An intelligent detection method for surface defects of fine abrasives

Abstract

The present invention relates to the technical field of image processing. More specifically, the present invention relates to an intelligent detection method for surface defects of fine abrasives, including: collecting a surface image of a diamond grinding wheel, segmenting the surface image using an image segmentation model to obtain a working layer image, and establishing a polar coordinate system with the center point of the working layer image as the origin, and obtaining a pixel point sequence composed of pixel points on the working layer image passed by the ray at preset angles in the polar coordinates. The present invention determines the importance of the positions of each pixel point in the working layer image. For example, the importance of a pixel point in the middle region or in the region close to the edge is different. According to the grinding characteristics that the diamond grinding wheel mainly utilizes the middle region of the working layer and diamond particles during work, the importance degree of each pixel point is obtained and used in the subsequent process of calculating the defects of the diamond grinding wheel, thereby improving the accuracy of the calculation result.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing. More specifically, the present invention relates to an intelligent detection method for surface defects of fine abrasives. Background Art

[0002] The silicon carbide substrate has a relatively high hardness, with a Mohs hardness of 9.25, second only to diamond. The silicon carbide substrate has the characteristics of high hardness, high brittleness, good wear resistance, and extremely stable chemical properties, making the precision machining of the silicon carbide substrate very difficult. Ordinary grinding wheels cannot meet the requirements of the chamfering process of the silicon carbide substrate in terms of processing life and processing efficiency. The diamond grinding wheel is a circular consolidated grinding tool with a through hole in the center, made of diamond abrasive and using metal powder, resin powder, ceramics, and electroplated metal as binders respectively. The diamond grinding wheel has a different structure from ordinary abrasive grinding wheels. Generally, it consists of a diamond abrasive layer, a transition layer, and a matrix. The working layer, also known as the diamond layer, is composed of abrasive, binder, and filler, which is the working part of the grinding wheel. The diamond grinding wheel can be used in the rough grinding and fine grinding stages, and has good cutting performance, which can effectively remove the irregularities and roughness on the surface of silicon carbide.

[0003] Currently, when detecting the defects of diamond grinding wheels, mainly the working layer of the diamond grinding wheel is detected. Since the working layer of the diamond grinding wheel is composed of abrasive, binder, and filler, when there are defects in the working layer, such as particle shedding or pits, the difference in gray values between the defective area and the normal area on the working layer is small and difficult to identify. Therefore, accurate detection of the defective area cannot be performed. Summary of the Invention

[0004] The present invention provides an intelligent detection method for surface defects of fine abrasives, aiming to solve the problem in the related art that when there are defects in the working layer, such as particle shedding or pits, the difference in gray values between the defective area and the normal area on the working layer is small and difficult to identify, so accurate detection of the defective area cannot be performed.

[0005] The present invention provides an intelligent detection method for surface defects of fine grinding tools, including: collecting the surface image of a diamond grinding wheel, segmenting the surface image using an image segmentation model to obtain a working layer image, establishing a polar coordinate system with the center point of the working layer image as the origin, and obtaining a pixel point sequence composed of pixel points on the working layer image passed by the ray at preset intervals in the polar coordinates; calculating the importance degree and defect possibility of the pixel points in the pixel point sequence at each angle in the working layer image to obtain the corresponding importance degree sequence and defect possibility sequence at each angle, where the importance degree of the pixel point reflects the distance between the position of the pixel point and the central region and the gray value of the pixel point, and the defect possibility of the pixel point reflects the change degree of the depth value of the pixel point and the depth values of its surrounding pixel points; calculating the difference degree between the defect possibility sequences at any two angles as the defect degree of the working layer image, and performing defect detection on the surface of the grinding tool based on the size of the defect degree, where the difference degree is also related to the importance degree of the pixel points. By using the image segmentation model to extract the working layer image and combining the polar coordinate system to process the pixel points, the entire working surface can be effectively decomposed into sequences at multiple angles, and the difference degree between the defect possibility sequences at different angles can be calculated. Through this difference degree, the defect degree of the diamond grinding wheel can be accurately evaluated.

