Thermal-state high-speed wire rod surface defect detection method

Abstract

The invention discloses a method for detecting surface defects of a thermal-state high-speed wire rod. The surface defect detection method comprises the steps that a thermal-state high-speed wire rod surface defect detection system is arranged, the thermal-state high-speed wire rod surface defect detection system comprises an imaging system, the imaging system comprises a surrounding type camera combination, and the surrounding type camera combination is arranged on an advancing route of a high-speed wire rod (2); in the rolling production process of the high-speed wire rod, the surrounding type camera combination continuously shoots and images the high-speed wire rod to obtain a wire rod surface image; the method comprises the following steps: preprocessing an obtained wire rod surface image; carrying out defect positioning on the preprocessed surface image of the wire rod to obtain a defect to be judged; and the preliminarily determined to-be-determined defect is subjected to counterfeit removal and true storage processing, and finally the true defect in the wire rod surface image is determined. According to the surface defect detection method, folding, scabbing, scratching and indentation on the surface of the wire can be effectively detected, and a defect generation area is effectively locked.

Description

Technical Field

[0001] The invention relates to a steel surface defect detection technology, and in particular to a hot high-speed wire surface defect detection method. Background Art

[0002] At present, high-end wire products have covered varieties such as cold heading steel, spring steel, bearing steel, and cord steel. With the continuous expansion and increase in the use of wire spring steel and cold heading steel in the automotive market, users have put forward higher quality requirements on the wire surface. Therefore, in the wire production process, the wire surface quality must be controlled, which requires comprehensive and strict detection of wire surface defects.

[0003] At present, the detection of wire surface defects is mainly achieved by manual visual observation. Not only is the efficiency of defect detection low, but the error rate and missed detection rate are also high. It is even more impossible to quantify the number of defects on the wire surface, and thus it is impossible to effectively judge the severity of defects in the finished wire and the final number of defects, and it is unable to meet the user's requirements for wire surface quality. Summary of the invention

[0004] The object of the present invention is to provide a method for detecting surface defects of hot high-speed wires, which can effectively detect defects such as folding, scarring, scratches, and indentations on the surface of the wires.

[0005] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0006] A method for detecting surface defects of hot high-speed wire, comprising:

[0007] S1, setting a hot high-speed wire surface defect detection system, wherein the set hot high-speed wire surface defect detection system includes an imaging system, and the imaging system includes a surround camera combination, and the surround camera combination is set on the travel route of the high-speed wire;

[0008] S2, during the high-speed wire rolling production process, the surround camera combination continuously shoots and images the high-speed wire to obtain the wire surface image;

[0009] S3, preprocessing the acquired wire surface image;

[0010] S4, performing defect location on the pre-processed wire surface image to obtain the defect to be determined;

[0011] S5, performing a process of eliminating false defects and retaining true defects on the initially determined defects to be determined, and finally determining the true defects in the wire surface image.

[0012] Furthermore, the process of eliminating false and retaining true for the initially determined defects to be determined includes:

[0013] S51, for the defect to be determined, the number of pixels whose grayscale value is greater than a certain value in the ROI area is counted, and then the ratio to the total number of pixels is calculated. When the ratio falls within a range, the defect to be determined is determined to be a true defect;

[0014] S52, for the defect to be determined, the average gray value of the defect in the ROI area and the average gray value of the non-defect area are calculated. If the difference between the two exceeds a certain range, the area is smaller than a certain range, the shape is close to a square, and the minimum gray value and the average gradient meet certain conditions, then the defect to be determined is determined to be a true defect;

[0015] S53, for the defect to be determined, the number of defects in the current ROI area is counted, and the distance between the defects is determined by standard deviation calculation. At the same time, it is determined whether the gray value of the defect area in the ROI is less than the gray value of the remaining area. If the distance between the defects is approximately equal, and the gray value of the defect area in the ROI is less than the gray value of the remaining area, the defect to be determined is determined to be a true defect;

[0016] S54, for the defect to be determined, a longitudinal expansion operation is performed on the edge map to connect the discontinuous lines, and then the aspect ratio, the grayscale mean of the defect and the grayscale mean of the non-defect are calculated. If a certain condition is met, it is a longitudinal defect, and then the gradient size is used to determine whether it is a black line. If it is determined to be a longitudinal defect or a black line, the defect to be determined is determined to be a true defect;

[0017] S55, for the defect to be determined, the grayscale contrast of the diagonal line is calculated on the edge map, and then the aspect ratio, the grayscale mean of the four corners of the ROI, the grayscale mean of the defect and the grayscale mean of the non-defect, and the number of defects in the ROI are calculated. If they meet certain conditions, it is determined to be an edge line, that is, the defect to be determined is determined to be a true defect;

[0018] S56, for the defect to be determined, the four corners of the ROI area, as well as the left 1 / 4 and right 1 / 4 areas of the ROI area are respectively intercepted, and the grayscale means of the two diagonal areas and the left and right 1 / 4 areas are respectively calculated. The ratio of the two can approximately describe the contrast between the left and right sides of the current ROI. The contrast of the three groups of areas is comprehensively considered. When this value is within a range, it is considered that the defect is a roller seam, that is, the defect to be determined is determined to be a true defect.

[0019] Furthermore, the image is preprocessed, including: edge trimming, illumination balancing, and smoothing and denoising.

[0020] Furthermore, the trimming is to extract the effective area to be detected, remove the background area, and use the gray gradient threshold search method to accurately determine the boundary of the finished product;

[0021] Due to the large variation range of finished product diameters and the noisy on-site environment, the imaging effects of different specifications or cameras vary greatly, which is directly manifested as uneven image brightness. It is necessary to balance the illumination, improve the overall brightness, and compensate the grayscale average of each column to 128;

[0022] Since there is a lot of background noise in the image to be detected, most of the noise is manifested as grayscale mutations, which will cause great interference to the detection. Therefore, it is necessary to perform noise reduction processing on it. Gaussian blur is selected, and the kernel size of Gaussian blur is determined by the parameter GaussSize.

