Electron beam welding quality detection method and system

Abstract

The present application belongs to the technical field of image data processing, and particularly relates to an electron beam welding quality detection method and system, which comprises the following steps: acquiring a weld seam image under irradiation of a fixed direction light source; recognizing and extracting a defect patch and its outwardly expanded neighborhood and reconstructing into a local three-dimensional surface; acquiring a shadow depth index representing the depth of the defect patch by analyzing the shadow characteristics of the defect patch under directional light irradiation; acquiring a stress concentration degree representing the sharpness of the micro-morphology of the defect patch and the consistency of the direction based on the maximum principal curvature and the direction of each point of the reconstructed three-dimensional surface; obtaining a comprehensive hazard index by the shadow depth index and the stress concentration degree, and detecting and evaluating the quality of the defect patch according to the index. The present application distinguishes high-risk defect patches from surface flaws by evaluating the depth and stress concentration degree of the defect patch, and provides a reliable basis for welding quality evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing. More particularly, the present application relates to a method and system for detecting the quality of electron beam welding. BACKGROUND

[0002] As a kind of high energy density precision welding technology, electron beam welding is applied to high-end manufacturing fields such as aerospace and nuclear industry, and the quality of its weld directly relates to the safety and reliability of the product. Therefore, it is crucial to accurately and efficiently detect and evaluate the defect patches of the weld. At present, the automatic detection method based on machine vision is the mainstream means to ensure the welding quality, but most of the existing methods are limited to the two-dimensional image level, and the defect patches are identified by analyzing the gray scale, texture and other information of the weld surface. This method can only determine the existence and planar size of the defect patch, but it is difficult to obtain the three-dimensional geometric information of the defect patch in the depth direction. The real degree of harm of the defect patch to the mechanical properties of the weld not only depends on its area, but also crucially depends on its depth and microscopic morphology. The existing technology lacks evaluation of three-dimensional features, which leads to inaccurate judgment of the actual harm level of the defect patch, and may misjudge harmless superficial flaws as serious defect patches or miss real dangerous deep cracks.

[0003] In order to solve the above problems, the industry has begun to try to introduce a three-dimensional reconstruction algorithm based on a single image, such as the Shape-from-Shading (SFS) technology, which uses the light and dark changes formed by the object surface under certain lighting to infer its three-dimensional shape, thereby estimating the macroscopic depth of the defect patch to some extent. However, when the SFS algorithm is applied to the detection of weld defect patches, it mainly focuses on the overall concave or convex degree of the defect patch, and is not sensitive to the microscopic sharp morphology which determines the key factor of stress concentration. A deep and flat pit and a crack with the same depth but a sharp bottom have very different harmfulness, but the macroscopic depth information inferred by the SFS algorithm alone cannot effectively distinguish them. Therefore, although the existing technology can estimate the depth of the defect patch, it ignores the decisive influence of the microscopic morphology on the harmfulness, so the evaluation result is still not comprehensive and accurate enough. SUMMARY

[0004] To solve the technical problem that the existing electron beam welding quality detection method cannot effectively evaluate the influence of the depth characteristics and microscopic morphology of the defect patch, the present application provides solutions in the following aspects.

[0005] In a first aspect, the present application provides an electron beam welding quality detection method, comprising: acquiring an electron beam weld image under illumination of a fixed direction light source, pre-processing and binarizing the image, and identifying potential defect patches; for each defect patch, performing connected domain aggregation and extracting an outer expansion neighborhood thereof; acquiring an average gray value of a normal weld area around the defect patch based on the outer expansion neighborhood; reconstructing a local three-dimensional surface of the defect patch based on the outer expansion neighborhood; setting a standard calibration object and determining a light source direction unit vector based on an image of the standard calibration object, acquiring a shadow depth index of the defect patch based on the light source direction unit vector, a geometric centroid of the defect patch, and the average gray value of the normal weld area around the defect patch; calculating a Hessian matrix of each pixel point on the surface of the defect patch for the local three-dimensional surface of the defect patch, extracting a maximum principal curvature and a direction vector thereof; acquiring a stress concentration degree of the defect patch based on the maximum principal curvature and the direction vector of each pixel point; multiplying the shadow depth index and the stress concentration degree of the defect patch to obtain a hazard index of the defect patch, and completing detection and quality evaluation of the defect patch based on the hazard index.

