear, but also may be misjudged by the analysis module as new flaws, or interfere with the accurate extraction of normal flaw features, thereby affecting the accuracy and reliability of the final detection result. Therefore, the scheme specifically combines Figure 5 、 Figure 6 and Figure 8 As shown, a plurality of cleaning pipes 504 are arranged in a ring array on the surface of the air bag ring 503. A plurality of guide cavities 6 corresponding to the rotating columns 501 are arranged in the guide frame 5. A main gear 601 and a secondary gear 602 meshing with the main gear 601 are arranged in each guide cavity 6. The side surfaces of the main gear 601 and the secondary gear 602 are always in contact with the side wall of the guide cavity 6, forming a gear pump structure. The secondary gear 602 is rotationally connected to the inner wall of the guide cavity 6. One end of the rotating column 501 extends into the corresponding guide cavity 6 and is coaxially keyed connected with the main gear 601. The guide cavities 6 are respectively provided with an air inlet and an air outlet.

[0079] The specific implementation process is as follows: the cable 1 moves uniformly along the guide groove of the guide frame 5 under the drive of the traction roller shaft 3, and rolls in contact with the rotating column 501 in the inner wall of the sliding groove, which drives the rotating column 501 to rotate synchronously. Since one end of the rotating column 501 is coaxially keyed connected with the main gear 601 in the flow guide cavity 6, the rotation of the rotating column 501 directly drives the main gear 601 to rotate, and in turn drives the reverse rotation of the secondary gear 602 meshing with the main gear 601. The side surfaces of the main gear 601 and the secondary gear 602 are always in close contact with the side wall of the flow guide cavity 6, forming a closed gear pump structure. With the rotation of the gear meshing, the volume in the flow guide cavity 6 changes periodically, and air is sucked from the outside through the air inlet to generate a high-pressure airflow. After the high-pressure airflow flows out of the air outlet of the flow guide cavity 6, it is delivered to the several cleaning pipes 504 on the surface of the air bag ring 503 through the pre-set air duct inside the mounting plate 4. The cleaning pipes 504 are arranged in a ring array, which can convert the high-pressure airflow into multiple uniform high-speed airflows, and spray them onto the surface of the cable 1 and the defect gap, so as to form a double cleaning effect with the flexible friction of the air bag ring 503: the air bag ring 503 first preliminarily sweeps the surface dust, and the high-speed airflow further penetrates into the defect gap such as micro-cracks and bubble depressions, completely flushing and stripping the stubborn stains such as oil stains, deep dust and the like, and the airflow impact force is gentle and will not damage the surface of the insulation layer, so that the surface of the cable 1 and the defect area are kept clean, which clears the obstacles for the image sensor 201 to accurately collect 2D texture images and 3D topographic data, reduces the stain shielding or interference with the defect feature recognition, and further improves the accuracy and reliability of the analysis module judgment.

[0080] Embodiment 3, which is different from the above-mentioned embodiments, further provides a mounting method of the mounting plate 4 in the adjacent arc-shaped path sub-unit: the included angle of the two mounting plates 4 in the same horizontal plane is not 180°. Specifically, in the embodiment 1, the mounting plate is set to have an included angle of 180°, and the cable 1 only forms unidirectional bending in the front-rear direction, and the stress is mainly concentrated on the insulation layers on the front and rear sides of the cable 1, which makes it difficult for the insulation layers on the upper and lower sides to be effectively and uniformly stressed, and the near-surface latent defects (such as micro-cracks and internal delamination) in such areas are difficult to be fully exposed, and there is a significant detection blind area. In order to realize stress coverage of the cable 1 in the whole circumferential direction and eliminate the detection blind area, the embodiment takes an included angle of 90° as an example for specific description:

[0081] Take two arc-shaped path sub-unit mounting plate 4, the flat end of two mounting plate 4 is close to each other, make two mounting plate 4 any import and export 502 coaxial alignment, then rotate one of the mounting plate 4 around the alignment, the included angle between the two mounting plate 4 is 90°, then through the circumferential uniform distribution of screw lock close two mounting plate 4 of the plane, make two arc-shaped guide frame 5 of their form continuous arc-shaped track in the vertical direction, constitute the multi-directional bending of the serpentine bending path, so that the cable 1 along the path movement, not only by the front and rear direction bending stress, but also in the upper and lower direction produces the corresponding deformation, make the cable 1 circumferential (front and rear, upper and lower) all areas of the insulation layer can bear stress evenly, the original 180° angle under the difficult to cover the upper and lower two sides of the implicit flaw, will be fully exposed under the stress, further eliminate the detection blind area.