[0023] Furthermore, the defect location includes Sobel edge finding, morphological opening and closing operations, filtering of small connected domains, and ROI search and merging.

[0024] Furthermore, the Sobel edge finding part uses the X-direction Sobel kernel and the Y-direction Sobel kernel. The Sobel X-direction and Y-direction kernels can respectively identify edge mutations in the X and Y directions, and thus can accurately describe edge conditions in the X and Y directions.

[0025] There are many noise points in the binarized result image, which will affect the final defect calibration. The two binary images are respectively subjected to morphological opening and closing operations to remove the noise points and weaken the influence of the noise, and then the two binary images are fused to obtain a fused binary image.

[0026] Since most wire surfaces are covered with iron oxide scale, there will still be some small dots in the binary image after morphological denoising. These small dots that are prone to false detection need to be filtered out.

[0027] The method used is to first obtain the number of pixels with a pixel value of 255 in each ROI area, and then compare it with the set small point threshold pointthres. The area below the threshold will be filtered out;

[0028] ROI (Region of interest) is the suspected defect area required by the algorithm. This step includes three steps: ROI search, merging, and screening.

[0029] Furthermore, the hot high-speed wire surface defect detection system also includes mechanical equipment, electrical systems, computer systems and software systems.

[0030] Further,

[0031] The mechanical equipment is used to fix the optoelectronic equipment, realize the focus adjustment of the inspection objects of different specifications to ensure accurate imaging parameters, and at the same time provide thermal protection for the equipment to adapt to the harsh and complex working conditions of the production line;

[0032] The electrical system is used to supply power to the entire system, synchronize the wire running speed with the camera acquisition speed, and access various production control signals;

[0033] The computer system includes an image acquisition and processing array, a system server, a database storage device, an operation terminal and a 10 Gigabit network switch;

[0034] The software system is used to control the camera to collect and acquire images for high-speed processing. Through defect detection algorithm research, it can eliminate defect-free backgrounds from massive data and save and extract defective surface images.

[0035] The surface defect detection method of the present invention aims to develop an economical and efficient hot high-speed wire rod detection method and equipment according to the variety and specification of hot high-speed wire rod rolling, so as to realize the real-time detection of online defects of hot high-speed wire rod. The developed hot high-speed wire rod defect detection system replaces the hot-rolled wire rod pickling process, takes green process as the main task, reduces environmental pollution, and actively meets the increasingly stringent environmental protection regulations.

[0036] Compared with the prior art, the surface defect detection method of the present invention has the following beneficial effects: it can effectively detect defects such as folding, scarring, scratches, indentations, etc. on the surface of the wire, effectively lock the defect generation area, and provide an effective means for timely investigation and effective resolution of on-site problems, and provide an objective and quantitative means for early preventive adjustment of on-site quality, quantification of product quality defects, and product grading judgment and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of a surround camera assembly involved in the method for detecting surface defects of hot high-speed wire rods of the present invention, wherein the high-speed wire rod shown in the figure is a diagram of its cross section;

[0038] Figure 2 A diagram of the computer system architecture involved in the surface defect detection method of the present invention;

[0039] Figure 3 A schematic diagram of white iron oxide scale on a high-speed wire involved in the surface defect detection method of the present invention;

[0040] Figure 4 A schematic diagram of dot-shaped black spots on a high-speed wire involved in the surface defect detection method of the present invention;

[0041] Figure 5A schematic diagram of zebra print black spots on a high-speed wire involved in the surface defect detection method of the present invention;

[0042] Figure 6 A schematic diagram of a black line on a high-speed wire involved in the surface defect detection method of the present invention;

[0043] Figure 7 A schematic diagram of an edge line on a high-speed wire involved in the surface defect detection method of the present invention;

[0044] Figure 8 A schematic diagram of a roller seam on a high-speed wire material involved in the surface defect detection method of the present invention;

[0045] Fig. 9 It is a schematic diagram of establishing different surface cleanliness groups of rolled products according to material grouping in the surface defect detection method of the present invention.

[0046] In the picture: 1- high-speed camera, 2- high-speed wire. DETAILED DESCRIPTION

[0047] The present invention will be further described below in conjunction with specific embodiments:

[0048] This embodiment provides a method for detecting surface defects of hot high-speed wires, which can effectively detect defects such as folding, scarring, scratches, and indentations on the surface of the wires.

[0049] It should be noted that the defects mentioned in this article refer to the abnormal surface morphology of the high-speed wire caused by various factors such as equipment, process, environment, etc. during the production process of the high-speed wire, and these defects will have a certain impact on the quality of downstream products. In this embodiment, the surface defects of the high-speed wire include scars, cracks, scratches, indentations, abrasions, ears, etc.

[0050] The surface defect detection method of this embodiment includes the following steps S1 to S5.

[0051] S1, first of all, the hardware preparation, as follows:

[0052] A hot high-speed wire surface defect detection system is set up, and the hot high-speed wire surface defect detection system includes an imaging system, mechanical equipment, an electrical system, a computer system and a software system.

[0053] The imaging system includes a surround camera assembly, a light source, a filter and a lens, and is used to obtain a full-circumference high-definition digital image of the wire surface at high speed for subsequent system processing and analysis.

[0054] The surround camera assembly is arranged on the travel route of the high-speed wire 2 .

