Stranded wire production quality monitoring method and system based on machine vision

Abstract

The invention relates to the technical field of image processing, and discloses a stranded wire production quality monitoring method and system based on machine vision. The method comprises the following steps: acquiring the surface of a stranded wire at multiple angles through a high-speed camera array to obtain an original image; filtering high reflection and noise to obtain a feature map; performing edge detection, segmenting and identifying defects, and generating a position mapping table; quantitatively calculating defect parameters to form a quality score; performing correlation analysis to adjust process parameters; and constructing a defect-process database to realize quality tracing early warning. According to the invention, full-process automatic control from defect identification, quantitative evaluation to process parameter optimization is realized, the defect-process relational database is constructed, and the tracing and early warning functions of production quality are supported, so that the quality stability and the intelligent level of stranded wire production are improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and particularly to a method and system for monitoring the production quality of stranded wires based on machine vision. Background Art

[0002] A stranded wire is a cable product formed by stranding multiple metal wires according to certain rules, and is widely used in fields such as elevators, bridges, and cranes. Its quality is directly related to structural safety and service life. Traditional quality inspection of stranded wire production mainly relies on manual sampling inspection or simple mechanical flaw detection equipment, usually adopting an off-line sampling detection method, that is, a certain length of sample is taken for inspection after the stranded wire production is completed. The main methods include visual inspection, tensile test, and bending test, etc. Although these methods can detect some obvious surface defects and problems with unqualified mechanical properties, the detection efficiency is low, the coverage rate is limited, and the detection results often lag behind the production process, resulting in the situation that once quality problems are found, a large number of products in the relevant batches may have been produced, causing serious economic losses. With the development of automation technology, some stranded wire production lines have begun to use simple optoelectronic sensors or laser diameter gauges and other equipment for on-line monitoring, but these equipment mainly detect single parameters such as diameter and cannot comprehensively evaluate various quality defects of the stranded wire.

[0003] However, there are still many deficiencies in the prior art in the aspect of monitoring the production quality of stranded wires. First, there is a lack of the ability to conduct all-round real-time monitoring of the surface of the stranded wire, especially it is difficult to capture complete surface detail information of the stranded wire in a moving state; second, the existing detection methods are difficult to effectively handle the high reflectivity problem on the metal surface of the stranded wire, resulting in unstable image quality and affecting the accuracy of defect recognition; third, there is a lack of systematic correlation analysis between defect detection and process parameters, and it is impossible to achieve closed-loop control from defect discovery to process optimization; fourth, the quality monitoring data fails to form effective knowledge accumulation with the production process, lacking the ability to learn from historical experience, resulting in repeated occurrence of similar defect problems; finally, the existing technology is mainly post-event detection, lacking a predictive quality risk warning mechanism and being difficult to achieve forward-looking control of the production quality of stranded wires. These deficiencies seriously limit the quality stability and intelligent level of stranded wire production and urgently require innovative technical solutions to solve them. Summary of the Invention

[0004] This application provides a method and system for monitoring the production quality of stranded wires based on machine vision, which is used to realize full-process automatic control from defect recognition, quantitative evaluation to process parameter optimization, and construct a defect-process relationship database to support the traceability and warning functions of production quality, thereby improving the quality stability and intelligent level of stranded wire production.

[0005] In a first aspect, the present application provides a method for monitoring the production quality of twisted wires based on machine vision. The method for monitoring the production quality of twisted wires based on machine vision includes: collecting the surface of the moving twisted wires from multiple angles through a high-speed industrial camera array to obtain the original image data of the twisted wires; filtering the highly reflective areas and random noises on the surface of the twisted wires according to the original image data of the twisted wires to obtain a surface feature map of the twisted wires; using the surface feature map of the twisted wires to identify broken wires, loose strands, deformation, and impurity attachment through edge detection and region segmentation to obtain a mapping table of the defect positions of the twisted wires; based on the mapping table of the defect positions of the twisted wires, quantitatively calculating the defect area, depth, and distribution density to generate a quality score report of the twisted wires; based on the quality score report of the twisted wires, performing correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine to form a corrected value of the stranding process parameters; according to the quality score report of the twisted wires and the corrected value of the stranding process parameters, constructing a database of the relationship between twisted wire defects and processes to achieve traceability and early warning of the production quality of twisted wires.

[0006] In a second aspect, the present application provides a system for monitoring the production quality of twisted wires based on machine vision. The system for monitoring the production quality of twisted wires based on machine vision includes: A collection module for collecting the surface of the moving twisted wires from multiple angles through a high-speed industrial camera array to obtain the original image data of the twisted wires; A filtering module for filtering the highly reflective areas and random noises on the surface of the twisted wires according to the original image data of the twisted wires to obtain a surface feature map of the twisted wires; An identification module for using the surface feature map of the twisted wires to identify broken wires, loose strands, deformation, and impurity attachment through edge detection and region segmentation to obtain a mapping table of the defect positions of the twisted wires; A quantization module for quantitatively calculating the defect area, depth, and distribution density based on the mapping table of the defect positions of the twisted wires to generate a quality score report of the twisted wires; An analysis module for performing correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine based on the quality score report of the twisted wires to form a corrected value of the stranding process parameters; A construction module for constructing a database of the relationship between twisted wire defects and processes according to the quality score report of the twisted wires and the corrected value of the stranding process parameters to achieve traceability and early warning of the production quality of twisted wires.

[0007] In a third aspect of the present invention, a computer device is provided, including: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned method for monitoring the production quality of twisted wires based on machine vision.

[0008] The fourth aspect of the present invention provides a computer-readable storage medium storing instructions which, when run on a computer, cause the computer to execute the above-mentioned method for monitoring the quality of twisted wire production based on machine vision.

[0009] In the technical solution provided by this application, the surface of the moving twisted wire is collected from multiple angles by a high-speed industrial camera array, realizing 360° all-round monitoring of the surface of the twisted wire, overcoming the blind area problem of traditional single-view detection, and significantly improving the defect detection rate. At the same time, aiming at the high-reflectivity characteristics of the metal surface of the twisted wire, the regional adaptive brightness mapping technology and the polar coordinate transformation method are used for light compensation, effectively solving the problem of reflective interference on the metal surface and improving the stability and reliability of the image quality. The application of the anisotropic diffusion filter effectively suppresses random noise while retaining the edge of the surface structure of the twisted wire, laying a foundation for subsequent accurate defect recognition. Through the multi-band wavelet decomposition and reconstruction technology, the selective enhancement of different scale features on the surface of the twisted wire is realized, improving the visibility of tiny defects. The refined application of edge detection and region segmentation algorithms realizes the accurate recognition of various types of defects such as broken wires, loose strands, deformation, and impurity adhesion. The generation of the defect position mapping table provides structured data support for subsequent quantitative analysis. Based on the quantitative calculation of defect area, depth, and distribution density, an objective twisted wire quality scoring system is formed, providing a data basis for quality assessment. By establishing the mapping relationship between defect characteristics and production process parameters through correlation analysis technology, a decision-making basis is provided for the precise adjustment of process parameters. Finally, the constructed twisted wire defect-process relationship database realizes the traceability and early warning functions of quality problems, transforming passive detection into active prevention. The innovation of the solution of the present invention lies in fully considering the specific application requirements of artificial intelligence algorithms in the field of twisted wire quality monitoring, such as the optimization of image enhancement algorithms for high-speed moving metal surfaces, the design of defect recognition models adapted to the unique structure of twisted wires, and the optimization algorithms of process parameters based on historical data. The organic combination of these algorithm features enables the solution to meet the requirements of high-speed, high-reflectivity, and high-precision detection in twisted wire production, and continuously optimizes production parameters in a data-driven manner to achieve quality closed-loop control. Compared with traditional methods, the present invention not only improves the accuracy and comprehensiveness of defect detection, but also establishes a causal relationship between defects and process parameters, transforming quality control from post-inspection to pre-prevention and process control, significantly reducing the production defect rate and material waste, and improving production efficiency and product consistency. Brief Description of the Drawings

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0011] Figure 1 It is a schematic diagram of an embodiment of the method for monitoring the production quality of twisted wires based on machine vision in the embodiments of the present application; Figure 2 It is a schematic diagram of an embodiment of the system for monitoring the production quality of twisted wires based on machine vision in the embodiments of the present application; Figure 3 It is a schematic block diagram of the structure of a computer device in the embodiments of the present invention. Detailed implementation manners

[0012] The embodiments of the present application provide a method and system for monitoring the production quality of twisted wires based on machine vision. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the following describes the specific process of the embodiments of the present application. Please refer to Figure 1 , an embodiment of the method for monitoring the production quality of twisted wires based on machine vision in the embodiments of the present application includes: Step S101: Multiangularly collect the surface of the moving twisted wire through a high-speed industrial camera array to obtain the original image data of the twisted wire; Step S102: Filter the highly reflective areas and random noises on the surface of the twisted wire according to the original image data of the twisted wire to obtain the surface feature map of the twisted wire; Step S103: Use the surface feature map of the twisted wire to identify broken wires, loose strands, deformation, and impurity attachment through edge detection and region segmentation, and obtain the mapping table of the defect positions of the twisted wire; Step S104: Based on the mapping table of the defect positions of the twisted wire, quantitatively calculate the defect area, depth, and distribution density, and generate a quality scoring report for the twisted wire; Step S105: Based on the stranding quality scoring report, perform correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine to form the corrected values of the stranding process parameters; Step S106: According to the stranding quality scoring report and the corrected values of the stranding process parameters, construct a stranding defect-process relationship database to realize the traceability and early warning of the stranding production quality.

[0014] It can be understood that the execution entity of this application can be a stranding production quality monitoring system based on machine vision, or a terminal or a server. Specifically, it is not limited here. In the embodiments of this application, the server is taken as an example of the execution entity for illustration.