[0006] The present application provides an electron beam welding quality detection method, which can comprehensively evaluate the macroscopic depth and microscopic geometric morphology of weld defect patches, so as to more accurately judge the actual hazard degree of the weld defect patches to the structural safety. By reconstructing two-dimensional image information into a local three-dimensional surface, the present application not only evaluates the depth information of the defect patches by shadow analysis, but also evaluates the stress concentration degree thereof by curvature analysis. Finally, the two key indicators are multiplied to obtain a comprehensive hazard index. This method can effectively distinguish between truly dangerous defect patches and defect patches with lower hazard, and provides a more profound and comprehensive physical basis for automatic evaluation of welding quality.

[0007] Preferably, the connected domain aggregation and the extraction of the outer expansion neighborhood of each defect patch comprise: defining a structure element; performing an inflation operation on a binary mask of each independent defect patch by applying the structure element to the binary mask, to form an outer expansion area containing the original defect patch.

[0008] Preferably, the reconstruction of the local three-dimensional surface of the defect patch based on the outer expansion neighborhood comprises: extracting all pixel points and corresponding gray values in the outer expansion neighborhood of the defect patch; taking the pixel coordinates as the bottom coordinates in a three-dimensional space and taking the gray values as the height values, each defect patch is reconstructed into a local three-dimensional surface.

[0009] Preferably, the shadow depth index of the defect patch satisfies the expression: ; wherein, is a defect patch index; is the i-th defect patch; is the j-th defect patch. is the shadow depth index of the defect patch is the shadow depth index of the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the light source direction unit vector is the dot product of two vectors is the sign function is the absolute value sign.

[0010] In the process of constructing the shadow depth index, the present application utilizes the gray scale difference between the light side and the back light side of the defect patch under the fixed direction light source, combines it with the position relationship of each point within the defect patch relative to the center of mass for evaluation; a deeper and steeper defect patch will produce more significant and asymmetric shadow, thereby obtaining a higher index value. This method provides a simple and robust index for evaluating the macroscopic severity of the defect patch, avoiding the complex true three-dimensional reconstruction process.

[0011] Preferably, the local three-dimensional surface of the defect patch calculates the Hessian matrix of each pixel point on the surface of the defect patch, extracts the maximum principal curvature and its direction vector of each pixel point, comprising: for each local three-dimensional surface corresponding to each defect patch, calculating the Hessian matrix of each pixel point on its surface, and obtaining the maximum principal curvature and its corresponding unit direction vector of each pixel point through eigenvalue decomposition.

[0012] Preferably, the stress concentration degree of the defect patch satisfies the expression: ; wherein, is the defect patch index is the defect patch is the defect patch is the stress concentration degree of the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the pixel index within the defect patch is the maximum principal curvature of the pixel point is the maximum principal curvature of the pixel point is the maximum principal curvature of the pixel point is the unit directional vector of the vector is the Frobenius norm of the matrix is the absolute value symbol denotes vector transposition.

[0013] In the process of constructing the stress concentration index, the application not only accumulates the sharpness of all points inside the defect patch through the maximum principal curvature, but also evaluates whether these sharp points tend to arrange along a certain direction in space through the outer product operation of the directional vector, so that the application can effectively identify those high-risk structures similar to micro-cracks that are prone to cause fracture, and provide a key mechanical basis for judging the real harmfulness of the defect patch.

[0014] Preferably, the harm index of the defect patch satisfies the expression: ; wherein, is the harm index of the defect patch ; is the shadow depth index of the defect patch ; is the stress concentration of the defect patch ; is the stress concentration of the defect patch .

[0015] The application obtains the harm index by multiplying the shadow depth index and the stress concentration, and the multiplication relationship ensures that only when a defect patch has significant depth and sharp and directional microstructure at the same time, the final harm index will be very high, which can effectively amplify the evaluation signal of the most dangerous defect patch, and at the same time suppress those defect patches that only show abnormalities in a single dimension, so that the final quality evaluation result is more accurate and more in line with the principles of fracture mechanics.