[0006] Further, performing defect detection on the surface of the grinding tool based on the size of the defect degree includes: if the defect degree of the working layer image is greater than or equal to a preset error, it is determined that there are defects on the surface of the diamond grinding wheel and an alarm is given.

[0007] Further, the calculation formula for the difference degree is: ; in the formula, is the difference degree between the defect possibility sequence at the th angle and the defect possibility sequence at the th angle, is the importance degree of the th pixel point in the importance degree sequence at the th angle, represents the importance degree of the th pixel point in the importance degree sequence at the th angle; is the number of pixel points passed by the ray at any angle, is the defect possibility of the th pixel point in the defect possibility sequence at the th angle, is the defect possibility of the th pixel point in the defect possibility sequence at the th angle, Represents a normalization function. This difference degree calculation formula makes it have higher sensitivity in the detection of local and tiny defects by refining the combination of defect possibility and importance degree. The consideration of importance degree improves the sensitivity to key areas and can more acutely reveal surface problems.

[0008] Furthermore, acquire the surface image of the diamond grinding wheel, including: using a depth camera to photograph the surface of the diamond grinding wheel to obtain the surface image of the diamond grinding wheel.

[0009] Furthermore, use an image segmentation model to segment the surface image, including: inputting the surface image into the image segmentation model to output the working layer image. By inputting the working layer image, the segmentation model can automatically distinguish the working layer from other irrelevant parts and extract the area related to defect detection. In this way, subsequent analysis can focus on the working layer without having to process all the information in the entire image, thus greatly improving the efficiency and accuracy of subsequent processing.

[0010] Furthermore, the training process of the image segmentation model includes: marking the pixel points belonging to the working layer in the surface image of the diamond grinding wheel as 1 and the pixel points belonging to the non-working layer as 0 to obtain a 0-1 marked image; inputting the 0-1 marked image into a semantic segmentation network model for training; completing the training when the loss function is less than a preset value or reaches a preset number of training times to obtain the image segmentation model.

[0011] Furthermore, obtain a pixel point sequence, including: establishing a ray in polar coordinates at every preset angle and forming a pixel point sequence by all the pixel points that the ray passes through in the working layer image, so as to obtain the pixel point sequence corresponding to each angle.

[0012] Furthermore, obtain the importance degree of each pixel point in the working layer image, including: performing edge detection on the working layer image to obtain two circular edges, which are the first circular edge and the second circular edge respectively; the importance degree of each pixel point in the working layer image is related to the difference between the distances of the pixel point from the first circular edge and the second circular edge respectively, and is also related to the difference between the gray value of the pixel point and the average gray value of all pixel points in the working layer image. By comprehensively considering the distance difference between the pixel point and the edge and the difference in gray value, the importance of each pixel point in the working layer image can be more comprehensively evaluated, enhancing the accuracy and reliability of detection.

[0013] Further, obtaining the defect possibility of each pixel point includes: establishing a sliding window in the working layer image, calculating the average value of the differences between the depth value of the pixel point and the depth values of all pixel points within the sliding window, and taking the product of the average value and the variance of the depth values of all pixel points within the sliding window as the defect possibility of the pixel point. The differences and variances of the local depth values reflect the surface morphology and possible damages (such as abrasion, cracks or depressions) of this area, and can reveal the subtle defects in the grinding wheel surface that are not easily captured by the global information, thereby improving the sensitivity of detection.

[0014] Further, the preset angle is 1 degree.

[0015] Beneficial effects:

[0016] (1) By the importance of the positions of the pixel points in the working layer image, for example: the pixel points are in the middle area or near the edge area, and according to the grinding characteristics that the diamond grinding wheel mainly uses the middle area of the working layer and the diamond particles during work, the importance degree of each pixel point is obtained and utilized in the subsequent steps of calculating the defects of the diamond grinding wheel, improving the accuracy of the calculation results.

[0017] (2) By calculating the defect possibility of each pixel point through the depth change of the pixel points in the working layer image, and using the importance degree of the pixel points as weights to weight the defect possibility of the pixel points, the accurate defect degree can be obtained, improving the accuracy of defect detection. Description of the drawings

[0018] Figure 1 is a schematic top view structural diagram of a diamond grinding wheel according to an embodiment of the present invention;

[0019] Figure 2 is a schematic flowchart of calculating the defect degree of the working layer image according to an embodiment of the present invention; Detailed implementation manners

[0020] The following will describe in detail the detailed implementation manners of the present invention with reference to the drawings.