[0055] See also Figure 1 The surround camera combination is composed of a combination of several high-speed cameras 1, which are arranged around the travel route of the high-speed wire 2. The several high-speed cameras 1 are evenly and discretely distributed around the travel route of the high-speed wire. Each high-speed camera 1 in the surround camera combination faces the travel route of the high-speed wire to perform high-speed photography, so that the high-speed wire 2 traveling at high speed can be photographed at 360 degrees, and the complete surface image of the high-speed wire 2 passing through the surround camera combination can be captured. In other words, the complete surface image of the high-speed wire 2 can be obtained through the surround camera combination.

[0056] The core function of the mechanical equipment is to fix the optoelectronic equipment, realize the focus adjustment of the inspection objects of different specifications to ensure accurate imaging parameters, and at the same time provide thermal protection for the equipment to adapt to the harsh and complex working conditions of the production line.

[0057] Specifically, the mechanical equipment is mainly composed of a mobile trolley, a camera box, a camera assembly, a camera group mounting seat, a cooling system and a guide groove. The trolley consists of four parts: a bottom plate, a slider, a guide rail and a limit block. The camera protection box is welded from stainless steel plates and is used to install and fix various system components (camera group, light source, etc.) and serve as a protective shell for the system. The inner wall of the protection box is pasted with an aluminum silicate fiber blanket for thermal insulation. The camera group mounting seat is driven by a motor to move up and down according to the diameter data of the inspected rolled piece until the central axis of the annular camera group is consistent with the axis of the steel bar, and each camera adjusts its focus under the drive of its own displacement stage. The camera box and the steel guide sleeve are designed to be separated, which effectively avoids the influence of mechanical vibration caused by steel passing on camera imaging. The system camera box adopts air conditioning internal circulation cooling. The air conditioner is installed on the trolley and connected to the camera box with a pipe. The inside of the camera box is welded with rectangular steel into a ventilation duct. Several ventilation holes are evenly distributed on one side of the ventilation duct to ensure that the air conditioning cold air is evenly delivered to all parts of the camera box.

[0058] The electrical system is mainly responsible for powering the entire system, synchronizing the wire running speed with the camera acquisition speed, and accessing various production control signals (such as HMD signals, speed signals, etc.).

[0059] Specifically,

[0060] The electrical system includes a transmission system and a PLC system, integrating corresponding mechanical equipment, light sources, imaging equipment, etc. In order to ensure seamless connection of various types of equipment and complete the set functions, the electrical integration will use special interface equipment to ensure the stability and reliability of long-distance, high-speed, and massive image data transmission. Special cables, optical cables, and connection equipment are selected to ensure the data connection between image acquisition equipment such as cameras and surface detection components. The design of special power cables for light sources ensures the effectiveness of long-distance power transmission between the on-site installation location and the constant current devices in the power distribution control cabinet. Before the wire reaches the detection position, the L2 system sends the wire information (specifications, steel types, materials, etc.) to the server computer in the form of a telegram through the network. The server computer calculates the corresponding movement distance of the camera horizontal slider and the distance of the camera group mounting seat moving up and down according to the diameter of the steel pipe, and sends it to the PLC through DP. The PLC sends the corresponding signal to control the camera and the camera group mounting seat to reach the specified position for shooting.

[0061] See also Figure 2 The computer system is mainly composed of image acquisition and processing array (IPU), system server, database storage device, operation terminal, 10 Gigabit network (Gigabit network) switch and other processing systems and network equipment. It is the operating platform and human-computer interaction interface of the core software of the entire system.

[0062] Specifically,

[0063] The computer system mainly includes the camera acquisition and calculation processing unit IPA, system server, data array, operation terminal, system switch, etc. The camera acquisition and calculation processing unit is responsible for the image acquisition, image processing and classification operation of each camera. The system server and data server are responsible for data integration, debugging, database and network management. The operation terminal is responsible for communicating with the system server and displaying HMI information.

[0064] The core functions of the software system include controlling the camera to collect and acquire images for high-speed processing, removing defect-free backgrounds from massive data through defect detection algorithm research, and saving and extracting defective surface images.

[0065] Specifically,

[0066] The software system is the core of the hot high-speed wire surface defect detection system, and defect detection is the core of the software system. After the images obtained by the system acquisition unit are processed by detection, the defect images and data are summarized to the system server, and the server is responsible for distributing the defect information to each operation terminal for display to the user.

[0067] S2, during the high-speed wire rolling production process, the rolled high-speed wire passes through the surround camera assembly, and the surround camera assembly continuously photographs and images the high-speed wire, thereby obtaining a continuous and complete wire surface image.

[0068] S3, for the acquired wire surface image, the image is preprocessed first, including:

[0069] Perform edge trimming on the image to extract the effective area to be detected and remove the background area;

[0070] Perform illumination balance on the image to make the image moderately bright and dark;

[0071] Perform smoothing and denoising on the image to remove noise in the image;

[0072] In summary, the purpose of preprocessing the wire surface image is to make the graphic elements (or "graphic features") in the wire surface image as clear and discernible as possible.

[0073] More specifically,

[0074] The trimming is to extract the effective area to be detected, remove the background area, and use the gray gradient threshold search method to accurately determine the boundary of the finished product. Due to the large range of diameter variation of the finished product, the noisy on-site environment, and the large differences in imaging effects of different specifications or different cameras, it is directly manifested as uneven image brightness. It is necessary to balance the illumination, improve the overall brightness, and compensate the grayscale average of each column to 128. Since there is a lot of background noise in the image to be detected, this noise is mostly manifested as a sudden change in grayscale, which will cause great interference to the detection. Therefore, it is necessary to perform noise reduction processing here, and Gaussian blur is selected. The kernel size of Gaussian blur is determined by the parameter GaussSize.

