Off-line roller surface morphology detection method and detection equipment based on machine vision

Abstract

The present invention belongs to the technical field of roller wear detection, and relates to a machine vision-based offline roller surface morphology detection method and detection equipment, the method comprising: S1 preliminary calibration of the detection equipment; S2 placing the detection equipment on one side of the roller changing track; S3 real-time acquisition of the roller surface image; S4 transmitting the acquired roller surface image to the server; S5 preprocessing the original roller surface image; S6 extracting the hot rolling texture feature quantity of the roller surface image; S7 selecting the hot rolling texture feature quantity of the roller surface image; S8 using the texture feature quantity selected in step S7 as a standard to grade the wear of the roller surface; S9 outputting the grade determination result to complete the detection. The present invention uses a camera to collect image data of the hot rolling working roller surface, combined with an efficient data processing algorithm, to achieve non-contact real-time detection of the roller surface, with the advantages of high detection efficiency and no damage to the roller surface.

Description

Technical Field

[0001] The invention relates to the technical field of roller wear detection, and in particular to a machine vision-based offline roller surface morphology detection method and detection equipment. Background Art

[0002] Rollers are tools that cause the rolled metal to undergo plastic deformation and are important consumables that determine the efficiency of the rolling mill and the quality of the rolled material. During hot rolling production, the surface of the roll contacts the hot rolled material and is affected by oxidizing gases such as water vapor, forming a dense oxide film on the surface of the roll. As the roll is continuously subjected to high temperature and high pressure cyclic loads during the rolling process, the roll surface wears and the oxide film peels off, resulting in a dynamic equilibrium process consisting of oxidation and friction wear. In this process, the oxide film is randomly destroyed, producing different shapes and colors such as meteors, lines, and loose sand, resulting in an uneven lateral wear roll shape. Therefore, improper use, operation, and maintenance of the roll will have a serious adverse effect on the roll itself, the surface quality of the strip, and the shape of the plate.

[0003] However, the current sensor detection technology cannot meet the urgent need for production line roll surface detection. The extreme environments such as high temperature, high speed, and special atmosphere unique to the steel production process make it impossible to track and monitor some physical state information of rolling equipment or rolls using traditional sensing technology. Nowadays, most of the production lines are mainly observed by naked eyes by workers after leaving the machine. This detection method requires workers to concentrate highly and has high requirements for on-site lighting conditions. The detection efficiency is low, and it is easy to have missed inspections and missed inspections. In addition, this manual detection method lacks a unified detection standard and mainly relies on the experience of the detection personnel to make judgments, which reduces the accuracy of the detection results. The detection accuracy of some contact instrument detection is affected by a variety of external conditions, and the contact measurement is in direct contact with the object being measured, which may have a certain impact on the contact surface.

[0004] Therefore, an efficient and accurate non-contact online detection technology is developed to solve the problems of difficulty in detecting the surface morphology of working rolls on the current hot rolling production line, lack of automated detection means, low detection reliability, low detection efficiency and lack of unified detection standards, so as to improve the production and management efficiency of the production line. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and a detection device for detecting the surface morphology of an offline rolling mill roll based on machine vision. A camera and a linear light source are used to collect image data of the surface of the hot-rolled working roll, and an efficient data processing algorithm is combined to realize real-time detection of the wear condition of the rolling mill surface. At the same time, during the detection process, based on the sensor technology controlled by an industrial computer, automated detection triggering and data processing procedures are realized, which significantly improves the detection efficiency. In addition, the detection equipment can quickly determine whether the rolling mill roll needs grinding or cooling during the rolling mill roll removal process, which helps to optimize the production process and improve the production line operation efficiency.

[0006] The technical solution adopted by the present invention is a method for detecting the surface morphology of an offline roller based on machine vision, which comprises the following steps:

[0007] S1. According to the roller system parameters to be detected, the offline roller surface morphology detection equipment is adjusted, and the offline roller surface morphology detection equipment is preliminarily calibrated. There is an angle between the first camera imaging center and the first linear light source and between the second camera imaging center and the second linear light source in the offline roller surface morphology detection equipment. , angle The expression is:

[0008] ;

[0009] In the formula, Indicates the installation height of the camera and linear light source from the ground. Indicates the horizontal distance between the camera and the linear light source;

[0010] S2. The offline roll surface morphology detection device is set on one side of the roll changing track. The roll changing vehicle drives the roll system to be detected to approach the offline roll surface morphology detection device, and the arc length of the roll surface morphology detection in the roll system to be detected is measured. , The expression is:

[0011] ;

[0012] In the formula, Indicates the distance between the center line of the offline roll surface morphology detection equipment and the center line of the roll changing track. Indicates the camera pixel height; Indicates the focal length of the camera; Indicates the radius of the roll system to be tested;

[0013] S3. When the roller system to be inspected enters the inspection range, the distance sensor and the temperature sensor on the side of the body of the offline roller surface morphology inspection device reach the working threshold, and the first camera and the second camera start to work, respectively collecting the original roller surface images of the upper roller and the lower roller in real time, and storing the original roller surface images in the control processor of the offline roller surface morphology inspection device, and the accuracy of the roller surface images collected by the first camera and the second camera is for:

[0014] ;

[0015] In the formula, Indicates the camera acquisition frequency, Indicates the moving speed of the roll system to be inspected driven by the roll changing vehicle;

[0016] S4, based on the SSH communication protocol, transmitting the original roller surface image stored in the control processor in step S3 to a remote server;

[0017] S5. Preprocessing the original roller surface image on the remote server to remove noise mixed in the original roller surface image and eliminate contrast differences caused by different roller surface morphologies;

[0018] S6, analyzing the visual differences of the roller surface images of different passes, and extracting the hot rolling texture feature quantity from the pre-processed roller surface images;

[0019] S7, using an ID3 decision tree method based on information theory to select hot rolling texture feature quantities of the roll surface image after image preprocessing;

[0020] S8, using the texture feature quantity selected in step S7 as a standard, taking the roller surface image and the roller surface temperature data collected by the offline roller surface morphology detection equipment as input, and using a lightweight convolutional neural network algorithm to grade the wear of the roller surface;

[0021] S9. Output the grading result of the surface wear state of the rollers in the roller system, and complete the detection of the surface wear state of the rollers.