[0021] As Figure 1 and Figure 2 shown, S101: Collect the grinding wheel image.

[0022] Specifically, use a camera device to photograph the diamond grinding wheel to obtain the surface image of the diamond grinding wheel. The camera device is a depth camera, and perform grayscale processing on the surface image of the diamond grinding wheel. The grayscale processing reduces the complexity of the subsequent step calculations and better extracts and analyzes the features in the image.

[0023] In one embodiment, during the grinding of silicon carbide using a diamond grinding wheel, it is the working layer that effects the grinding of silicon carbide. When there are defects in the working layer ( Figure 1 the shaded area in), it will directly affect the grinding effect of silicon carbide. Therefore, in this embodiment, it is necessary to segment the working layer to detect the defects of the working layer. Specifically, the surface image of the diamond grinding wheel is input into the trained image segmentation model, and the working layer image and the non-working layer image are output. Thus, the working layer image of the diamond grinding wheel can be obtained. Among them, the process of training the image segmentation model is as follows: The pixel points belonging to the working layer in the surface image of the diamond grinding wheel are marked as 1, and the pixel points belonging to the non-working layer are marked as 0, so as to obtain a 0-1 labeled image; then the 0-1 labeled image is input into the semantic segmentation network model for training; when the loss function is less than the preset value or reaches the preset number of training times, the training is completed, and the trained image segmentation model is obtained.

[0024] In one embodiment, a polar coordinate system is established with the midpoint of the working layer image as the origin. In the polar coordinate system, a ray is constructed in the working layer image at every preset angle from 0 degrees to 360 degrees, where the preset angle is 1 degree. In other embodiments, the preset angle can be 2 degrees or 5 degrees, etc., and can be adjusted manually according to the specific situation. And a pixel point sequence is constructed based on the pixel points of the working surface passed by the ray at each angle.

[0025] S102: Calculate the importance degree of each pixel point.

[0026] In one embodiment, during the use of the diamond grinding wheel, different regions on the working layer of the diamond grinding wheel play different roles in actual use, so their importance is different. For example, the central region of the grinding wheel usually bears the greatest cutting pressure, so its importance is greater; while the edge region may bear less pressure, so its importance is smaller. The reason is that for the central region: usually the grinding effect is the most important. If there are defects in the central region, it will affect the stability and accuracy of the entire grinding process. Therefore, the importance degree of the pixel points in this region is greater. For the edge region: although edge defects may affect the stability of the grinding wheel, they usually have less impact on the grinding effect. Therefore, the importance degree of the pixel points in this region is smaller. Further explanation is that the working layer of the diamond grinding wheel is composed of abrasive, binder and filler. The abrasive (especially diamond abrasive) is the most important component in the diamond grinding wheel, and its hardness and cutting ability are usually very strong. Therefore, in the working layer image, the abrasive usually corresponds to a region with a higher gray value. Therefore, the higher the gray value of the pixel point, the greater the possibility that it belongs to the abrasive, and the greater the importance of the pixel point; the lower the gray value of the pixel point, the smaller the possibility that it belongs to the abrasive, and the smaller the importance of the pixel point. Therefore, the importance degree of the pixel point can be calculated according to the distance between the pixel point and the central region and the gray value of each pixel point.

[0027] In one embodiment, to determine whether each pixel point in the working layer image is close to the edge, it is first necessary to perform edge detection on the working layer image to obtain two edges. For the convenience of subsequent calculations, the above two edges need to be distinguished according to their sizes, and are divided into a first circular edge and a second circular edge, where the first circular edge is the largest edge. Among them, the canny edge detection algorithm can be used to perform edge detection on the working layer image. It should be noted that the geometric center point of the first circular edge can be used as the midpoint of the working layer image.