[0075] It should be noted that the aforementioned edge trimming, illumination balancing and smoothing denoising are all image processing techniques in the prior art, and their specific implementation processes are common knowledge known to those skilled in the art.

[0076] S4, performing defect location on the pre-processed wire surface image, thereby preliminarily determining the positions of all possible defective patterns in the wire surface image, or in simpler terms, preliminarily screening out the defects in the wire surface image.

[0077] The defect location includes Sobel edge finding, morphological opening and closing operations, filtering of small connected domains, ROI search and merging and other processes, and mainly locates the position of the suspected defect part through the edge gradient operator, mainly including Sobel edge finding, morphological opening and closing operations, filtering of small connected domains, ROI search and merging and other processes.

[0078] Sobel uses the X-direction Sobel kernel and the Y-direction Sobel kernel to find the edge. The Sobel X-direction and Y-direction kernels can identify edge mutations in the X and Y directions respectively, so they can accurately describe the edge conditions in the X and Y directions.

[0079] There are many noise points in the binarized result image, which will affect the final defect calibration. The two binary images are respectively subjected to morphological opening and closing operations to remove the noise points and weaken the influence of the noise, and then the two binary images are fused to obtain a fused binary image.

[0080] Since most wire surfaces are covered with iron oxide scale, there will still be some small dots in the binary image after morphological denoising. These small dot areas that are prone to false detection need to be filtered out.

[0081] The method used is to first obtain the number of pixels with a pixel value of 255 in each ROI area, and then compare it with the set small point threshold pointthres. The areas below the threshold will be filtered out.

[0082] ROI (Region of interest) is the suspected defect area required by the algorithm. This step includes three steps: ROI search, merging, and screening.

[0083] It should be noted that the defects initially located cannot be identified as real defects. They need to be further judged and identified before they can be finally determined as defects. Therefore, the defects initially located should be treated as defects to be determined.

[0084] S5, performing a process of eliminating false defects and retaining true defects on the initially determined defects to be determined, thereby finally determining the true defects in the wire surface image.

[0085] Here, the defects to be determined initially are processed to eliminate false defects and retain true ones. In simple terms, false defects are eliminated from all defects to be determined, that is, "false elimination". After eliminating false defects, the defects to be determined that remain can be determined as true defects, that is, "retaining the true ones".

[0086] It should be noted that the composition of graphic elements in the wire surface image is relatively complex, and there are many graphic elements that are relatively close to the graphics of defects, such as black spots, iron oxide scale, black lines, roller seams, etc. These graphic elements are very similar to the graphics of real defects. If complex and sophisticated identification methods are not used, it is difficult to distinguish them from real defects. Since the defect location previously performed is a relatively simple and rough identification process, this leads to the fact that many of the defects to be determined obtained by preliminary positioning are not real defects, but other graphic elements similar to the graphics of real defects, that is, pseudo-defects. The purpose of performing true-retaining and false-removing processing on the defects to be determined is to eliminate the pseudo-defects that are similar to the graphics of real defects in the defects to be determined, so as to finally determine the true defects in the defects to be determined.

[0087] This method of initial screening followed by "eliminating the false and retaining the true" can greatly save computer computing power.

[0088] In this embodiment, the method of "eliminating false and retaining true for the initially determined defects to be determined" includes the following steps.

[0089] S51, for the defect to be determined, count the number of pixels in the ROI area whose grayscale value is greater than a certain value, and then calculate the ratio to the total number of pixels. When the ratio falls within a certain range, the defect to be determined is determined to be a true defect.

[0090] Specifically,

[0091] White flaky iron oxide is a common pseudo defect in iron oxide. Its boundary gradient is high and is often retained by the Sobel operator, such as Figure 3 As shown. For white-skinned iron oxide, first count the number of pixels in the ROI area whose grayscale value is greater than a certain value, then calculate the ratio to the total number of pixels. When the ratio falls within a certain range, discard the current defect. In addition to the above operations, filter the white-skinned iron oxide through the edge map.

[0092] S52, for the defect to be determined, the average gray value of the defect in the ROI area and the average gray value of the non-defect area are calculated. If the difference between the two exceeds a certain range, the area is smaller than a certain range, the shape is close to a square, and the minimum gray value and the average gradient meet certain conditions, then the defect to be determined is determined to be a true defect.

[0093] Specifically,

[0094] The dot-shaped black spots are small in area and mostly circular in shape. The defect frame is generally close to a square. The defect area is clearly distinguished from the non-defect area, and the grayscale difference is obvious. For example, Figure 4The removal method is to first calculate the average gray value of the defect in the ROI area and the average gray value of the non-defect area. If the difference between the two exceeds a certain range, the area is smaller than a certain range, the shape is close to a square, and the minimum gray value and average gradient meet certain conditions, it is a dot-like black spot.

[0095] S53, for the defects to be determined, the number of defects in the current ROI area is counted, and the distances between the defects are determined by standard deviation calculation. At the same time, it is determined whether the gray value of the defective area in the ROI is less than the gray value of the remaining areas.

[0096] If the distances between defects are approximately equal, and the grayscale value of the defective area in the ROI is smaller than the grayscale value of the remaining areas, the defect to be determined is determined to be a true defect.

[0097] Specifically,

[0098] Zebra print black spots are a type of black spots that often appear periodically on the entire roll of steel, such as Figure 5 . The black spots of zebra prints are often equally spaced and mostly appear in cleanliness level 1. They also often appear near the roller seam. Zebra prints are a type of pseudo-defects that are very similar to string scars. The removal method is to count the number of defects in the current ROI area, and then use the standard deviation calculation to determine whether the distances between defects are approximately equal. Secondly, determine whether the grayscale value of the defective area in the ROI is less than the grayscale value of the rest of the area. If all the above conditions are met, the current ROI area is determined to be a zebra print black spot and is discarded.

