A dynamic position on-line measurement system for cold-rolled copper strip

Abstract

This invention belongs to the field of rolling process detection in metallurgical rolling technology. It discloses an online dynamic position measurement system for cold-rolled copper strip. Its core advantage lies in its avoidance of complex deep learning models, instead relying on improved traditional machine vision algorithms, thus achieving excellent portability, extremely fast processing speed, and low memory overhead. The method acquires images using a high-resolution industrial camera and quickly identifies the presence of copper strip using simple pixel thresholding, effectively avoiding unnecessary computational resource waste. Subsequently, it combines optimized Canny edge detection and Hough transform algorithms to accurately locate the copper strip edge and calculate the bandwidth and deviation in real time. This lightweight design eliminates the need for high-performance hardware, making the system easy to deploy and maintain in various industrial scenarios, providing an efficient, reliable, and cost-effective deviation monitoring solution for cold rolling production.

Description

Technical Field

[0001] This invention belongs to the field of rolling process detection in metallurgical rolling technology, and relates to a measurement system, specifically an online dynamic position measurement system for cold-rolled copper strip. Background Technology

[0002] In the continuous rolling process of cold-rolled copper strip, the strip is prone to lateral deviation during roller conveying and multi-pass rolling. The industry standard defines the phenomenon where the actual centerline of the strip does not coincide with the theoretical rolling reference centerline as a deviation fault. Deviation is a high-frequency negative condition in the cold rolling process, and its causes include multiple factors such as roll parallelism deviation, wedge thickness difference in the incoming strip, uneven tension distribution, roller wear, and warping of the incoming strip surface. The production losses caused by deviation have a layered characteristic: mild deviation only causes uneven thickness at the edge of the finished copper strip, strip wavy surface, and side burrs, significantly reducing the finished product grade and market competitiveness; if the deviation continues to accumulate and is not intervened in time, the edge of the copper strip will scrape hard against the mill guides and looper sidewalls, and in severe cases, directly cause copper strip tearing, strip breakage, and shutdown accidents. This not only interrupts the continuous rolling production rhythm but also damages core equipment such as rolls and guide rolls, resulting in high spare parts replacement costs and long-term downtime losses. Currently, most copper cold rolling production lines in China still rely on manual monitoring, with operators visually judging the degree of copper strip deviation based on real-time video feeds displayed on control panels. This traditional detection method has several inherent drawbacks: First, the accuracy of human vision is limited, making it difficult to quickly detect minute deviations in the copper strip. Fluctuations in lighting, workshop oil mist, and the mirror-like reflection of the copper strip can all interfere with visual judgment, leading to a high probability of misjudgment and missed detection. Second, prolonged screen monitoring significantly increases the workload of frontline operators, which is inconsistent with the industry trend of intelligent manufacturing to reduce manpower and increase efficiency. Third, there is a significant time lag between manual judgment and the issuance of correction commands and the mill's implementation of adjustments, allowing even minor deviations to easily develop into serious shape defects or even strip breakage accidents within this lag window. Against this backdrop of industry pain points, developing an online visual measurement system that is adapted to the special working conditions of high reflectivity in cold-rolled copper strip and can calculate the dynamic lateral position and actual bandwidth of the copper strip in real time, automatically, and with high precision has crucial engineering application value for closed-loop control of copper rolling shape, improving yield, and reducing production line operation and maintenance costs.

[0003] Currently, domestic and international researchers and equipment manufacturers have conducted extensive theoretical research and equipment development around online detection of metal strip misalignment, forming multiple technical routes including machine vision, photoelectric sensing, and ultrasonic detection. Related journal articles and patent achievements are gradually being applied, providing a theoretical and technical foundation for automated strip monitoring. Existing representative research and patent achievements can be divided into two main categories: literature and authorized patents. In the literature section, the paper "Strip Operation Monitoring System Based on Image Processing" (Proceedings of the 2009 China Instrumentation and Measurement & Control Technology Conference) proposes a strip edge positioning scheme integrating the Canny edge detection algorithm, the region seed growth segmentation algorithm, and the Hough line fitting algorithm. Through the joint use of multiple algorithms, it achieves automatic strip edge identification and real-time early warning of misalignment faults, and can output quantitative offset monitoring data for the rolling mill control system. Another paper, "Application Research of Misalignment Detection by CCD Sensors" (Silicon Valley, Issue 3, 2011), relies on a CCD image sensor to build a simple visual acquisition unit. It determines the left and right edges of the strip by the location of sudden grayscale changes in the acquired image, and calculates the difference between the real-time edge coordinates and the equipment's preset standard reference coordinates to determine whether the strip has deviated. This is an early basic research scheme for machine vision misalignment detection. Two highly relevant prior art documents exist in the patent field. One is CN114758008A, "A System and Method for Detecting Strip Deviation in Finishing Rolling Based on a Binocular Linear Array Camera." This patent adopts a dual-camera linear array CMOS camera architecture. First, the internal and external parameters of the two cameras are calibrated using a black-and-white calibration board to ensure that the two imaging fields of view overlap. The mill bite signal serves as the image acquisition trigger source. The acquired images are transmitted to a dedicated image processing server. The Canny operator combined with a sub-pixel algorithm is used to extract the pixel coordinates of the left and right edges of the strip. Combined with a binocular geometric model, the actual bandwidth and real-time deviation of the strip are calculated. Finally, the detected data are processed. The value is pushed to the terminal display for visualization; the second is CN102755998A "Method for Centerline Detection in Metal Strip Rolling". This patent adopts a multi-position linear array image sensor layout, with at least three acquisition devices arranged on both sides of the strip conveyor roller. The gray-scale change point of the image pixel is used as the basis for edge judgment. Multiple calculation formulas are established for the roller reference width and the strip edge distance. The center offset and bandwidth deviation of the strip are calculated simultaneously. When the width deviation is less than 1 mm, the conveying condition is judged to be normal. If it exceeds the threshold, an abnormal signal is output. The whole solution is applicable to the centerline monitoring scenario of hot-rolled and ordinary cold-rolled carbon steel strip.