[0015] Specifically, a high-speed industrial camera array is used to collect the surface of the moving stranded wire from multiple angles to obtain the original image data of the stranded wire. During this process, multiple high-speed industrial cameras are installed at key positions on the stranded wire production line to form a circular array layout, ensuring a 360° omnidirectional shooting of the stranded wire surface. These cameras are equipped with high-resolution sensors (such as 4096×3072 pixels), high frame rates (above 120fps), and use global shutter technology to avoid image smear during high-speed movement. At the same time, a combined lighting technology is adopted, including a 45° direct light source for main lighting, a low-angle side light source to enhance the surface texture features of the stranded wire, and a backlight for contour outlining. To ensure the synchronization of image acquisition and the movement of the stranded wire, the system uses a high-precision clock synchronization control unit to dynamically adjust the camera exposure time and trigger interval according to the speed of the stranded wire. When the stranded wire runs at a speed of 8m / min, the trigger interval of the camera will be set to about 100 milliseconds, ensuring a 30% overlap area between adjacent images for subsequent image stitching. After obtaining the original image data, the system filters out the highly reflective areas and random noise on the surface of the stranded wire to obtain the surface feature map of the stranded wire. The processing flow first divides the different reflective areas on the surface of the stranded wire through regional adaptive brightness mapping technology and establishes a reflective intensity distribution map. This technology divides the image into multiple sub-blocks (such as 64×64 pixels), calculates the brightness mean and variance of each sub-block, and marks the areas with brightness values exceeding the threshold (such as relative average brightness higher than 150%) as highly reflective areas. Subsequently, the polar coordinate transformation method is used to perform light compensation on these high-light areas. This method maps the elliptical high-light area formed by the reflection of a point light source into a rectangular area through polar coordinate mapping, facilitating the application of directional filtering and reducing the interference of surface specular reflection. Then, an anisotropic diffusion filter is used for processing. This filter retains the structural edges of the stranded wire surface while smoothing random noise. The filtering operation is adaptively adjusted by calculating the pixel gradient direction, retaining details in the edge direction and enhancing the smoothing effect in the non-edge direction. Through multi-band wavelet decomposition and non-local mean filtering, the different scale features on the surface of the stranded wire are enhanced and fused to generate the surface feature map of the stranded wire. Using the surface feature map of the stranded wire obtained by the above processing, the system identifies four types of typical defects, namely broken wires, loose strands, deformation, and impurity attachment, through edge detection and region segmentation, and obtains the mapping table of the stranded wire defect positions. The Sobel operator is used to calculate the image gradient and extract the edge features of the stranded wire surface. The Sobel operator calculates the gradients in the horizontal and vertical directions by applying two 3×3 convolution kernels to the image respectively, and then synthesizes the total gradient magnitude and direction. The double-threshold method is applied to the obtained gradient image for edge connection. The high threshold (such as gradient value > 100) is used to determine the strong edge pixels, and the low threshold (such as gradient value > 50) is used to connect adjacent strong edges. Then, the marker-watershed algorithm is used for region segmentation. This algorithm regards the edge topology map as a topographic map and divides the image region by simulating the process of rising water level.After obtaining the partition marking map, the system identifies different types of defects based on geometric features: broken wire defects are detected by the interruption of straight line segments in the stranded wire through the Hough line transform, loose strand defects are identified by measuring the abnormal spacing between adjacent strands (more than 1.5 times the normal spacing), deformation features are discriminated based on the disproportion of the regional shape ratio (such as the aspect ratio deviation exceeding 20%), and impurity adhesion is determined through texture contrast and regional connectivity analysis.

[0016] Based on the defect location mapping table, the system quantifies and calculates the defect area, depth, and distribution density, and generates a stranded wire quality scoring report. The area calculation is performed by counting the pixels of the defect area marked in the mapping table and converting it into the actual area (square millimeters) in combination with the camera resolution and the physical size relationship. The depth calculation is based on the stereo reconstruction technology. Using the images of the same defect area captured by cameras at different angles, the defect depth is calculated through the principle of triangulation. In specific implementation, the corresponding point matching adopts an algorithm based on phase correlation to reduce the matching ambiguity caused by the repeatability of the stranded wire surface texture. For the distribution density calculation, the system counts the occurrence frequency of various types of defects on the stranded wire per unit length (such as 1 meter) and generates a defect density distribution curve along the length direction of the stranded wire. According to industry standards and empirical thresholds, various types of defects are divided into three levels: minor (such as area < 1 mm², depth < 0.2 mm), general (such as area 1 - 3 mm², depth 0.2 - 0.5 mm), and severe (such as area > 3 mm², depth > 0.5 mm), and the comprehensive quality score is obtained through weighted calculation.

[0017] Based on the quality scoring report, the system conducts a correlation analysis and adjustment of the traction speed, stranding tension, and pre-twist angle of the stranding machine to form the corrected values of process parameters. First, a defect-process impact matrix is established to quantify the correlation strength between various defects and process parameters. For example, the wire breakage defect has a high positive correlation with the stranding tension (correlation coefficient > 0.7), while it only has a moderate correlation with the traction speed (correlation coefficient about 0.4). The influence coefficient of tension on wire breakage is determined through regression analysis, and a mathematical model is established to predict the wire breakage probability under different tensions. For the loose strand defect, the hierarchical clustering method is used to analyze its relationship pattern with the traction speed, identify the inflection point in the speed-loose strand curve, and determine the optimal speed range. In addition, combined with the relationship between the defect depth distribution data and the pre-twist angle, the partial least squares calculation is used to determine the optimal pre-twist angle to minimize the defect occurrence rate while maintaining production efficiency. Based on the above analysis results and historical data, the system constructs a database of stranding defect-process relationships to achieve quality traceability and early warning for stranding production. Specifically, through time-series correlation processing, the defect data and corresponding process parameters of each batch of stranding products are recorded, and through multi-dimensional factor analysis, the root causes of defects are identified. For example, during a certain production process, it is observed that the loose strand defects are concentrated in a specific section of the stranded wire. By querying the process parameter records, it is found that the traction speed corresponding to this section has suddenly increased by 15%, and the tension adjustment is not timely, resulting in uneven tension between wire strands, thus triggering the loose strand defect. The system records this pattern in the defect feature dictionary and issues an early warning when a similar process parameter change trend is detected in future production. Through continuous learning and data accumulation, the accuracy of the early warning mechanism is continuously improved, significantly reducing the occurrence of defects and improving the production efficiency and product quality of stranding.

[0018] In the embodiments of the present application, a high-speed industrial camera array is used to collect the surface of the moving stranded wire from multiple angles, achieving 360° omnidirectional monitoring of the surface of the stranded wire, overcoming the blind spot problem of traditional single-view detection, and significantly improving the defect detection rate. At the same time, aiming at the high-reflectivity characteristics of the metal surface of the stranded wire, a region adaptive brightness mapping technology and a polar coordinate transformation method are used for light compensation, effectively solving the problem of reflective interference on the metal surface and improving the stability and reliability of the image quality. The application of the anisotropic diffusion filter effectively suppresses random noise while retaining the structural edges of the stranded wire surface, laying a foundation for subsequent accurate defect recognition. Through multi-band wavelet decomposition and reconstruction technology, the selective enhancement of different scale features on the surface of the stranded wire is carried out, improving the visibility of micro defects. The refined application of edge detection and region segmentation algorithms realizes the accurate recognition of various types of defects such as broken wires, loose strands, deformation, and impurity attachment. The generation of the defect position mapping table provides structured data support for subsequent quantitative analysis. Based on the quantitative calculation of defect area, depth, and distribution density, an objective stranded wire quality scoring system is formed, providing a data basis for quality assessment. By establishing a mapping relationship between defect features and production process parameters through correlation analysis technology, a decision-making basis is provided for the precise adjustment of process parameters. Finally, the constructed stranded wire defect-process relationship database realizes the traceability and early warning functions of quality problems, transforming passive detection into active prevention. The innovation of the solution of the present invention lies in fully considering the specific application requirements of artificial intelligence algorithms in the field of stranded wire quality monitoring, such as the optimization of image enhancement algorithms for high-speed moving metal surfaces, the design of defect recognition models adapted to the unique structure of stranded wires, and the optimization algorithms of process parameters based on historical data. The organic combination of these algorithm features enables the solution to meet the high-speed, high-reflectivity, and high-precision detection requirements of stranded wire production, and continuously optimizes production parameters in a data-driven manner to achieve quality closed-loop control. Compared with traditional methods, the present invention not only improves the accuracy and comprehensiveness of defect detection, but also establishes a causal relationship between defects and process parameters, transforming quality control from post-inspection to pre-prevention and process control, greatly reducing production defect rates and material waste, and improving production efficiency and product consistency.

[0019] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The surface of the stranded wire is photographed omnidirectionally by a high-speed industrial camera installed at a key position on the stranded wire production line to obtain original multi-angle images of the stranded wire; The surface of the stranded wire is combinedly illuminated by configuring a 45° direct light source, a low-angle side light source, and a backlight source to obtain an image with enhanced surface features of the stranded wire; The acquisition of the original multi-angle images of the stranded wire and the speed of the stranded wire production line are synchronously processed by a high-precision clock signal synchronization control unit to obtain clear moving images of the stranded wire; The environmental adaptability of the parameters of the high-speed industrial camera is adjusted by a temperature and humidity compensation device, and the multi-angle original image of the stranded wire, the enhanced image of the stranded wire surface features, and the moving image of the stranded wire are transmitted to an image processing unit through a gigabit network to form original image data of the stranded wire.

[0020] Specifically, a high-speed industrial camera is installed at key positions on the stranded wire production line to take an all-round picture of the stranded wire surface, and a multi-angle original image of the stranded wire is obtained. A high-speed industrial camera refers to an industrial-grade camera with a frame rate exceeding 100 frames per second, having characteristics such as high resolution, low noise, and anti-vibration, and is suitable for use in the high-speed moving environment of the stranded wire. In practical applications, usually 4-8 high-speed cameras are evenly arranged in the circumferential direction of the stranded wire to form an annular array structure. Each camera is responsible for collecting a field of view range of about 45°-90° on the stranded wire surface to ensure that the entire circumference of 360° of the stranded wire surface is covered without dead angles. There is a 10%-15% overlapping area between the field of view areas of adjacent cameras, which is convenient for subsequent image stitching. The camera uses a global shutter technology to eliminate image blurring during the high-speed movement of the stranded wire and is equipped with a high-resolution sensor to ensure the detection ability for tiny defects. The multi-angle original image of the stranded wire obtained through this layout contains basic information such as the geometric shape, color, and texture of the stranded wire surface. After obtaining the multi-angle original image of the stranded wire, the stranded wire surface is illuminated in combination by configuring a 45° direct light source, a low-angle side light source, and a backlight to obtain an enhanced image of the stranded wire surface features. The 45° direct light source refers to a light source that is incident at an angle of 45° relative to the normal direction of the stranded wire surface, mainly providing basic illumination to highlight the color and basic contour of the stranded wire surface; the low-angle side light source refers to a light source that is nearly parallel to the stranded wire surface (about 10°-15° angle), and its function is to enhance the shadow effect of tiny structures such as unevenness on the stranded wire surface, which helps to detect defects parallel to the surface such as loose strands and fine scratches; the backlight is placed behind the stranded wire, and the silhouette effect is used to highlight the contour of the stranded wire, which is convenient for accurately measuring the diameter of the stranded wire and detecting contour abnormalities such as broken wires and burrs. The three light sources cooperate with each other to enhance specifically for different types of defects. In the enhanced image of the stranded wire surface features collected, the broken wire defect shows an obvious contour interruption under the illumination of the backlight, the loose strand defect shows irregular shadows under the illumination of the side light source, the deformation defect shows contour abnormalities under the combined action of multi-angle light sources, and impurities show color and texture abnormalities under the illumination of the 45° direct light.

