An online detection method and system for brake disc defects based on machine vision

Abstract

The present invention provides an online detection method and system for brake disc defects based on machine vision. The method includes: collecting a first image on a finished product conveying device through a first image acquisition device; performing a reduction process on the first image according to the conveying direction and speed of the finished product conveying device to obtain a second image; performing a positioning and extraction operation on the second image to obtain a regional image corresponding to the brake disc; preprocessing the regional image and performing defect extraction, and outputting a first alarm message after defects are extracted. The online detection method and system for brake disc defects based on machine vision of the present invention realizes dynamic online detection and improves the detection efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly relates to an on-line detection method and system for brake disc defects based on machine vision. Background Art

[0002] With the rapid progress of technology and the continuous development of industrial production, quality control plays a crucial role in the manufacturing industry. However, in large-scale production processes, defects and failures inevitably occur, which have a negative impact on product quality and production efficiency. Traditional manual inspection cannot meet the requirements of high-efficiency and high-precision quality control, so machine vision defect detection technology has emerged.

[0003] When the deep learning method is applied in practice, the problem of the movement of the brake disc on the production line also needs to be considered, that is, how to achieve dynamic on-line detection. Summary of the Invention

[0004] One of the purposes of the present invention is to provide an on-line detection method and system for brake disc defects based on machine vision, which realizes dynamic on-line detection and improves the detection efficiency.

[0005] An on-line detection method for brake disc defects based on machine vision provided by an embodiment of the present invention includes:

[0006] Collect a first image on a finished product conveying device through a first image acquisition device;

[0007] According to the conveying direction and conveying speed of the finished product conveying device, perform reduction processing on the first image to obtain a second image;

[0008] Perform a positioning extraction operation on the second image to obtain a regional image corresponding to the brake disc;

[0009] Perform preprocessing on the regional image and perform defect extraction. After extracting the defect, output a first alarm message.

[0010] Preferably, the step of performing reduction processing on the first image includes:

[0011] According to the pixel sampling interval, conveying direction and conveying speed of the first image, query a preset deviation parameter determination table to retrieve deviation parameters;

[0012] Construct a reduction reconstruction map according to the starting pixel point and pixel sampling order in the first image, and resample the pixels after the starting pixel point from the first image according to the deviation parameters to obtain a second image.

[0013] Preferably, the steps of performing a positioning extraction operation on the second image are as follows:

[0014] After binarizing the first image, edge extraction is performed to obtain an edge-extracted image;

[0015] The least circumscribed circle fitting method is used to perform circle fitting on the edge-extracted image, a mask is set, and the working surface is extracted to obtain a region image.

[0016] Preferably, the preprocessing of the region image includes: median filtering for image denoising and image enhancement using the Laplacian operator.

[0017] Preferably, the defect extraction steps are as follows:

[0018] The OTSU algorithm is used to segment and extract defects, the improved Canny operator is used for scratch detection, the method of combining probabilistic Hough transform is used for scratch connection, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

[0019] The present invention also provides an on-line brake disc defect detection system based on machine vision, including: an acquisition unit, a restoration unit, a positioning and extraction unit, and a defect detection unit;

[0020] Among them, the acquisition unit acquires the first image on the finished product conveying device through the first image acquisition device; the restoration unit performs restoration processing on the first image according to the conveying direction and conveying speed of the finished product conveying device to obtain a second image; the positioning and extraction unit performs positioning and extraction operations on the second image to obtain the region image corresponding to the brake disc; the defect detection unit preprocesses the region image and performs defect extraction, and outputs the first alarm information after extracting the defect.

[0021] Preferably, the steps of the restoration unit performing restoration processing on the first image include:

[0022] According to the pixel sampling interval, conveying direction and conveying speed of the first image, query the preset deviation parameter determination table to retrieve the deviation parameter;

[0023] A restoration reconstruction diagram is constructed based on the starting pixel point and pixel sampling order in the first image, and pixels after the starting pixel point are resampled from the first image according to the deviation parameter to obtain a second image.

[0024] Preferably, the steps of the positioning and extraction unit performing positioning and extraction operations on the second image are as follows:

[0025] After binarizing the first image, edge extraction is performed to obtain an edge-extracted image;

[0026] The least circumscribed circle fitting method is used to perform circle fitting on the edge-extracted image, a mask is set, and the working surface is extracted to obtain a region image.