[0016] Preferably, the detection and quality evaluation of the defect patch based on the harm index include: collecting and analyzing the harm indexes of all detected defect patches of the same process and material in a stable production period; setting a threshold value of the harm index; in response to the harm index of the defect patch being greater than the threshold value, triggering an alarm.

[0017] The application sets the alarm threshold of the harm index by analyzing historical data, and automatically finds a critical value that can clearly distinguish between ordinary defect patches and serious defect patches as a judgment standard using data in a stable production period, avoiding the subjectivity and uncertainty of manually setting the threshold value, so that the alarm mechanism can closely match the actual situation of a specific process and material, thereby providing an objective, consistent and reliable basis for whether to alarm in subsequent real-time detection.

[0018] ​Preferably, the setting standard calibration object and determining the light source direction unit vector based on the image of the standard calibration object comprises: placing a standard calibration object on the detection focal plane where the weld to be detected is located under the same camera and light source fixed layout as the actual detection working condition; collecting a clear image of the standard calibration object; locating the contour of the standard calibration object in the two-dimensional image through an image algorithm; calculating the geometric center point coordinates of the standard calibration object; searching for the area with the highest brightness value within the area defined by the contour of the standard calibration object, and calculating the centroid of the area with the highest brightness value; and normalizing the vector from the geometric center point to the centroid to obtain the light source direction unit vector.

[0019] In a second aspect, the present application provides an electron beam welding quality detection system, comprising a processor and a memory, wherein the memory stores computer program instructions, and the computer program instructions, when executed by the processor, implement the above-mentioned electron beam welding quality detection method.

[0020] By adopting the above technical solution, the above-mentioned electron beam welding quality detection method is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is manufactured according to the memory and the processor, and use is facilitated.

[0021] The present application has the following beneficial effects:

[0022] The present application can more accurately judge the actual harm degree of the weld defect patch to the structure safety by comprehensively evaluating the macroscopic depth and microscopic geometric shape of the weld defect patch, not only uses shadow analysis to evaluate the depth information of the defect patch, but also evaluates the stress concentration degree thereof through curvature analysis, and the harm index formed can distinguish the truly dangerous defect patch from the surface defect with lower harm, thereby providing a more profound and comprehensive physical basis for the automatic evaluation of the welding quality.

[0023] The present application realizes the evaluation of the sharpness and direction consistency of the defect patch by reconstructing the two-dimensional gray-scale image into a local three-dimensional surface and performing differential geometry analysis thereon, identifies whether there is a structure with high stress concentration degree similar to a micro-crack inside the defect patch, so that the detection is no longer limited to the size or depth of the defect patch, but goes deep to the level of the influence of the defect patch on the mechanical properties of the component, thereby improving the accuracy of the detection.

[0024] The application establishes a complete quality evaluation process from defect patch identification to automatic alarm, provides technical support for intelligent production, sets an alarm threshold by statistical analysis of historical data to avoid subjectivity of manual setting, and automatically triggers an alarm when the defect patch hazard index in real-time detection exceeds the threshold, which provides an objective, consistent and reliable basis for whether intervention is needed, thereby improving the efficiency and automation level of quality control. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flow chart schematically showing a method for detecting the quality of electron beam welding in the application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0027] The specific embodiments of the application will be described in detail below with reference to the drawings.

[0028] The embodiments of the application disclose a method for detecting the quality of electron beam welding, referring to Figure 1 , comprising steps S1-S4:

[0029] S1, acquire an electron beam weld image under illumination of a fixed direction light source, pre-process and binarize the image, and identify potential defect patches; for each defect patch, perform connected domain aggregation and extract its outer expansion neighborhood; acquire the average gray value of the normal weld area around the defect patch based on the outer expansion neighborhood; and reconstruct the local three-dimensional surface of the defect patch based on the outer expansion neighborhood.

[0030] It should be noted that, in order to make up for the lack of perception and evaluation of depth information in the prior art, the application identifies and segments all potential defect patches from the original weld image, and converts the two-dimensional gray information thereof into a local three-dimensional surface model that can be subjected to geometric analysis.