[0028] In one embodiment, a calculation method is provided to calculate the importance degree of each pixel point, and the calculation formula is: 。

[0029] In the formula, represents the importance degree of the th pixel point in the working layer image, represents the th pixel point in the working layer image, represents the pixel point on the first circular edge of the working layer image that is closest to the th pixel point , represents the pixel point on the second circular edge of the working layer image that is closest to the th pixel point , represents the th pixel point and the th pixel point represents the th pixel point and the th pixel point represents the th pixel point and the th pixel point represents the gray value of the th pixel point in the working layer image, represents the average value of the gray values of all pixel points in the working layer image, is the exponential function with the natural constant as the base. Among them, the distance between any two pixel points can be calculated by the Euclidean distance algorithm.

[0030] Among them, represents the difference between the shortest distances from the th pixel point to the first circular edge and the second circular edge respectively. The larger the difference, the more the th pixel point is biased towards the two edges, and then the The smaller the importance of a pixel point; the smaller the difference indicates that the more the pixel point tends to the middle area between the two edges, then the greater the importance of the pixel point. When the value is equal to 0, the value of is 1, indicating that the smaller the value of the smaller the gray value of the pixel point, the less likely the pixel point belongs to the abrasive, and then the smaller the importance of the pixel point. Thus, the importance of each pixel point in the pixel point sequence can be obtained, and then the importance sequence can be obtained.

[0031] Through the above steps, by combining the distance difference and gray level difference of pixel points to evaluate the importance of pixel points, two important visual features - the region where the pixel point is located and the gray level difference - can be comprehensively considered. The distance difference of the edge reflects the change in physical form, while the gray level difference reflects the change in the brightness of the image or the surface material. Combining the two can more accurately evaluate the importance of each region in the image, especially for those defect regions that are both located in the central region and have significant gray level changes, higher attention can be obtained, thereby improving the accuracy of defect detection.

[0032] S103: Calculate the defect possibility of each pixel point.

[0033] In one embodiment, since the working surface of the diamond grinding wheel is composed of a mixture of abrasive, binder, and filler, when the working surface of the diamond grinding wheel is damaged, the texture of the damaged area is similar to that of the normal area. Therefore, it is difficult to identify the defect area on the working surface of the diamond grinding wheel. However, the depth of the defect area changes compared with the normal area. Therefore, it is necessary to calculate the possibility of each pixel point inputting a defect according to the depth information of each pixel point. And the depth information can be obtained from the depth image of the diamond grinding wheel surface.

[0034] In one embodiment, obtaining the defect possibility of each pixel point includes: establishing a sliding window in the working layer image, calculating the average value of the differences between the depth values of the pixel point and all pixel points in the sliding window, and taking the product of the average value and the variance of the depth values of all pixel points in the sliding window as the defect possibility of the pixel point. Among them, using the sliding window method can focus on each local area in the image, so as to capture local details and small surface changes.

[0035] Specifically, set a sliding window centered on any pixel point in the working layer image, and the size of the sliding window is , in other embodiments, the size of the sliding window can also be and so on. The defect possibility of a pixel point is positively correlated with the difference between the depth value of any pixel point in the sliding window and the average value of the depth values of all pixel points in the sliding window, and is also positively correlated with the variance of the depth values of all pixel points in the sliding window.

[0036] In one embodiment, the calculation formula for the defect possibility of each pixel point is: . In the formula, is the defect possibility of the th pixel point in the working layer image, is the depth value of the th pixel point in the working layer image, is the average value of the depth values of all pixel points in the window, represents the number of all pixel points in the window, is the standard normalization function.

[0037] Among them, represents the difference between the th pixel point in the sliding window and the average value of the depth values of all pixel points in the sliding window. The greater the difference, the greater the defect possibility of the th pixel point. represents the variance of the depth values of all pixel points in the window. The larger this value, the greater the degree of fluctuation of the depth values in the window, and the greater the defect possibility of the th pixel point. Thus, the defect possibility of each pixel point in the pixel point sequence can be obtained, and then the defect possibility sequence can be obtained.

[0038] Exemplarily, when the angle is 1 degree, the sequence of pixel points passed by the ray through the working surface is , is the number of pixel points passed by the ray through the working surface. By calculating the importance degree and defect possibility of each pixel point according to the above method, the importance degree sequence composed of each pixel point at an angle of 1 degree can be obtained as , and the defect possibility sequence composed of each pixel point at an angle of 1 degree can be obtained as .