[0021] The high-precision clock signal synchronization control unit synchronizes the acquisition of the multi-angle original images of the stranded wire with the speed of the stranded wire production line to obtain clear moving images of the stranded wire. The high-precision clock signal synchronization control unit is a highly stable clock generator based on a crystal oscillator and a PLL circuit, with nanosecond-level precision and extremely low jitter. This unit calculates the trigger delay and exposure time according to the speed of the stranded wire production line measured by the encoder in real time, and accurately triggers the corresponding camera to expose when the stranded wire faces the same detection area. When the speed of the stranded wire changes, the synchronization control unit will adjust the trigger timing of the camera in real time to ensure that the image acquisition is synchronized with the movement of the stranded wire. For the stranded wire moving at high speed, the motion compensation function of the camera is usually also activated to make the direction of the electronic reading of the camera sensor consistent with the movement direction of the stranded wire during the exposure process, further eliminating motion blur. In the processed moving images of the stranded wire, the edges of the stranded wire are clear and the texture details are well preserved, providing high-quality image data for subsequent accurate defect detection. The environmental adaptability of the parameters of the high-speed industrial camera is adjusted by the temperature and humidity compensation device, and the multi-angle original images of the stranded wire, the enhanced images of the surface features of the stranded wire, and the moving images of the stranded wire are transmitted to the image processing unit through a gigabit network to form the original image data of the stranded wire. The temperature and humidity compensation device monitors the environmental parameters in real time through temperature and humidity sensors, and dynamically adjusts the imaging parameters of the camera such as the aperture size, gain value, and exposure time according to the preset parameter correction algorithm. When the environmental temperature rises, the dark current of the camera increases, resulting in more noise. The compensation device will appropriately reduce the camera gain and adjust the white balance parameters to adapt to the change of the light source spectrum at the same time. When the humidity changes, the compensation device will adjust the aperture to deal with the possible problem of lens fogging. These compensation measures ensure that the contrast, brightness, and color of the stranded wire images remain consistent under different environmental conditions, avoiding interference from environmental factors on defect detection. The adjusted images are transmitted to the central image processing unit through an industrial-grade gigabit Ethernet. This unit sorts and integrates the images taken by different cameras according to the physical position and time tags of the stranded wire, eliminates the image overlap between multiple cameras, and performs preliminary image preprocessing such as denoising, correction, and normalization, finally forming the standardized original image data of the stranded wire.

[0022] For example, a wire stranding production line produces steel wire strands with a diameter of 2.5 mm at a production speed of 10 meters per minute. Six high-speed industrial cameras are configured on this production line, evenly distributed in a ring, with a field of view overlap rate of 15%. The field of view of each camera covers approximately 80 mm of the stranded wire segment, and the frame rate is set at 120 fps. The system first acquires the original images and records the basic features of the stranded wire surface. Immediately afterwards, the combined lighting system is activated. The 45° direct light source provides the main lighting, the low-angle side light source strengthens the surface texture from a nearly parallel angle, and the back light source outlines the profile of the stranded wire. Based on the 10-meter-per-minute stranded wire speed, the clock synchronization control unit calculates that the camera trigger interval is 0.34 seconds and the exposure time is 1 / 1000 seconds, ensuring sufficient overlap between adjacent images and no motion blur. When the factory temperature changes, the temperature and humidity compensation device automatically adjusts the camera parameters to maintain stable image quality. The final images are transmitted to the image processing unit via a gigabit network, where spatial stitching and temporal alignment are performed to form a dataset of the stranded wire surface.

[0023] In a specific embodiment, the process of performing step S102 may specifically include the following steps: For the original image data of the stranded wire, different regions with different degrees of reflectivity on the metal surface of the stranded wire are segmented through regional adaptive brightness mapping technology, and a reflectivity intensity distribution map of the stranded wire is established; For the highlight regions in the reflectivity intensity distribution map of the stranded wire, light compensation is performed through the polar coordinate transformation method to reduce the interference of specular reflection on the surface of the stranded wire and form a light balance map of the stranded wire; The light balance map of the stranded wire is processed through an anisotropic diffusion filter to retain the edge of the surface structure of the stranded wire while suppressing random noise, and an edge-preserving map of the stranded wire is obtained; The edge-preserving map of the stranded wire is subjected to multi-band wavelet decomposition and reconstruction to selectively enhance different scale features on the surface of the stranded wire, and a multi-scale feature set of the stranded wire is generated; The multi-scale feature set of the stranded wire is subjected to structural similarity fusion through non-local mean filtering to retain the texture information on the surface of the stranded wire while eliminating residual noise, and a surface feature map of the stranded wire is obtained.

[0024] Specifically, through the regional adaptive brightness mapping technology, different reflective regions on the surface of the stranded wire metal are segmented, and a distribution map of the reflective intensity of the stranded wire is established. The regional adaptive brightness mapping technology refers to the process of dividing the entire image into multiple sub-blocks, calculating the brightness statistical characteristics of each sub-block separately, and then locally adjusting the pixel values based on these characteristics. In the image processing of the stranded wire, first, the original image is divided into small blocks of 64×64 pixels, and the average brightness value and standard deviation of each block are calculated to form a brightness distribution matrix. Then, a threshold is set for each block, usually the average brightness plus 1.5 times the standard deviation. The regions exceeding this threshold are marked as high-reflective regions. In this way, the surface of the stranded wire is divided into high-reflective regions, medium-reflective regions, and low-reflective regions, generating a distribution map of the reflective intensity of the stranded wire. This distribution map records the reflective characteristics of the surface of the stranded wire with different marker values, providing a basis for subsequent elimination of reflective interference. The highlight regions in the distribution map of the reflective intensity of the stranded wire are subjected to illumination compensation through the polar coordinate transformation method to reduce the specular reflection interference on the surface of the stranded wire and form an illumination equilibrium map of the stranded wire. The polar coordinate transformation method is a technology that converts the image in the rectangular coordinate system to the polar coordinate system for processing and then converts it back to the rectangular coordinate system. Specifically, first, the center point of the highlight region is identified, and then the polar coordinate values of each pixel point in the highlight region relative to the center point are calculated, including the distance to the center point and the angle with the horizontal axis. In the polar coordinate space, the highlight region usually appears as an approximate ellipse, and it can be stretched into a more regular shape through coordinate transformation for convenient application of directional filtering. During specific processing, different intensities of brightness attenuation are applied to the pixels in different directions, so that the brightness of the highlight region smoothly transitions to the surrounding area. After processing, the polar coordinates are remapped back to the rectangular coordinates to obtain the image with balanced illumination. This processing weakens the strong specular reflection interference on the metal surface while retaining the surface texture information contained in the reflection pattern.

[0025] The illumination balance map of the stranded wire is processed by an anisotropic diffusion filter to retain the structural edge of the stranded wire surface while suppressing random noise, and obtain the stranded wire edge preservation map. Anisotropic diffusion filtering is a nonlinear filtering technology that dynamically adjusts the filtering strength and direction according to the local structural features of the image. In the stranded wire image processing, the filter first calculates the gradient value and direction of each pixel of the image. The area with a large gradient value usually corresponds to the structural edge, and the gradient direction is perpendicular to the edge direction. Then, a diffusion tensor is constructed based on the gradient information, allowing strong diffusion (smoothing) in the edge direction and suppressing diffusion in the direction perpendicular to the edge, thereby retaining edge details. This filter is applied iteratively, usually 10-20 times, and the time step of each iteration is controlled between 0.1-0.25 to prevent over-smoothing. In the processed stranded wire edge preservation map, random noise is significantly suppressed, while the structural edges of the stranded wire surface (such as the spiral texture of the stranded wire, the boundary between strands, etc.) are completely preserved, laying the foundation for subsequent feature extraction and defect detection. The edge-preserving image of the stranded wire is decomposed and reconstructed by multi-band wavelet, and the different scale features of the stranded wire surface are selectively enhanced to generate a multi-scale feature set of the stranded wire. Multi-band wavelet decomposition is a technique for decomposing an image into different frequency components, and is often used to extract multi-scale features. In this method, a wavelet basis function (such as Daubechies wavelet or Gabor wavelet) suitable for the texture features of the stranded wire is first selected, and then the edge-preserving image of the stranded wire is decomposed in multiple levels, usually 3-5 levels. After decomposition, a series of sub-band images are obtained, including low-frequency approximate sub-bands and high-frequency detail sub-bands. The low-frequency sub-band contains the overall structural information of the image, while the high-frequency sub-band contains detail information such as edges and textures. For the stranded wire image, different types of defects appear differently at different scales: wire breaks are usually obvious in low-frequency sub-bands, while surface fine scratches are more obvious in high-frequency sub-bands. Therefore, different enhancement coefficients are applied to different sub-bands, such as appropriately enhancing the mid-frequency sub-band (corresponding to the spiral texture of the stranded wire) and appropriately suppressing or selectively enhancing the high-frequency sub-band (corresponding to noise and minor defects). The enhanced sub-bands are reconstructed by inverse wavelet transform to form a multi-scale feature set of the stranded wire, which contains the structure and texture information of the stranded wire surface at different scales.

[0026] The structural similarity fusion of the multi-scale feature set of stranded wires is performed through non-local mean filtering, which retains the surface texture information of the stranded wires while eliminating residual noise, resulting in a surface feature map of the stranded wires. Non-local mean filtering is a noise reduction method based on the weighted average of similar regions in an image. It uses the similarity between non-local regions in the image for filtering. In the image processing of stranded wires, a search window (usually 21×21 pixels) and a comparison window (usually 7×7 pixels) are first defined in each image of the multi-scale feature set of stranded wires. For each pixel to be processed, similar regions to its surrounding comparison window are found within the search window, and the similarity weight is calculated. The similarity is usually based on the Euclidean distance between the two comparison windows. The smaller the distance, the higher the similarity, and a greater weight is assigned. Then, the weighted average of these similar regions is used to replace the original pixel value. This method can effectively remove random noise while retaining the image structure and texture, especially suitable for processing objects with regular textures such as stranded wires. Since the surface of stranded wires usually has repetitive spiral patterns, non-local mean filtering can find a large number of similar texture regions for averaging, effectively eliminating noise while retaining texture details. The resulting surface feature map of the stranded wires has clear structural edges, uniform illumination distribution, and a low noise level, providing a high-quality feature image for subsequent defect recognition.

