Turbocharger shell surface defect detection method

Abstract

The invention relates to the technical field of image processing, in particular to a turbocharger shell surface defect detection method, which comprises the following steps: collecting a plurality of turbocharger shell surface grey-scale maps, and carrying out connected domain division on each grey-scale map to obtain a plurality of connected domains; calculating a circumscribed rectangle of the connected domains, taking a central pixel point of the circumscribed rectangle as a central point of an initial neighborhood matrix, and performing layer-by-layer expansion from the central point to the outside according to the size of a preset neighborhood matrix to obtain a plurality of neighborhood matrixes; calculating a density equilibrium index of each connected domain; and traversing the density equilibrium indexes of all the connected domains, and when the minimum value of the density equilibrium indexes is smaller than a preset threshold value, judging that the surface of the turbocharger shell has defects. Through connected domain analysis and density equilibrium index calculation, the influence of external factors on an image analysis result can be reduced, and the accuracy and reliability of defect detection are improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to a method for detecting surface defects of a turbocharger housing. Background Art

[0002] A turbocharger is a device that uses the inertial force of the exhaust gas discharged by the engine to drive the turbine in the turbine chamber, thereby driving the coaxial impeller to compress the air sent from the air filter pipe and pressurize it into the cylinder. By increasing the engine's intake volume, the turbocharger can significantly increase the engine's power and torque, giving the vehicle stronger power output and acceleration performance.

[0003] Prior art, such as a patent application document with publication number CN115018847A, discloses a method for automatic identification and classification of surface defects of metal sheets, the method comprising: obtaining a visible light image of the surface of the metal sheet, and converting the visible light image of the surface of the metal sheet into a grayscale image of the surface of the metal sheet; using the Otsu threshold method to obtain a segmentation threshold corresponding to the grayscale image of the surface of the metal sheet, and using the segmentation threshold to convert the grayscale image of the surface of the metal sheet into a binary image of the surface of the metal sheet; using the minimum circumscribed rectangle method to locate and extract defects in the binary image of the metal sheet, and obtain defect feature maps corresponding to each defect; using a trained deep convolutional neural network to classify the defects corresponding to each defect feature map, and obtain the defect type corresponding to each defect feature map.

[0004] The above method uses the minimum circumscribed rectangle to locate and extract defects in the binary image. However, due to external factors such as the smooth and reflective surface of metal materials, problems such as high light noise and uneven lighting are generated during the image acquisition process, resulting in inaccurate defect detection results when the image is subjected to defect detection and analysis. Summary of the invention

[0005] In order to solve the above technical problem of inaccurate defect detection results, the present invention provides the following solution.

[0006] A method for detecting surface defects of a turbocharger housing, comprising: Collecting multiple grayscale images of the shell surfaces of turbochargers, dividing each of the grayscale images into connected domains, and obtaining multiple connected domains; Calculate the circumscribed rectangle of the connected domain, take the central pixel point of the circumscribed rectangle as the central point of the initial neighborhood matrix, and expand outward layer by layer from the central point according to the preset neighborhood matrix size to obtain multiple neighborhood matrices; Calculate the density balance index of each connected domain, the density balance index is: ; In the formula, Connected domain The density balance index, For the The gradient fluctuation value of the neighborhood matrix, represents standard deviation; The density balance indexes of all connected domains are traversed, and when the minimum value of the density balance index is less than a preset threshold, it is determined that there is a defect on the surface of the turbocharger housing.

[0007] The present invention first divides the collected grayscale image into connected domains, and calculates the circumscribed rectangles of each connected domain to determine the approximate position and range of the shell surface features. Then, the central pixel point of the circumscribed rectangle is used as the central point of the initial neighborhood matrix, and multiple neighborhood matrices are obtained by layer-by-layer expansion. Then, the pixel information around each feature is analyzed in more detail, and the density balance index of the connected domain is calculated. Based on this, it is determined whether there are defects on the shell surface, thereby improving the accuracy and reliability of defect detection.