[0027] Preferably, the preprocessing of the region image by the defect detection unit includes: median filtering for image denoising and image enhancement using the Laplacian operator.

[0028] Preferably, the defect extraction steps of the defect detection unit are as follows:

[0029] The OTSU algorithm is used to segment and extract the defects, the improved Canny operator is used for scratch detection, the method of combining probabilistic Hough transform is used for scratch connection, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

[0030] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structure specifically pointed out in the written specification and the drawings.

[0031] Next, through the drawings and embodiments, the technical solutions of the present invention will be further described in detail. Description of the Drawings

[0032] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0033] Figure 1 It is a schematic diagram of an on-line brake disc defect detection method based on machine vision in an embodiment of the present invention;

[0034] Figure 2 It is a schematic diagram of working surface positioning in an embodiment of the present invention;

[0035] Figure 3 It is a schematic diagram of the principle of the Canny operator during the extraction of the working surface in an embodiment of the present invention;

[0036] Figure 4 It is a schematic diagram of linear interpolation of pixel gradient direction in an embodiment of the present invention;

[0037] Figure 5 It is a convolution template diagram in an embodiment of the present invention;

[0038] Figure 6 It is a program flow chart of the Canny operator during the defect extraction and recognition in an embodiment of the present invention;

[0039] Figure 7 It is a schematic diagram of an on-line brake disc defect detection system based on machine vision in an embodiment of the present invention. Detailed Embodiments

[0040] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for illustrating and explaining the present invention, and are not used to limit the present invention.

[0041] An embodiment of the present invention provides an online detection method for brake disc defects based on machine vision, as Figure 1 shown, including:

[0042] Step 1: Collect a first image on the finished product conveying device through a first image acquisition device;

[0043] Step 2: According to the conveying direction and conveying speed of the finished product conveying device, perform reduction processing on the first image to obtain a second image;

[0044] Step 3: Perform a positioning extraction operation on the second image to obtain a regional image corresponding to the brake disc;

[0045] Step 4: Preprocess the regional image and perform defect extraction. After extracting the defect, output a first alarm message.

[0046] Among them, the first image acquisition device includes a gantry arranged above the conveyor belt of the finished product conveying device, a spherical integral light source arranged on the gantry, a camera arranged in the spherical integral light source, a control unit, and a communication unit; the control unit is electrically connected to the spherical integral light source, the camera, and the communication unit respectively; the control unit controls the opening and closing of the spherical integral light source, the control unit obtains the first image collected by the camera, and sends the first image to the industrial big data platform through the communication unit. Since this application is for photographing and detecting the brake discs transported on the conveyor belt; when photographing, the brake discs on the conveyor belt are moving and will be distorted; therefore, it is necessary to restore the distortion. Therefore, the steps for the industrial big data platform to perform reduction processing on the first image include:

[0047] According to the pixel sampling interval, conveying direction, and conveying speed of the first image, query the preset deviation parameter determination table to retrieve the deviation parameter; in the deviation parameter determination table, the pixel acquisition interval, conveying speed, conveying direction, and deviation parameter are correspondingly associated; the pixel sampling interval is the minimum value of the imaging time difference between two adjacent pixels in the first image, and the conveying speed is the distance that the items on the conveyor belt move per unit time;

[0048] Construct a restoration reconstruction map according to the starting pixel point and pixel sampling order in the first image, and re-sample the pixels after the starting pixel point from the first image according to the deviation parameter to obtain a second image.

[0049] A deviation parameter determination table is used to determine the deviation parameters of pixels when the pixel sampling interval, the conveying direction and the conveying speed of the finished product conveying device are known. The deviation parameters include a lateral deviation value and a longitudinal deviation value. During resampling and reconstruction, the coordinates of the pixel points after the starting pixel point in the restored and reconstructed image are respectively added with the lateral deviation value and the longitudinal deviation value to obtain the extracted coordinates, and the pixel values ​​of the extracted coordinates in the first image are used as the values ​​of the pixel points in the restored and reconstructed image. In addition, it is also possible to determine whether there is an abnormality based on the deviation parameters. When the lateral deviation value in the deviation parameters is less than or equal to a preset first threshold value or the longitudinal deviation value is less than or equal to a preset second threshold value, it is determined that the camera setting is abnormal, and an alarm can be issued by a first alarm device set next to the first image acquisition device to remind the staff to make adjustments.

