CN106600593A - Aluminum ceramic ball surface detect detection method - Google Patents

Abstract

The invention relates to a method for detecting defects on the surface of a medium-alumina ceramic ball: collecting an optical image of the surface of a medium-alumina ceramic ball; performing gray scale transformation on the optical image; constructing a linear smoothing filter to filter the binary image to remove high Frequency components and sharpening details; the high contrast preservation algorithm is used to enhance the filtered image; the threshold edge description method is used to segment the filtered image for the first time to obtain preliminary defect information on the surface of the aluminum ceramic ball; the area is used as a feature for statistical classification, And filter; expand and merge the selected areas into domains, and segment the areas separately; perform linear smoothing operations on the local initial images obtained in the previous step, and use dynamic threshold method to perform image segmentation accurately to obtain accurate medium-aluminum porcelain Defect information on the surface of the ball; perform area statistics and calculate its pixel points. If the pixel point is greater than 0, it is determined that the corresponding medium-aluminum ceramic ball is unqualified. The invention improves the detection accuracy and automation degree, and improves the detection efficiency.

Description

A method for detecting surface defects of medium-aluminum ceramic balls

technical field

The invention relates to a method for detecting surface defects of medium-alumina ceramic balls.

Background technique

Porcelain balls are widely used in petroleum, chemical, fertilizer, natural gas and environmental protection industries, as covering support materials for catalysts in reactors and tower packing. It has the characteristics of high temperature and high pressure resistance, low water absorption and stable chemical properties. It can withstand the corrosion of acid, alkali and other organic solvents, and can withstand the temperature changes in the production process. For aluminum ceramic balls of various sizes, in order to ensure product usability, it is necessary to detect micro-defects on the surface (such as micro-cracks, micro-damages). At present, the surface defect detection of such parts mainly relies on human visual inspection. Due to the influence of the inspector's technology, experience, working environment and visual fatigue, it is easy to have false detection and missed detection, and the efficiency of manual visual inspection is low and lacks accuracy. And normalization, stability and reliability are relatively poor. In order to solve the problems of high difficulty, low efficiency and high missed detection rate of manual visual inspection, it is necessary to introduce an automatic detection technology, which can not only reduce labor costs but also achieve strict control of product quality.

At present, computer vision technology is relatively mature, and has many advantages such as non-contact, fast speed, high precision, and strong anti-interference ability. Requirements for reliability and sensitivity, and easy maintenance.

Contents of the invention

In view of this, the object of the present invention is to provide a method for detecting surface defects of medium-aluminum ceramic balls, which improves the detection accuracy and automation, and improves the detection efficiency.

In order to achieve the above object, the present invention adopts the following technical solution: a method for detecting surface defects of medium-aluminum ceramic balls, which is characterized in that it includes the following steps:

Step S1: collecting an optical image of the surface of the aluminum ceramic ball;

Step S2: performing grayscale transformation on the optical image to obtain a binary image;

Step S3: Construct a linear smoothing filter to filter the binary image, remove high-frequency components and sharpen details, and obtain a filtered image;

Step S4: using a high-contrast preservation algorithm to enhance the filtered image, highlighting the contrast between the cracks on the surface of the medium-alumina ceramic ball and the background;

Step S5: Segment the filtered image for the first time by using the threshold edge description method to obtain several regions, and filter the background in combination with image morphology operations to obtain preliminary defect information on the surface of the medium-aluminum ceramic ball;

Step S6: Statistically classify the several regions based on the area, set the area values M1 and M2, and filter out the regions whose area values are between M1 and M2;

Step S7: Set the expansion coefficient for the screened area to expand the corresponding image, merge the expanded image into a domain, use the contrast method to separate the area separately, and obtain the local area with the same characteristics as the expanded image initial image;

Step S8: Perform linear smoothing operation on the local initial image obtained in step S7 to filter again, and use the dynamic threshold method to perform accurate image segmentation on the re-filtered image to obtain accurate defect information on the surface of the aluminum ceramic ball;

Step S9: Perform area statistics on the accurate defect information on the surface of the medium-aluminum ceramic ball, and calculate its pixel points. If the pixel point is greater than 0, it is determined that the corresponding medium-aluminum ceramic ball is unqualified.