[0099] S54, for the defect to be determined, a longitudinal expansion operation is performed on the edge map to connect the discontinuous lines, and then the aspect ratio, the grayscale mean of the defect and the grayscale mean of the non-defect are calculated. If certain conditions are met, it is a longitudinal defect. Then, the size of the gradient is used to determine whether it is a black line. If it is determined to be a longitudinal defect or a black line, the defect to be determined is determined to be a true defect.

[0100] Specifically,

[0101] Black line is a black line on the surface of the wire, with obvious depth, usually intermittent or continuous, and with a long longitudinal distance, such as Figure 6 The removal method is to first perform a longitudinal expansion operation on the edge map to connect the discontinuous lines, and then calculate the aspect ratio, the grayscale mean of the defect and the grayscale mean of the non-defect. If certain conditions are met, it is a longitudinal defect, and then determine whether it is a black line based on the size of the gradient.

[0102] S55, for the defect to be determined, the diagonal grayscale value contrast is calculated on the edge map, and then the aspect ratio, the grayscale mean of the four corners of the ROI (1 / 4ROI length * 1 / 4ROI width), the grayscale mean of the defect and the grayscale mean of the non-defect and the number of defects in the ROI are calculated. If they meet certain conditions, it is determined to be an edge line, that is, the defect to be determined is determined to be a true defect.

[0103] Specifically,

[0104] The edge line is a pseudo defect caused by the tilt of the steel coil due to shaking. It usually includes part of the background area and has a large length and width ratio, such as Figure 7 The removal method is to first calculate the diagonal grayscale contrast on the edge map, then calculate the aspect ratio, the grayscale mean of the four corners of the ROI (1 / 4ROI length * 1 / 4ROI width), the grayscale mean of the defect and the grayscale mean of the non-defect, and the number of defects in the ROI. If they meet certain conditions, it is determined to be an edge line.

[0105] S56, for the defect to be determined, the four corners of the ROI area (1 / 4 length * 1 / 4 width) are intercepted respectively, as well as the left 1 / 4 and right 1 / 4 areas of the ROI area, and the grayscale mean of the two diagonal areas and the left and right 1 / 4 areas are calculated respectively. The ratio of the two can approximately describe the contrast between the left and right sides of the current ROI. The contrast of the three groups of areas is comprehensively considered. When this value is within a certain range, it is considered that the defect is a roller seam, that is, the defect to be determined is determined to be a true defect.

[0106] Specifically,

[0107] The roll seam line is a light-dark boundary line on the surface of the wire, usually running through the surface of the steel body. Its characteristics are that one side is dark and the other side is bright, and the edge is shallow, which is similar to the real shallow longitudinal defect. Figure 8 shown.

[0108] The biggest feature of the roller seam is the large difference in grayscale on both sides. To address this feature, we first intercept the four corners of the ROI area (1 / 4 length * 1 / 4 width), as well as the left 1 / 4 and right 1 / 4 areas of the ROI area, and calculate the grayscale mean of the two diagonal areas and the left and right 1 / 4 areas. The ratio of the two can approximately describe the contrast between the left and right sides of the current ROI. Taking the contrast of the three groups of areas into comprehensive consideration, when this value is within a certain range, the defect is considered to be a roller seam and is discarded.

[0109] It should be noted that

[0110] For step S5, due to the relatively complex working conditions in the production process, on the one hand, due to the small diameter of the wire, the minimum diameter is Φ5.5mm, the fastest speed can reach 110m / s, and the surface defect morphology changes quickly; on the other hand, there are many specifications and steel types produced, with nearly 40 specifications covering Φ5.5-Φ28mm, and nearly 600 steel types, including cold heading steel, spring steel, bearing steel, galvanized steel wire, welding wire steel, cord steel, etc. Due to the different characteristics of steel types, in addition to the real defects, there are often thousands of pseudo defects in different forms in the surface inspection of wires. It is particularly necessary to track the state of the surface of different specifications and steel types, defect detection conditions and a large number of detection pictures for a long time. The focus is to judge the rules, points and distribution of pseudo defects detected by surface conditions according to the changes in steel types and specifications, so as to identify, sort out, summarize and summarize pseudo defects.

[0111] After analyzing a large number of images, it was found that the main false defects include black spots, iron oxide scale, black lines, roller seams and other false defects. Fig. 9 As shown in the figure. Since the main algorithm uses the edge detection algorithm, the gradient of these pseudo defects is relatively obvious and they are easily detected during the inspection. Therefore, in order to reduce the false detection rate, it is very necessary to filter these pseudo defects. Therefore, the algorithm first needs to filter pseudo defects such as black spots of different shapes, iron oxide scales of different shapes, and roller seams.

[0112] Based on the above-mentioned judgment that the surface conditions of rolled products vary greatly and change rapidly, a single algorithm is used. For surface iron oxide scale, black spots, black lines and roll gap changes, the algorithm detection will result in a sharp increase in the number of false defects. This not only leads to a large number of false defect misdetections, but also brings a lot of burden to operators in checking defects. It takes a lot of time to check defects, and it will also cause the online defect detection to be unable to be checked in time, affecting the use of the entire defect detection system.

[0113] According to the degree of iron oxide scale formation on the surface of each type of steel and the different surface temperature cooling of red steel during the rolling process of red steel, the major types of steel are classified and sorted, and then different types of steel are classified and grouped. Different types of steel are divided into three categories according to the steel-out marks, as shown in Table 1. The system's detection algorithm sets three sets of wire algorithms for the three major types of steel, and selects different algorithms to process the images according to different cleanliness.