[0004] A comprehensive review of the aforementioned literature and the technical solutions in the four comparative patent documents reveals that while existing strip misalignment detection technologies have achieved basic automated monitoring functions, they still exhibit significant technical limitations in the complex conditions of high-reflectivity, narrow-width, high-speed rolling of cold-rolled copper strip. Each technical approach has its own specific shortcomings. The first type of monitoring scheme, based on general image processing, such as the Canny+Hough joint monitoring method presented in the conference paper, is designed only for low-reflectivity carbon steel strip and does not consider the specular reflection characteristics of copper strip. During rolling, the reflective highlights on the copper strip surface cause severe image grayscale distortion, directly leading to numerous false contours in edge extraction and a significant decrease in edge positioning accuracy. Furthermore, this scheme lacks an image distortion correction module, and perspective stretching errors caused by camera tilt installation cannot be eliminated, resulting in insufficient stability in dynamic offset calculation. The second type of CCD / multi-line array hardware detection solution, represented by patent CN102755998A, requires the deployment of three or more image acquisition devices. It involves a large number of hardware devices, complex wiring, and high costs for equipment procurement, on-site installation, and daily maintenance. It relies solely on simple grayscale abrupt changes to identify edges without adaptive image enhancement methods. The probability of detection failure is high under interference from workshop oil mist and stray light. Furthermore, the entire system is only compatible with carbon steel and does not optimize the discrimination threshold for the characteristics of copper strip materials. The third type of binocular linear array vision solution, namely the patented technology CN114758008A, adopts a dual-line linear array camera hardware architecture. The equipment is large in size, and it is difficult to install and deploy it in the narrow space of the rolling mill. The system is equipped with water cooling and gas sealing protection structures, which further increases the cost of the whole machine. Its image preprocessing only uses the fixed parameter Canny operator, which cannot adjust the brightness and gamma parameters according to the difference in reflection between the left and right areas of the copper strip. When facing the copper strip surface with uneven brightness, the edge on one side is easily lost. At the same time, this solution is developed for hot-rolled fine-rolled strip steel and is not adapted to the real-time requirements of continuous high-speed rolling of cold-rolled copper strip. The algorithm has a high computational load and the processing time of a single frame image is long, which is difficult to match the rolling line speed of copper strip of tens of meters per second. In addition, photoelectric and ultrasonic correction sensors in the industry also have obvious shortcomings. The sensor body needs to be equipped with a complex protective housing that shields against light, dust and high temperature, which imposes strict requirements on the installation space of the rolling mill. Moreover, single-point sensing can only collect local point data and cannot fully obtain the full width and overall offset information of the copper strip. It can only achieve single-point correction and cannot complete the full-width strip shape data acquisition.

[0005] Existing machine vision-based deviation detection technologies generally suffer from common defects: First, the imaging system lacks sealed light protection, allowing ambient light and rolling oil mist to continuously interfere with image quality. Second, the image processing algorithm uses globally uniform parameters, failing to adaptively optimize for regional differences in reflectivity between the left and right sides of the copper strip, resulting in significant edge recognition failures in high-brightness areas. Third, most solutions employ multi-camera, binocular hardware architectures, leading to excessively high hardware investment and installation space costs. Fourth, the lack of perspective distortion correction processes means that geometric errors caused by oblique camera shots cannot be compensated, resulting in low accuracy in dynamic position measurement. Fifth, all existing technologies focus on carbon steel strip, lacking a dedicated integrated measurement solution adapted to the mirror reflection of cold-rolled copper strip and the high-speed operation of thin strips.

[0006] Based on the current technological gaps in the industry and the multiple limitations exposed by the four existing comparative documents, this invention focuses on the cold-rolled copper strip rolling process as a specific optimization target. It designs a single CMOS area array sealed and light-shielding imaging device, complemented by a lightweight image processing workflow including four-point perspective distortion correction, adaptive brightness gamma enhancement for left and right ROI partitioning, directional multi-search line edge sampling, and tilt angle constraint-improved Hough line fitting. Relying on single-camera hardware, it achieves dynamic lateral offset of the copper strip and real-time bandwidth synchronous online measurement. The entire system has a simplified hardware structure, small installation space, and requires no complex protective components. The algorithm is developed based on traditional vision operators, eliminating the need for high-performance GPU hardware support. It balances measurement accuracy, real-time performance, and equipment investment costs, effectively solving many pain points of existing technologies such as susceptibility to light, copper strip reflection, imaging distortion interference, high hardware costs, and poor adaptability. It fills the gap in dedicated dynamic position online visual measurement technology for cold-rolled copper strip and can directly connect to the rolling mill's upper-level control system to achieve closed-loop correction of deviation, possessing extremely high industrial application value and promising prospects for widespread adoption. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention aims to provide an online dynamic position measurement system for cold-rolled copper strip. This system aims to adapt to the special rolling conditions of high-reflectivity mirror surfaces in cold-rolled copper strip, eliminate perspective imaging distortion caused by tilted camera installation, automatically identify the state of the copper strip, adaptively optimize image quality by region, precisely extract copper strip edges using lightweight design, calculate the actual bandwidth and dynamic deviation of the copper strip in real time with high precision, reduce equipment installation and maintenance costs with single-CMOS array hardware, meet the real-time detection requirements of high-speed rolling without high-performance computing hardware, allow direct access of detection data to the rolling mill's upper-level shape system for closed-loop correction, minimize manual intervention throughout the process, reduce missed detections, misjudgments, and detection lag, ultimately improving copper strip yield, reducing strip breakage and shape defects, simplifying on-site deployment and maintenance, and enabling continuous online monitoring of cold-rolled copper strips of various widths.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A dynamic online position measurement system for cold-rolled copper strip specifically includes the following steps:

[0010] a: On-site cold-rolled copper strip image data acquisition and image preprocessing:

[0011] b: Based on step a above, establish a strip presence determination model:

[0012] c: Based on step a above, an adaptive image enhancement module is established by adopting a strategy of coordinated adjustment of regional contrast, gamma value, and brightness, and parameters are optimized for the left and right ROI regions of the strip respectively:

[0013] d: Based on step c above, establish an edge detection module that integrates the Canny algorithm and the improved Hough transform:

[0014] e: Based on step d above, establish a dynamic position calculation model for the strip:

[0015] f: The dynamic position detection model of cold-rolled copper strip is embedded into the upper control system of strip shape measurement and control. The dynamic position of the strip on the strip shape roller is monitored in real time during copper strip production, and the real-time bandwidth and deviation amount are fed back to the upper control system.

[0016] As a limitation of this invention, step a specifically includes the following steps:

[0017] a1: Construct a system for image data acquisition, including selecting a high-resolution, low-noise, and fast-readout CMOS industrial array camera, and a suitable area light source. Construct a sealed enclosure to house the camera and light source. Mount the camera at a specific tilt angle above the plate roller. This angle needs to be precisely calculated and repeatedly calibrated to ensure comprehensive and accurate coverage of the acquisition area on the copper strip surface, guaranteeing rich and complete image information.

[0018] a2: The intensity of the configured surface light source is finely adjusted. By adjusting the intensity distribution of the light source, the illumination of the copper strip surface is made uniform, minimizing the loss of image information caused by poor lighting conditions such as shadows or overexposure. At the same time, the sealed enclosure is optimized to ensure a stable internal environment and good light-shielding performance, preventing external stray light from interfering with the image acquisition process, thereby improving the accuracy and reliability of image acquisition.

[0019] a3: Addressing the spatial distortion caused by perspective transformation due to the camera's inherent characteristics and installation tilt angle, a perspective transformation algorithm is used to accurately correct the image, restoring it to its true spatial shape and size proportions, thus providing a high-quality, distortion-free image foundation for subsequent strip edge detection.

[0020] As a further limitation of the present invention, perspective transformation processing is performed on the acquired image. First, four fixed points in the image are selected as reference points, and the original coordinates of the four fixed points in the Cartesian coordinate system of the image are denoted as follows: , , , The target coordinates of the transformed image in Cartesian coordinates are denoted as . , , , Based on these eight points, the perspective transformation matrix is ​​obtained according to the principle of perspective transformation. Finally, perspective transformation is performed on all points in the image.

[0021] As another limitation of the present invention, step b specifically includes the following steps:

[0022] b1: Use a camera to collect images at fixed time intervals under conditions with and without copper strips, and construct image datasets dataset1 and dataset2 for the two types of working conditions respectively, to provide a basis for subsequent analysis;

[0023] b2: Based on the mill structure and copper strip path, the ROI selection tool is used to locate the areas where copper strip may appear in the image. Stable ROI areas are determined through statistical analysis to improve processing efficiency.

[0024] b3: Based on the difference in reflective properties of copper strip, calculate the average pixel value of the ROI area and compare it with the threshold. If it is higher than the threshold, it is determined that there is copper strip rolling state, otherwise it is determined that there is no copper strip state, thus realizing the accurate judgment of the presence of copper strip.

[0025] As a further limitation of the present invention, step b3 specifically includes the following steps:

[0026] b31: Calculate the mean pixel value within the ROI region of the image datasets dataset1 and dataset2, and determine the threshold μ based on the difference between the two;

[0027] b32: Calculate the baseline value of the average pixel value of the ROI region under the condition of no copper strip. The average pixel value of the ROI region when copper bands are present. ,when The presence of copper strips is detected and the detection program is triggered.

[0028] As a third limitation of the present invention, step c specifically includes the following steps:

[0029] c1: Select an ROI region on the left and an ROI region on the right of the copper strip image to ensure that the edges of copper strips of different widths are included;

[0030] c2: Perform enhancement processing on the ROI regions on both sides respectively, and the specific steps are as follows:

[0031] As a further limitation of the present invention, step c2 specifically includes the following steps:

[0032] c21: Statistic the edge strength and edge length in the ROI regions on both sides in real time, which are denoted as S and L;

[0033] c22: Adjust the gamma value, brightness value and contrast in the ROI region until a higher edge strength and a shorter edge length are obtained, which are denoted as S>s and L<l, where s and l represent the edge strength threshold and the edge length threshold respectively;

[0034] As the fourth limitation of the present invention, step d specifically includes the following steps:

[0035] d1: Use the Canny edge detection algorithm to perform copper strip edge extraction on the image after image enhancement processing.

[0036] d2: Based on the above step d1, perform edge search in the left and right ROI regions respectively;

[0037] d3: Perform Hough transform on the first edge points obtained after the above d2 processing.