[0022] Furthermore, in step S1, the offline roller surface morphology detection equipment is adjusted and preliminarily calibrated, which specifically includes the following steps:

[0023] S11, installing the first surface topography detection component on the upper part of the machine body, and installing the second surface topography detection component on the lower part of the machine body;

[0024] S12, adjusting the height of the first camera and the first linear light source in the first surface topography detection assembly according to the height of the upper roller in the roller system to be detected, and adjusting the height of the second camera and the second linear light source in the second surface topography detection assembly according to the height of the lower roller in the roller system to be detected;

[0025] S13, adjusting the irradiation angles of the first linear light source and the second linear light source, so that the first linear light source can irradiate the upper roller in the roller system to be inspected through the transparent plate, and the second linear light source can irradiate the lower roller in the roller system to be inspected through the transparent plate;

[0026] S14, adjusting the shooting angles and shooting parameters of the first camera and the second camera;

[0027] S15, adjusting the working thresholds of the distance sensor and the temperature sensor according to the distance between the machine body and the roller system to be detected;

[0028] S16. Debugging the roller surface profile detection equipment;

[0029] S17. After debugging is completed, start the power supply and prepare to collect the roller surface image.

[0030] Furthermore, in step S5, the collected roller surface image is preprocessed on the remote server, which specifically includes the following steps:

[0031] S51, the shape of the roll in the roll system to be inspected is a large curvature cylinder, the black edges in the roll surface image are removed by an image segmentation algorithm, and the ROI area is extracted;

[0032] S52, cutting the entire roller surface image into multiple images with the same pixel size, so as to refine the roller surface detection range;

[0033] S53, performing image enhancement processing on the cut multiple images by using an adaptive histogram equalization method based on contrast limitation;

[0034] S54. The image is denoised by using Gaussian blur and morphological operations, so that the image can present clear surface wear features.

[0035] Furthermore, the hot rolling texture feature in step S6 includes roughness feature, contrast feature, directionality feature, linearity feature, regularity feature and coarseness feature of the roll surface image.

[0036] Another aspect of the present invention provides a machine vision-based offline roller surface morphology detection device, the offline roller surface morphology detection device is used in the above-mentioned machine vision-based offline roller surface morphology detection method, the offline roller surface morphology detection device comprises a body, a first surface morphology detection component, a second surface morphology detection component and a control processor,

[0037] The machine body is arranged on one side of the roller changing track of the hot rolling production line. The rollers to be inspected are transported along the roller changing track by the roller changing vehicle after being unloaded from the finishing rolling production line, and pass through the machine body. The machine body is a frame structure, and side panels are arranged on each side of the machine body. The side panel on the first side of the machine body is a transparent panel, and a support column is arranged inside the machine body. The control processor is integrated in the machine body, and the control processor is connected to the remote server for communication;

[0038] The first surface morphology detection component is arranged in the upper part of the body, and the second surface morphology detection component is arranged in the lower part of the body; the first surface morphology detection component includes a first camera, a first linear light source, a first steering connecting plate and a first connecting column, the first camera and the first linear light source are respectively arranged on both sides of the support column, the first camera is arranged on the first camera fixing plate, the first camera fixing plate is rotatably connected to the first side of the support column, and the first camera can adjust the imaging center of the first camera through the first camera fixing plate, the first steering connecting plate is arranged on the second side of the support column, the two ends of the first linear light source are fixed to the first connecting column through a connecting frame, the first connecting column is provided with a steering pin shaft, the steering pin shaft is arranged in the steering limit groove of the first steering connecting plate, a light source controller is arranged at the bottom of the body, the first linear light source is connected to the light source controller for communication, and the first linear light source can adjust the irradiation angle of the light source through the first steering connecting plate; the second surface morphology detection component includes a second camera, a second A linear light source, a second steering connecting plate and a second connecting column, a second camera and a second linear light source are respectively arranged on both sides of the support column, the second camera is arranged on a second camera fixing plate, the second camera fixing plate is rotatably connected to the second side of the support column, and the second camera can adjust the imaging center of the second camera through the second camera fixing plate, the second steering connecting plate is arranged on the first side of the support column, the two ends of the second linear light source are fixed to the second connecting column through a connecting frame, the second connecting column is provided with a steering pin shaft, the steering pin shaft is arranged in the steering limit groove of the second steering connecting plate, the second linear light source is communicatively connected with the light source controller, and the second linear light source can adjust the irradiation angle of the light source through the second steering connecting plate; the first camera and the first linear light source in the first surface morphology detection component can irradiate the upper roller in the roller system to be detected through the transparent plate, and the second camera and the second linear light source in the second surface morphology detection component can irradiate the lower roller in the roller system to be detected through the transparent plate.

[0039] Furthermore, the first camera and the second camera can be adjusted on the support column so that the imaging center of the first camera in the first surface morphology detection component is consistent with the geometric height of the upper roller in the roller system to be detected, and the imaging center of the second camera in the second surface morphology detection component is consistent with the geometric height of the lower roller in the roller system to be detected; the first linear light source and the second linear light source can be adjusted on the support column so that the geometric center of the first linear light source in the first surface morphology detection component is consistent with the imaging center of the first camera, and the geometric center of the second linear light source in the second surface morphology detection component is consistent with the imaging center of the second camera.