Further, in the step S1, the device for collecting the optical image includes a conveyor belt and a grabbing device for rotating the aluminum ceramic ball, and the grabbing device moves with the conveyor belt above the conveyor belt; There are a number of dark boxes with light sources and industrial cameras used to obtain images inside the dark boxes. The dark boxes are provided with openings for the passage of medium-aluminum ceramic balls in the direction of the conveyor belt; A collection frame, a qualified collection frame is provided below the tail end of the conveyor belt; it also includes a controller connected to the industrial camera, the grabbing device and a motor that controls the movement of the conveyor belt, and is used to receive the optical data collected by the industrial camera. Image, control the grabbing device to loosen or clamp the aluminum ceramic ball and control the movement of the conveyor belt; the controller is connected with the upper computer, transmits the optical image to the upper computer and receives the control command of the upper computer.

Further, the light source is a combined strip light source, which is arranged around the top of the dark box; the inner wall of the dark box is provided with a diffusion plate, the top of the dark box is provided with a camera opening, and an industrial camera is arranged directly above the camera opening.

Further, in the step S3, the linear smoothing filter adopts a local mean value operation, and the gray value of each pixel is replaced by the weight of all values in its local neighborhood, and the calculation formula is:

Among them, M is the total number of pixels in the neighborhood N, h[i,j] is the gray value of the filtered pixel [i,j], f[k,l] is the pre-filtered pixel [k,l] The gray value of the neighboring pixels.

Further, the specific content of the step S4 is as follows:

Step S41: Using the Gaussian kernel to blur the filtered image, the selected 3*3 blur approximation template is:

The formula for Gaussian kernel approximation calculation is:

Step S42: using the weight matrix to convolve the optical image to obtain a blurred image;

Step S43: subtracting the original image from the blurred image to obtain a high-pass image;

Step S44: adding dynamic weights to the original image and the high-pass image, the expression is:

AdImg=BlurImg+Amount*(RawImg-BlurImg+127)

Among them: AdImg is an enhanced image, BlurImg is a blurred image, RawImg-BlurImg+127 is a high-pass image, RawImg is an original image, Amount is a weight, and the value range of the weight is [0,1].

Further, in the step S5, the specific content of the threshold edge description method is as follows:

Step S51: Select an initial approximate threshold value T1 and T2, wherein T1 is smaller than T2;

Step S52: Use the estimated values T1 and T2 to divide the image into three groups of regions R ₁ , R ₂ and R ₃ according to whether the gray value is smaller than T1, larger than T2, and between T1 and T2;

Step S53: merge the regions R ₁ and R ₂ , and re-divide the image into regions L1 and L2;

Step S54: use the derivation method to obtain the second derivation at the junction of the regions L1 and L2, and save the result in matrix form;

Step S55: Set T1' and T2' again for the above-mentioned saved image, and obtain the boundary information of all regions satisfying the above-mentioned conditions and save them in matrix form.

Further, in the step S8, the specific content of the dynamic threshold method is as follows:

Step S81: record the threshold of the image point set after filtering again as g{e}, record the threshold of the point set of the local image as g{o}, and set the reference difference t;

Step S82: Subtract the reference difference t from the value in g{e}, and compare with g{o}, if g{o}<g{e}-t is satisfied, save the point set on the image Otherwise, it is eliminated, and finally the accurately filtered surface defect information of the medium-alumina ceramic ball is obtained and marked.