[0114] Table 1, classification of different steel grades according to steel tapping marks

[0115]

[0116] Among them, the surface of the wire in the first category is relatively clean, and the pseudo defects are mainly black spots; the surface of the wire in the second category is relatively complex, and the pseudo defects are mainly black spots and iron oxide scale; the surface of the wire in the third category is the most complex, and the black spots and iron oxide scale are very serious. Then, according to the material grouping, different surface cleanliness groups of rolled pieces are established, and then the algorithm of different cleanliness groups is studied.

[0117] Based on the previous steps, different clean grouping algorithm parameters are finally set. In the process of adjusting parameters, a large number of pictures are needed to judge and adjust the parameters. The wire rolling speed is fast, and the amount of online detection data is relatively large. If you rely on online adjustment of parameters, it will bring a lot of uncertainty. The adjustment direction and results may be completely opposite, and the adjustment effect is not good. In order to solve this problem, it is proposed to save the camera images of each steel from the beginning to the end after the sizing mill, and give priority to the camera images for wire scarring, folding, indentation, scratches and other defects. After the defect images are selected, the defect type and quantity are counted, and the camera recording of the whole steel is detected through the optimization algorithm. The detection results are compared with the type and quantity of the selected defect images, and the detection parameters are continuously cycled and optimized until the system detection accuracy is close to the manual confirmation accuracy. After finding the detection method and effect, hundreds of thousands of camera images of different specifications and steel types are used to adjust and optimize the detection parameters of all specifications and steel types.

[0118] Through material grouping, cleanliness grouping and parameter setting, it can be realized in the system in a windowed manner, specifically:

[0119] (1) Users need to create material groups and corresponding cleanliness selections in the system;

[0120] (2) The parameter maintenance interface will list the current newly created material grouping algorithm parameters to enable system detection;

[0121] (3) Users can modify algorithm parameters according to defect detection requirements;

[0122] (4) At the same time, users can also use the steel grade maintenance function to operate all steel grades in the current newly created material group, mainly including adding, moving to other material groups, deleting material groups, etc. After successfully completing the material group management operation, the system will synchronize the relevant information to the server for use by the detection algorithm.

[0123] The surface defect detection method of this embodiment aims to develop an economical and efficient hot high-speed wire rod detection method and equipment according to the variety and specifications of hot high-speed wire rod rolling, and realize the real-time detection of online defects of hot high-speed wire rod. The developed hot high-speed wire rod defect detection system replaces the hot-rolled wire rod pickling process, takes green process as the main task, reduces environmental pollution, and actively meets the increasingly stringent environmental protection regulations.

[0124] In this embodiment, the surface detection system of the hot high-speed wire surface defect algorithm detection effectively detects defects such as folding, scarring, scratches, and indentations on the wire surface, and effectively locks the defect generation area, providing an effective means for timely investigation and effective resolution of on-site problems, and provides an objective and quantitative means for early preventive adjustment of on-site quality, quantification of product quality defects, and product grading judgment and control.

[0125] The following specifically provides parameter selection and optimization for the defects to be determined:

[0126] 1. Parameter Description

[0127] In the detection algorithm of the finished product meter tester, different parameter values ​​are set for different cleanliness algorithms, and five sets of parameters are preset. The parameters used in the algorithm are detailed in Table 2:

[0128] Table 2, detailed table of algorithm parameters

[0129]

[0130] Parameter Description:

[0131] (1) Skip detection height: This parameter is mainly used to detect the finished product head. After the trimming algorithm is completed, if the image height is found to be less than 60, it is considered that the current finished product part is too small and will not be detected.

[0132] (2) Skip width detection: If the width of the image is less than 200 after trimming, it is considered that there is a problem with the finished product and will not be detected.

[0133] (3) X-direction merging coefficient: Search in the X direction with a certain value as the radius. If other defects are found within the range and the grayscale values ​​are similar, they will be merged. The larger the parameter value, the wider the search radius, and the smaller the parameter value, the smaller the search radius.

[0134] (4) Y-direction merging coefficient: Search in the Y direction with a certain value as the radius. If other defects are found within the range and the grayscale values ​​are similar, they will be merged. The larger the parameter value, the wider the search radius, and the smaller the parameter value, the smaller the search radius.

[0135] (5) Binarization coefficient: This parameter is mainly used to obtain a binary image of the image after trimming, which will be used in the later stage of filtering defects. This parameter is a value less than 1 and greater than 0. This value is multiplied by the global mean of the original image, and the resulting value is used as the threshold to perform binary processing on the image.

[0136] (6) X-direction left Gaussian coefficient, X-direction right Gaussian coefficient: Gaussian filtering is a commonly used denoising process in image processing, in which the Gaussian kernel has a significant impact on the denoising result. The larger the parameter value, the more obvious the denoising effect, but it should not be too large, otherwise important image information may be lost. X-direction denoising can mainly target some horizontally distributed noise, such as iron oxide.

[0137] (7) Left Gaussian coefficient in Y direction, right Gaussian coefficient in Y direction: The Gaussian kernel is a matrix. The adjustment strategy in the Y direction is the same as that in the X direction, but the Y direction mainly targets the noise distributed in the longitudinal direction.

[0138] (8) Expansion coefficient: The obtained defects are expanded and magnified. There are two purposes: first, to make defects that are very close to each other directly connected, and second, to make the defect frame larger. The larger the value, the greater the original defect is enlarged. At the same time, the value should not be too large, because if the defect is connected with the surrounding pseudo-defects, it will mislead the filtering and screening steps.

[0139] (9) Transverse defect parameters and longitudinal defect parameters: The method mainly uses the Sobel operator to detect defects. The gradient map after convolution is binarized in the X and Y directions using the above two parameters, and the parts with higher gradients in the gradient map are retained. The larger the values ​​of the above two parameters, the stricter the detection, and the smaller the values, the looser the detection.