[0040] Furthermore, a distance sensor and a temperature sensor are provided on the side panel on the first side of the body, and a display screen is provided on the side panel on the second side of the body. Heat dissipation holes are provided on the side panels on the third and fourth sides of the body, a cooling fan is provided on the top of the body, and rollers are provided at the corners of the bottom of the body.

[0041] The characteristics and beneficial effects of the present invention are:

[0042] 1. The present invention provides a machine vision-based off-line roller surface morphology detection method and detection equipment, the detection equipment is arranged on one side of the roller changing roller in the roller off-line area, and the detection area of ​​the roller is usually arranged in the roller placement cooling area, the roller grinding area and the roller off-line area. Among them, although the detection in the roller placement cooling area can observe the surface state of the roller after cooling, the detection feedback time is long after the cooling is completed, and it is difficult to guide production decisions in time; while the detection in the grinding area is usually carried out on the premise that it has been determined that the roller needs to be ground, which will not only delay the normal progress of the grinding work, but also reduce the practical significance of the detection, and it is difficult to achieve the goal of early judgment and optimized decision-making. In contrast, the detection in the roller off-line area can make full use of the time window between the roller coming off the production line and transferring cooling or grinding, and quickly judge whether the roller needs grinding or direct cooling without affecting the production process. Such a detection method not only improves the detection efficiency, but also provides a real-time basis for the rapid adjustment of the production line, meeting the needs of modern production for efficient detection and rapid decision-making.

[0043] 2. The machine vision-based off-line roller surface morphology detection method and detection equipment provided by the present invention have a first surface morphology detection component and a second surface morphology detection component arranged on a machine body. When the roller system passes through the detection equipment, the irradiation angles of the first linear light source and the second linear light source can be adjusted respectively through a steering connecting plate fixed on the machine body. At the same time, the surface images of the upper roller and the lower roller in the roller system to be detected can be collected in real time through the first camera and the second camera respectively, which is beneficial to increase the detection accuracy and enhance the adaptability and flexibility of the detection equipment.

[0044] 3. The machine vision-based off-line roller surface morphology detection method and detection equipment provided by the present invention can adjust the height of the first camera and the first linear light source on the machine body support column according to the height of the upper roller in the roller system to be detected, and at the same time adjust the height of the second camera and the second linear light source on the machine body support column according to the height of the lower roller in the roller system to be detected, which is beneficial for the detection equipment to perform surface wear detection on rollers of different sizes and models in the roller system.

[0045] 4. The machine vision-based offline roller surface morphology detection method and detection equipment provided by the present invention are provided with side panels on each side of the machine body, and each side panel is provided with heat dissipation holes, heat dissipation fans and other components, which effectively protect the detection equipment from interference from harsh working environments such as high temperature, water mist, dust, etc., greatly extend the service life of the equipment, and improve the stability and reliability of detection.

[0046] 5. The machine vision-based offline roller surface morphology detection method and detection equipment provided by the present invention are placed on one side of the hot rolling production line finishing roller change track. By real-time detection of the hot rolling roller surface, hot rolling production and roller surface detection can be synchronized, effectively solving the problem of feedback lag and production efficiency impact in traditional detection methods. Compared with detection in the cooling zone or grinding zone, the present invention can quickly determine whether the roller needs grinding or direct cooling within the time window from the roller unloading from the production line to the transfer, thereby avoiding unnecessary processing procedures and saving production time.

[0047] 6. The machine vision-based off-line roller surface morphology detection method and detection equipment provided by the present invention use machine vision and sensor technology to perform automated non-contact detection of the roller surface, have high detection efficiency and will not cause damage to the roller surface, will not affect the normal production process, and have remote communication function and user interaction interface function. Users can monitor and use the overall detection equipment through the device control interface and image detection interface on the remote server PC, which is conducive to enhancing the user experience and improving the detection efficiency.

[0048] 7. The machine vision-based off-line roller surface morphology detection method and detection equipment provided by the present invention can pre-process the collected roller surface image, extract the hot rolling texture feature quantity from the pre-processed roller surface image, and analyze the texture feature selection result according to the classification result of the decision tree, so as to effectively improve the discrimination effect of the roller wear condition. In addition, a lightweight convolutional neural network based on wear characteristics is adopted to perform wear grading judgment on the roller surface wear defects and form a visual detection report, which can significantly improve the production efficiency of the production line and reduce production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1It is a schematic flow chart of the method for detecting the surface morphology of an offline roller based on machine vision of the present invention;

[0050] Figure 2 It is a schematic diagram of the process of adjusting the surface morphology detection device of the offline roller in the detection method of the present invention;

[0051] Figure 3 It is a data communication workflow diagram in the detection method of the present invention;

[0052] Figure 4 It is a schematic diagram of the process of preprocessing the roller surface image in the detection method of the present invention;

[0053] Figure 5 It is a software detection workflow diagram in the detection method of the present invention;

[0054] Figure 6 is a roller surface wear classification diagram of the present invention;

[0055] Figure 7 It is a schematic diagram of the detection device of the offline roller surface morphology detection method based on machine vision of the present invention being arranged on one side of the roller system to be detected;

[0056] Figure 8 It is a front view of the detection device of the offline roller surface morphology detection method based on machine vision of the present invention;

[0057] Fig. 9 It is a rear view of the detection device of the offline roller surface morphology detection method based on machine vision of the present invention;

[0058] Fig.10 It is a right view of the detection device of the offline roller surface profile detection method based on machine vision of the present invention;

[0059] Fig.11 It is a first isometric view of the detection device of the offline roller surface morphology detection method based on machine vision of the present invention;

[0060] Fig.12 It is a second isometric diagram of the detection device of the offline roller surface profile detection method based on machine vision of the present invention;

[0061] Fig.13 It is a schematic diagram of the installation of the first surface topography detection component of the present invention in the machine body;

[0062] Fig.14 is a schematic diagram of the angle between the camera and the linear light source of the present invention;

[0063] Fig.15 It is a schematic diagram of the parameters of the roll object to be measured in the present invention;

[0064] Fig.16This is a working principle diagram of image acquisition by the linear array camera of the present invention;

[0065] Fig.17 is a roller surface image collected by an embodiment of the present invention;

[0066] Fig.18 is one of the images preprocessed by the embodiment of the present invention;

[0067] Fig.19 is the second image after preprocessing in the embodiment of the present invention;

[0068] Fig. 20 This is a graph showing the results of the roller surface rating according to an embodiment of the present invention.