[0038] As a further limitation of the present invention, step d1 specifically includes the following steps:

[0039] d11: First perform Gaussian filtering processing on the image to smooth the image;

[0040] d12: Use the Sobel operator to calculate the gradient magnitude and direction of each pixel point of the denoised image;

[0041] After the gradient and direction calculation is completed, use the linear interpolation algorithm to extract the maximum value in the gradient neighborhood, divide the image into a high threshold and a low threshold. If the gradient value of a pixel exceeds the high threshold, it is marked as a strong edge; if the gradient value is between the low threshold and the high threshold, it is marked as a weak edge; otherwise, it is eliminated;

[0042] Compare the gradient of the points from the repeated d13 process with the set double thresholds: if the value is greater than the high threshold, it is regarded as an edge point and retained; if it is less than the low threshold, it is not retained; if it is between the two thresholds, the adjacent point slightly higher than the high threshold is retained.

[0043] As a further limitation of the present invention, step d2 specifically includes the following steps:

[0044] d21: Set M search lines in the left ROI area to detect the strip edge from left to right, and perform M edge searches. Set N search lines in the right ROI area to detect the strip edge from right to left, and perform N edge searches.

[0045] d22: The first edge point determined by each search line within the left region is denoted as (x m ,y m The first edge point determined by each search line within the right-hand region is denoted as (x). n ,y n ), count the number of first edge points on the left and right sides respectively C m and C n ;

[0046] d23: If C m and C n If the respective thresholds are met, proceed to step d3; otherwise, proceed to step c.

[0047] As a further limitation of the present invention, step d3 specifically includes the following steps:

[0048] d31: First, initialize the parameter space. The parameterized representation uses the polar coordinate system line parameterization formula: ρ=xcosθ+ysinθ, where ρ represents the perpendicular distance from the line to the origin, and θ represents the angle between the line normal and the x-axis (0≤θ<π).

[0049] d32: Discretize the parameters and construct an accumulator, θ is discretized into N θ Given N intervals, ρ is discretized into N. ρ Establish a two-dimensional accumulator matrix A[ρ] for each interval. j ,θ k (Size N) ρ ×N θ ), initialized to 0;

[0050] d33: Iterate through all θ values ​​for the left edge point obtained in step d22 above, for each discrete angle θ k ∈[0,π), calculate the result ρ mk =x m cosθ k +y m sinθ k Round to the nearest discrete value ρ mj The right edge point iterates through all θ values, for each discrete angle θ. k ∈[0,π), calculate the result ρ nk =x n cosθ k +y n sinθ kRound to the nearest discrete value ρ nj Increment 1 at the corresponding position of the accumulator, A[ρ j ,θ k ]=A[ρ j ,θ k +1;

[0051] d34: Set threshold T on the left side m Search in accumulator A for A(ρ) jm ,θ km )>T m The unit is used as a parameter for the candidate line, and a threshold T is set on the right side. n Search in accumulator A for A(ρ) jn ,θ kn )>T n The elements are used as parameters for candidate lines;

[0052] d35: Reconstruct the corresponding linear equation for each detected peak point, with the left side being l. m The right side is l n Calculate the angle of inclination θ of line l;

[0053] d36: Determine whether θ satisfies a≤θ≤b, where a and b are thresholds for the angle of the line. If satisfied, retain the line l. Determine the start and end points of the line segment by the intersection of the line and the boundary of the ROI in the image, and obtain the edge line segment within the ROI area. If not satisfied, discard.

[0054] As a further limitation of the present invention, step e specifically includes the following steps:

[0055] e1: Establish a pixel coordinate system with the left side of the plate roller as the origin and the central axis as the horizontal coordinate;

[0056] e2: The pixel width of the plate roller in the image is denoted as W. roll If the actual width is W, then the actual width W of the strip in the image is... strip =W pixel_strip ×W roll / W, scaling factor P=W roll / W:

[0057] e3: Select the x-coordinates of the fitted line segment within the left ROI region at the two endpoints, one-quarter, one-half, and three-quarters of its length, and denote them as X. m0 -X m4 Select the two endpoints, one-quarter, one-half, and three-quarters of the fitted line segment within the ROI region on the right, and denote the x-coordinates as X. n0 -X n4 ;

[0058] e4: Calculate the average pixel value of the x-coordinate of each selected position. and ;

[0059] e5: Pixel distance between the strip and the left edge of the sheet roll The pixel distance between the strip and the right edge of the plate roll ;

[0060] e6: Calculate strip width Strip misalignment E=P×(D) right_edge -D lefrt_edge ) / 2, where a deviation to the left is defined as positive;

[0061] By adopting the above-described technical solution, the beneficial effects achieved by this invention compared to the prior art are as follows:

[0062] (1) The present invention uses a single CMOS area array camera with a small sealed light-shielding box, abandoning the existing binocular, multi-line array multi-camera architecture. The equipment is small in size and can be installed in the narrow space of the rolling mill frame. There is no water cooling or complex gas sealing protection structure. The cost of hardware procurement, wiring and maintenance is reduced by more than 40%, solving the defects of limited installation space and high cost of existing equipment.

[0063] (2) The present invention sets up a four-point homogeneous perspective distortion correction process to accurately compensate for the image stretching deformation caused by the tilted installation of the camera and eliminate fixed geometric measurement errors. Compared with the existing carbon steel detection equipment without correction, the edge positioning pixel error is reduced from 5~8 pixels to 1~2 pixels, and the bandwidth and deviation measurement accuracy are significantly improved.

[0064] (3) In view of the uneven reflection of the left and right mirror surfaces of cold-rolled copper strip, the present invention sets up an independent ROI partition adaptive enhancement logic for the left and right sides, and dynamically adjusts the gamma, brightness and contrast of each area in real time according to the edge intensity and edge length, so as to avoid the problem of loss of one side edge and high-gloss pseudo edge caused by the global uniform parameters, and adapts to the special working conditions of high reflectivity of copper strip. The existing carbon steel detection scheme does not have a regional collaborative enhancement mechanism.