[0008] Preferably, the connected domain division includes: Perform mixed Gaussian modeling on the grayscale image, determine the center point set based on all the pixel points within the preset number of standard deviation intervals of the Gaussian distribution of each unit, and select the extreme value as the starting pixel point in the center point set; Starting from the starting pixel point, start to build a connected domain. Every time a neighboring pixel point is added to the connected domain, the equalization coefficient of the corresponding neighboring pixel point is calculated. If the equalization coefficient of the neighborhood pixel point exceeds the ratio of the maximum value to the minimum value in the center point set, it is considered that the neighborhood pixel point does not belong to the current connected domain. The traversal process continues until the equalization coefficients of the neighborhood pixel points of all edge pixels in the connected domain exceed the ratio of the maximum value to the minimum value in the center point set, and the division of a connected domain is completed.

[0009] By calculating the equalization coefficient of the neighborhood pixels and comparing it with the ratio of the maximum to the minimum in the center point set, it is possible to determine whether the neighborhood pixels belong to the current connected domain, which helps to exclude pixels that do not belong to the current connected domain and ensure the purity and consistency of the connected domain.

[0010] Preferably, the equalization coefficient satisfies the relationship: ; In the formula, is the first The equilibrium index of the neighboring pixels is For the The pixel value of a pixel, For the The pixel value of the neighboring pixel points, is the total number of neighborhood pixels added.

[0011] Under the influence of external factors, the grayscale distribution of the image may change. Calculating the equalization coefficient helps to enhance the robustness to external factors and improve the consistency and accuracy of the connected domain division.

[0012] Preferably, the size of the preset neighborhood matrix is ​​3×3.

[0013] Preferably, the initial neighborhood matrix further includes: Extract all the pixels contained in a protrusion from the grayscale image to form a protrusion area; The number of pixels in the initial neighborhood matrix is ​​less than or equal to the number of pixels in the protrusion area.

[0014] By extracting the protruding areas in the grayscale image, the initial neighborhood matrix can capture the local features in the image; limiting the number of pixels in the initial neighborhood matrix to not exceed the number of pixels in the protruding areas can reduce the computational complexity in the image processing and analysis process.

[0015] Preferably, the gradient of each pixel in the grayscale image is calculated, and for each pair of continuous pixels in the neighborhood matrix, the corresponding gradient change is calculated, and the number of continuous points within each gradient change range is counted to construct a gradient continuity matrix; then the gradient fluctuation value satisfies the relationship: ; In the formula, For the The gradient fluctuation value of the neighborhood matrix, For the The median value of the gradient continuous matrix of the neighborhood matrix is The probability of is a logarithmic function.

[0016] By calculating the negative logarithm of the probability of occurrence of each gradient value in the gradient continuity matrix multiplied by its probability itself, the uncertainty of the gradient change is quantified. In defect detection, the gradient fluctuation value can help identify abnormal areas in the image. For example, if the gradient fluctuation value of a certain area is significantly higher than that of the surrounding area, it may indicate that there is a defect or abnormality in the area.

[0017] Preferably, a region growing algorithm is used to divide the connected domain.

[0018] Preferably, the width of the narrowest edge among all the edges in all the collected grayscale images is used as the size of the neighborhood matrix.

[0019] When determining the size of the neighborhood matrix, considering the width of the narrowest edge can optimize the use of computing resources. Instead of using a fixed-size neighborhood matrix for the entire image, it can be dynamically adjusted according to the actual edge width, which can reduce unnecessary calculations while maintaining the detection effect.