[0050] The steps of performing a positioning extraction operation on the second image are as follows:

[0051] Binarizing the first image and then extracting edges to obtain an edge-extracted image;

[0052] The minimum circumscribed circle fitting method is used to fit the edge extraction image, set the mask, extract the working surface, and obtain the regional image.

[0053] The brake disc is a regular circular shape. Through the illumination of a circular spherical integral light source, it is found that there is a significant difference in the grayscale value between the brake disc area and the non-brake disc area. Based on this, the edge detection and binarization algorithms can be used to locate the brake disc. However, when the edge detection algorithm is used to locate and extract the brake disc, the real-time recognition is not enough, and the pixels of different brake disc edge images are different. The threshold needs to be re-debugged, and a step of least squares fitting circle operation needs to be added, which invisibly increases the workload, so this method cannot meet actual needs. When the binarization algorithm is used to locate and extract the brake disc, it is found that this method can quickly and accurately extract the brake disc area based on the grayscale difference, so the binarization algorithm is used to locate and extract the brake disc.

[0054] The weighted average method is used to grayscale the brake disc. The formula is as follows:

[0055] Gray(i,j)=0.299R(i,j)+0.578G(i,j)+0.114B(i,j)

[0056] Among them, R(i,j), G(i,j), B(i,j) are the pixel values ​​of the red, green, and blue color channels respectively, and Gray(i,j) is the grayscale value.

[0057] Then use the threshold operator to binarize the image, traverse the brake disc contour, and use the maximum contour to intercept the brake disc area.

[0058] This application mainly focuses on the scratch defects on the working surface of the brake disc. Accurately extracting the working surface can reduce the complexity of subsequent inspections and improve the inspection speed. The implementation process of working surface positioning is as follows Figure 2 shown.

[0059] A circle fitting algorithm based on the Canny operator is used to extract the working surface. The Canny operator is also known as the optimal edge detection algorithm. It uses a dual-threshold method to determine potential boundaries, that is, setting an upper threshold and a lower threshold. The schematic diagram of the Canny operator is as follows Figure 3 shown,

[0060] Among them, enoising: The Gaussian kernel used in Gaussian filtering is a Gaussian function with two dimensions, x and y, and the standard deviation σ is generally the same in both dimensions, and the form is:

[0061]

[0062] Gradient: Calculate the gradient magnitude and direction according to the Sobel operator to find the gradient of the image. The Sobel operator is two 3×3 matrices, namely S x and S y . The former is used to calculate the pixel gradient matrix G x in the x direction of the image, and the latter is used to calculate the pixel gradient matrix G y in the y direction of the image. The specific form is:

[0063]

[0064] Among them, I is the grayscale image matrix, and the * here represents the cross-correlation operation. Then from:

[0065]

[0066] the gradient intensity matrix G xy can be calculated.

[0067] Non-maximum suppression: Compare the current pixel gradient intensity with the gradient intensities of adjacent pixels along the positive and negative gradient directions. If it is the largest (i.e., an extreme value), then retain this pixel as an edge point. If it is not the largest, then suppress it and do not use it as an edge point. For more accurate calculation, in this paper, linear interpolation is used between two adjacent pixels across the gradient direction to obtain the pixel gradient to be compared, as shown in Figure 4 shown: The adjacency situation of pixels can be divided into 4 regions, and each region contains upper and lower parts. If the gradient intensity of the central pixel point in the x direction is g x (i,j), and the gradient intensity in the y direction is g y (i,j), and the gradient intensity is gxy (i,j), then according to g x (i,j) and g y The positive and negative and size of (i, j) can determine the area to which its gradient direction belongs, and then the two gradient intensities g involved in the comparison of the positive and negative gradient directions are obtained according to the linear interpolation of its pixel gradient direction and the pixel gradient of the adjacent points. up (i,j) and g down (i,j). The formula is as follows:

[0068] g up (i,j)=(1-t)·g xy (i,j+1)+t·g xy (i-1,j+1)

[0069] g down (i,j)=(1-t)·g xy (i,j-1)+t·g xy (i+1,j-1)

[0070] The calculation methods for the other three regions are similar. It should be noted that if g x (i,j)=g y (i, j) = 0, indicating that the pixel has no pixel gradient and is not an edge.