[0027] For example, the original image captured by a high-speed industrial camera contains obvious uneven reflective areas. Through the regional adaptive brightness mapping technology, the system first divides the image into 236 64×64 pixel sub-blocks, calculates the average brightness value of each sub-block, and finds that the brightness values of 57 of them exceed the threshold of 200 (8-bit grayscale image, maximum value 255), and these areas are marked as high-reflective areas. The polar coordinate transformation method is applied to these high-reflective areas, and each elliptical highlight area is converted to polar coordinate space. The brightness reduction transformation is applied in the radial direction, so that the brightness value of the brightest area in the center gradually transitions from 240 to about 165 in the surrounding area. After converting back to rectangular coordinates, the illumination balance map is obtained. Next, anisotropic diffusion filtering is applied to the illumination balance map, and the number of iterations is set to 15 times, the time step is 0.15, and edge preservation filtering is performed. In the obtained strand edge preservation map, random noise is suppressed, while the spiral texture and strand outline on the strand surface are retained. Then, the edge-preserving image is decomposed by 4-level wavelet, and a low-frequency approximate subband and three groups of directional detail subbands are obtained. The intermediate frequency part (corresponding to the gap between the strands of the twisted wire) in the detail subband is enhanced with an enhancement factor of 1.5, and the high frequency part (mainly containing noise) is suppressed with a suppression factor of 0.7. After reconstruction, a multi-scale feature set is obtained. Non-local mean filtering is applied to the multi-scale feature set, with the search window set to 21×21 pixels, the comparison window to 7×7 pixels, and the noise variance estimation value to 15. In the surface feature map of the twisted wire obtained after processing, the defect features such as broken wire, loose strands, deformation and impurity adhesion of the twisted wire are clearly retained, while the background noise and illumination inhomogeneity are effectively suppressed.

[0028] In a specific embodiment, the process of executing step S103 may specifically include the following steps: The Sobel operator is used to calculate the gradient of the stranded wire surface feature map, extract the stranded wire surface edge features, and generate a stranded wire edge enhancement map; The edge enhancement map of the twisted wire is connected by the double threshold method to construct the complete outline of the twisted wire and form the edge topology map of the twisted wire; According to the strand edge topology map, the marker-watershed algorithm is used to perform region segmentation, and different characteristic areas on the strand surface are divided to obtain a strand partition marker map; The broken wire features in the stranded wire partition marking diagram are identified by line segment detection, the broken wire position information is marked, and the broken wire defect data is generated; The loose strand features in the stranded wire partition marking diagram are identified by spacing measurement, the loose strand position information of the stranded wire is marked, and the loose strand defect data is generated; The deformation features and impurity attachment features in the stranded wire partition marking diagram are identified through morphological analysis, the deformation and impurity attachment position information of the stranded wire is marked, and the broken wire defect data, loose strand defect data, deformation and impurity attachment position information are integrated into a stranded wire defect position mapping table.

[0029] Specifically, the surface feature map of the twisted pair is subjected to gradient calculation through the Sobel operator to extract the edge features of the twisted pair surface, and a twisted pair edge enhancement map is generated. The Sobel operator is a classic edge detection operator that detects edges by calculating the gradient values of image pixels in the horizontal and vertical directions. In the image processing of the twisted pair, the calculation process of the Sobel operator can be expressed by the following formula:

[0030]

[0031]

[0032] Among them, represents the surface feature map of the twisted pair, and are the Sobel convolution kernels in the horizontal and vertical directions respectively, and are the gradient maps in the horizontal and vertical directions respectively, is the comprehensive gradient magnitude map, is the gradient direction map, and * represents the convolution operation. In this way, the edge features of the stranded wire surface are extracted to form an edge enhancement map, in which the pixel values at the edge are higher and the pixel values at the non-edge are lower. In practical applications, when processing a stranded wire surface feature map, the image is first converted to a grayscale image, and then a 3×3 or 5×5 Sobel convolution kernel is applied to calculate the gradient value of each pixel. For example, for the spiral texture on the stranded wire surface, the horizontal and vertical gradients reflect the rate of change of the texture in two directions respectively. By calculating the comprehensive gradient amplitude, the edge of the spiral texture can be highlighted. The stranded wire edge enhancement map is edge-connected by the double threshold method to construct the complete outline of the stranded wire and form a stranded wire edge topology map. The double threshold method is an edge connection technique that uses two thresholds (high threshold and low threshold) to determine whether a pixel belongs to an edge. The specific operation is to first use a high threshold to filter out strong edge points, and then start from these strong edge points to connect those points that exceed the low threshold and are adjacent to the strong edge points to finally form an edge. In the image processing of stranded wires, the high threshold is usually set to the 70%-80% quantile of the gradient histogram, and the low threshold is set to 40%-50% of the high threshold. After this processing, the stranded wire edge topology map contains the complete contour information of the stranded wire surface, including structural features such as strand boundaries and spiral patterns. Even if the gradient value is weak in some areas (such as areas with uneven illumination), as long as it is connected to a strong edge, it can be correctly retained, thereby ensuring the continuity and integrity of the stranded wire contour. According to the stranded wire edge topology map, the marker-watershed algorithm is used to perform regional segmentation, divide the stranded wire surface into different feature areas, and obtain the stranded wire partition marker map. The marker-watershed algorithm is an image segmentation method based on mathematical morphology. It regards the image as a topographic map, in which the pixel gray value represents the height, and the image is segmented by simulating the water level rise process. In the stranded wire edge topology map, the edge line is marked as a "watershed" (i.e., the region boundary), and then different "seed points" are marked in the non-edge area. Next, starting from these seed points, the region is “watered” and expanded in the order of pixel gradient values from low to high. When the water in different regions is about to merge, a watershed line is established at the merge point, and the entire image is finally segmented into multiple non-overlapping regions. In the segmentation of twisted wire images, seed points are usually selected as representative points of different structural features of twisted wires (such as different strands, spiral pattern gaps, etc.). In this way, the surface of the twisted wire is divided into multiple regions with different characteristics, forming a twisted wire partition labeling map, and each region is marked with a unique label value.

[0033] The broken wire feature in the partition marking diagram of twisted pairs is identified through line segment detection, the position information of the broken wire in the twisted pair is marked, and the broken wire defect data is generated. Line segment detection refers to the process of identifying and extracting line segment structures in an image. In the detection of broken wires in twisted pairs, the commonly used method is the Hough line transform, which can identify straight lines in the image in the parameter space. For the partition marking diagram of twisted pairs, first, the region boundary is extracted to form a binary edge map, and then the Hough line transform is applied to detect line segments. The line segments in the normal structure of the twisted pair usually show a regular spiral arrangement, while line segment interruptions or abnormal directions will occur at the broken wire. By setting parameters such as line segment length, angle, and spacing, abnormal line segment patterns can be identified, thus locating the position of the broken wire. Once a broken wire is found, characteristic parameters such as its position coordinates, line segment length, and fracture angle are recorded to form broken wire defect data, which contains the detailed position and severity information of the broken wire.

[0034] The loose strand feature in the partition marking diagram of twisted pairs is identified through spacing measurement, the position information of the loose strand in the twisted pair is marked, and the loose strand defect data is generated. Loose strand refers to the phenomenon that the spacing between strands in the twisted pair increases abnormally, and this defect can be accurately identified through spacing measurement. The calculation process of spacing measurement can be expressed by the following formula:

[0035]

[0036]

[0037] where represents the average minimum distance between region and region , is the Euclidean distance from point p to point q, is the number of point pairs used in the calculation, is the region and its adjacent region 's average spacing, is the reference spacing value of the normal twisted pair, is the percentage of spacing deviation. When exceeds the threshold (usually set to 20% - 30%), the region is marked as a loose strand region. In this way, the position of the loose strand in the partition marking diagram of the twisted pair is accurately identified and recorded, including information such as the coordinate range of the loose strand region and the spacing deviation value, forming loose strand defect data.

[0038] The deformation features and impurity attachment features in the twisted pair partition marking diagram are identified through morphological analysis, and the position information of twisted pair deformation and impurity attachment is marked. The broken wire defect data, loose strand defect data, and the position information of deformation and impurity attachment are integrated into a twisted pair defect position mapping table. Morphological analysis refers to a method of identification by studying the shape and structural characteristics of the target. In the identification of twisted pair defects, regional shape parameters (such as area, perimeter, aspect ratio, circularity, etc.) and texture features (such as gray level co-occurrence matrix, local binary pattern, etc.) are mainly used to distinguish deformation and impurity attachment. The deformation features usually manifest as abnormal changes in the contour of the twisted pair, and can be identified by calculating the deviation of the shape parameters of the region from the standard template. Impurity attachment usually manifests as abnormal local texture and color, and can be detected by analyzing the texture features and gray level distribution of the region. The identified positions of deformation and impurity attachment, together with the previously obtained broken wire and loose strand position information, are recorded in the twisted pair defect position mapping table. This mapping table is a comprehensive data structure that contains information such as the position coordinates, type identification, geometric parameters, and severity of various types of defects, providing a data basis for subsequent defect quantification assessment and process parameter adjustment.

[0039] For example, the visual inspection system on a steel wire stranding production line captured a stranding image, and obtained the stranding surface feature map after preliminary processing. The Sobel operator was applied to the feature map for gradient calculation to extract edge features. A 3×3 Sobel convolution kernel was used in the calculation to obtain the gradient maps in the horizontal and vertical directions, respectively. Then, the comprehensive gradient amplitude and direction were calculated to generate the stranding edge enhancement map. In the edge enhancement map, the spiral texture and strand boundary of the stranded wire appear as high gradient value areas, which are clearly distinguishable. Next, the double threshold method was applied for edge connection, and the high threshold was set to the 75% quantile of the gradient histogram (value 120), and the low threshold was set to 45% of the high threshold (value 54). In this way, the edges of the stranded wire surface were effectively connected to form a continuous contour line, and the stranded wire edge topology map was constructed. Then, the marker-watershed algorithm was used to segment the stranded wire surface. First, seed points were set in each obvious closed area, and a total of 24 different regional seeds were marked. Then, regional growth was performed, and finally a stranded wire partition labeling map was obtained, in which each area was represented by a different label value (1-24). Based on the partition marking map, a special analysis is conducted for different types of defects: for broken wire detection, Hough line transform is applied to detect straight line segments, the minimum line length is set to 20 pixels, and the angle tolerance is ±15 degrees. A broken wire position is successfully identified, which is located in the upper right corner of the image; for loose strand detection, the average spacing between adjacent areas is calculated, and it is found that the spacing deviations of the two areas are 32% and 27% respectively, exceeding the preset 20% threshold, and are determined to be loose strand areas; for deformation and impurity attachment detection, by calculating the shape parameters and texture features of the area, a deformation (roundness deviation exceeds 30%) and two impurity attachments (local binary pattern features are obviously abnormal) are identified. All detected defect information is integrated into the defect location mapping table, which records the type, location, geometric features and severity of each defect in detail, providing accurate data support for subsequent quality assessment and process adjustment.