[0140] (10) Noise coefficient: In the edge map, if the area of ​​a connected domain is smaller than this value, it is considered that the area may be a noise point and the connected domain is directly removed. The larger the value, the stricter the noise removal step; the smaller the value, the looser the noise removal step.

[0141] (11) Distance coefficient: This parameter is an auxiliary parameter used to judge when there are two or more defects in a complete defect frame. If the distance between the defects is greater than a certain value, it is considered that the defect frame has a pseudo defect and the current defect is discarded. This value is a decimal greater than 0 and less than 1. It is used together with the defect frame height. The larger the value, the less likely the current defect is to be discarded, and the smaller the value, the more likely it is to be discarded.

[0142] (12) Defect coefficient: This parameter is an auxiliary parameter, which indicates the total number of pixels in a connected domain in the edge map. It is usually combined with the gray value to filter out pseudo defects. The larger the value, the stricter the removal condition.

[0143] (13) Black spot coefficient: This parameter is a decimal number and is used to multiply the grayscale mean. If the grayscale mean in the current defect frame is less than the product value, the current defect is considered to be a black spot.

[0144] (14) Minimum number of pixels: used at the end of the algorithm. If the number of pixels in the defect box is smaller than this value, it will be discarded. The larger the value, the more likely it is to be discarded.

[0145] (15) Severe defect height: The height threshold of a relatively serious real defect. If it exceeds this value, it is considered a severe defect. The larger the value, the stricter the severe defect requirement.

[0146] (16) Severe defect width: The width threshold of a relatively serious real defect. A defect exceeding this value is considered a severe defect. The larger the value, the stricter the severe defect requirement.

[0147] 2. Parameter adjustment guidance

[0148] 1) Scarring and indentation missed detection: This type of defect is a common defect. If missed detection is found, the following parameters should be considered: (sorted by priority)

[0149] Noise coefficient / Oxide noise coefficient Decrease in steps of 5 or 10 Transverse defect parameters / Oxide transverse defect parameters Decrease in steps of 5 or 10 Gaussian coefficient in the left direction of X / Gaussian coefficient in the right direction of X Decrease in steps of 2 (must be an odd number)

[0150] 2) Missed detection of longitudinal and longitudinal scratches: This type of defect is a longitudinal defect. If missed detection is found, the following parameters should be considered: (sorted by priority)

[0151] Noise coefficient / Oxide noise coefficient Decrease in steps of 5 or 10 Longitudinal defect parameters / Oxide longitudinal defect parameters Decrease in steps of 5 or 10 Gaussian coefficient in the left direction of Y / Gaussian coefficient in the right direction of Y Decrease in steps of 2 (must be an odd number)

[0152] 3) If it is found that the defects that should be merged are not merged together, if they are in the X direction, adjust the X direction merging coefficient; if they are in the Y direction, adjust the Y direction merging coefficient. Usually increase it in steps of 20 and observe the effect.

[0153] 4) If the image false positive rate is too high and it is judged that the false positives are mostly iron oxide scale, the following parameters should be considered: (sorted by priority)

[0154]

[0155]

[0156] 5) If the image false positive rate is too high and it is judged that most of the false positives are zebra prints and black spots, the following parameters should be considered: (sorted by priority)

[0157] Dark spot coefficient Increase in steps of 0.1, with a maximum value of 1 Distance coefficient Decrease in steps of 0.1 Noise coefficient Increase in steps of 10 Transverse defect parameters Increase in steps of 10

[0158] 6) If it is found that too many roller seams are detected, adjust the high roller seam parameters and low roller seam parameters to make them close to 1.

[0159] 7) If large scars are missed, start with the following two parameters:

[0160] X-direction merging coefficient / Y-direction merging coefficient Increase in steps of 5 Noise coefficient Decrease in steps of 10

[0161] 3. Parameter Optimization

[0162] Since the surface of some images in some wires is relatively rough and does not completely match the cleanliness to which it belongs, it can be controlled by adding material group parameters. The newly added material group will run the cleanliness three algorithms by default and add new material group parameters. When adding new parameters, most parameters should first refer to the cleanliness parameter settings of the wire before, and then some more important parameters (noise coefficient / Oxide noise coefficient, transverse defect parameter / Oxide transverse defect parameter, longitudinal defect parameter / Oxide longitudinal defect parameter) should be set according to the surface conditions of the wire.

[0163] 1) Scarring and indentation missed detection: This type of defect is a common defect. If missed detection is found, the following parameters should be considered: (sorted by priority)

[0164]

[0165] 2) Missed detection of longitudinal and longitudinal scratches: This type of defect is a longitudinal defect. If missed detection is found, the following parameters should be considered: (sorted by priority)

[0166]

[0167]

[0168] 3) If the image false positive rate is too high and it is judged that the false positives are mostly iron oxide scale, the following parameters should be considered: (sorted by priority)

[0169]

[0170] 4) If the false positive rate is too high and it is judged that most of the false positives are black spots, the following parameters should be considered: (sorted by priority)

[0171] Dark spot coefficient Increase in steps of 0.1, with a maximum value of 1 Distance coefficient Decrease in steps of 0.1 Noise coefficient Increase in steps of 10 Transverse defect parameters Increase in steps of 10

[0172] 5) If it is found that too many roller seams are detected, adjust the high roller seam parameters and low roller seam parameters to make them close to 1, and reduce the distance coefficient in steps of 0.1.