[0069] Main reference numerals:

[0070] 1. Machine body; 11. Transparent plate; 12. Support column; 13. Distance sensor; 14. Temperature sensor; 15. Display screen; 16. Roller; 2. First surface morphology detection component; 21. First camera; 22. First linear light source; 23. First camera fixing plate; 24. First steering connecting plate; 25. First connecting column; 26. Light source controller; 3. Second surface morphology detection component; 31. Second camera; 32. Second linear light source; 33. Second camera fixing plate; 34. Second steering connecting plate; 35. Second connecting column; 4. Upper roller; 5. Lower roller; 6. Hot rolling production line; 7. Roller changing vehicle; 8. Roller changing track. DETAILED DESCRIPTION

[0071] In order to fully describe the technical content, structural features, objectives and effects of the present invention, the following will be described in detail with reference to the accompanying drawings.

[0072] The present invention provides a method for detecting the surface morphology of a roll off the production line based on machine vision. The technical term off the production line refers to the detection of the surface morphology of the roll off the production line during its movement. Figure 1 As shown, it includes the following steps:

[0073] S1. According to the roller system parameters to be detected, the offline roller surface morphology detection equipment is adjusted, and the offline roller surface morphology detection equipment is preliminarily calibrated. There is an angle between the first camera imaging center and the first linear light source and between the second camera imaging center and the second linear light source in the offline roller surface morphology detection equipment. , The expression is:

[0074] ;

[0075] In the formula, Indicates the installation height of the camera and linear light source from the ground. Indicates the horizontal distance between the camera and the linear light source.

[0076] like Figure 2 As shown, in step S1, the offline roller surface morphology detection equipment is adjusted and preliminarily calibrated, which specifically includes the following steps:

[0077] S11, installing the first surface topography detection component 2 and the second surface topography detection component 3 on the upper part and the lower part of the machine body respectively.

[0078] S12. Adjust the height of the first camera 21 and the first linear light source 22 in the first surface morphology detection component 2 according to the height of the upper roller 4 in the roller system to be detected. At the same time, adjust the height of the second camera 31 and the second linear light source 32 in the second surface morphology detection component 3 according to the height of the lower roller 5 in the roller system to be detected, so as to ensure that the height position of the first camera 21 and the first linear light source 22 meets the height of the upper roller 4 in the roller system to be detected, and the height position of the second camera 31 and the second linear light source 32 meets the height of the lower roller 5 in the roller system to be detected.

[0079] S13, adjust the irradiation angles of the first linear light source 22 and the second linear light source 32, so that the first linear light source 22 can irradiate the upper roller 4 in the roller system to be detected through the transparent plate 11, and the second linear light source 32 can irradiate the lower roller 5 in the roller system to be detected through the transparent plate 11.

[0080] S14, adjusting the shooting angles and shooting parameters of the first camera 21 and the second camera 31 to achieve clear image acquisition and ensure that the images acquired by the first camera 21 and the second camera 31 have high brightness.

[0081] S15, adjusting the working thresholds of the distance sensor 13 and the temperature sensor 14 according to the distance between the machine body and the roller system to be detected.

[0082] S16. Debug the surface morphology detection equipment of the offline rolling mill rolls.

[0083] S17. After debugging is completed, start the power supply and prepare to collect the roller surface image.

[0084] S2. The offline roll surface morphology detection device is set on one side of the roll changing track. The roll changing vehicle 7 drives the roll system to be detected to approach the offline roll surface morphology detection device, and the arc length of the roll surface morphology detection in the roll system to be detected is measured. , The expression is:

[0085] ;

[0086] In the formula, Indicates the distance between the center line of the offline roll surface morphology detection equipment and the center line of the roll changing track. Indicates the camera pixel height; Indicates the focal length of the camera; Indicates the radius of the roller system to be tested.

[0087] S3, when the roll to be inspected enters the inspection range, the distance sensor 13 and the temperature sensor 14 on the side of the body 1 of the offline roll surface morphology inspection device reach the working threshold, and the first camera 21 and the second camera 31 start to work, respectively collecting the original roll surface images of the upper roll 4 and the lower roll 5 in real time, such as Fig.16 As shown, the original roll surface image is stored in the control processor of the offline roll surface morphology detection device, and the roll surface image accuracy collected by the first camera 21 and the second camera 31 is for:

[0088] ;

[0089] In the formula, Indicates the camera acquisition frequency, Indicates the moving speed of the roll system to be inspected driven by the roll changing vehicle.

[0090] At this time, the sensor data is transmitted to the single-chip microcomputer, and the single-chip microcomputer is connected to the industrial computer through a serial port communication method. The industrial computer communicates with the camera 21 through a POE data line, thereby realizing the external hard triggering of the camera to collect images and display the results on the display screen 15.

[0091] S4, the roller surface image collected in real time by the first camera 21 and the second camera 31 is transmitted to the remote server. The detection device where the industrial computer is located is the server. The remote server PC and the industrial computer perform data communication and transmission through the SSH communication protocol. A session request is sent to the server at a certain interval to query whether there is a new image data file. If so, data is transmitted and the newly generated image data file is transmitted to the remote server. The data communication transmission workflow is as follows: Figure 3 shown.