[0065] (4) This invention automatically identifies the presence or absence of copper strip by using the average threshold of the ROI pixels of the copper strip. It automatically stops complex operations such as edge and fitting during periods without strip, which greatly reduces the computing power of the industrial control computer. It does not require high-performance GPU hardware and can be run stably by ordinary industrial control computers. Compared with the existing technology that performs full-frame operation, the processing time of a single frame is reduced by 60%, which meets the real-time detection requirements of high-speed rolling of copper strip at tens of meters per second.

[0066] (5) The present invention uses 10 search lines for partitioned orientation to extract edge points, and improves the Hough transform with a -10°~10° tilt angle constraint to filter out invalid pseudo edges generated by oil stains and roller noise, resulting in stronger edge extraction stability; the mean of multiple sampling points is used to fit the edge line, further suppressing the measurement fluctuation caused by single-point noise.

[0067] (6) The detection data of this invention is directly connected to the upper control system of the rolling mill shape through the standardized TCP / IP protocol, realizing real-time closed-loop correction of deviation, eliminating the lag of manual judgment, reducing copper strip edge damage, strip breakage, and poor strip shape, and improving the copper strip yield; at the same time, the full amount of detection data is stored locally, supporting production quality traceability, and the degree of automation is far higher than that of existing detection equipment that only displays visual information.

[0068] (7) The entire solution of this invention is based on traditional machine vision operators, does not rely on deep learning large models, has strong portability, is compatible with cold-rolled copper strip production lines of various widths from 800 to 1200 mm, has a short production line transformation cycle, and has broad industrial promotion value.

[0069] In summary, this invention addresses the industry pain points of high reflectivity in cold-rolled copper strip, complex rolling conditions, and the high cost and insufficient measurement accuracy of existing multi-camera vision inspection equipment. It innovatively adopts a lightweight hardware architecture with a single CMOS area array camera and a sealed light-shielding enclosure. This is complemented by a complete integrated technical solution including perspective distortion correction, adaptive image enhancement with left and right ROI regions, intelligent strip presence / absence detection, directional search line combined with angle constraint-improved Hough transform, and closed-loop linkage of the rolling mill shape system. This effectively overcomes multiple shortcomings of traditional carbon steel-specific inspection devices, such as poor adaptability, large geometric measurement errors, high computational power consumption, and inability to automatically close-loop corrective actions. It balances hardware investment costs, online inspection accuracy, high-speed real-time computing performance, and industrial intelligent control requirements. The overall solution is easy to deploy, maintain, and highly versatile, significantly reducing strip breakage and defective scrap rates in copper strip rolling, and improving the automation level and overall production efficiency of the production line.

[0070] This invention is applicable to various continuous rolling production lines for cold-rolled copper strip with a width of 800~1200mm. It is used for fully automatic online real-time measurement of the dynamic lateral position and actual width of the strip during the cold-rolled copper strip rolling process, and outputs quantitative deviation data to drive the rolling mill to complete closed-loop deviation correction control. Attached Figure Description

[0071] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0072] Figure 1 Diagram of the preprocessing process for raw image data;

[0073] Figure 2 Select ROI regions with and without images;

[0074] Figure 3 Strip edge ROI region selection and image enhancement;

[0075] Figure 4 Schematic diagram of edge strength and edge length;

[0076] Figure 5 Schematic diagram of point-line duality of Hough transform;

[0077] Figure 6 Canny edge detection flowchart;

[0078] Figure 7 Edge line fitting plot;

[0079] Figure 8 Flowchart of the detection procedure start;

[0080] Figure 9 Schematic diagram of strip location and width calculation

[0081] Figure 10 Schematic diagram of a position detection system. Detailed Implementation

[0082] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0083] Example: An online dynamic position measurement system for cold-rolled copper strip

[0084] like Figures 1 to 10 As shown, this embodiment relates to an online dynamic position measurement system for cold-rolled copper strip, which includes the following execution steps:

[0085] a: On-site cold-rolled copper strip image data acquisition and image preprocessing:

[0086] a1: Construct a system for image data acquisition, including selecting a high-resolution, low-noise, and fast-readout CMOS industrial array camera, and a suitable LED surface light source. Construct a sealed enclosure (248mm × 248mm × 250mm) to house the camera and light source. The camera will be mounted diagonally above the plate roller, at a horizontal distance of 300mm and a vertical height of 800mm. This distance requires precise calculation and repeated calibration to ensure comprehensive and accurate coverage of the copper strip surface acquisition area, guaranteeing rich and complete image information.

[0087] a2: The light source needs to illuminate a copper strip with a straight-line distance of approximately 1.2m. The intensity of the configured surface light source is adjusted to ensure uniform illumination of the copper strip surface by adjusting the intensity distribution. Simultaneously, the sealed enclosure is optimized to ensure a stable internal environment and good light-shielding performance, preventing external stray light from interfering with the image acquisition process, thereby improving the accuracy and reliability of image acquisition.