[0039] S104: Calculate the defect degree of the working layer image.

[0040] It should be noted that when there are no defects on the working surface of the diamond grinding wheel, the difference degree between each defect possibility sequence is basically 0. Therefore, the difference degree between two defect possibility sequences can be used to characterize the defect degree of the working layer image. If the difference degree between two defect possibility sequences is smaller, the possibility of defects on the working surface of the diamond grinding wheel is smaller, and the defect degree of the working layer image is smaller. Also, considering the importance of the position of the pixel points, the importance degree sequence of each pixel point is used to weight it, thereby improving the accuracy of the finally calculated defect degree.

[0041] In one embodiment, the formula for calculating the difference degree between any two defect possibility sequences is as follows: ; where is the difference degree between the defect possibility sequence at the th angle and the defect possibility sequence at the th, is the importance degree of the th pixel point in the importance degree sequence at the th angle, represents the importance degree of the th pixel point in the importance degree sequence at the th angle; is the number of pixel points passed by the ray at any angle, is the th defect possibility of the th pixel point in the defect possibility sequence at the th angle, is the th defect possibility of the th pixel point in the defect possibility sequence at the represents the th comprehensive weight of the th pixel point, is the difference between the th defect possibility of the th pixel point in the defect possibility sequence at the th angle and the

[0042] th defect possibility of the th pixel point in the defect possibility sequence at the th angle. The larger the difference, the greater the difference degree between the two defect possibility sequences, and the greater the possibility of defects. It should be noted that the At this angle, a sequence composed of the defect probabilities of all pixel points on the working surface passed by the ray. Thus, the difference degree between any two defect probability sequences can be calculated, and the difference degree between any two defect probability sequences is used as the defect degree of the working layer image to determine whether there is a defect in the diamond grinding wheel.

[0043] According to the above steps, when calculating the defect degree of the working layer image, not only the difference between the defect probability sequences is considered, but also the sequence of the importance degrees of the pixel points is introduced, which can make the detection result more refined and comprehensive. For example, pixel points with a higher importance degree (pixel points close to the central area and with a higher gray value) occupy a greater weight in the calculation of the difference degree, ensuring that these pixel points that are more critical to the performance of the grinding tool receive more attention, thereby improving the accuracy of defect evaluation.

[0044] S105: Detect the defect based on the size of the defect degree of the working layer image.

[0045] Specifically, if the defect degree of the working layer image is greater than or equal to the preset error, it is determined that there is a defect in the diamond grinding wheel. At this time, an alarm is given to remind the staff to replace or repair the diamond grinding wheel. If the defect degree of the working layer image is less than the preset error, it is determined that there is no defect in the diamond grinding wheel, and the diamond grinding wheel can be continued to be used. The empirical value of the preset error is 2. In other embodiments, the empirical value of the preset error can be 3 or 5, etc., which can be adjusted according to the actual situation. It should be noted that the difference degree between the defect probability sequences corresponding to any two angles is used as the defect degree of the working layer image. When a defect is determined, the defect area can also be located to facilitate the inspection by the staff.

[0046] In one embodiment, locating the defect area includes constructing a matrix based on the difference degrees between all defect probability sequences. The matrix is shown in the following table.

[0047]

[0048] Exemplarily, in the matrix, the difference degree between the defect probability sequence corresponding to Angle 1 and the defect probability sequence corresponding to Angle 359 is 2, indicating that the defect degree of the working layer image is also equal to 2. At this time, the defect degree of the working layer image is equal to the preset error. At this time, it is determined that there is a defect in the diamond grinding wheel, and an alarm is given to remind the staff to replace or repair the diamond grinding wheel. Thus, the defect detection of the diamond grinding wheel is completed.