[0040] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Count the pixels of the broken wire, loose strand, deformation and impurity adhesion defect areas in the stranded wire defect position mapping table, calculate the area parameters of various defects, and form a stranded wire defect area statistics table; Extract the grayscale gradient information of the defect area from the strand defect position mapping table, calculate the defect depth value through stereo reconstruction technology, and construct a strand defect depth distribution map; According to the defect coordinate information in the strand defect position mapping table, the number of defects per unit length of the strand is calculated to generate a strand defect density distribution curve; Compare the strand defect area statistics table and strand defect depth distribution diagram with the strand quality standard threshold, determine the severity level of each defect, and output the strand defect severity assessment table; Based on the stranded wire defect severity assessment table and the stranded wire defect density distribution curve, the comprehensive score of the stranded wire quality is obtained through weighted calculation, and the stranded wire quality scoring result is formed; Integrate the stranded wire quality scoring result with the defect detailed information, and file it according to the stranded wire product batch and production time to generate a stranded wire quality scoring report.

[0041] Specifically, pixel counting is performed on the defect areas of broken wires, loose strands, deformation, and impurity attachment in the stranded wire defect position mapping table, and the area parameters of various defects are calculated to form a stranded wire defect area statistics table, which is the primary link of quality assessment. Pixel counting refers to the process of counting the number of pixels in the marked area of each defect in the mapping table and then converting the pixel number into the actual physical area according to the camera calibration parameters. Specifically, first, the binary mask images of various defects are extracted from the defect position mapping table, where the pixel value of the defect area is 1 and the non-defect area is 0. Then, count the number of 1s in each mask image to obtain the number of pixels in the defect area. Next, calculate the actual area according to the correspondence between pixels and actual physical dimensions obtained by camera calibration (usually expressed in millimeters / pixel).

[0042] For example, if the camera calibration relationship is 0.02 mm / pixel and a broken wire defect area contains 850 pixels, its actual area is 850×0.02² = 0.34 square millimeters. Similar calculations are performed on all defect areas, and finally a stranded wire defect area statistics table is generated, which includes the ID, type, position coordinates, and area parameters of each defect.

[0043] Extract the gray gradient information of the defect area from the strand defect position mapping table, calculate the defect depth value through stereo reconstruction technology, and construct the strand defect depth distribution map, which is a key step in evaluating the three-dimensional characteristics of the defect. The gray gradient information refers to the direction and amplitude of the change in pixel gray values in the image, while stereo reconstruction technology is based on images taken by multi-angle cameras and calculates the three-dimensional coordinates of objects in the scene through the principle of triangulation. In the calculation of the strand defect depth, first extract the same defect area from the images collected by the multi-angle cameras, and find the corresponding point pairs through feature matching algorithms. Feature matching usually uses algorithms such as SIFT or SURF, which can stably identify corresponding feature points under different viewing angles and lighting conditions. Then, use the known internal and external camera parameters (obtained through camera calibration) and the disparity of the corresponding points to calculate the depth values of each point in the defect area. For reflective metal surfaces such as strands, a sub-pixel-level matching algorithm based on phase correlation is usually used to improve the accuracy of depth calculation. The finally generated defect depth distribution map is a three-dimensional data structure, containing the spatial coordinates and depth values of each point in the defect area, intuitively showing the three-dimensional shape and severity of the defect. According to the defect coordinate information in the strand defect position mapping table, calculate the number of defects on the strand per unit length, and generate the strand defect density distribution curve, which is an important means to measure the defect distribution law. The defect density refers to the number of defects appearing on the strand per unit length, usually measured in defects per meter. In the calculation process, first divide the strand length direction into several equal-length intervals (such as each 10 cm as an interval), and then count the number of defects contained in each interval. When counting, the corresponding relationship between the position coordinates of the defects and the strand length direction needs to be considered, which is usually achieved through the conversion between the camera coordinate system and the strand physical coordinate system. After the statistics are completed, divide the number of defects by the interval length to obtain the defect density value of each interval. Taking the strand length as the abscissa and the defect density as the ordinate, draw the strand defect density distribution curve, which intuitively reflects the distribution of defects along the strand length direction, helps to identify defect-prone areas and analyze the regularity of defect generation.

[0044] Comparing the stranded wire defect area statistics table, the stranded wire defect depth distribution map with the stranded wire quality standard threshold values to determine the severity levels of various defects and outputting the stranded wire defect severity assessment form is the core link of quality rating. The stranded wire quality standard threshold values are a series of reference values set based on industry standards and production experience, used to judge the defect severity. Generally, the defect severity is divided into three levels: minor, general, and severe. During the comparison process, first, the area value of each defect is extracted from the defect area statistics table and compared with the area threshold value of the corresponding type of defect; then, the maximum depth value and average depth value of each defect are extracted from the defect depth distribution map and compared with the depth threshold value. For example, for the broken wire defect, if the area is less than 0.5 square millimeters and the depth is less than 0.1 millimeter, it is determined as a minor defect; if the area is between 0.5 - 2 square millimeters or the depth is between 0.1 - 0.3 millimeters, it is determined as a general defect; if the area is greater than 2 square millimeters or the depth is greater than 0.3 millimeter, it is determined as a severe defect. Similarly, defects such as loose strands, deformation, and impurity attachment are rated according to their respective standards. The finally generated stranded wire defect severity assessment form contains information such as the ID, type, area, area, depth, and severity level of each defect.

[0045] Based on the stranded wire defect severity assessment form and the stranded wire defect density distribution curve, the comprehensive score of the stranded wire quality is obtained through weighted calculation to form the stranded wire quality scoring result. Weighted calculation refers to a method of assigning different weights to defects of different types and severities and comprehensively considering the influencing factors of the defects for scoring. In the calculation process, first, the basic weights are set according to the severity of the influence of the defects on the stranded wire performance. Usually, broken wire > loose strand > deformation > impurity attachment; second, the grade weights are set according to the severity level of the defects. Usually, severe > general > minor; then, considering the distribution density of the defects, the weights of the defects in the high-density area are appropriately increased. The weighted deduction value is calculated for each defect, and the deduction values of all defects are accumulated and deducted from the full score (usually 100 points) to obtain the comprehensive score of the stranded wire quality. This weighted calculation method can comprehensively reflect the quality status of all aspects of the stranded wire, considering both the type and severity of the defects and taking into account the distribution characteristics of the defects, and the obtained scoring result is more objective and reasonable. Integrating the stranded wire quality scoring result with the detailed defect information and filing it according to the stranded wire product batch and production time, generating the stranded wire quality scoring report is the final output of quality monitoring. The quality scoring report is a comprehensive document, including the basic information of the stranded wire product (such as batch number, specification, production date, etc.), the quality scoring result, defect statistical analysis, and a detailed defect list, etc. When generating the report, first, the quality scoring result is associated with the stranded wire product information; then, the defect data is statistically analyzed to calculate the quantity, proportion, and distribution law of various types of defects; then, the detailed defect information is integrated, including the type, location, size, severity, and image screenshots of each defect; it is organized and sorted according to the stranded wire product batch and production time to form a standardized quality scoring report. This report not only intuitively shows the overall quality status of the stranded wire product but also provides detailed defect information for subsequent analysis and processing reference, and is an important basis for the quality control and traceability of stranded wire production.

[0046] For example, the production line uses a machine vision system to monitor the quality of stranded wires. During the detection process, the system first generates a mapping table of the defect positions of the stranded wires, recording multiple defect areas. Pixel counting is performed on these defect areas, and it is found that a broken wire defect area contains 932 pixels. The area is calculated to be 0.302 square millimeters through conversion using the camera calibration parameter (0.018 mm / pixel); two loose strand defect areas contain 1580 and 2145 pixels respectively, and the converted areas are 0.512 and 0.695 square millimeters; a deformed defect area contains 3276 pixels, and the converted area is 1.063 square millimeters; the pixel numbers of three impurity attachment defect areas are 425, 531, and 688 respectively, and the converted areas are 0.138, 0.172, and 0.223 square millimeters respectively. These calculation results are recorded in the defect area statistical table. Then, the system extracts the corresponding point pairs of these defect areas from the images collected by the multi-angle cameras, and calculates the depth values through stereo matching and triangulation. For example, for the broken wire defect, the maximum depth is calculated to be 0.18 mm, and the average depth is 0.12 mm; the maximum depths of the two loose strand defects are 0.22 and 0.26 mm respectively, and the average depths are 0.15 and 0.19 mm. Then, the system divides the length direction of the stranded wire into intervals of every 10 cm, counts the number of defects in each interval, and finds that there are 2 defects in the 0 - 10 cm interval, 1 defect in the 10 - 20 cm interval, 3 defects in the 20 - 30 cm interval, and 1 defect in the 30 - 40 cm interval, and generates a defect density distribution curve accordingly. Subsequently, the system compares these defect data with the preset quality standard threshold values, determines that the broken wire defect is of a general grade (area < 0.5 square millimeters but depth between 0.1 - 0.3 mm), the two loose strand defects are of general and severe grades respectively, the deformed defect is of a severe grade, and the three impurity attachment defects are all of a minor grade, and generates a defect severity assessment form. Based on these assessment results, the system calculates the total deduction score to be 25.2 points according to the set weights (broken wire weight 5, loose strand weight 4, deformed weight 3, impurity attachment weight 2; severe grade coefficient 1.0, general grade coefficient 0.6, minor grade coefficient 0.3), and after deducting from the full score of 100 points, the comprehensive quality score of the stranded wire is 74.8 points. The system integrates this score result with the detailed defect information and archives it according to the production time of this batch of stranded wires, generating a quality scoring report.

[0047] In a specific embodiment, the process of executing step S105 may specifically include the following steps: Analyze the correlation between the defect types and the production process parameters of the stranded wire in the quality scoring report of the stranded wire through the association mapping technology, and construct a stranded wire defect - process influence matrix; Perform a regression analysis on the broken wire defect and the stranding tension data according to the stranded wire defect - process influence matrix, calculate the stranding tension influence coefficient, and form a tension regulation reference value; Analyze the relationship pattern between the loose strand defect and the drawing speed of the stranding machine in the stranding quality scoring report through the hierarchical clustering method, obtain the critical threshold of the drawing speed, and establish the parameter of the speed control range. Perform partial least squares calculation based on the stranding defect depth distribution map and the historical data of the pre-twist angle, identify the optimal pre-twist angle range, and generate the angle correction amount. Based on the tension regulation reference value, the speed control range parameter, and the angle correction amount, adopt the comprehensive weighted decision tree algorithm to optimize the combination of process parameters, and calculate the ideal adjustment amplitude of each process parameter. Smooth the ideal adjustment amplitude of the process parameters through the dynamic response function, reduce the production fluctuations caused by parameter mutations, and output the corrected values of the stranding process parameters.