[0088] a3: Addressing the perspective distortion caused by the camera's inherent characteristics and mounting tilt angle, a perspective transformation algorithm is used to accurately correct the image, restoring it to its true spatial shape and size proportions. This provides a high-quality, distortion-free image foundation for subsequent strip edge detection. Specifically, this includes the following steps:

[0089] a31: Perform perspective transformation on the acquired image. First, select four fixed points in the image as reference points, and determine the original coordinates of these four fixed points in the Cartesian coordinate system of the image, denoted as [reference points]. , , , The target coordinates of the transformed image in Cartesian coordinates are denoted as . , , , Based on these eight points, the perspective transformation matrix is ​​obtained according to the principle of perspective transformation. The perspective transformation process is as follows: First, the original coordinate points in the Cartesian coordinate system are transformed... and target coordinates Add a coordinate dimension to each coordinate system and transform them to a homogeneous coordinate system, denoted as . and The transformation relationship is denoted as For ease of subsequent calculations, let Then the original coordinate points in the homogeneous coordinate system can be transformed into Then, according to the perspective transformation formula... ( This is the perspective transformation matrix. And generally make ) can be obtained According to the formula It can be obtained ,Right now Based on the given 4 original coordinate points and 4 target coordinate points, 8 systems of equations can be obtained. The perspective transformation matrix can then be obtained. The eight positional parameters are used to perform perspective transformation on all points in the image.

[0090] a32: Perform grayscale processing on the image after perspective transformation, and use the grayscale image for all subsequent processing.

[0091] b: Establish a strip presence determination model based on the above step a, the specific steps include:

[0092] b1: Use a camera to collect images in the states with copper strip and without copper strip at fixed time intervals, respectively construct image datasets dataset1 and dataset2 under the two types of working conditions, to provide a basis for subsequent analysis;

[0093] b2: According to the rolling mill structure and the copper strip path, use a ROI selection tool to locate the area where the copper strip may appear in the image, and determine the stable ROI area through statistical analysis to improve processing efficiency;

[0094] b3: Based on the difference in the reflective characteristics of the copper strip, calculate the mean pixel value of the ROI area and compare it with a threshold. If the value is higher than the threshold, it is determined as the copper strip rolling state, otherwise it is the no-copper-strip state, so as to realize accurate judgment of the presence state of the copper strip, which specifically includes the following steps:

[0095] b31: Statistically calculate the mean pixel values in the ROI areas of the image datasets dataset1 and dataset2, and determine the threshold of 50 according to the difference between the two;

[0096] b32: Calculate that the mean pixel value of the ROI area in the no-copper-strip state is in the range of 32 to 56, and the mean pixel value of the ROI area in the copper strip presence state is in the range of 120 to 200. When the difference between the two is greater than the set threshold 50, the presence of the copper strip is identified and the detection procedure is triggered.

[0097] c: Based on the above step a, adopt a coordinated adjustment strategy of zonal contrast, gamma value and brightness to establish an adaptive image enhancement module, and perform parameter optimization on the ROI areas on the left and right sides of the strip respectively, the specific steps include:

[0098] c1: Select one ROI area on the left side of the copper strip image and one ROI area on the right side, to ensure that the edges of copper strips of different widths can be included;

[0099] c2: Perform enhancement processing on the ROI areas on both sides respectively, the specific steps are as follows:

[0100] c21: Statistically calculate the edge intensity and edge length in the ROI areas on both sides in real time, denoted as S and L;

[0101] c22: Adjust the gamma value, brightness value and contrast in the ROI area until a higher edge intensity and a shorter edge length are obtained, which are denoted as S>s and L<l, where s and l represent the edge intensity threshold and edge length threshold respectively, here the thresholds s and l are normalized and set to 0.8 and 0.2 respectively;

[0102] d: Based on step c above, establish an edge detection module integrating the Canny algorithm and the improved Hough transform. The specific steps are as follows:

[0103] d1: The Canny edge detection algorithm is used to extract copper strip edges from the enhanced image. The specific steps are as follows:

[0104] d11: First, Gaussian filtering is applied to the image to smooth it.

[0105] d12: Using the Sobel operator, the gradient magnitude and direction of each pixel in the denoised image are calculated. Calculate the gradient of the image and direction , respectively ;

[0106] d13: After the gradient and direction calculations are completed, the maximum value in the gradient neighborhood is extracted using a linear interpolation algorithm. The image is then divided into high-threshold and low-threshold categories. If the gradient value of a pixel exceeds the high threshold, it is marked as a strong edge. If the gradient value is between the low and high thresholds, it is marked as a weak edge. Otherwise, it is eliminated.

[0107] d14: The gradient of the point when repeating d13 is compared with the set double threshold. If the value is greater than the high threshold, it is considered an edge point and is retained. If it is less than the low threshold, it is not retained. If it is between the two, it is retained as a neighboring point with a value slightly higher than the high threshold.

[0108] d2: Based on step d1 above, perform edge search within the left and right ROI regions respectively. The specific steps are as follows:

[0109] d21: Set 10 search lines in the left ROI area to detect the strip edge from left to right, and perform 10 edge searches. Set 10 search lines in the right ROI area to detect the strip edge from right to left, and perform 10 edge searches.

[0110] d22: The first edge point determined by each search line within the left region is denoted as (x m ,y m The first edge point determined by each search line within the right-hand region is denoted as (x). n ,y n ), count the number of first edge points on the left and right sides respectively C m and C n ;

[0111] d23: If C m and C n If the threshold of 10 is met, proceed to step d3; otherwise, proceed to step c.

[0112] d3: Perform a Hough transform on the first edge point obtained after processing d2 above. The specific steps are as follows:

[0113] d31: First, initialize the parameter space. The parameterized representation uses the polar coordinate system line parameterization formula: ρ=xcosθ+ysinθ, where ρ represents the perpendicular distance from the line to the origin, and θ represents the angle between the line normal and the x-axis (0≤θ<π).