[0020] The beneficial effects of the present invention are: The present invention realizes accurate detection of surface defects of turbocharger housing by collecting multiple grayscale images, dividing connected domains, calculating circumscribed rectangles and neighborhood matrices, calculating density balance indexes, and traversing and determining defects. This method not only considers the overall information of the image, but also improves the accuracy of detection through detailed local analysis, especially when interfered by external factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 It is a method flow chart of steps S1 to S4 in a method for detecting surface defects of a turbocharger housing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0023] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0024] Reference Figure 1 A method for detecting surface defects of a turbocharger housing includes steps S1 to S4, which are specifically as follows: S1: collecting grayscale images of the shell surfaces of multiple turbochargers, dividing each of the grayscale images into connected domains, and obtaining multiple connected domains.

[0025] In an embodiment of the present invention, a high-resolution industrial camera or other imaging device is selected to ensure that the details of the shell surface can be captured, and a suitable light source, such as an LED light bar or a ring light source, is configured to provide stable and uniform lighting conditions. Image acquisition parameters are set based on experience, so as to capture images of multiple turbocharger shell surfaces.

[0026] In order to facilitate image analysis, the acquired image of the turbocharger housing surface is converted into a grayscale image.

[0027] Due to the uneven light and shadow on the surface of the turbocharger, the amount of information expressed in different areas of the image is inconsistent. Specifically, some areas may be too bright, while others may be too dark. This phenomenon of uneven light and shadow will directly affect the quality of the image, making the judgment results of overly bright or dark areas inaccurate when these images are directly used for defect judgment.

[0028] In order to solve the above problems, it is necessary to divide the surface area of ​​the turbocharger. Taking a grayscale image as an example, the specific steps are as follows: A mixed Gaussian model is performed on the grayscale image, and a central point set is determined based on all the pixel points within a preset number of standard deviation intervals of each unit Gaussian distribution, and an extreme value is selected as a starting pixel point in the central point set; in an embodiment of the present invention, all the pixel points within three standard deviation intervals of each unit Gaussian distribution are set to form a central point set, and the maximum and minimum values ​​of the central point set are respectively used as the starting pixel points for connecting domain division, and the connected domain is constructed starting from the starting pixel point. Each time a neighboring pixel point is added to the connected domain, the equalization coefficient of the corresponding neighboring pixel point is calculated. If the equalization coefficient of the neighboring pixel point exceeds the ratio of the maximum value to the minimum value in the central point set, it is considered that the neighboring pixel point does not belong to the current connected domain, and the traversal process continues until the equalization coefficients of the neighboring pixel points of all edge pixels in the connected domain exceed the ratio of the maximum value to the minimum value in the central point set, that is, the division of a connected domain is completed, and then multiple connected domains are obtained.

[0029] Among them, the balance coefficient satisfies the relationship:

[0030] In the formula, is the first The equilibrium index of the neighboring pixels is For the The pixel value of a pixel, For the The pixel value of the neighboring pixel points, is the total number of neighborhood pixels added.

[0031] In other embodiments, a region growing algorithm may be used to divide the connected domain.

[0032] S2: Calculate the circumscribed rectangle of the connected domain, take the central pixel point of the circumscribed rectangle as the central point of the initial neighborhood matrix, and expand outward layer by layer from the central point according to the preset neighborhood matrix size to obtain multiple neighborhood matrices.

[0033] On the surface of the turbocharger housing, the granular protrusions (which may be tiny defects, impurities or other surface features generated during the manufacturing process) are densely distributed, and these protrusions will form a specific pixel pattern in the image. In a connected domain (i.e., a continuous area in image processing, usually composed of pixels with similar properties), these granular protrusions will cause the existence of some pixels with fixed gradient changes. In other words, the brightness or color changes of these pixels are relatively consistent in the image, and there will be no significant differences due to different positions.

[0034] At the same time, since the density of these granular protrusions is similar on the entire shell surface, the pixel value changes of each corresponding protrusion area will also show similarity.

[0035] Specifically, first, traverse each pixel in the grayscale image to find all the pixels belonging to the same connected domain, and calculate the coordinates of the upper left corner and lower right corner of the bounding rectangle based on the coordinate information of these pixels, so as to determine the position and size of the bounding rectangle.