[0092] S5. During the use of the roller, as the rolling tonnage and rolling mileage increase, it will experience different wear stages, in which the wear will cause the reflectivity of the surface light source to decrease, thus affecting the clarity and accuracy of image acquisition. Therefore, image preprocessing becomes very important, aiming to remove the noise and interference caused by wear, enhance the contrast and clarity of the image, and thus accurately obtain the wear information of the roller surface. The collected original roller surface image is preprocessed on the remote server, such as Figure 4 and Figure 5 As shown, the specific steps include:

[0093] S51. Since the shape of the roller to be inspected is a large-curvature cylinder and the optical path is particularly sensitive to changes in surface curvature, the original roller surface image captured by the linear array camera always inevitably contains a black edge portion with a large brightness difference. The black edge in the roller surface image is removed through an image segmentation algorithm, and the ROI area is extracted to avoid damage to the roller surface texture characteristics caused by uneven image brightness due to surface curvature problems.

[0094] S52. The surface morphology of the working roll after the rolling process is complex and changeable. In order to refine the surface wear assessment scale and facilitate the subsequent neural network operation, the entire roll surface image is cut into multiple images with the same pixel size.

[0095] S53. In the work roll surface inspection, wear defects can be classified as structural texture surfaces due to their significant texture features. Different hot rolling work roll surfaces have different texture distributions, and on the same hot rolling work roll, the size and grayscale of the texture primitives may also vary significantly. This complex texture greatly increases the difficulty of hot rolling work roll surface defect detection. In order to highlight the texture and color features of the wear image, the image enhancement processing is performed on the multiple images after cutting by the adaptive histogram equalization method based on contrast limitation. The histogram equalization sets the image to Gray level, each gray level The probability of occurrence is , the information entropy of this gray level is:

[0096] .

[0097] The entropy contained in the image is:

[0098] ;

[0099] In the formula, Indicates the grayscale level, Indicates Grayscale, Indicates The probability of grayscale occurrence.

[0100] When the histogram of an image is uniformly distributed, the information entropy The image quality can be enhanced by histogram equalization technology, but its main disadvantage is that it may lead to contrast enhancement and noise amplification, which will destroy the texture features in the roll surface image. To overcome these disadvantages, contrast adaptive histogram equalization controls the contrast enhancement in local histogram equalization by setting contrast limits.

[0101] Adaptive histogram equalization uses a local method to process the image, where the local sub-image The average gray value , whose expression is:

[0102] ;

[0103] In the formula, and Represent the pixel length and pixel width of the sub-image respectively, and Indicates the position of the pixel in the sub-image.

[0104] According to the above formula, the central pixel value of the sub-image after the change can be calculated, and its expression is:

[0105] ;

[0106] In the formula, , are the center pixel values ​​before and after the transformation respectively. When Then it makes On the contrary, Thus, the contrast in the local window is changed, and the image contrast is enhanced while the influence of noise is effectively suppressed, so as to obtain a clearer and more natural image effect.

[0107] To prevent excessive contrast enhancement, the local histogram peak is limited. The image has grayscale levels. Assume that the total number of pixels before and after shearing is and , The gray level in the image is The number of pixels, The gray value of the image after cutting is The number of pixels, Represents the defined contrast limit coefficient, then the image gray value shear coefficient can be defined, and its coefficient calculation expression is:

[0108] , .

[0109] The image histogram is sheared by applying the shearing function to suppress the peak part of the pixel concentration, and then the histogram is averaged so that the gray levels that account for a large proportion will not be over-stretched, resulting in a large number of gray levels being merged, thereby avoiding the appearance of unnatural phenomena in the image due to excessive enhancement.

[0110] S54. The image is denoised by using Gaussian blur and morphological operations, so that the image can present clear surface wear characteristics. This improves the accuracy of subsequent image analysis and recognition, and provides reliable data support for monitoring and evaluation of roller wear.

[0111] S6. At present, the evaluation of roll wear mainly relies on manual judgment. This method is greatly affected by work experience and prone to individual errors. In order to improve the accuracy and objectivity of the wear condition judgment, the hot rolling texture feature quantity, namely HRTM (Hot Rolling Texture Matrix), is extracted from the pre-processed roll surface image, the morphological characteristics of the oxide film are analyzed, and a feature sample library and corresponding judgment classification standards are established. The texture feature quantity includes the roughness feature, contrast feature, orientation feature, linearity feature, regularity feature and coarseness feature of the roll surface image.

[0112] Among them, the roughness It can be obtained by calculating the average value of the entire image, and its expression is:

[0113] ;

[0114] In the formula, Indicates the optimal size value that can maximize the average intensity value of the pixels in the active window.

[0115] For each pixel position in the image , The calculation formula is as follows:

[0116] ;

[0117] In the formula, is a set of predefined candidate window sizes, Indicates Centered at window. is the number of pixels in the window. is the image at position The pixel intensity value. Indicates taking the parameter value that maximizes the expression in the brackets.

[0118] Contrast Through the statistics of pixel intensity distribution, the expression is:

[0119] ;

[0120] In the formula, Represents the standard deviation of pixel grayscale values, Indicates the kurtosis of the pixel gray value. Represents the maximum pixel gray value in the image. Represents the minimum pixel gray value in the image. is a weighting coefficient used to balance the influence of the standard deviation term and the brightness contrast term.

[0121] Entropy-based directionality It is an indicator based on the concept of information entropy and is used to measure the global directional characteristics of a given texture area. Its expression is:

[0122] ;

[0123] In the formula, is the histogram of the image; Represents the peak in the histogram; are all the peaks in the histogram; Peak All discrete regions included; The center position of the wave crest.