[0071] Hysteresis threshold: define a high threshold and a low threshold. Pixels with gradient strength lower than the low threshold are suppressed and not considered as edge points; pixels with gradient strength higher than the high threshold are defined as strong edges and retained as edge points; pixels between the high and low thresholds are defined as weak edges and reserved for further processing. Afterwards, due to uneven illumination and interference problems that cannot be solved by preprocessing of the extracted grayscale image itself, an accurate circle fitting algorithm helps to separate the working area from the non-working area. At present, random Hough transform circle detection, least squares circle fitting, random sampling consistent circle fitting, and minimum circumscribed circle fitting are all commonly used circle fitting methods. Since there are fewer free interfering pixels on the edge of the disk after threshold segmentation, this application uses the minimum circumscribed circle fitting method to perform circle fitting on the threshold segmented image, set a mask, and thus extract the working surface.

[0072] The minimum circumscribed circle fitting uses the Welzl algorithm, which is a recursive randomized algorithm. The specific steps of the algorithm are as follows: Assume that there is a set of points {P1, P2, …, P n}, where P i =(x i ,y i ) represents a point on the plane. The center of the smallest circumscribed circle (x c ,y c ), with radius R. Then:

[0073]

[0074] The minimum circumscribed circle can be regarded as the smallest circle that encloses a given set of points, such that all points in the point set are inside or on the circle. The algorithm finally finds the minimum circumscribed circle of the point set and returns the center and radius, filters out the required circles according to the radius size, sets the mask area, and locates and extracts the working surface of the brake disc.

[0075] Among them, the preprocessing of the regional image includes: median filtering for image denoising and image enhancement using the Laplacian operator.

[0076] Among them, the defect extraction steps are as follows:

[0077] The OTSU algorithm is used to segment and extract the defects, the improved Canny operator is used for scratch detection, the method of combining probabilistic Hough transform is used for scratch connection, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

[0078] When collecting the brake disc image, due to the influence of the surrounding environment and the camera itself, inevitable noise interference will be generated on the image. Therefore, it is necessary to perform denoising processing on the image. The median filtering algorithm has a good effect, so it is selected; due to uneven illumination, the inner ring of the denoised image is brighter. In order to make the scratches clearer, image enhancement processing is also required;

[0079] The Laplacian operator can well highlight the scratches. It can be seen that no matter what color the scratches are, they will appear as prominent white after being processed by the Laplacian operator, which is beneficial to the subsequent scratch extraction. Therefore, the Laplacian operator is selected for image enhancement.

[0080] The definition of the Laplacian operator is:

[0081]

[0082] In the x direction:

[0083]

[0084] In the y direction:

[0085]

[0086] Therefore, the difference approximation of the Laplacian operator is:

[0087]

[0088] This application selects Figure 5 the convolution template for convolution.

[0089] The central coefficient of the template is negative. It is necessary to subtract the image after Laplace transform from the original image to obtain a sharpening effect. Therefore, the basic method of Laplace image enhancement is as follows:

[0090]

[0091] In the formula, f(x,y) is the input image, g(x,y) is the sharpened image, and c = -1.

[0092] Due to uneven illumination, the surface texture of the brake disc, or environmental dust, the segmented image not only contains scratch features but also other noise and interference, which increases the difficulty of scratch extraction. Therefore, it is necessary to segment the scratch area from the complex background. The Otsu method (OTSU) can better extract scratches and can suppress the surrounding texture as much as possible. Therefore, the OTSU algorithm is used for scratch extraction.