[0031] Specifically, an industrial camera acquires a grayscale image of the electron beam weld under test under a fixed, non-perpendicular directional light source at a certain angle to the weld surface. The image is then subjected to adaptive thresholding to obtain a binarized image, and regions with pixel values ​​below a threshold are identified as potential defect patches. A connected component algorithm is used to process the binarized image, aggregating spatially adjacent defect patch pixels into independent defect patches. In this embodiment, 45° is used as the angle between the industrial camera and the weld surface, and the Otsu's method is employed for adaptive thresholding. The operator can select the appropriate angle and thresholding algorithm based on the specific requirements.

[0032] Furthermore, in the grayscale image, for each segmented defect patch, a neighborhood is selected, and all pixels in the neighborhood and their corresponding grayscale values ​​are extracted; the pixel coordinates are used as the bottom coordinates in three-dimensional space, and the grayscale values ​​are used as the height values, so each defect patch is reconstructed into a local three-dimensional surface.

[0033] More specifically, for the specific operation of expanding each segmented defect patch, a structuring element is defined, typically a structuring element with a radius of [missing information]. The square core of the pixel, empirically A value of 5 to 10 pixels can be used to ensure a complete transition area including the edges of the defective patches. This structuring element is applied to the binary mask of each individual defective patch. The dilation operation slides the center of the structuring element across every pixel in the mask and sets the entire area covered by the structuring element to 1, causing the boundaries of the original defective patches to expand outwards uniformly. Each pixel is used to form an extended region mask that includes the original defect patch but is slightly larger. Using the newly generated extended region mask, all corresponding pixels and their grayscale values ​​are extracted from the original grayscale image, thus obtaining a local neighborhood containing the complete defect patch and the normal weld area surrounding the defect patch. Subtracting the defect patch from the extended local neighborhood of each defect patch yields the normal weld area surrounding each defect patch, which is used for comparison with the defect patch.

[0034] S2. Set up a standard calibration object and determine the unit vector of the light source direction based on the image of the standard calibration object. Obtain the shadow depth index of the defect patch based on the unit vector of the light source direction, the geometric centroid of the defect patch, and the average gray value of the normal weld area around the defect patch.

[0035] It should be noted that the ultimate goal of this invention is to assess the true extent of the harm caused by defect patches to the mechanical properties of welded joints. The true harm of a defect patch depends on its depth and shape extending inwards. For example, a defect patch with the same surface area, if it is a shallow, flat, bowl-shaped pit, is far less harmful than a deep crack with a sharp bottom, because the latter is more likely to accumulate stress at the tip, leading to the fracture of the entire structure. Assessing the depth of the defect patch plays a crucial role in assessing the true extent of its harm to the mechanical properties of the welded joint.

[0036] When a light source with a fixed direction illuminates a surface with three-dimensional undulations, the surface geometry directly affects the reflection and shading of light. For a concave defect patch, such as a pore or pit, its inner wall facing away from the light source cannot be directly illuminated due to the shading of its own edges, thus forming an internal shadow area. The area, shape, and gray value of the shadow are physically related to the depth, slope, and incident angle of the light source of the defect patch. A deep and steep defect patch will produce a large and dark shadow under the same lighting conditions. This invention infers the depth information of the defect patch by analyzing its projection characteristics.

[0037] Specifically, with a fixed camera and light source layout identical to the actual inspection conditions, a standard calibration object is placed on the inspection plane where the weld to be inspected is located; a clear image of the standard calibration object is acquired; the outline of the standard calibration object in the two-dimensional image is located using an image algorithm; the pixel coordinates of its geometric center point are calculated; within the area defined by the outline of the standard calibration object, the region with the highest brightness value is searched, and the centroid of this highlight region is calculated; the vector from the geometric center point to the centroid represents the projection direction of the light source direction onto the image plane; the vector is normalized to obtain the unit vector of the light source direction. This embodiment uses Hough circle transform to locate the outline of the standard calibration object in the two-dimensional image. The implementer can select an image algorithm to locate the outline of the standard calibration object in the two-dimensional image according to the actual situation.

[0038] Furthermore, the geometric centroid of each defect patch is calculated, and the shadow depth index of each defect patch is obtained based on the unit vector of the light source direction and the geometric centroid of each defect patch:

[0039]

[0040] in, It is a defect plaque index; It is the first One defective patch; It is a defective patch. The shadow depth index; It is the pixel index within the defect patch; It is the first The first defective patch 1 pixel; It is a pixel. grayscale value; It is the first The average gray value of the normal weld area surrounding the defective patch; From the first The first defective patch The pixel points to the first The vector of the centroid of a defective patch; It is the unit vector of the light source direction; It is the dot product of two vectors; It is a symbolic function; It is the absolute value symbol.