[0114] d32: Discretize the parameters and construct an accumulator, θ is discretized into N θ Given N intervals, ρ is discretized into N. ρ Establish a two-dimensional accumulator matrix A[ρ] for each interval. j ,θ k (Size N) ρ ×N θ ), initialized to 0;

[0115] d33: Iterate through all θ values ​​for the left edge point obtained in step d22 above, for each discrete angle θ k ∈[0,π), calculate the result ρ mk =x m cosθ k +y m sinθ k Round to the nearest discrete value ρ mj The right edge point iterates through all θ values, for each discrete angle θ. k ∈[0,π), calculate the result ρ nk =x n cosθ k +y n sinθ k Round to the nearest discrete value ρ nj Increment 1 at the corresponding position of the accumulator, A[ρ j ,θ k ]=A[ρ j ,θ k +1;

[0116] d34: Set threshold T on the left side m Search in accumulator A for A(ρ) jm ,θ km )>T m The unit is used as a parameter for the candidate line, and a threshold T is set on the right side. n Search in accumulator A for A(ρ) jn ,θ kn )>T n The elements are used as parameters for candidate lines;

[0117] d35: Reconstruct the corresponding linear equation for each detected peak point, with the left side being l. m The right side is l n Calculate the angle of inclination θ of line l;

[0118] d36: Determine if θ satisfies -10°≤θ≤10°. If it does, keep the line l. Determine the start and end points of the line segment through the intersection of the line and the boundary of the ROI in the image to obtain the edge line segment within the ROI area. If it does not satisfy the condition, discard it.

[0119] e: Based on step d above, establish a dynamic position calculation model for the strip. The specific steps are as follows:

[0120] e1: Establish a pixel coordinate system with the left side of the plate roller as the origin and the central axis as the horizontal coordinate;

[0121] e2: The pixel width of the plate roller in the image is denoted as W. roll =1200 (pixels), the actual width is W=800mm, then the actual width W of the strip in the image strip =W pixel_strip ×W roll / W=900 (pixels), scaling factor P=W roll / W=1.5;

[0122] e3: Select the x-coordinates of the fitted line segment within the left ROI region at the two endpoints, one-quarter, one-half, and three-quarters of its length, and denote them as X. m0 -X m4 Select the two endpoints, one-quarter, one-half, and three-quarters of the fitted line segment within the ROI region on the right, and denote the x-coordinates as X. n0 -X n4 ;

[0123] e4: Calculate the average pixel value of the x-coordinate of the selected position. and ;

[0124] e5: Pixel distance between the strip and the left edge of the sheet roll The pixel distance between the strip and the right edge of the plate roll ;

[0125] e6: Calculate strip width Strip deviation E = P × (D) right_edge -D lefrt_edge ) / 2, where a deviation to the left is defined as positive;

[0126] f: The dynamic position detection model of cold-rolled copper strip is embedded into the upper control system of strip shape measurement and control. The dynamic position of the strip on the strip shape roller is monitored in real time during copper strip production, and the real-time bandwidth and deviation amount are fed back to the upper control system.

[0127] It should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still modify the technical solutions described in the above embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic online position measurement system for cold-rolled copper strip, characterized in that: Specifically, the following steps are included: a: On-site cold-rolled copper strip image data acquisition and image preprocessing: b: Based on step a above, establish a strip presence determination model: c: Based on step a above, an adaptive image enhancement module is established by adopting a strategy of coordinated adjustment of regional contrast, gamma value, and brightness, and parameters are optimized for the left and right ROI regions of the strip respectively: d: Based on step c above, establish an edge detection module that integrates the Canny algorithm and the improved Hough transform: e: Based on step d above, establish a dynamic position calculation model for the strip: f: The dynamic position detection model of cold-rolled copper strip is embedded into the upper control system of strip shape measurement and control. The dynamic position of the strip on the strip shape roller is monitored in real time during copper strip production, and the real-time bandwidth and deviation amount are fed back to the upper control system.

2. The online dynamic position measurement system for cold-rolled copper strip according to claim 1, characterized in that: Step a specifically includes the following steps: a1: Construct a system for image data acquisition, including selecting a high-resolution, low-noise, and fast-readout CMOS industrial array camera, and a suitable area light source. Construct a sealed enclosure to house the camera and light source. Mount the camera at a specific tilt angle above the plate roller. This angle needs to be precisely calculated and repeatedly calibrated to ensure comprehensive and accurate coverage of the acquisition area on the copper strip surface, guaranteeing rich and complete image information. a2: The intensity of the configured surface light source is finely adjusted. By adjusting the intensity distribution of the light source, the illumination of the copper strip surface is made uniform, minimizing the loss of image information caused by poor lighting conditions such as shadows or overexposure. At the same time, the sealed enclosure is optimized to ensure a stable internal environment and good light-shielding performance, preventing external stray light from interfering with the image acquisition process, thereby improving the accuracy and reliability of image acquisition. a3: Addressing the spatial distortion caused by perspective transformation due to the camera's inherent characteristics and installation tilt angle, a perspective transformation algorithm is used to accurately correct the image, restoring it to its true spatial shape and size proportions, thus providing a high-quality, distortion-free image foundation for subsequent strip edge detection.

3. The online dynamic position measurement system for cold-rolled copper strip according to claim 2, characterized in that: The acquired image is subjected to perspective transformation. First, four fixed points in the image are selected as reference points, and the original coordinates of the four fixed points in the Cartesian coordinate system of the image are denoted as follows: , , , The target coordinates of the transformed image in Cartesian coordinates are denoted as . , , , Based on these eight points, the perspective transformation matrix is ​​obtained according to the principle of perspective transformation. Finally, perspective transformation is performed on all points in the image.