[0048] Specifically, the association mapping technology is a data mining method used to discover the association rules and correlation patterns between different data sets. In stranding production, first collect the process parameter data (including stranding tension, drawing speed, pre-twist angle, etc.) in the historical production process and various defect data generated during the corresponding time period. Then calculate the correlation coefficient between each defect type and each process parameter to form a correlation coefficient matrix. The correlation coefficient usually adopts the Pearson correlation coefficient or the Spearman rank correlation coefficient, and its value range is between -1 and 1. The larger the absolute value, the stronger the correlation. For example, the correlation coefficient between the broken wire defect and the stranding tension is relatively high, indicating a strong correlation between the two; the correlation coefficient between the loose strand defect and the drawing speed is also relatively high, indicating an obvious association. Organize all the correlation coefficients into a matrix form, with rows representing different types of defects (broken wire, loose strand, deformation, impurity attachment), columns representing different process parameters, and the matrix element values representing the degree of correlation, thus forming the stranding defect-process influence matrix. This matrix intuitively shows the association strength between various defects and each process parameter, providing a data basis for subsequent targeted process parameter adjustment. Performing regression analysis on the broken wire defect and the stranding tension data according to the stranding defect-process influence matrix and calculating the stranding tension influence coefficient to form the tension regulation reference value is the key data processing step. Regression analysis is a statistical analysis method used to determine the quantitative relationship between variables. When dealing with the relationship between the broken wire defect and the stranding tension, a polynomial regression model is used for fitting, and the model can be expressed as:

[0049] where represents the broken wire frequency at the tension of T, , , , is the polynomial coefficient representing the basic broken wire trend, is the jth characteristic weight coefficient, is the influence factor of the j-th process feature, is the attenuation coefficient, is the reference tension point, and n is the number of features. The regression model not only considers the basic polynomial relationship between tension and wire breakage, but also introduces an exponential adjustment term based on specific process conditions, which can more accurately describe the complex situation in actual production. By using optimization algorithms such as the least squares method or maximum likelihood estimation, the model is fitted with historical data to obtain the values of each coefficient. Then, by taking the derivative and setting it to zero, the optimal tension value that minimizes the wire breakage frequency can be found as the tension control reference value. This reference value is the ideal set point for the stranding tension in wire stranding production, which can minimize the occurrence of wire breakage defects to the greatest extent.

[0050] Analyzing the relationship pattern between the loose strand defect and the traction speed of the stranding machine in the stranding quality scoring report through the hierarchical clustering method, obtaining the critical threshold of the traction speed, and establishing the speed control interval parameters are important steps to achieve precise speed control. Hierarchical clustering is a clustering algorithm based on graph theory. It regards data points as nodes in a graph, constructs an adjacency matrix by analyzing the similarity between data points, and then uses the spectral properties (eigenvalues and eigenvectors) of the matrix for clustering. In stranding production, first, the loose strand defect data at different traction speeds are organized in the form of data points, and each data point contains the traction speed value and the corresponding loose strand defect characteristics (such as quantity, area, severity, etc.). Then, the similarity between data points is calculated to construct an adjacency matrix. Next, the adjacency matrix is eigen-decomposed, and the eigenvectors corresponding to the first k smallest eigenvalues are used to form a low-dimensional representation. In this low-dimensional space, traditional clustering algorithms such as K-means are applied for grouping. By analyzing the clustering results, the relationship pattern between the traction speed and the loose strand defect can be found, especially the speed intervals where the loose strand defect increases significantly. The boundary points of these intervals are the critical thresholds of the traction speed. Based on these critical thresholds, the safe control interval of the traction speed is determined as the speed control interval parameter, providing a clear boundary for the speed setting of the stranding machine. Calculating the partial least squares based on the stranding defect depth distribution map and the historical data of the pre-twist angle, identifying the optimal pre-twist angle range, and generating the angle correction amount are scientific methods to optimize the pre-twist parameters. Partial least squares (PLS) is a multivariate statistical analysis method, especially suitable for situations where there is multicollinearity among predictive variables. In stranding production, first, the defect depth characteristics (such as average depth, maximum depth, depth variance, etc.) are extracted from the defect depth distribution map as the dependent variable matrix Y; the pre-twist angle and other relevant process parameters are extracted from the historical data as the independent variable matrix X. Then, the PLS algorithm is applied to decompose X and Y into linear combinations of latent variables and find the latent structure that can simultaneously explain the variations of X and Y. Specifically, the PLS algorithm extracts the components of X in an iterative manner to maximize its covariance with Y, thus establishing a prediction model. By analyzing the model coefficients, the influence law of the pre-twist angle on the defect depth can be determined, and the optimal angle range that minimizes the defect depth can be found. Based on the difference between this range and the current actual angle, the angle correction amount is generated to guide the adjustment of the pre-twist angle in production to reduce the defect depth and improve the stranding quality.

[0051] Based on the tension control reference value, speed control range parameters, and angle correction amount, the integrated weighted decision tree algorithm is used to optimize the combination of process parameters. Calculating the ideal adjustment amplitude of each process parameter is the core step of integrating multiple parameters. The integrated weighted decision tree is a composite algorithm that combines the decision tree structure and parameter weights and can handle multi-objective optimization problems. In the optimization of stranding process parameters, first, a decision tree structure is constructed. Each layer of the tree represents a process parameter (for example, the first layer is the stranding tension, the second layer is the traction speed, and the third layer is the pre-twist angle), and each node represents a possible value of the parameter. Then, weights are assigned to the branches of the tree, and the weight values are determined according to the influence degree of the parameter on different types of defects (from the defect-process influence matrix). Next, algorithms such as dynamic programming or Monte Carlo tree search are used to find the optimal path in the decision tree, that is, the parameter combination that can minimize the comprehensive defects to the greatest extent. By comparing the difference between the current process parameters and the optimal parameter combination, the ideal adjustment amplitude of each parameter is calculated to guide the corresponding adjustment of the production equipment. The ideal adjustment amplitude of the process parameter is smoothed through a dynamic response function to reduce the production fluctuations caused by parameter mutations. Outputting the corrected value of the stranding process parameter is the last link to ensure stable production. The dynamic response function is a mathematical function that describes the response process of the system to input changes and usually adopts the response model of a first-order or second-order system. In the adjustment of stranding process parameters, in order to avoid the impact of parameter mutations on production, it is necessary to smooth the ideal adjustment amplitude. The specific approach is to use the ideal adjustment amplitude as the target value and calculate a smooth parameter change curve considering the current actual parameter value and the dynamic characteristics of the production line. This smoothing process takes into account the inertia and stability requirements of the stranding mechanical system, avoids overshoot, oscillation, and instability phenomena caused by parameter mutations, ensures a smooth transition of the production process, and reduces new defects that may occur during the adjustment process. The parameter change value after smoothing is the finally output corrected value of the stranding process parameter and is directly used to guide the parameter adjustment of the production equipment.

[0052] For example, in the production line for manufacturing a batch of steel wire strands with a standard structure, during the production process, the visual inspection system found that there were relatively many defects such as broken wires and loose strands. By using the correlation mapping technology to analyze the defect data and process parameter records of the recent production, a defect-process impact matrix was constructed to clarify the correlation intensity between various defects and process parameters. According to this matrix, a regression analysis was performed on the broken wire defect and the stranding tension data. By applying a complex regression model, the model parameters were fitted to determine the optimal tension value. Through the hierarchical clustering method, the relationship between the loose strand defect and the traction speed was analyzed, and the recent data points were clustered in the feature space to find the safe range of the traction speed. At the same time, through the partial least squares calculation, the relationship between the defect depth and the pre-twist angle was analyzed, and a prediction model was established to find the optimal range of the pre-twist angle. Considering the adjustment requirements of these three parameters, the comprehensive weighted decision tree algorithm was applied for optimal combination. Taking into account the different degrees of influence of various defects on the product quality, different weights were assigned to each parameter, and the ideal adjustment amplitude was calculated. To ensure a smooth transition in production, the dynamic response function was used to smooth these adjustment amplitudes, and a step-by-step progressive adjustment strategy was designed.

[0053] In a specific embodiment, the process of executing step S106 may specifically include the following steps: Perform time-series correlation processing on the defect data in the strand quality scoring report and the corrected values of the strand process parameters to generate a corresponding relationship table of strand defects-process parameters; Based on the corresponding relationship table of strand defects-process parameters, through multi-dimensional factor analysis, correlate and attribute the defect generation mechanism and process parameter deviation to construct a root cause map of strand defects; Through the hierarchical clustering method, perform feature induction and grouping on the root cause map of strand defects, extract the typical process patterns of strand defects, and form a feature dictionary of strand defects; Perform defect trend prediction on the feature dictionary of strand defects and historical production data through time series analysis to establish a defect development pattern library of strand defects; Based on the defect development pattern library of strand defects and real-time production parameters, evaluate the potential defect risks through similarity matching to generate a quality warning index for strand defects; Integrate and combine the corresponding relationship table of strand defects-process parameters, the root cause map of strand defects, the feature dictionary of strand defects, and the quality warning index of strand defects to construct a database of strand defect-process relationships.