Compared with the prior art, the present invention has the following beneficial effects: the present invention adopts the combination of computer vision and combined light source diffuse light technology to realize the detection and screening of defects on the surface of aluminum ceramic balls of various sizes, especially to solve the problem of It is difficult to detect small cracks on the surface of ceramic balls. In addition, the invention can realize multi-station detection and solve the problems of large calculation amount and slow detection speed of common algorithms, and can truly reflect the surface defect information of medium-alumina ceramic balls, and can accurately judge the detection object Whether it is qualified or not, and the amount of data is small, the detection efficiency is high, and it has strong practicability and broad application prospects.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention.

Fig. 2 is a schematic diagram of a device for collecting optical images according to an embodiment of the present invention.

Fig. 3 is a partial schematic diagram of a dark box and an industrial camera according to an embodiment of the present invention.

Fig. 4 is a cross-sectional view along line A-A of Fig. 3 .

Fig. 5 is a B-B sectional view of Fig. 3 .

Fig. 6 is a C-C sectional view of Fig. 3 .

In the figure: 1-conveyor belt; 2-grabbing device; 3-medium aluminum ceramic ball; 4-black box; 5-industrial camera; 6-controller; 7-host computer; 8-box; 9-unqualified collection box ; 10-qualified collection frame; 41-combined strip light source; 42-diffusion plate; 43-camera opening; 44-opening.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Please refer to Fig. 1, the present invention provides a kind of surface defect detection method of medium-aluminum ceramic ball, it is characterized in that, comprises the following steps:

Step S1: collecting an optical image of the surface of the aluminum ceramic ball;

Please refer to Fig. 2 to Fig. 6, the device of collecting described optical image comprises conveyer belt 1 and the grasping device 2 that is used for rotating medium aluminum ceramic ball 3, and described grasping device 2 moves with conveyer belt 1 above conveyer belt 1; The conveyor belt is also provided with some dark boxes 4 with light sources and an industrial camera 5 for obtaining images inside the dark box 4, and the dark box 4 is provided with an opening 44 for the passage of the aluminum ceramic ball 3 in the direction in which the conveyor belt 1 advances; While the conveyor belt 1 is moving, the medium-aluminum ceramic balls 3 enter the dark box sequentially with the grasping device 2, and the optical image of the surface of the medium-aluminum ceramic balls 3 is obtained by the industrial camera 5; before entering the next dark box, the grasping device 2 drives the medium The aluminum ceramic ball 3 rotates at a certain angle, and the angle of each rotation is determined by the number of dark boxes, as long as it is ensured that the aluminum ceramic ball rotates 360° in the same plane after passing through all the dark boxes, for example, in this embodiment , is provided with 6 obscura, each time the aluminum ceramic ball rotates 60 °.

Also includes a controller 6, the controller 6 is connected with the industrial camera 5, the grasping device 2 and the motor that controls the movement of the conveyor belt 1, and is used to receive the optical image collected by the industrial camera 5 and control the grasping device 2 to release Or clamp the aluminum ceramic ball and control the movement of the conveyor belt 1; the controller 6 is connected with the host computer 7, transmits the optical image to the host computer 7 and receives the control command from the host computer. An unqualified collection frame 9 is arranged directly below the dark box 4 with a light source, and a qualified collection frame 10 is arranged below the tail end of the conveyor belt 1; Box 8 with opening.

In this embodiment, the upper computer is a computer. After the computer receives the optical image transmitted by the controller, it judges the qualification of the corresponding aluminum ceramic ball according to the following steps S2 to S9. If it is judged that the aluminum ceramic ball is not If it is qualified, the computer sends a control command to the controller to loosen the corresponding grabbing device, and then the aluminum ceramic balls fall to the unqualified collection box below. The device drives the order of the aluminum ceramic ball to rotate, so that the aluminum ceramic ball enters the next dark box. After the aluminum ceramic ball passes through all the dark boxes, the aluminum ceramic ball is qualified as a whole, and the controller controls the grabbing device to release the aluminum ceramic ball at the end of the conveyor belt. The ceramic balls will be dropped into the qualified collection box separately.