[0124] Linearity Describes the line similarity of elements within the image, and its expression is:

[0125] ;

[0126] In the formula, is the directional co-occurrence matrix, where an element Defined as the distance along the edge The relative frequency of two adjacent units appearing on the image, where the direction of one unit is , and the direction of the other unit is .

[0127] Regularity It is the global feature of the given image texture. The four independent characteristics defined above are used as parameters, and the sum of the changes of these four characteristics is used as a measure of regularity. Its expression is:

[0128] ;

[0129] In the formula, , , and They are , , , The standard deviation of .

[0130] Roughness It is based on the results of psychological experiments on vision and is used to characterize the effects of roughness and contrast on vision. Its expression is:

[0131] .

[0132] S7. The hot rolling texture feature quantity of the roll surface image is selected by using the decision tree method. The ID3 algorithm is based on information theory and uses information entropy and information gain as the measurement criteria to achieve data classification. According to the contribution of different features to the ID3 decision tree, six cases were studied as shown in Table 1. By comparing the classification of different cases, roughness, contrast, and orientation were selected as the grading criteria for judging the surface wear of the working roll.

[0133] Table 1 Contribution of different feature quantities to decision tree

[0134]

[0135] S8. Using the texture feature quantity selected in step S7 as the standard, a lightweight convolutional neural network algorithm is used to grade the wear of the roller surface. The specific grading standards are shown in Table 2. The roller surface wear images manually rated by on-site workers are used as training sets and verification sets. Figure 6 As shown. Specifically: MobileNetV2-WDC uses a linear bottleneck inverse residual component with deep separable convolution as the basic unit to expand the dimension of the input information, and uses deep separable convolution for feature extraction, and finally linearly reduces the dimension of the features. The core of MobileNetV2-WDC is the Bottleneck layer composed of Inverted Residuals blocks containing deep separable convolutions and Linear Bottlenecks blocks. This deep separable convolution divides the traditional convolution operation into deep convolution (DW) operation and point-by-point convolution (PW) operation. Compared with traditional convolutional neural networks, this method greatly reduces the number of parameters and calculations, while improving the accuracy of classification.

[0136] Table 2 Hot rolling site roll surface rating standards

[0137]

[0138] S9. Output the grading result of the surface wear state of the rollers in the roller system, and complete the detection of the surface wear state of the rollers.

[0139] In addition, a visual inspection report of the wear status of the roll surface can be generated on the remote server screen. Specifically, the final inspection results, including high-definition images, wear levels, temperature data, maintenance suggestions, etc., are presented in a visual form on the screen of the remote control center. The system generates a detailed inspection report, including high-definition images, wear levels, temperature data, maintenance suggestions, etc., for real-time viewing and decision-making by managers and technicians; the inspection report is uploaded to the database to manage, trace back, query and compare and analyze all roll data information in the roll system.

[0140] Another aspect of the present invention provides a detection device for a method for detecting the surface morphology of an offline roller based on machine vision, such as Figure 7 As shown, it includes a machine body 1, a first surface morphology detection component 2, a second surface morphology detection component 3 and a control processor. The machine body 1 is arranged on one side of the finishing roll changing track 8 of the hot rolling production line 6. The rolls to be detected are transported along the roll changing track 8 by the roll changing vehicle 7 after being unloaded from the finishing rolling production line, and pass through the machine body 1.

[0141] like Figure 8~Figure 10 As shown, the body 1 is a frame structure, and each side of the body 1 is provided with a side panel, a transparent panel 11 is provided on the side panel of the first side of the body 1, and a support column 12 is provided inside the body 1, a control processor is integrated in the body 1, and the control processor is connected to the remote server under the same WIFI component, and the control processor includes an industrial computer and a single-chip microcomputer. A distance sensor 13 and a temperature sensor 14 are also provided on the side panel of the first side of the body 1, and a display screen 15 is also provided on the side panel of the second side of the body 1, and heat dissipation holes are provided on the side panels of the third and fourth sides of the body 1, a heat dissipation fan is provided on the top of the body 1, and rollers 16 are provided at the corners of the bottom of the body 1.

[0142] like Figure 11~Figure 13As shown, the first surface topography detection assembly 2 and the second surface topography detection assembly 3 are respectively arranged at the upper and lower parts of the body 1. The first surface topography detection assembly 2 includes a first camera 21, a first linear light source 22, a first steering connection plate 24 and a first connection column 25. The first camera 21 and the first linear light source 22 are respectively arranged on both sides of the support column 12, the first camera 21 is arranged on the first camera fixing plate 23, the first camera fixing plate 23 is rotatably connected to the first side of the support column 12, the first camera 21 can adjust the imaging center of the first camera 21 through the first camera fixing plate 23, the first steering connection plate 24 is arranged on the second side of the support column 12, the two ends of the first linear light source 22 are fixed to the first connection column 25 through a connecting frame, the first connection column 25 is provided with a steering pin shaft, and the steering pin shaft is arranged in the steering limit groove of the first steering connection plate 24, and a light source controller 26 is arranged at the bottom of the body, the light source controller 26 is connected to the first linear light source 22 in communication, and the first linear light source 22 can adjust the irradiation angle of the light source through the first steering connection plate 24. The second surface morphology detection component 3 includes a second camera 31, a second linear light source 32, a second steering connecting plate 34 and a second connecting column 35. The second camera 31 and the second linear light source 32 are respectively arranged on both sides of the support column 12. The second camera 31 is arranged on the second camera fixing plate 33. The second camera fixing plate 33 is rotatably connected to the second side of the support column 12. The second camera 31 can adjust the imaging center of the second camera 31 through the second camera fixing plate 33. The second steering connecting plate 34 is arranged on the first side of the support column 12. Both ends of the second linear light source 32 are fixed to the second connecting column 35 through a connecting frame. The second connecting column 35 is provided with a steering pin shaft, which is arranged in the steering limit groove of the second steering connecting plate 34. The light source controller 26 is communicatively connected to the second linear light source 32. The second linear light source 32 can adjust the irradiation angle of the light source through the second steering connecting plate 34. The first camera 21 and the first linear light source 22 in the first surface morphology detection component 2 can illuminate the upper roller 4 in the roller system to be inspected through the transparent plate 11, and the second camera 31 and the second linear light source 32 in the second surface morphology detection component 3 can illuminate the lower roller 5 in the roller system to be inspected through the transparent plate 11.