[0093] The scratches segmented by OTSU have fracture situations, which affect the correctness of the scratch size and quantity. Therefore, it is necessary to connect the fractured scratches. The fractured scratches can be connected through morphological operations. On the basis of opening operation, an improved Canny operator is proposed for scratch detection, and the method of combining probabilistic Hough transform is used for scratch connection, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

[0094] Among them, the improved Canny operator. The basic steps of the Canny edge detection operator are as follows: first, smooth the image with a Gaussian filter; then calculate the magnitude and direction of the gradient using the finite difference of the first-order partial derivative; then perform non-maximum suppression on the gradient magnitude; finally, use the double-threshold algorithm to detect and connect the edges. In the traditional Canny operator, hysteresis threshold connection is used to connect strong edge pixels with adjacent weak edge pixels to form a continuous edge line. The implementation process is as Figure 6 shown.

[0095] However, only using forward connection may lead to incomplete connection of some edge lines. Therefore, this paper introduces an improvement of bidirectional connection in the Canny operator. In bidirectional connection, first start from strong edge pixels, connect along the edge direction to adjacent weak edge pixels, and mark them as strong edge pixels. Then, continue to start from the new strong edge pixels and connect in the opposite direction to further enhance the continuity of the edge line. In this way, weak edge pixels can be connected from two directions to form a complete edge line, improving the accuracy of Canny edge detection and reducing false alarms.

[0096] Specifically: G x and G y are the gradients of the image I in the horizontal and vertical directions respectively. The Sobel operator is used for gradient calculation, which can be mathematically expressed as: Gx = Sobel x (I); G y = Sobel y = Sobel(I). The magnitude M and direction D of the gradient can be calculated by the following formulas:

[0097]

[0098] D = arctan2(G y , G x )

[0099] The gradient is calculated as M, D = calculate_gradient(I), and then non-maximum suppression is performed: S = non_max_suppression(M, D). The gradient intensity of the current pixel is compared with the gradient intensities of adjacent pixels along the positive and negative gradient directions. If it is the maximum (i.e., an extreme value), the pixel is retained as an edge point; if not, it is suppressed and not considered an edge point. Subsequently, a high threshold and a low threshold are defined for double-thresholding, trong_threshold = max(S) × threshold2; weak_threshold = strong_threshold × threshold1. Pixel points with gradient intensities lower than the low threshold are suppressed; pixel points higher than the high threshold are defined as strong edges and retained as edge points. Finally, two-way connection is implemented. edges[strong_i, strong_j] = 255. If a weak-edge pixel is adjacent to a strong-edge pixel within its 3x3 neighborhood, it is promoted to a strong edge. The coordinates of these newly marked strong-edge pixels are added to strong_i and strong_j. edges[weak_i, weak_j] = 100. If a weak-edge pixel is adjacent to a pixel marked 100 within its 3x3 neighborhood, it is marked 100, which represents the reverse connection and ensures connectivity in the opposite direction. Two-way connection improves the continuity of the detected edges by considering both directions, and this method makes the algorithm more robust to edge breaks or gaps. In addition, Canny edge detection involves computationally intensive tasks such as gradient calculation and non-maximum suppression. ThreadPoolExecutor is used to execute calculate_gradient and non_max_suppression in parallel. Parallel processing aims to accelerate the entire process by executing these tasks in parallel and make full use of the multi-core architecture, effectively utilizing the available resources.

[0100] The improved Canny operator is compared with the directly called Canny operator to quantitatively evaluate the precision, recall, and F1-Score metrics of edge detection.

[0101] Precision:

[0102] Precision = TP / (TP + FP)

[0103] Recall:

[0104] Recall = TP / (TP + FN)

[0105] F1 - score: The harmonic mean of precision and recall.

[0106] F1 - score = 2 * (Precision * Recall) / (Precision + Recall)

[0107] Among them, TP is the number of pixels correctly detected as edges; FP is the number of pixels detected as edges but actually non - edges; TN is the number of pixels correctly detected as non - edges; FN is the number of pixels detected as non - edges but actually edges.

[0108] In terms of the three evaluation metrics of precision, recall, and F1 - score, the improved Canny has achieved positive effects. On average, the improvements in these three metrics are 3.2%, 13.2%, and 6.0% respectively. Specifically: The improvement range of precision is between 1% and 6%. The recall has increased significantly, with the improvement range between 12% and 16%. The F1 - score has increased, with the improvement range between 4% and 10%. These improvement ranges reflect that the optimization of the algorithm has achieved good results in improving accuracy and comprehensiveness. Especially the improvement in recall is the most significant, indicating that the improved algorithm can capture scratch information more comprehensively and effectively, reducing missed detections. Such improvements have practical significance for the application of scratch detection, being able to identify scratches more reliably and providing a more reliable basis for further applications.