[0054] Specifically, time-series correlation processing refers to a data processing method that establishes a mapping relationship between defect data and process parameter changes in chronological order. In specific operations, first, information such as the defect occurrence time, type, and severity recorded in the quality score report is extracted. At the same time, the process parameter correction records of the stranding machine during the corresponding time period are collected, including the adjustment time and amplitude of parameters such as stranding tension, traction speed, and pre-twist angle. Then, a time-window analysis mechanism is established. Usually, a forward time window (such as from 1 hour before parameter adjustment to the adjustment time) and a backward time window (such as from 1 hour to 24 hours after parameter adjustment) are set, and the change in defect occurrence within each time window is statistically analyzed. The change in defect data before and after process parameter adjustment is correlated with the specific adjusted parameters and recorded in the stranding defect-process parameter correspondence table. This table contains fields such as time stamps, process parameter adjustment content, and defect change situations, clearly showing the direct impact of process parameter adjustment on defect occurrence. Through multi-dimensional factor analysis of the stranding defect-process parameter correspondence table, correlating and attributing the defect generation mechanism and process parameter deviation is a key link in deeply understanding the defect causes. Multi-dimensional factor analysis is a statistical method for exploring the internal structure of data and the relationships between variables, and potential factors are discovered through dimensionality reduction and variable clustering. In stranding production, first, key variable data is extracted from the defect-process parameter correspondence table to form a multi-dimensional data matrix. Then, through principal component analysis or factor analysis methods, the main factors that can explain the data variation are extracted. In actual operations, after standardizing the data, the correlation matrix or covariance matrix is calculated, the eigenvalues and eigenvectors are solved, and the first few main factors are extracted (generally, factors with a cumulative contribution rate of more than 80% are selected). Then, through factor rotation (such as orthogonal rotation or oblique rotation), the factor structure is optimized so that each factor has a clear process physical meaning. Based on the factor analysis results, a causal relationship network between defect types and process parameter deviations is established to form a stranding defect root cause map. This map takes the defect type as the core node, the process parameters as the influencing nodes, the thickness of the connection line represents the influence intensity, and the direction represents the causal relationship, intuitively showing the influence mechanism of different process parameter deviations on various types of defects. By using the hierarchical clustering method to summarize and group the characteristics of the stranding defect root cause map, extracting the typical process patterns of stranding defects and forming a stranding defect feature dictionary is an effective means to achieve knowledge precipitation. Hierarchical clustering is a method of constructing a clustering hierarchy from bottom to top or from top to bottom, suitable for discovering hierarchical patterns in data. In stranding defect analysis, first, the feature vectors of the defect-process parameter relationship are extracted from the root cause map, and each vector contains information such as defect type, severity, associated process parameter set, and its influence intensity. Then, a similarity measurement standard between feature vectors is defined, and commonly used ones include Euclidean distance, Manhattan distance, or cosine similarity, etc.Next, the bottom-up agglomerative hierarchical clustering algorithm is adopted. Initially, each defect case is regarded as an independent cluster, and then the most similar clusters are gradually merged until the preset number of clusters or similarity threshold is reached. During the clustering process, the distance between clusters can be calculated by methods such as single-link, complete-link, or average-link. After clustering, feature extraction and summarization are performed on the defect cases within each cluster to generalize the typical process patterns of this type of defect, including the main influencing parameters, parameter critical values, defect feature descriptions, etc. These typical process patterns are organized into structured knowledge entries to form a dictionary of stranded wire defect features, providing knowledge base support for subsequent defect prediction and prevention.

[0055] The defect trend prediction of twisted pair cables is carried out through time series analysis on the defect feature dictionary and historical production data. Establishing a defect development pattern library for twisted pair cables is a scientific method for predicting future defect risks. Time series analysis is a statistical technique that studies a sequence of data points arranged in chronological order, explores its internal laws, and makes predictions. In the defect trend prediction of twisted pair cables, first, sequence data of various defects changing over time are extracted from historical production data, and at the same time, the typical process pattern information in the defect feature dictionary is combined. Then, the stationarity of the time series data is tested. If it is not stationary, it is made stationary through methods such as differencing and logarithmic transformation. Next, methods such as autoregressive moving average model (ARMA), autoregressive integrated moving average model (ARIMA), or seasonal ARIMA model are used to model the processed data. The model parameters are determined by maximum likelihood estimation or least squares method, and the model selection is based on information criteria such as AIC and BIC. After verifying the effectiveness of the model, the model is used to predict the defect occurrence trend in the future for a period of time. In addition, process parameter variables can be introduced to establish a multivariate time series model, such as vector autoregressive model (VAR) or state space model, to improve the prediction accuracy. Based on the time series models and prediction results of different defect types, a defect development pattern library for twisted pair cables is formed, recording the development laws, change cycles, and trend characteristics of various defects. Evaluating potential defect risks through similarity matching based on the defect development pattern library of twisted pair cables and real-time production parameters and generating a quality warning index for twisted pair cables is the core link to achieve preventive quality control. Similarity matching is a method of predicting future development by calculating the similarity between the current state and historical patterns. In the quality warning of twisted pair cables, first, real-time production parameter data are obtained, including process parameter values such as the current stranding tension, traction speed, and pre-twist angle. Then, these parameters are compared with the historical patterns in the defect development pattern library, and the similarity is calculated. Similarity calculation methods include Euclidean distance, cosine similarity, DTW (dynamic time warping) distance, etc., and a measurement standard suitable for the characteristics of twisted pair cable production is selected. Next, the historical patterns ranked among the top in terms of similarity are analyzed, and the defect development trends and risk levels under these patterns are extracted. Based on the deviation degree between the defect risks of similar patterns and the current production state, risk indices of various defects are calculated to form a risk assessment matrix. According to the risk assessment results, warning levels (such as low risk, medium risk, high risk) are set, a quality warning index for twisted pair cables is generated, operators are timely reminded to pay attention to potential quality problems, and targeted adjustment suggestions are given.

[0056] Integrating the relationship table of stranding defects and process parameters, the root cause map of stranding defects, the defect feature dictionary of stranding, and the quality warning indicators of stranding to construct a stranding defect - process relationship database is the ultimate goal of quality traceability and warning. Database construction first requires designing a reasonable data structure to store the above four core data sets in the form of a relational or non-relational database. In the design of a relational database, usually the relationship table of stranding defects and process parameters is used as the main table to record basic information such as defect ID, process parameter ID, and timestamp; while the root cause map of defects is stored as a node table and an edge table to record defect types, process parameters, and influence relationships; the defect feature dictionary is used as a knowledge base table to store information such as defect type codes, typical process mode descriptions, and parameter ranges; the quality warning indicators are used as a dynamically updated table to record the current risk assessment results. The database also needs to design query interfaces to support multi-dimensional queries and analyses by time, defect type, process parameters, etc. In addition, the database management system needs to have a data update mechanism to continuously enrich and optimize the content of the knowledge base with new production data and analysis results. The finally formed stranding defect - process relationship database is a dynamic learning and continuously optimizing knowledge system, which can not only support the traceability of the causes of historical defects but also conduct risk warnings based on the current production status, realizing the whole-process intelligent monitoring of the production quality of stranded wires.

[0057] For example, the production line adjusted process parameters multiple times within a month and recorded the corresponding defect data. Through time-series correlation processing, it was found that after adjusting the stranding tension from 12 N to 11 N on a certain occasion, the number of wire breakage defects decreased by 40% within 24 hours; and after increasing the traction speed from 10 m / min to 13 m / min on another occasion, the number of loose strand defects increased significantly within the subsequent 4 hours. These time-series correlation data were recorded in the defect-process parameter correspondence table. Then, by performing multi-dimensional factor analysis on these data, three main factors were extracted: the first factor was mainly related to the stranding tension and wire breakage defects, explaining 45% of the total variation; the second factor was highly correlated with the traction speed and loose strand defects, explaining 30% of the variation; the third factor was closely related to the pre-twist angle and wire strand deformation, explaining 15% of the variation. Based on these factors, a defect root cause map was constructed, clearly showing the relationship network between process parameter deviations and various types of defects. Then, the hierarchical clustering method was used to analyze the defect cases in the root cause map, and five typical defect-process patterns were identified: the wire breakage pattern caused by high tension, the loose strand pattern caused by rapid traction, the deformation pattern caused by excessive angle, the mixed defect pattern caused by unstable tension, and the seasonal defect pattern related to temperature fluctuations. These patterns were detailedly recorded in the defect feature dictionary, including parameter critical values and typical defect feature descriptions. Based on the feature dictionary and historical data, prediction models for various types of defects were established through time series analysis. For example, it was found that the wire breakage defects had obvious short-term fluctuations related to the tension change, while the loose strand defects showed a cumulative effect related to the equipment operation time. In actual production, when the current process parameter combination (tension 11.5 N, speed 12.8 m / min, angle 15 degrees) was detected, through similarity matching, it was found that the similarity between this parameter combination and the pattern that caused loose strand defects in history reached 85%, and a medium-level loose strand risk warning was immediately generated, suggesting reducing the traction speed to below 11 m / min. This data processing and analysis process was finally integrated into the wire strand defect-process relationship database, realizing the whole-process intelligent quality monitoring from historical data analysis to real-time warning, effectively improving the quality stability and problem-solving efficiency of wire strand production.

[0058] The above describes the method for monitoring the quality of wire strand production based on machine vision in the embodiments of the present application. Next, the system for monitoring the quality of wire strand production based on machine vision in the embodiments of the present application will be described. Please refer to Figure 2 One embodiment of the system for monitoring the quality of wire strand production based on machine vision in the embodiments of the present application includes: An acquisition module, configured to collect multi-angle data of the surface of the moving wire strand through a high-speed industrial camera array to obtain the original image data of the wire strand; A filtering module, configured to filter high-reflection regions and random noises on the surface of the stranded wire according to the original image data of the stranded wire, so as to obtain a surface feature map of the stranded wire; An identification module, configured to use the surface feature map of the stranded wire to identify broken wires, loose strands, deformation, and impurity attachment through edge detection and region segmentation, and obtain a mapping table of the defect positions of the stranded wire; A quantization module, configured to perform quantization calculations on the defect area, depth, and distribution density based on the mapping table of the defect positions of the stranded wire, and generate a quality score report of the stranded wire; An analysis module, configured to perform correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine according to the quality score report of the stranded wire, and form a correction value of the stranding process parameters; A construction module, configured to construct a database of the relationship between stranded wire defects and processes according to the quality score report of the stranded wire and the correction value of the stranding process parameters, so as to realize the traceability and early warning of the production quality of the stranded wire.

[0059] Through the collaborative cooperation of the above-mentioned various components, the surface of the stranded wire in motion is collected from multiple angles by a high-speed industrial camera array, realizing 360° all-round monitoring of the surface of the stranded wire, overcoming the blind spot problem of traditional single-view detection, and significantly improving the defect detection rate. At the same time, aiming at the high specular reflection characteristics of the metal surface of the stranded wire, the regional adaptive brightness mapping technology and the polar coordinate transformation method are used for light compensation, effectively solving the problem of specular reflection interference on the metal surface and improving the stability and reliability of the image quality. The application of the anisotropic diffusion filter effectively suppresses random noise while retaining the structural edges of the stranded wire surface, laying a foundation for subsequent accurate defect identification. Through the multi-band wavelet decomposition and reconstruction technology, the selective enhancement of different scale features on the surface of the stranded wire is realized, improving the visibility of tiny defects. The refined application of edge detection and region segmentation algorithms realizes the accurate identification of various types of defects such as broken wires, loose strands, deformation, and impurity attachment. The generation of the defect position mapping table provides structured data support for subsequent quantitative analysis. Based on the quantitative calculation of defect area, depth, and distribution density, an objective stranded wire quality scoring system is formed, providing a data basis for quality assessment. By establishing the mapping relationship between defect features and production process parameters through correlation analysis technology, a decision-making basis is provided for the precise adjustment of process parameters. Finally, the constructed stranded wire defect-process relationship database realizes the traceability and early warning functions of quality problems, transforming passive detection into active prevention. The innovation of the solution of the present invention lies in fully considering the specific application requirements of artificial intelligence algorithms in the field of stranded wire quality monitoring, such as the optimization of image enhancement algorithms for high-speed moving metal surfaces, the design of defect recognition models adapted to the unique structure of stranded wires, and the process parameter optimization algorithms based on historical data. The organic combination of these algorithm features enables the solution to adapt to the high-speed, high specular reflection, and high-precision detection requirements of stranded wire production, and continuously optimizes production parameters in a data-driven manner to achieve quality closed-loop control. Compared with traditional methods, the present invention not only improves the accuracy and comprehensiveness of defect detection, but also establishes a causal relationship between defects and process parameters, transforming quality control from post-inspection to pre-prevention and process control, significantly reducing production defect rates and material waste, and improving production efficiency and product consistency.