[0143] In a preferred embodiment, the first camera 21 and the second camera 31 can be adjusted on the support column 12 so that the imaging center of the first camera 21 in the first surface morphology detection component 2 is consistent with the geometric height of the upper roller 4 in the roller system to be detected, and the imaging center of the second camera 31 in the second surface morphology detection component 3 is consistent with the geometric height of the lower roller 5 in the roller system to be detected; the first linear light source 22 and the second linear light source 32 can be adjusted on the support column 12 so that the geometric center of the first linear light source 22 in the first surface morphology detection component 2 is consistent with the imaging center of the first camera 21, and the geometric center of the second linear light source 32 in the second surface morphology detection component 3 is consistent with the imaging center of the second camera 31.

[0144] like Fig.14 and Fig.15 As shown, there is an angle between the imaging center of the first camera 21 and the first linear light source 22, and between the imaging center of the second camera 21 and the second linear light source 32. , and the angle expression is:

[0145] ;

[0146] In the formula, Indicates the installation height of the camera and linear light source from the ground. Indicates the horizontal distance between the camera and the linear light source.

[0147] Example

[0148] This embodiment uses a 2250 hot rolling production line as an embodiment object, whose standard roller diameter is 800mm, and uses the above-mentioned detection equipment and detection method. The horizontal distance between the geometric center of the fixed camera and the geometric center of the linear light source is 180mm, the angle between the camera optical axis and the illumination direction of the linear light source is 12.68°, and the distance between the center of the detection equipment body and the center of the roller to be tested is 800mm. Install and adjust the equipment according to the above steps. The roller replacement vehicle drags the roller to be tested off the machine and passes through the detection equipment body. The equipment automatically collects the surface image of the roller, such as Fig.17 Then, according to the steps in the method, the roller surface image collected by the machine body is transmitted to the server where the production line control center is located, and the roller surface image is processed in the server. The pre-processed image is as shown in Fig.18 and Fig.19 The hot rolling texture feature is extracted from the preprocessed image, and analyzed by the ID3 decision tree algorithm. Finally, the image roughness, contrast and orientation are selected as the wear judgment and grading criteria, and their contribution in the decision tree algorithm is 0.506, 0.138 and 0.188 respectively. The above wear judgment criteria and image data are passed into the MobileNetV2-WDC neural network to finally realize the wear grading. The final result is shown in the figure below. Fig. 20As shown in the figure. The accuracy of the roller surface wear classification model can reach 98.57%, the average classification time for a single wear defect image is 71.7ms, and the model parameter volume and floating point operation are 2.27M and 612.74M respectively. It meets the actual industrial production needs and realizes the automatic image acquisition from the front-end equipment to the image data transmission, and finally to the remote server end. Roller surface wear rating software interface development and roller data processing work.

[0149] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for detecting the surface morphology of an offline roller based on machine vision, characterized in that: It includes the following steps: S1. According to the roller system parameters to be detected, the offline roller surface morphology detection equipment is adjusted, and the offline roller surface morphology detection equipment is preliminarily calibrated. There is an angle between the first camera imaging center and the first linear light source and between the second camera imaging center and the second linear light source in the offline roller surface morphology detection equipment. , angle The expression is: ； In the formula, Indicates the installation height of the camera and linear light source from the ground. Indicates the horizontal distance between the camera and the linear light source; S2. The offline roll surface morphology detection device is set on one side of the roll changing track. The roll changing vehicle drives the roll system to be detected to approach the offline roll surface morphology detection device, and the arc length of the roll surface morphology detection in the roll system to be detected is measured. , The expression is: ； In the formula, Indicates the distance between the center line of the offline roll surface morphology detection equipment and the center line of the roll changing track. Indicates the camera pixel height; Indicates the focal length of the camera; Indicates the radius of the roll system to be tested; S3. When the roller system to be inspected enters the inspection range, the distance sensor and the temperature sensor on the side of the body of the offline roller surface morphology inspection device reach the working threshold, and the first camera and the second camera start to work, respectively collecting the original roller surface images of the upper roller and the lower roller in real time, and storing the original roller surface images in the control processor of the offline roller surface morphology inspection device, and the accuracy of the roller surface images collected by the first camera and the second camera is for: ； In the formula, Indicates the camera acquisition frequency, Indicates the moving speed of the roll system to be inspected driven by the roll changing vehicle; S4, based on the SSH communication protocol, transmitting the original roller surface image stored in the control processor in step S3 to a remote server; S5. Preprocessing the original roller surface image on the remote server to remove noise mixed in the original roller surface image and eliminate contrast differences caused by different roller surface morphologies; S6, analyzing the visual differences of the roller surface images of different passes, and extracting the hot rolling texture feature quantity from the pre-processed roller surface images; S7, using an ID3 decision tree method based on information theory to select hot rolling texture feature quantities of the roll surface image after image preprocessing; S8, using the texture feature quantity selected in step S7 as a standard, taking the roller surface image and the roller surface temperature data collected by the offline roller surface morphology detection equipment as input, and using a lightweight convolutional neural network algorithm to grade the wear of the roller surface; S9. Output the grading result of the surface wear state of the rollers in the roller system, and complete the detection of the surface wear state of the rollers.