[0109] Probabilistic Hough transform. Although the standard Hough transform can accurately detect straight lines in an image, it needs to traverse the entire parameter space and perform accumulations at each point, resulting in a very high computational complexity. Therefore, a new Hough transform, namely the probabilistic Hough transform, is adopted. Suppose the feature points in the original image are

[0110] D = {d i = {x, y i}|i = 1, 2, …}

[0111] Randomly select a pair of feature points (d i , d j ), where d i = (x i, y i ), d j = (x j , y j ), so that the parameters (a, b) can be obtained through the following formula.

[0112]

[0113] The array H(a, b) accumulates continuously according to the obtained results. When calculating until a certain array exceeds a specified maximum value, this maximum H(a, b) array is the detected straight line.

[0114] By proposing an improved Canny edge detection algorithm, bidirectional connection is introduced, and the method of combining probabilistic Hough transform is used to connect scratches, which can better handle complex images and broken edges, improve the accuracy and stability of scratch connection, and suppress the interference of the texture on the surface of the brake disc itself.

[0115] Moreover, based on the improved Canny operator, parallel processing is introduced in this paper to calculate gradients and non-maximum suppression in parallel. By using a thread pool for parallel calculation, the performance of multi-core processors can be fully utilized to accelerate the execution efficiency of the algorithm. It can better connect broken scratches and suppress the interference of other information.

[0116] In one embodiment, the camera of the first image acquisition device is arranged on the lower end surface of the rotating platform, and the rotating platform can drive the camera to rotate in the horizontal plane with the central axis of the camera as the rotation axis;

[0117] In the previous embodiment, when the pixel sampling interval, the conveying direction and the conveying speed of the finished product conveying device are known, the deviation parameters of the pixels are determined through the deviation parameter determination table. It analyzes through three known parameters. By reducing the number of known parameters for analysis, the error in determining the known parameters can be reduced, and the accuracy of restoring the image can be further improved. Therefore, in this embodiment, the rotating platform is used to rotate and adjust the camera, and it is only necessary to ensure that the horizontal scanning direction during image acquisition is perpendicular to the conveying direction of the finished product conveying device. The determination of the deviation parameters can be determined through the second deviation parameter determination table, and the pixel sampling interval, the conveying speed and the deviation parameters in the second deviation parameter determination table are corresponding and associated;

[0118] In order to realize the automatic calibration of the position of the camera, a standard object is conveyed on the finished product conveying device, and the camera performs multiple image acquisitions of the standard object at each rotation position under the rotation of the rotating platform, and measures the length of the item image in the longitudinal scanning direction of the acquired images; The rotation position with the largest measured value is used as the target position after automatic adjustment to rotate and adjust the camera.

[0119] In one embodiment, a second image acquisition device is further provided beside the finished product conveying device. The second image acquisition device includes: a detection platform, a spherical integral light source arranged above the detection platform, a camera arranged in the spherical integral light source, a control unit, and a communication unit; a finished product pushing device (or a grasping robot can also be configured) is arranged on the other side of the finished product conveying device to push the brake disc from the finished product conveying device onto the detection platform. The second image acquisition device performs spot checks on the finished product conveying device by means of spot checks and adopts a static measurement method. The second image collected by the second image acquisition device can be used to check whether the longitudinal scanning direction of the camera of the first image acquisition device is parallel to the conveying direction of the finished product conveying device. Specifically, it is to detect whether the deviation between the length of the first image in the longitudinal scanning direction after reduction and the corresponding length of the second image is within a preset deviation range. When it is not, a prompt for calibrating the first image acquisition device is output.

[0120] The present invention also provides an on-line defect detection system for brake discs based on machine vision, as Figure 7 shown, including: an acquisition unit 1, a reduction unit 2, a positioning and extraction unit 3, and a defect detection unit 4;

[0121] Among them, the acquisition unit 1 acquires a first image on the finished product conveying device through a first image acquisition device; the reduction unit 2 performs reduction processing on the first image according to the conveying direction and conveying speed of the finished product conveying device to obtain a second image; the positioning and extraction unit 3 performs positioning and extraction operations on the second image to obtain a regional image corresponding to the brake disc; the defect detection unit 4 preprocesses the regional image and performs defect extraction, and outputs a first alarm message after extracting the defect.