[0060] Referring to Figure 3 , an embodiment of the present invention further provides a computer device, which may be a server, and its internal structure may be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0061] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0062] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0063] Those of ordinary skill in the art can understand that all or part of the processes in the above-described embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-described method embodiments. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0064] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A quality monitoring method for stranding production based on machine vision, characterized in that, The method for monitoring the quality of stranded wire production based on machine vision includes: Collecting the surface of the moving stranded wire from multiple angles through a high-speed industrial camera array to obtain the original image data of the stranded wire; Filtering the highly reflective areas and random noises on the surface of the stranded wire according to the original image data of the stranded wire to obtain the surface feature map of the stranded wire; Using the surface feature map of the stranded wire to identify broken wires, loose strands, deformation, and impurity attachment through edge detection and region segmentation, and obtaining the mapping table of the defect positions of the stranded wire; Quantitatively calculating the defect area, depth, and distribution density based on the mapping table of the defect positions of the stranded wire to generate a quality score report of the stranded wire; Based on the quality score report of the stranded wire, performing correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine to form a correction value of the stranding process parameters; According to the quality score report of the stranded wire and the correction value of the stranding process parameters, constructing a database of the relationship between stranded wire defects and processes to achieve traceability and early warning of the quality of stranded wire production.

2. The method for monitoring the production quality of twisted wires based on machine vision according to claim 1, wherein The step of collecting the surface of the moving stranded wire from multiple angles through a high-speed industrial camera array to obtain the original image data of the stranded wire includes: Taking an omni-directional photograph of the surface of the stranded wire through a high-speed industrial camera installed at a key position on the stranded wire production line to obtain the original multi-angle image of the stranded wire; Performing combined illumination on the surface of the stranded wire by configuring a 45° direct light source, a low-angle side light source, and a backlight source to obtain an enhanced image of the surface features of the stranded wire; Synchronously processing the acquisition of the original multi-angle image of the stranded wire and the speed of the stranded wire production line through a high-precision clock signal synchronization control unit to obtain a clear moving image of the stranded wire; Adjusting the parameters of the high-speed industrial camera adaptively to the environment through a temperature and humidity compensation device, and transmitting the original multi-angle image of the stranded wire, the enhanced image of the surface features of the stranded wire, and the moving image of the stranded wire to an image processing unit through a gigabit network to form the original image data of the stranded wire.

3. The method for monitoring the quality of twisted wire production based on machine vision according to claim 1, wherein The step of filtering the highly reflective areas and random noises on the surface of the stranded wire according to the original image data of the stranded wire to obtain the surface feature map of the stranded wire includes: Segmenting the areas with different degrees of reflectivity on the metal surface of the stranded wire through regional adaptive brightness mapping technology for the original image data of the stranded wire, and establishing a reflectivity distribution map of the stranded wire; Performing light compensation on the high-light areas in the reflectivity distribution map of the stranded wire through the polar coordinate transformation method to reduce the interference of specular reflection on the surface of the stranded wire and form an illumination-equalized map of the stranded wire; Processing the illumination-equalized map of the stranded wire through an anisotropic diffusion filter to retain the edge of the surface structure of the stranded wire while suppressing random noises, and obtaining an edge-preserved map of the stranded wire; Performing multi-band wavelet decomposition and reconstruction on the edge-preserved map of the stranded wire to selectively enhance the features of different scales on the surface of the stranded wire and generate a multi-scale feature set of the stranded wire; Performing structural similarity fusion on the multi-scale feature set of the stranded wire through non-local mean filtering, retaining the texture information on the surface of the stranded wire while eliminating residual noises, and obtaining the surface feature map of the stranded wire.

4. The method for monitoring the quality of stranded wire production based on machine vision according to claim 1, characterized in that, The step of using the surface feature map of the stranded wire to identify broken wires, loose strands, deformation, and impurity attachment through edge detection and region segmentation, and obtaining the mapping table of the defect positions of the stranded wire includes: Perform gradient calculation on the surface feature map of the stranded wire through the Sobel operator, extract the edge features of the stranded wire surface, and generate an edge-enhanced map of the stranded wire; Perform edge connection on the edge-enhanced map of the stranded wire through the double-threshold method, construct the complete contour of the stranded wire, and form a topological map of the stranded wire edge; Perform region segmentation on the topological map of the stranded wire edge through the marker-watershed algorithm, divide different feature regions on the stranded wire surface, and obtain a partitioned label map of the stranded wire; Identify the broken wire features in the partitioned label map of the stranded wire through line segment detection, mark the position information of the broken wire on the stranded wire, and generate broken wire defect data; Identify the loose strand features in the partitioned label map of the stranded wire through spacing measurement, mark the position information of the loose strand on the stranded wire, and generate loose strand defect data; Identify the deformation features and impurity attachment features in the partitioned label map of the stranded wire through morphological analysis, mark the position information of the deformation and impurity attachment on the stranded wire, and integrate the broken wire defect data, loose strand defect data, and the position information of deformation and impurity attachment into a defect position mapping table of the stranded wire.

5. The method for monitoring the quality of stranded wire production based on machine vision according to claim 1, characterized in that, Based on the defect position mapping table of the stranded wire, perform quantitative calculations on the defect area, depth, and distribution density, and generate a quality score report for the stranded wire, including: Perform pixel counting on the broken wire, loose strand, deformation, and impurity attachment defect regions in the defect position mapping table of the stranded wire, calculate the area parameters of various defects, and form a statistical table of the stranded wire defect area; Extract the gray gradient information of the defect regions from the defect position mapping table of the stranded wire, calculate the defect depth value through stereo reconstruction technology, and construct a depth distribution map of the stranded wire defects; According to the defect coordinate information in the defect position mapping table of the stranded wire, calculate the number of defects per unit length of the stranded wire, and generate a defect density distribution curve of the stranded wire; Compare the statistical table of the stranded wire defect area, the depth distribution map of the stranded wire defects with the quality standard threshold of the stranded wire, determine the severity level of various defects, and output an evaluation table of the stranded wire defect severity; Based on the evaluation table of the stranded wire defect severity and the defect density distribution curve of the stranded wire, obtain the comprehensive quality score of the stranded wire through weighted calculation, and form the quality score result of the stranded wire; Integrate the quality score result of the stranded wire with the detailed defect information, file it according to the batch and production time of the stranded wire products, and generate a quality score report for the stranded wire.

6. The method for monitoring the quality of twisted wire production based on machine vision according to claim 5, wherein Based on the quality score report of the stranded wire, perform correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine, and form a corrected value of the stranding process parameters, including: Analyze the correlation between the defect types and the stranding process parameters in the quality score report of the stranded wire through correlation mapping technology, and construct a defect-process influence matrix of the stranded wire; Perform regression analysis on the broken wire defect and the stranding tension data according to the defect-process influence matrix of the stranded wire, calculate the influence coefficient of the stranding tension, and form a reference value for tension control; Analyze the relationship pattern between the loose strand defect and the traction speed of the stranding machine in the quality score report of the stranded wire through hierarchical clustering method, obtain the critical threshold of the traction speed, and establish the parameter of the speed control range; Perform partial least squares calculation according to the depth distribution map of the stranded wire defects and the historical data of the pre-twist angle, identify the optimal pre-twist angle range, and generate an angle correction amount; Based on the above-mentioned tension regulation reference value, speed control interval parameter, and angle correction amount, the comprehensive weighted decision tree algorithm is used to optimize and combine the process parameters, and calculate the ideal adjustment range of each process parameter; The ideal adjustment range of the process parameters is smoothed through the dynamic response function to reduce the production fluctuations caused by parameter mutations, and the corrected value of the stranding process parameters is output.

7. The method for monitoring the production quality of twisted wires based on machine vision according to claim 6, characterized in that, According to the stranding quality scoring report and the corrected value of the stranding process parameters, a stranding defect-process relationship database is constructed to realize the traceability and early warning of the stranding production quality, including: Perform time-series correlation processing on the defect data in the stranding quality scoring report and the corrected value of the stranding process parameters to generate a corresponding relationship table of stranding defects-process parameters; Based on the corresponding relationship table of stranding defects-process parameters, through multi-dimensional factor analysis, the correlation attribution of the defect generation mechanism and process parameter deviation is carried out, and a stranding defect root cause map is constructed; Through the hierarchical clustering method, the characteristics of the stranding defect root cause map are summarized and grouped, and the typical process patterns of stranding defects are extracted to form a stranding defect feature dictionary; Through time series analysis of the stranding defect feature dictionary and historical production data, the defect trend is predicted, and a stranding defect development pattern library is established; Based on the stranding defect development pattern library and real-time production parameters, the potential defect risks are evaluated through similarity matching, and a stranding quality early warning index is generated; Integrate and integrate the corresponding relationship table of stranding defects-process parameters, the stranding defect root cause map, the stranding defect feature dictionary, and the stranding quality early warning index to construct a stranding defect-process relationship database.

8. A quality monitoring system for stranding production based on machine vision, which is used to implement the quality monitoring method for stranding production based on machine vision as described in any one of claims 1-7, characterized in that, The stranding production quality monitoring system based on machine vision includes: An acquisition module for multi-angle acquisition of the surface of the moving stranding through a high-speed industrial camera array to obtain the original image data of the stranding; A filtering module for filtering the high-reflection area and random noise on the surface of the stranding according to the original image data of the stranding to obtain a stranding surface feature map; An identification module for using the stranding surface feature map to identify broken wires, loose strands, deformations, and impurity attachments through edge detection and region segmentation, and obtaining a stranding defect position mapping table; A quantization module for quantitatively calculating the defect area, depth, and distribution density based on the stranding defect position mapping table to generate a stranding quality scoring report; An analysis module for performing correlation analysis and adjustment on the traction speed, stranding tension, and pre-twist angle of the stranding machine according to the stranding quality scoring report to form a corrected value of the stranding process parameters; A construction module for constructing a stranding defect-process relationship database according to the stranding quality scoring report and the corrected value of the stranding process parameters to realize the traceability and early warning of the stranding production quality.

9. A computer device, characterized in that, It includes a memory and a processor, and the memory stores a computer program that can run on the processor. The characteristic is that when the processor executes the computer program, it realizes the machine vision-based stranding production quality monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and when the computer program is run by a processor, the processor is caused to execute the machine vision-based quality monitoring method for stranding production according to any one of claims 1 to 7.

Images