[0049] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. An intelligent detection method for surface defects of fine grinding tools, characterized in that: include: The surface image of the diamond grinding wheel is collected, and the surface image is segmented using an image segmentation model to obtain a working layer image, and a polar coordinate system is established with the center point of the working layer image as the origin, and a pixel point sequence composed of pixel points on the working layer image passed by the ray at every preset angle in the polar coordinates is obtained; The importance and defect probability of the pixels in the pixel sequence at each angle in the working layer image are calculated to obtain the corresponding importance sequence and defect probability sequence at each angle, wherein the importance of the pixel reflects the distance between the position of the pixel and the central area and the gray value of the pixel, and the defect probability of the pixel reflects the degree of change between the depth value of the pixel and the depth values ​​of the surrounding pixels; the calculation formula for the importance of each pixel is: ； In the formula, Indicates the first The importance of each pixel, Indicates the first pixels, Indicates the distance from the first circular edge to the first circular edge in the working layer image. The nearest pixel , Indicates the distance from the second circular edge to the first The nearest pixel , Indicates Pixels and The distance between pixels, Indicates Pixels and The distance between pixels, Indicates Pixels and The distance between pixels, Indicates the first The gray value of a pixel, Represents the average gray value of all pixels in the working layer image. The natural constant An exponential function with base ; Establishing a sliding window in the working layer image, calculating an average value of the difference between the depth values ​​of the pixel point and all the pixel points in the sliding window, and taking the product of the average value and the variance of the depth values ​​of all the pixel points in the sliding window as the defect possibility of the pixel point; The calculation formula for the defect probability of each pixel is: ; In the formula, For the working layer image The probability of pixel defects is For the working layer image The depth value of each pixel, is the mean depth value of all pixels in the window, Indicates the number of all pixels in the window. is the standard normalization function; The difference between the defect possibility sequences at any two angles is calculated as the defect degree of the working layer image, and defect detection is performed on the mold surface based on the magnitude of the defect degree, wherein the difference is also related to the importance of the pixel point.

2. The intelligent detection method for surface defects of fine grinding tools according to claim 1 is characterized in that: Defect detection is performed on the surface of the mold based on the degree of the defect, including: If the defect degree of the working layer image is greater than or equal to the preset error, it is determined that a defect occurs on the surface of the diamond grinding wheel and an alarm is issued.

3. The intelligent detection method for surface defects of fine grinding tools according to claim 1 is characterized in that: The calculation formula of the difference is: ； In the formula, For the The defect possibility sequence under the angle and the The difference between the defect probability sequences is For the The first in the importance sequence under each angle The importance of each pixel, Indicates The first in the importance sequence under each angle The importance of each pixel; is the number of pixels that the ray passes through at any angle, For the The first in the defect possibility sequence under the angle The probability of pixel defects is For the The first in the defect possibility sequence under the angle The probability of pixel defects is Represents the normalization function.

4. The intelligent detection method for surface defects of fine grinding tools according to claim 1 is characterized in that: Acquire surface images of diamond grinding wheels, including: The surface of the diamond grinding wheel is captured by a depth camera to obtain a surface image of the diamond grinding wheel.

5. The intelligent detection method for surface defects of fine grinding tools according to claim 1 is characterized in that: Surface image segmentation using image segmentation models, including: The surface image is input into an image segmentation model, and a working layer image is output.

6. The intelligent detection method for surface defects of fine grinding tools according to claim 5 is characterized in that: The training process of the image segmentation model includes: The pixel points belonging to the working layer in the surface image of the diamond grinding wheel are marked as 1, and the pixel points belonging to the non-working layer are marked as 0, so as to obtain a 0-1 marked image; Inputting the 0-1 labeled image into a semantic segmentation network model for training; In response to the loss function being less than a preset value or reaching a preset number of training times, the training is completed to obtain an image segmentation model.

7. The intelligent detection method for surface defects of fine grinding tools according to claim 1 is characterized in that: Get pixel sequence, including: A ray is established in polar coordinates at every preset angle, and all the pixels of the working layer image through which the ray passes are combined into a pixel point sequence, thereby obtaining a pixel point sequence corresponding to each angle.

8. The intelligent detection method for surface defects of fine grinding tools according to claim 1 is characterized in that: Get the defect possibility of each pixel, including: A sliding window is established in the working layer image, the average value of the difference between the pixel point and the depth values ​​of all the pixels in the sliding window is calculated, and the product of the average value and the variance of the depth values ​​of all the pixels in the sliding window is used as the defect possibility of the pixel point.

9. The intelligent detection method for surface defects of fine grinding tools according to claim 1, characterized in that: The preset angle is 1 degree.

Images