In this embodiment, the light source is a combination strip light source 41, which is arranged on the peripheral side of the top of the dark box 4; the inner wall of the dark box 4 is provided with a diffusion plate 42, and the top of the dark box 4 is provided with a camera opening 43. An industrial camera 5 is arranged directly above the camera opening 43; preferably, in order to avoid strong reflections on the surface of the aluminum ceramic ball in the direct light irradiation, the combined strip light source 41 is inclined to the outside at a certain angle during installation, and the light is irradiated on the diffuser plate. On the surface, direct irradiation on the surface of the medium-alumina ceramic ball through diffuse reflection can not only effectively avoid the shadow of the light, but also effectively avoid strong reflection, so as to obtain a clear image of crack defects on the surface of the medium-alumina ceramic ball.

Step S2: performing grayscale transformation on the optical image to obtain a binary image;

Step S3: Construct a linear smoothing filter to filter the binary image, remove high-frequency components and sharpen details, and obtain a filtered image; the linear smoothing filter uses a local mean value operation, and the gray value of each pixel uses its local The weight replacement of all values in the neighborhood, the calculation formula is:

Among them, M is the total number of pixels in the neighborhood N, h[i,j] is the gray value of the filtered pixel [i,j], f[k,l] is the pre-filtered pixel [k,l] The gray value of the neighborhood pixel; for example, take a 3×3 neighborhood at the pixel [i,j], get

The linear smoothing filter can remove high-frequency components and sharp details in the image. The present invention adopts a 9×9 smoothing filter, and its weight template is as follows:

Step S4: using a high-contrast preservation algorithm to enhance the filtered image, so as to highlight the contrast between the cracks on the surface of the medium-aluminum ceramic ball and the background, and facilitate subsequent segmentation processing. The specific method is as follows;

Step S41: Use the Gaussian kernel to blur the filtered image. According to the existing situation, a good effect can be obtained by selecting a weight factor (blurring radius) of 3*3. The selected blurring approximation template of 3*3 is:

The formula for Gaussian kernel approximation calculation is:

Step S42: using the weight matrix to convolve the optical image to obtain a blurred image;

Step S43: Subtracting the original image from the blurred image to obtain a high-pass image, that is, retaining the perfect edge, which is the place where there is a surface defect in this example;

Step S44: adding dynamic weights to the original image and the high-pass image, the expression is:

AdImg=BlurImg+Amount*(RawImg-BlurImg+127)

Among them: AdImg is an enhanced image, BlurImg is a blurred image, RawImg-BlurImg+127 is a high-pass image, RawImg is an original image, Amount is a weight, and the value range of the weight is [0,1].

Step S5: Use the threshold edge description method to perform image segmentation on the filtered image for the first time to obtain several regions, and combine the image morphology operation to filter out the background to obtain preliminary defect information on the surface of the aluminum ceramic ball; the specific content of the threshold edge description method is as follows :

Step S51: Select an initial approximate threshold value T1 and T2, wherein T1 is smaller than T2;

Step S52: Use the estimated values T1 and T2 to divide the image into three groups of regions R ₁ , R ₂ and R ₃ according to whether the gray value is smaller than T1, larger than T2, and between T1 and T2;

Step S53: merge the regions R ₁ and R ₂ , and re-divide the image into regions L1 and L2;

Step S54: Use the derivation method to conduct a second derivation at the junction of the regions L1 and L2. Since the edge features at the junction of the regions L1 and L2 are significantly transformed, the changed extreme value obtained by the second derivation can accurately determine the position of the boundary , the result is saved in matrix form;

Step S55: Set T1' and T2' again for the above-mentioned saved image, obtain the boundary information of all regions that meet the above conditions, and display the boundary with a specific color to distinguish it from the original image, and save the result in matrix form.

Step S6: Statistically classify the qualified areas with the area as the feature, calculate the size of all the screened areas, set the area values M1 and M2, and filter out the area values between M1 and M2 area, which can remove the interference noise due to the characteristics of the surface of the medium-alumina ceramic ball.