4. The online dynamic position measurement system for cold-rolled copper strip according to claim 1, characterized in that: Step b specifically includes the following steps: b1: Use a camera to collect images at fixed time intervals under conditions with and without copper strips, and construct image datasets dataset1 and dataset2 for the two types of working conditions respectively, to provide a basis for subsequent analysis; b2: Based on the mill structure and copper strip path, the ROI selection tool is used to locate the areas where copper strip may appear in the image. Stable ROI areas are determined through statistical analysis to improve processing efficiency. b3: Based on the difference in reflective properties of copper strip, calculate the average pixel value of the ROI area and compare it with the threshold. If it is higher than the threshold, it is determined that there is copper strip rolling state, otherwise it is determined that there is no copper strip state, thus realizing the accurate judgment of the presence of copper strip.

5. The online dynamic position measurement system for cold-rolled copper strip according to claim 1, characterized in that: Step c specifically includes the following steps: c1: Select an ROI region on the left and an ROI region on the right of the copper strip image to ensure that the edges of copper strips of different widths are included; c2: Enhance the ROI regions on both sides respectively.

6. The online dynamic position measurement system for cold-rolled copper strip according to claim 1, characterized in that: Step d specifically includes the following steps: d1: The Canny edge detection algorithm is used to extract copper strip edges from the image after image enhancement. d2: Based on step d1 above, perform edge search within the left and right ROI regions respectively; d3: Perform Hough transform on the first edge point obtained after processing d2 above.

7. The online dynamic position measurement system for cold-rolled copper strip according to claim 6, characterized in that: Step d1 specifically includes the following steps: d11: First, Gaussian filtering is applied to the image to smooth it. d12: Using the Sobel operator, the gradient magnitude and direction of each pixel in the denoised image are calculated; d13: After the gradient and direction calculations are completed, the maximum value in the gradient neighborhood is extracted using a linear interpolation algorithm. The image is then divided into high-threshold and low-threshold categories. If the gradient value of a pixel exceeds the high threshold, it is marked as a strong edge. If the gradient value is between the low and high thresholds, it is marked as a weak edge. Otherwise, it is eliminated. d14: The gradient of the point when repeating d13 is compared with the set double threshold. If the value is greater than the high threshold, it is considered an edge point and is retained. If it is less than the low threshold, it is not retained. If it is between the two, it is retained as a neighboring point with a value slightly higher than the high threshold.

8. The online dynamic position measurement system for cold-rolled copper strip according to claim 6, characterized in that: Step d2 specifically includes the following steps: d21: Set M search lines in the left ROI area to detect the strip edge from left to right, and perform M edge searches. Set N search lines in the right ROI area to detect the strip edge from right to left, and perform N edge searches. d22: The first edge point determined by each search line within the left region is denoted as (x m ,y m The first edge point determined by each search line within the right-hand region is denoted as (x). n ,y n ), count the number of first edge points on the left and right sides respectively C m and C n ; d23: If C m and C n If the respective thresholds are met, proceed to step d3; otherwise, proceed to step c.

9. The online dynamic position measurement system for cold-rolled copper strip according to claim 6, characterized in that: Step d3 specifically includes the following steps: d31: First, initialize the parameter space. The parameterized representation uses the polar coordinate system line parameterization formula: ρ=xcosθ+ysinθ, where ρ represents the perpendicular distance from the line to the origin, and θ represents the angle between the line normal and the x-axis (0≤θ<π). d32: Discretize the parameters and construct an accumulator, θ is discretized into N θ Given N intervals, ρ is discretized into N. ρ Establish a two-dimensional accumulator matrix A[ρ] for each interval. j ,θ k (Size N) ρ ×N θ ), initialized to 0; d33: Iterate through all θ values ​​for the left edge point obtained in step d22 above, for each discrete angle θ k ∈[0,π), calculate the result ρ mk =x m cosθ k +y m sinθ k Round to the nearest discrete value ρ mj The right edge point iterates through all θ values, for each discrete angle θ. k ∈[0,π), calculate the result ρ nk =x n cosθ k +y n sinθ k Round to the nearest discrete value ρ nj Increment 1 at the corresponding position of the accumulator, A[ρ j ,θ k ]=A[ρ j ,θ k +1; d34: Set threshold T on the left side m Search in accumulator A for A(ρ) jm ,θ km )>T m The unit is used as a parameter for the candidate line, and a threshold T is set on the right side. n Search in accumulator A for A(ρ) jn ,θ kn )>T n The elements are used as parameters for candidate lines; d35: Reconstruct the corresponding linear equation for each detected peak point, with the left side being l. m The right side is l n Calculate the angle of inclination θ of line l; d36: Determine whether θ satisfies a≤θ≤b, where a and b are thresholds for the angle of the line. If satisfied, retain the line l. Determine the start and end points of the line segment by the intersection of the line and the boundary of the ROI in the image, and obtain the edge line segment within the ROI area. If not satisfied, discard.

10. The online dynamic position measurement system for cold-rolled copper strip according to any one of claims 1-4, characterized in that: Step e specifically includes the following steps: e1: Establish a pixel coordinate system with the left side of the plate roller as the origin and the central axis as the horizontal coordinate; e2: The pixel width of the plate roller in the image is denoted as W. roll If the actual width is W, then the actual width W of the strip in the image is... strip =W pixel_strip ×W roll / W, scaling factor P=W roll / W: e3: Select the x-coordinates of the fitted line segment within the left ROI region at the two endpoints, one-quarter, one-half, and three-quarters of its length, and denote them as X. m0 -X m4 Select the two endpoints, one-quarter, one-half, and three-quarters of the fitted line segment within the ROI region on the right, and denote the x-coordinates as X. n0 -X n4 ; e4: Calculate the average pixel value of the x-coordinate of the selected position. and ; e5: Pixel distance between the strip and the left edge of the sheet roll The pixel distance between the strip and the right edge of the plate roll ; e6: Calculate strip width Strip deviation Here, it is defined as veering to the left as positive.

Images