[0122] Among them, the steps of the reduction unit 2 performing reduction processing on the first image include:

[0123] According to the pixel sampling interval, conveying direction and conveying speed of the first image, query the preset deviation parameter determination table to retrieve the deviation parameters;

[0124] Construct a reduction reconstruction diagram based on the starting pixel point and pixel sampling order in the first image, and resample the pixels after the starting pixel point from the first image according to the deviation parameters to obtain a second image.

[0125] Among them, the steps of the positioning and extraction unit 3 performing positioning and extraction operations on the second image are as follows:

[0126] Perform binarization processing on the first image and then perform edge extraction to obtain an edge extraction image;

[0127] Adopt the minimum circumscribed circle fitting method to perform circle fitting on the edge extraction image, set a mask, and extract the working surface to obtain a regional image.

[0128] Among them, the preprocessing of the defect detection unit 4 for the regional image includes: median filtering for image denoising and image enhancement using the Laplacian operator.

[0129] Among them, the defect extraction steps of the defect detection unit 4 are as follows:

[0130] The OTSU algorithm is used to segment and extract the defects, the improved Canny operator is used for scratch detection, the method combining the probabilistic Hough transform is used for scratch connection, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

[0131] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A brake disc defect online detection method based on machine vision, characterized in that: include: Capturing a first image on the finished product conveying device by a first image acquisition device; According to the conveying direction and conveying speed of the finished product conveying device, the first image is restored to obtain a second image; Performing a positioning extraction operation on the second image to obtain an area image corresponding to the brake disc; Preprocessing the regional image and extracting defects, and outputting the first alarm information after the defects are extracted; The first image acquisition device includes a gantry arranged above the conveyor belt of the finished product conveying device, a spherical integral light source arranged on the gantry, a camera arranged in the spherical integral light source, a control unit and a communication unit; the control unit is electrically connected to the spherical integral light source, the camera and the communication unit respectively; the control unit controls the opening and closing of the spherical integral light source, the control unit obtains the first image collected by the camera through the camera, and sends the first image to the industrial big data platform through the communication unit; The camera of the first image acquisition device is arranged on the lower end surface of the rotating platform, and the rotating platform can drive the camera to rotate in a horizontal plane with the central axis of the camera as the rotating axis; The standard object is transported on the finished product conveying device, and the camera collects images of the standard object multiple times at various rotation positions under the rotation of the rotating platform, and measures the length of the object image in the longitudinal scanning direction of the collected image; the rotation position with the largest measured value is used as the target position after automatic adjustment to rotate the camera; A second image acquisition device is arranged next to the finished product conveying device, and the second image acquisition device includes: a detection platform, a spherical integrating light source arranged above the detection platform, a camera arranged in the spherical integrating light source, a control unit and a communication unit; a finished product pushing device is arranged on the other side of the finished product conveying device to push the brake disc from the finished product conveying device to the detection platform, and the second image acquisition device conducts random sampling from the finished product conveying device by static measurement. The second image acquired by the second image acquisition device can be used to check whether the longitudinal scanning direction of the camera of the first image acquisition device is parallel to the conveying direction of the finished product conveying device; specifically, it is to detect whether the deviation between the length of the first image in the longitudinal scanning direction after restoration and the length corresponding to the second image is within a preset deviation range. If not, a prompt that the first image acquisition device needs to be calibrated is output.

2. The online detection method for brake disc defects based on machine vision according to claim 1 is characterized in that: The step of restoring the first image includes: According to the pixel sampling interval, conveying direction and conveying speed of the first image, a preset deviation parameter determination table is searched to retrieve the deviation parameter; A restored and reconstructed image is constructed according to the starting pixel point and the pixel sampling order in the first image, and the pixels after the starting pixel point are resampled from the first image according to the deviation parameter to obtain the second image.

3. The online detection method for brake disc defects based on machine vision according to claim 1 is characterized in that: The steps of performing a positioning extraction operation on the second image are as follows: Binarizing the first image and then extracting edges to obtain an edge-extracted image; The minimum circumscribed circle fitting method is used to fit the edge extraction image, set the mask, extract the working surface, and obtain the regional image.