2. The machine vision-based method for detecting the surface morphology of an offline roller according to claim 1, characterized in that: In step S1, the surface morphology detection equipment of the offline roller is adjusted and preliminarily calibrated, which specifically includes the following steps: S11, installing the first surface topography detection component on the upper part of the machine body, and installing the second surface topography detection component on the lower part of the machine body; S12, adjusting the height of the first camera and the first linear light source in the first surface topography detection assembly according to the height of the upper roller in the roller system to be detected, and adjusting the height of the second camera and the second linear light source in the second surface topography detection assembly according to the height of the lower roller in the roller system to be detected; S13, adjusting the irradiation angles of the first linear light source and the second linear light source, so that the first linear light source can irradiate the upper roller in the roller system to be inspected through the transparent plate, and the second linear light source can irradiate the lower roller in the roller system to be inspected through the transparent plate; S14, adjusting the shooting angles and shooting parameters of the first camera and the second camera; S15, adjusting the working thresholds of the distance sensor and the temperature sensor according to the distance between the machine body and the roller system to be detected; S16. Debugging the roller surface profile detection equipment; S17. After debugging is completed, start the power supply and prepare to collect the roller surface image.

3. The method for detecting the surface morphology of the offline roller based on machine vision according to claim 1, characterized in that: In step S5, the collected roller surface image is preprocessed on the remote server, which specifically includes the following steps: S51, the shape of the roll in the roll system to be inspected is a large curvature cylinder, the black edges in the roll surface image are removed by an image segmentation algorithm, and the ROI area is extracted; S52, cutting the entire roller surface image into multiple images with the same pixel size, so as to refine the roller surface detection range; S53, performing image enhancement processing on the cut multiple images by using an adaptive histogram equalization method based on contrast limitation; S54. The image is denoised by using Gaussian blur and morphological operations, so that the image can present clear surface wear features.

4. The method for detecting the surface morphology of the offline roller based on machine vision according to claim 1, characterized in that: The hot rolling texture feature in step S6 includes the roughness feature, contrast feature, directionality feature, linearity feature, regularity feature and coarseness feature of the roll surface image.

5. A machine vision-based offline roller surface morphology detection device, the offline roller surface morphology detection device is used for the machine vision-based offline roller surface morphology detection method according to any one of claims 1 to 4, characterized in that: The offline roller surface profile detection device comprises a body, a first surface profile detection component, a second surface profile detection component and a control processor. The machine body is arranged on one side of the roller changing track of the hot rolling production line. The rollers to be inspected are transported along the roller changing track by the roller changing vehicle after being unloaded from the finishing rolling production line, and pass through the machine body. The machine body is a frame structure, and side panels are arranged on each side of the machine body. The side panel on the first side of the machine body is a transparent panel, and a support column is arranged inside the machine body. The control processor is integrated in the machine body, and the control processor is connected to the remote server for communication; The first surface morphology detection component is arranged in the upper part of the body, and the second surface morphology detection component is arranged in the lower part of the body; the first surface morphology detection component includes a first camera, a first linear light source, a first steering connecting plate and a first connecting column, the first camera and the first linear light source are respectively arranged on both sides of the support column, the first camera is arranged on the first camera fixing plate, the first camera fixing plate is rotatably connected to the first side of the support column, and the first camera can adjust the imaging center of the first camera through the first camera fixing plate, the first steering connecting plate is arranged on the second side of the support column, the two ends of the first linear light source are fixed to the first connecting column through a connecting frame, the first connecting column is provided with a steering pin shaft, the steering pin shaft is arranged in the steering limit groove of the first steering connecting plate, a light source controller is arranged at the bottom of the body, the first linear light source is connected to the light source controller for communication, and the first linear light source can adjust the irradiation angle of the light source through the first steering connecting plate; the second surface morphology detection component includes a second camera, a second A linear light source, a second steering connecting plate and a second connecting column, a second camera and a second linear light source are respectively arranged on both sides of the support column, the second camera is arranged on a second camera fixing plate, the second camera fixing plate is rotatably connected to the second side of the support column, and the second camera can adjust the imaging center of the second camera through the second camera fixing plate, the second steering connecting plate is arranged on the first side of the support column, the two ends of the second linear light source are fixed to the second connecting column through a connecting frame, the second connecting column is provided with a steering pin shaft, the steering pin shaft is arranged in the steering limit groove of the second steering connecting plate, the second linear light source is communicatively connected with the light source controller, and the second linear light source can adjust the irradiation angle of the light source through the second steering connecting plate; the first camera and the first linear light source in the first surface morphology detection component can irradiate the upper roller in the roller system to be detected through the transparent plate, and the second camera and the second linear light source in the second surface morphology detection component can irradiate the lower roller in the roller system to be detected through the transparent plate.

6. The machine vision-based off-line roller surface morphology detection device according to claim 5 is characterized in that: The first camera and the second camera can be adjusted on the support column so that the imaging center of the first camera in the first surface topography detection assembly is consistent with the geometric height of the upper roller in the roller system to be detected, and the imaging center of the second camera in the second surface topography detection assembly is consistent with the geometric height of the lower roller in the roller system to be detected; The first linear light source and the second linear light source can be adjusted on the supporting column so that the geometric center of the first linear light source in the first surface morphology detection component is consistent with the imaging center of the first camera, and the geometric center of the second linear light source in the second surface morphology detection component is consistent with the imaging center of the second camera.

7. The machine vision-based off-line roller surface morphology detection device according to claim 5, characterized in that: A distance sensor and a temperature sensor are also provided on the side panel of the first side of the body, and a display screen is also provided on the side panel of the second side of the body. Heat dissipation holes are provided on the side panels of the third and fourth sides of the body, a cooling fan is provided on the top of the body, and rollers are provided at the corners of the bottom of the body.

Images