[0036] The geometric center of the circumscribed rectangle of the connected domain is calculated to determine the central pixel point of the circumscribed rectangle, and the central pixel point is used as the central point of the initial neighborhood matrix. In an embodiment of the present invention, the size of the neighborhood matrix is ​​set to 3×3. In other embodiments, the width of the narrowest edge among all edges in all grayscale images collected can also be used as the size of the neighborhood matrix, and then expanded outward layer by layer from the center point to obtain multiple neighborhood matrices, such as 5×5, 7×7, etc.

[0037] Among them, all the pixels contained in a protrusion are extracted from the grayscale image to form a protrusion area, and the number of pixels in the initial neighborhood matrix constructed above is less than or equal to the number of pixels in the protrusion area. This is because when the size of the neighborhood matrix is ​​large enough to include or exceed the number of pixels contained in a protrusion point, the neighborhood matrix may "average" or "blur" the influence of the protrusion point. In other words, since the neighborhood matrix contains a large number of non-protrusion pixels, the values ​​of these pixels may reduce the significance of the protrusion point in the overall neighborhood matrix.

[0038] By obtaining neighborhood matrices of different sizes, they are used to analyze the feature changes within the connected domain.

[0039] S3: Calculate the density balance index of each connected domain.

[0040] After multiple neighborhood matrices are obtained in step S2, in order to further evaluate the uniformity or consistency of the pixels inside the connected domain, the following steps are performed: Calculate the gradient of each pixel in the grayscale image. For each pair of consecutive pixels in the neighborhood matrix, calculate the corresponding gradient change. Count all the calculated gradient changes, classify them according to the range of the change (such as 1-2, 3-4, ...), and count the number of consecutive points in each range.

[0041] These statistics are organized into a matrix called the gradient continuity matrix.

[0042] Then the gradient fluctuation value of the neighborhood matrix is ​​calculated to satisfy the relationship:

[0043] In the formula, For the The gradient fluctuation value of the neighborhood matrix, For the The median value of the gradient continuous matrix of the neighborhood matrix is The probability of is a logarithmic function.

[0044] The gradient fluctuation value is a statistic calculated based on the gradient continuity matrix, which reflects the instability or volatility of the grayscale changes in the image. Specifically, the gradient fluctuation value is obtained by calculating the negative entropy of the probability of each element in the gradient continuity matrix, where the probability represents the ratio of the number of continuous points within a certain gradient change range to the total number of continuous points. The larger the gradient fluctuation value, the stronger the volatility of the grayscale changes in the image, that is, the image contains more details and texture information; conversely, it means that the image is smoother and has fewer details.

[0045] According to the above calculation The gradient fluctuation value of each neighborhood matrix can be calculated in the same way to obtain the gradient fluctuation value of all neighborhood matrices.

[0046] Then the density balance index of the connected domain satisfies the relationship:

[0047] In the formula, Connected domain The density balance index, For the The gradient fluctuation value of the neighborhood matrix, Represents standard deviation.

[0048] By quantifying the density variation within the connected domain, when the density distribution within the connected domain is relatively uniform, the standard deviation of the gradient fluctuation value of each neighborhood matrix will be smaller, resulting in a larger standard deviation value, that is, a higher density balance index. On the contrary, if the density distribution within the connected domain is uneven, the standard deviation of the gradient fluctuation value will be larger, resulting in a lower density balance index.

[0049] Through the above calculation domain The density balance index of all connected domains can be calculated in the same way.

[0050] S4: traversing the density balance indexes of all connected domains, and when the minimum value of the density balance index is less than a preset threshold, determining that there is a defect on the surface of the turbocharger housing.

[0051] The density balance indexes of all connected domains are traversed to find the minimum value of the density balance indexes of all connected domains, and a threshold is set (the threshold is set to 0.5 in the embodiment of the present invention, and can be adjusted according to the specific situation in other embodiments). If the minimum value is less than the set threshold, it is determined that there is a defect on the surface of the turbocharger housing.