4. The online detection method for brake disc defects based on machine vision according to claim 1 is characterized in that: The preprocessing of the regional image includes: median filtering image denoising and Laplacian operator image enhancement.

5. The online detection method for brake disc defects based on machine vision according to claim 1 is characterized in that: The defect extraction steps are as follows: The OTSU algorithm is used to segment and extract defects, the improved Canny operator is used for scratch detection, the probabilistic Hough transform method is combined to connect scratches, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

6. A brake disc defect online detection system based on machine vision, characterized in that: include: Acquisition unit, restoration unit, positioning extraction unit and defect detection unit; The acquisition unit acquires a first image on the finished product conveying device through a first image acquisition device; the restoration unit restores the first image according to the conveying direction and conveying speed of the finished product conveying device to acquire a second image; the positioning extraction unit performs a positioning extraction operation on the second image to obtain a regional image corresponding to the brake disc; the defect detection unit pre-processes the regional image and extracts defects, and outputs a first alarm message after the defects are extracted; The first image acquisition device includes a gantry arranged above the conveyor belt of the finished product conveying device, a spherical integral light source arranged on the gantry, a camera arranged in the spherical integral light source, a control unit and a communication unit; the control unit is electrically connected to the spherical integral light source, the camera and the communication unit respectively; the control unit controls the opening and closing of the spherical integral light source, the control unit obtains the first image collected by the camera through the camera, and sends the first image to the industrial big data platform through the communication unit; The camera of the first image acquisition device is arranged on the lower end surface of the rotating platform, and the rotating platform can drive the camera to rotate in a horizontal plane with the central axis of the camera as the rotating axis; The standard object is transported on the finished product conveying device, and the camera collects images of the standard object multiple times at various rotation positions under the rotation of the rotating platform, and measures the length of the object image in the longitudinal scanning direction of the collected image; the rotation position with the largest measured value is used as the target position after automatic adjustment to rotate the camera; A second image acquisition device is arranged next to the finished product conveying device, and the second image acquisition device includes: a detection platform, a spherical integrating light source arranged above the detection platform, a camera arranged in the spherical integrating light source, a control unit and a communication unit; a finished product pushing device is arranged on the other side of the finished product conveying device to push the brake disc from the finished product conveying device to the detection platform, and the second image acquisition device conducts random sampling from the finished product conveying device by static measurement. The second image acquired by the second image acquisition device can be used to check whether the longitudinal scanning direction of the camera of the first image acquisition device is parallel to the conveying direction of the finished product conveying device; specifically, it is to detect whether the deviation between the length of the first image in the longitudinal scanning direction after restoration and the length corresponding to the second image is within a preset deviation range. If not, a prompt that the first image acquisition device needs to be calibrated is output.

7. The machine vision-based brake disc defect online detection system according to claim 6, characterized in that: The restoration unit performs restoration processing on the first image, including: According to the pixel sampling interval, conveying direction and conveying speed of the first image, a preset deviation parameter determination table is searched to retrieve the deviation parameter; A restored and reconstructed image is constructed according to the starting pixel point and the pixel sampling order in the first image, and the pixels after the starting pixel point are resampled from the first image according to the deviation parameter to obtain the second image.

8. The machine vision-based brake disc defect online detection system according to claim 6, characterized in that: The steps of the positioning extraction unit performing a positioning extraction operation on the second image are as follows: Binarizing the first image and then extracting edges to obtain an edge-extracted image; The minimum circumscribed circle fitting method is used to fit the edge extraction image, set the mask, extract the working surface, and obtain the regional image.

9. The machine vision-based brake disc defect online detection system according to claim 6, characterized in that: The defect detection unit performs preprocessing on the regional image including: median filtering image denoising and using Laplace operator for image enhancement.

10. The machine vision-based brake disc defect online detection system according to claim 6, characterized in that: The defect extraction steps of the defect detection unit are as follows: The OTSU algorithm is used to segment and extract defects, the improved Canny operator is used for scratch detection, the probabilistic Hough transform method is combined to connect scratches, and parallel processing is introduced to shorten the detection time of the improved Canny algorithm.

Images