[0173] The above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting surface defects of hot high-speed wire, characterized in that: include: S1, setting a hot high-speed wire surface defect detection system, wherein the set hot high-speed wire surface defect detection system includes an imaging system, and the imaging system includes a surround camera combination, and the surround camera combination is set on the travel route of the high-speed wire (2); S2, during the high-speed wire rolling production process, the surround camera combination continuously shoots and images the high-speed wire to obtain the wire surface image; S3, preprocessing the acquired wire surface image; S4, performing defect location on the pre-processed wire surface image to obtain the defect to be determined; S5, performing a process of eliminating false defects and retaining true defects on the initially determined defects to be determined, and finally determining the true defects in the wire surface image.

2. The method for detecting surface defects of hot high-speed wire according to claim 1, characterized in that: The processing of eliminating false and retaining true of the initially determined defects to be determined includes: S51, for the defect to be determined, the number of pixels whose grayscale value is greater than a certain value in the ROI area is counted, and then the ratio to the total number of pixels is calculated. When the ratio falls within a range, the defect to be determined is determined to be a true defect; S52, for the defect to be determined, the average gray value of the defect in the ROI area and the average gray value of the non-defect area are calculated. If the difference between the two exceeds a certain range, the area is smaller than a certain range, the shape is close to a square, and the minimum gray value and the average gradient meet certain conditions, then the defect to be determined is determined to be a true defect; S53, for the defect to be determined, the number of defects in the current ROI area is counted, and the distance between the defects is determined by standard deviation calculation. At the same time, it is determined whether the gray value of the defect area in the ROI is less than the gray value of the remaining area. If the distance between the defects is approximately equal, and the gray value of the defect area in the ROI is less than the gray value of the remaining area, the defect to be determined is determined to be a true defect; S54, for the defect to be determined, a longitudinal expansion operation is performed on the edge map to connect the discontinuous lines, and then the aspect ratio, the grayscale mean of the defect and the grayscale mean of the non-defect are calculated. If a certain condition is met, it is a longitudinal defect, and then the gradient size is used to determine whether it is a black line. If it is determined to be a longitudinal defect or a black line, the defect to be determined is determined to be a true defect; S55, for the defect to be determined, the grayscale contrast of the diagonal line is calculated on the edge map, and then the aspect ratio, the grayscale mean of the four corners of the ROI, the grayscale mean of the defect and the grayscale mean of the non-defect, and the number of defects in the ROI are calculated. If they meet certain conditions, it is determined to be an edge line, that is, the defect to be determined is determined to be a true defect; S56, for the defect to be determined, the four corners of the ROI area, as well as the left 1 / 4 and right 1 / 4 areas of the ROI area are respectively intercepted, and the grayscale means of the two diagonal areas and the left and right 1 / 4 areas are respectively calculated. The ratio of the two can approximately describe the contrast between the left and right sides of the current ROI. The contrast of the three groups of areas is comprehensively considered. When this value is within a range, it is considered that the defect is a roller seam, that is, the defect to be determined is determined to be a true defect.

3. The method for detecting surface defects of hot high-speed wire according to claim 1, characterized in that: The image preprocessing includes: edge trimming, illumination balancing and smoothing denoising.

4. The method for detecting surface defects of hot high-speed wire according to claim 3, characterized in that: The trimming is to extract the effective area to be detected, remove the background area, and use the gray gradient threshold search method to accurately determine the boundary of the finished product; Due to the large variation range of finished product diameters and the noisy on-site environment, the imaging effects of different specifications or cameras vary greatly, which is directly manifested as uneven image brightness. It is necessary to balance the illumination, improve the overall brightness, and compensate the grayscale average of each column to 128; Since there is a lot of background noise in the image to be detected, most of the noise is manifested as grayscale mutations, which will cause great interference to the detection. Therefore, it is necessary to perform noise reduction processing on it. Gaussian blur is selected, and the kernel size of Gaussian blur is determined by the parameter GaussSize.

5. The method for detecting surface defects of hot high-speed wire according to claim 1, characterized in that: The defect location includes Sobel edge finding, morphological opening and closing operations, filtering of small connected domains, and ROI search and merging.

6. The method for detecting surface defects of hot high-speed wire according to claim 5, characterized in that: The Sobel edge finding part uses the X-direction Sobel kernel and the Y-direction Sobel kernel. The Sobel X-direction and Y-direction kernels can identify edge mutations in the X and Y directions respectively, so the edge conditions in the X and Y directions can be accurately described. There are many noise points in the binarized result image, which will affect the final defect calibration. The two binary images are respectively subjected to morphological opening and closing operations to remove the noise points and weaken the influence of the noise, and then the two binary images are fused to obtain a fused binary image. Since most wire surfaces are covered with iron oxide scale, there will still be some small dots in the binary image after morphological denoising. These small dots that are prone to false detection need to be filtered out. The method used is to first obtain the number of pixels with a pixel value of 255 in each ROI area, and then compare it with the set small point threshold pointthres. The area below the threshold will be filtered out; ROI (Region of interest) is the suspected defect area required by the algorithm. This step includes three steps: ROI search, merging, and screening.

7. The method for detecting surface defects of hot high-speed wire according to claim 1, characterized in that: The hot high-speed wire surface defect detection system also includes mechanical equipment, electrical system, computer system and software system.

8. The method for detecting surface defects of hot high-speed wire according to claim 7, characterized in that: The mechanical equipment is used to fix the optoelectronic equipment, realize the focus adjustment of the inspection objects of different specifications to ensure accurate imaging parameters, and at the same time provide thermal protection for the equipment to adapt to the harsh and complex working conditions of the production line; The electrical system is used to supply power to the entire system, synchronize the wire running speed with the camera acquisition speed, and access various production control signals; The computer system includes an image acquisition and processing array, a system server, a database storage device, an operation terminal and a 10 Gigabit network switch; The software system is used to control the camera to collect and acquire images for high-speed processing. Through defect detection algorithm research, it can eliminate defect-free backgrounds from massive data and save and extract defective surface images.

Images