Step S7: Set the expansion coefficient for the screened area to expand the corresponding image, merge the expanded image into a domain, use the contrast method to separate the area separately, and obtain the local area with the same characteristics as the expanded image Initial image; the contrast method is as follows:

Perform logical AND operation on the image after the initial gray scale transformation and the merged point set after the above-mentioned expansion processing, and obtain the local initial image after difference with the same characteristics as the image after expansion, which is the segmented image containing the defects of medium-alumina ceramic balls .

Step S8: Perform linear smoothing operation on the local initial image obtained in step S7 to filter again, and use the dynamic threshold method to perform image segmentation on the re-filtered image to obtain accurate defect information on the surface of the aluminum ceramic ball; the dynamic threshold method The specific content is as follows:

Step S81: record the threshold of the image point set after filtering again as g{e}, record the threshold of the point set of the local image as g{o}, and set the reference difference t;

Step S82: Subtract the reference difference t from the value in g{e}, and compare with g{o}, if g{o}<g{e}-t is satisfied, save the point set on the image Otherwise, it is eliminated, and finally the accurately filtered surface defect information of the medium-alumina ceramic ball is obtained and marked.

Step S9: Perform area statistics on the accurate defect information on the surface of the medium-aluminum ceramic ball, and calculate its pixel points. If the pixel point is greater than 0, it is determined that the corresponding medium-aluminum ceramic ball is unqualified.

The above descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

Claims (7)

1. a kind of middle aluminum porcelain ball detection method of surface flaw, it is characterised in that comprise the following steps：

Step S1：The optical imagery on aluminum porcelain ball surface in collection；

Step S2：Greyscale transformation is carried out to the optical imagery, bianry image is obtained；

Step S3：Construction linear smoothing filter is filtered to the bianry image, removes radio-frequency component and sharpens details, obtains To filtering image；

Step S4：Using high contrast retain algorithm the filtering image is strengthened, highlight middle aluminum porcelain ball surface crack with Contrast between background；

Step S5：Image segmentation first is carried out to the filtering image using threshold skirt descriptive method and obtains some regions, with reference to Morphological image computing wiping out background, obtains the defect information on preliminary middle aluminum porcelain ball surface；

Step S6：Statistical classification is carried out for some regions are characterized with area, area value M1 and M2 is set, screening is appeared Region of the product value between M1 and M2；

Step S7：The region setting coefficient of expansion to filtering out expands to corresponding image, and the image after expansion is merged For domain, using to being compared to differ from method, region is individually split, obtained and the characteristics of image identical local and initial figure after expansion Picture；

Step S8：The local and initial image that step S7 is obtained carries out again linear smoothing computing and is filtered, and will filter again Imagery exploitation dynamic thresholding method afterwards carries out image and precisely splits, and obtains the defect information on accurately middle aluminum porcelain ball surface；

Step S9：Area statistics are carried out to the defect information on the accurate middle aluminum porcelain ball surface, its pixel is calculated, if pixel Point is more than 0, then judge that aluminum porcelain ball is unqualified in correspondence.

2. middle aluminum porcelain ball detection method of surface flaw according to claim 1, it is characterised in that：In step S1, adopt Collecting the device of the optical imagery includes conveyer belt and the grabbing device for aluminum porcelain ball in rotation, and the grabbing device is in transmission Band top is with transmission Tape movement；Some camera bellows with light source are additionally provided with directly over the conveyer belt and for obtaining inside camera bellows The industrial camera of image, the camera bellows offers the opening passed through for middle aluminum porcelain ball on the direction of transmission tape travel；It is described Unqualified collection frame is provided with immediately below camera bellows with light source, below the conveyer belt tail end competent collection frame is provided with； Also include controller, the controller and the motor connection of the industrial camera, grabbing device and control conveyer belt, be used for Receive during the optical imagery, control grabbing device that industrial camera collects unclamps or clamp aluminum porcelain ball and control the fortune of conveyer belt It is dynamic；The controller is connected with host computer, by optical image transmission is to host computer and receives the control command of host computer.