[0052] The present invention first divides the collected grayscale image into connected domains, and calculates the circumscribed rectangles of each connected domain to determine the approximate position and range of the shell surface features. Then, the central pixel point of the circumscribed rectangle is used as the central point of the initial neighborhood matrix, and multiple neighborhood matrices are obtained by layer-by-layer expansion. Then, the pixel information around each feature is analyzed in more detail, and the density balance index of the connected domain is calculated. Based on this, it is determined whether there are defects on the shell surface, thereby improving the accuracy and reliability of defect detection.

[0053] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0054] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for detecting surface defects of a turbocharger housing, characterized in that: include: Collecting multiple grayscale images of the shell surfaces of turbochargers, dividing each of the grayscale images into connected domains, and obtaining multiple connected domains; Calculate the circumscribed rectangle of the connected domain, take the central pixel point of the circumscribed rectangle as the central point of the initial neighborhood matrix, and expand outward layer by layer from the central point according to the preset neighborhood matrix size to obtain multiple neighborhood matrices; Calculate the density balance index of each connected domain, the density balance index is: ; In the formula, Connected domain The density balance index, For the The gradient fluctuation value of the neighborhood matrix, represents standard deviation; Traversing the density balance indexes of all connected domains, and when the minimum value of the density balance index is less than a preset threshold, determining that there is a defect on the surface of the turbocharger housing; Among them, the gradient of each pixel in the grayscale image is calculated, and for each pair of continuous pixels in the neighborhood matrix, the corresponding gradient change is calculated, and the number of continuous points within each gradient change range is counted to construct a gradient continuity matrix; then the gradient fluctuation value satisfies the relationship: ; In the formula, For the The gradient fluctuation value of the neighborhood matrix, For the The median value of the gradient continuous matrix of the neighborhood matrix is The probability of is a logarithmic function.

2. A method for detecting surface defects of a turbocharger housing according to claim 1, characterized in that: The connected domain division includes: Perform mixed Gaussian modeling on the grayscale image, determine the center point set based on all the pixel points within the preset number of standard deviation intervals of the Gaussian distribution of each unit, and select the extreme value as the starting pixel point in the center point set; Starting from the starting pixel point, start to build a connected domain. Every time a neighboring pixel point is added to the connected domain, the equalization coefficient of the corresponding neighboring pixel point is calculated. If the equalization coefficient of the neighborhood pixel point exceeds the ratio of the maximum value to the minimum value in the center point set, it is considered that the neighborhood pixel point does not belong to the current connected domain. The traversal process continues until the equalization coefficients of the neighborhood pixel points of all edge pixels in the connected domain exceed the ratio of the maximum value to the minimum value in the center point set, and the division of a connected domain is completed.

3. A method for detecting surface defects of a turbocharger housing according to claim 2, characterized in that: The equalization coefficient satisfies the relationship: ; In the formula, is the first The equilibrium index of the neighboring pixels is For the The pixel value of a pixel, For the The pixel value of the neighboring pixel, is the total number of neighborhood pixels added.

4. A method for detecting surface defects of a turbocharger housing according to claim 2, characterized in that: The size of the preset neighborhood matrix is ​​3×3.

5. A method for detecting surface defects of a turbocharger housing according to claim 4, characterized in that: The initial neighborhood matrix also includes: Extract all the pixels contained in a protrusion from the grayscale image to form a protrusion area; The number of pixels in the initial neighborhood matrix is ​​less than or equal to the number of pixels in the protrusion area.

6. A method for detecting surface defects of a turbocharger housing according to claim 1, characterized in that: The connected domain is divided using the region growing algorithm.

7. A method for detecting surface defects of a turbocharger housing according to claim 6, characterized in that: The width of the narrowest edge among all the edges in all the collected grayscale images is used as the size of the neighborhood matrix.

Images