[0082] Obviously, the above examples are merely examples for clarity, and not limited to the embodiments. For those of ordinary skill in the art, other different forms of changes or variations can be made on the basis of the above description. Here, it is not necessary and impossible to enumerate all the embodiments. The obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A machine vision-based cable insulation defect detection device, characterized in that, include: A curved conveying module is used to convey the cable (1) in a non-linear manner; An image module is used to acquire optical images of each arcuate convex side of the cable (1) in a non-linear configuration; The optical image coverage angle is greater than 180°; The analysis and control module is used to identify the surface and near-surface defects of the insulation layer of the cable (1) based on optical images and output the identification results; The curved conveyor module includes a traction subunit and several arc-shaped path subunits; The arc-shaped path sub-units are connected end to end in sequence to construct the detection path for the non-linear transport of the cable (1); The traction subunit is used to pull the cable (1) in the arc path subunit, so that the cable (1) is transported along the detection path; Each arc-shaped path subunit includes a mounting plate (4), and each mounting plate (4) has symmetrical inlets and outlets (502) on its surface. Each mounting plate (4) has an arc-shaped guide frame (5) fixedly connected to its surface. Each guide frame (5) has a guide groove along its length on its side surface away from the mounting plate (4), and the two ends of the guide groove correspond to the two inlets and outlets (502) respectively. Each mounting plate (4) has an arc-shaped sealing plate symmetrically arranged on its surface near the guide frame (5). The traction subunit includes a traction roller (3) for traction cable (1), and the drive of the traction roller (3) is connected to the analysis and control module for signal connection; The image module includes several protective covers (2), each of which is hinged to a mounting plate (4) and corresponds to a sealing plate; several image sensors (201) that are signal-connected to the analysis and control module are fixedly connected to the side of the protective cover (2) near the guide frame (5), and the acquisition end of the image sensor (201) is facing the arc peak area of ​​the guide frame (5); Each mounting plate (4) is equipped with a pump assembly for supplying gas. The pump assembly is connected to the analysis and control module. The inner walls of the inlet and outlet (502) are equipped with air bladder rings (503) that communicate with the pump assembly. The inner wall of the guide chute is linearly arrayed with several rotating columns (501) that are rotatably connected to the guide frame (5) along the length of the guide frame (5). The surface of the airbag ring (503) is arranged with several cleaning tubes (504) in a ring array; the guide frame (5) is provided with several guide cavities (6) corresponding to the rotating column (501) one by one. The guide cavity (6) is provided with a main gear (601) and a secondary gear (602) meshing with the main gear (601). The side surfaces of the main gear (601) and the secondary gear (602) are always in contact with the side wall of the guide cavity (6). The secondary gear (602) is rotatably connected to the inner wall of the guide cavity (6). One end of the rotating column (501) extends into the corresponding guide cavity (6) and is coaxially fixedly connected to the main gear (601); the two sides of the guide cavity (6) are respectively opened with an air inlet communicating with the outside and an air outlet communicating with the cleaning tube (504); the sealing plates are provided with an arc-shaped diffused strip light source on the side close to each other; The included angle between two adjacent mounting plates (4) in the same horizontal plane is not 180°.

2. The cable insulation layer defect detection device based on machine vision according to claim 1, characterized in that, The image module is used to acquire two-dimensional images of the texture, color and planar defects of the raised side surface of the cable (1) based on 2D images; and then to acquire three-dimensional morphological data of the raised side surface of the cable (1) based on 3D images.

3. The cable insulation layer defect detection device based on machine vision according to claim 1, characterized in that, The analysis and control module is used to receive two-dimensional images and three-dimensional morphological data, perform image preprocessing, and then extract the defect feature parameters of the arc-shaped protrusion side area of ​​the cable (1) based on the preprocessed image. The extracted feature parameters are compared with the preset defect feature library, and the defect type and normal deformation are distinguished by combining the machine learning model. The severity of each defect is then quantitatively analyzed, and the insulation layer of the cable (1) to be tested is judged to be qualified according to the preset quality standard. An analysis report containing defect type, location, size parameters and qualification judgment results is output. Among them, the defect feature parameters include the shape, area, and grayscale distribution in 2D dimension, as well as the depth, height, and slope variations in 3D dimension.

Images