3. middle aluminum porcelain ball detection method of surface flaw according to claim 2, it is characterised in that：The light source is combobar Shape light source, the week side of boss being arranged at the top of camera bellows；The inwall of the camera bellows is provided with diffusing panel, camera is provided with the top of camera bellows and is opened Mouthful, the camera opening is arranged above industrial camera.

4. middle aluminum porcelain ball detection method of surface flaw according to claim 1, it is characterised in that：In step S3, institute Linear smoothing filter is stated using local mean value computing, each grey scale pixel value is put with the weights of all values in its local neighborhood Change, computing formula is：

$h\&lsqb;i,j\&rsqb;=\frac{1}{M}\underset{(k,l)\&Element;N}{\&Sigma;}f\&lsqb;k,l\&rsqb;$

Wherein, M is the pixel sum in neighborhood N, and h [i, j] is the gray value of Filtered Picture vegetarian refreshments [i, j], and f [k, l] is filter The gray value of the neighborhood territory pixel point of wavefront pixel [k, l].

5. middle aluminum porcelain ball detection method of surface flaw according to claim 1, it is characterised in that：Step S4 it is concrete Content is as follows：

Step S41：Filtering image is obscured using gaussian kernel, the fuzzy close template of the 3*3 of selection is：

$\left[\begin{array}{ccc}0.9474& 0.11831& 0.09474\\ 0.11831& 0.14776& 0.11831\\ 0.09474& 0.11831& 0.09474\end{array}\right]$

The formula of wherein gaussian kernel approximation computation is：

$G(X,Y)=\frac{1}{2{\mathrm{\&pi;\&sigma;}}^2}{e}^{-(x^2+y^2)/2{\mathrm{\&sigma;}}^2}$

Step S42：Convolution is carried out to optical imagery using weight matrix, broad image is obtained；

Step S43：Original image is subtracted each other with broad image, high-pass image is obtained；

Step S44：Original image and high-pass image are carried out into Dynamic Weights to be added, expression formula is：

AdImg=BlurImg+Amount* (RawImg-BlurImg+127)

Wherein：To strengthen image, BlurImg is broad image to AdImg, and RawImg-BlurImg+127 is high-pass image, RawImg is original image, and Amount is weights, and the span of weights is [0,1].

6. middle aluminum porcelain ball detection method of surface flaw according to claim 1, it is characterised in that：In step S5, threshold The particular content of value edge-description method is as follows：

Step S51：The estimated value T1 and T2 of an initial approximation threshold value are selected, wherein, T1 is less than T2；

Step S52：Using estimated value T1 and T2 image according to gray value whether less than T1, more than T2 and between T1 and T2 it Between be divided into three groups of region R₁、R₂And R₃；

Step S53：Combined region R₁And R₂, image is divided into region L1 and L2 again；

Step S54：Using method of derivation to the secondary derivation of region L1 and L2 region intersection, as a result preserve in the matrix form；

Step S55：T1 ' and T2 ' are once again set up for the image after above-mentioned preservation, all regions of above-mentioned condition are met Boundary information is preserved in the matrix form.

7. middle aluminum porcelain ball detection method of surface flaw according to claim 1, it is characterised in that：In step S8, move The particular content of state threshold method is as follows：

Step S81：Filtered image point set threshold value g { e } will be denoted as again, the threshold value of the point set of topography will be denoted as into g { o }, and set reference difference t；

Step S82：Numerical value in g { e } is individually subtracted into reference difference t, and is compared with g { o }, if meeting g { o }<G { e }-t, then Point set on the image is stayed, is otherwise rejected, finally obtain the middle aluminum porcelain ball surface defect information after accurate filtering, and to it It is marked.