[0041] Represents pixels The sign of the dot product between the centroid's position vector and the unit vector of the light source direction indicates whether the pixel is located on the light-facing or light-receding side of the defect patch. To retain only directional information and ignore distance, this invention uses a sign function to... Convert to +1 or -1; if the pixel is located on the light-facing side of the defect patch, the dot product... If positive, the sign function outputs +1; if the pixel is located on the backlight side of the defect patch, the dot product... If the dot product is negative, the sign function outputs -1; if the dot product is exactly zero, the sign function outputs 0.

[0042] Represents pixels The grayscale deviation relative to the normal weld area, i.e., the difference between the actual grayscale value of that point and the average grayscale value of the surrounding normal weld area, reflects the pixel's grayscale deviation. The greater the deviation in darkness compared to a normal weld area, the more significant the pixel difference. The more likely it is to belong to the shaded area.

[0043] For a concave defect patch, the backlight side is usually darker, the gray scale deviation is large, and the sign function gives -1, and the product is a large negative value; the light side is brighter, the gray scale deviation is small or even negative, the sign is +1, the absolute value of the product is small, and the sum is a large negative number, and the index is large after taking the absolute value; for a convex defect patch, the light side is brighter, but the gray scale is high due to the highlight, and the backlight side forms a shadow, the overall contribution is biased to the positive direction, the sum is a large positive number, and the absolute value is also large; for a flat surface defect, the light and dark distribution is basically symmetrical, the positive and negative contributions of the light side and the backlight side cancel each other out, the sum is close to zero, and the index is also small. Therefore, the larger the shadow depth index, the more significant the three-dimensional morphology of the defect patch in the depth direction, the deeper the defect patch relative to the plane or the more prominent the defect patch relative to the plane, and the greater the possible harm; otherwise, the possible harm is small, such as surface defects.

[0044] S3, calculating the Hessian matrix of each pixel point on the surface of the defect patch for the local three-dimensional surface of the defect patch, extracting the maximum principal curvature and the direction vector of each pixel point; obtaining the stress concentration degree of the defect patch based on the maximum principal curvature and the direction vector of each pixel point.

[0045] It should be noted that the harmfulness of a defect patch is not only determined by its macroscopic depth, but also by its microscopic geometric morphology. According to the principle of fracture mechanics, the more acute points of the overall morphology of the defect patch, such as the tip of a V-shaped crack, are more likely to produce stress concentration when the component is under stress, thereby becoming the starting point of the entire structure that is most likely to produce a crack and cause a fracture; if a large number of independent sharp points are arranged in the same direction in space, they are no longer independent stress concentration points, but will be connected into a continuous structure similar to a micro-crack, which will greatly promote the connection and propagation of the crack along these neatly arranged weak points, and the stress concentration effect of the structure tip will be greatly amplified, posing a serious threat to the overall strength and fatigue life of the structure. Therefore, after evaluating the macroscopic depth of the defect patch, the present application further analyzes the microscopic geometric characteristics of the defect patch, not only evaluating the overall sharpness of the defect patch, but also judging whether the most sharp regions have directional consistency in space.

[0046] Specifically, the Hessian matrix of each point on the local three-dimensional surface corresponding to each defect patch is calculated, and the maximum principal curvature and the corresponding unit direction vector are obtained by eigenvalue decomposition. The absolute value of the maximum principal curvature represents the degree of sharpness of the surface curvature at that point, and the larger the value, the sharper the point; the direction vector represents the direction of the most severe surface curvature at that point.

[0047] Further, the stress concentration degree of each defect patch is obtained:

[0048]

[0049] in, It is a defect plaque index; It is the first One defective patch; It is a defective patch. Stress concentration; It is the pixel index within the defect patch; It is the first The first defective patch 1 pixel; It is a pixel. Maximum principal curvature; It is a pixel. The unit direction vector; It is the Frobenius norm of the matrix; It is the absolute value symbol; This represents the transpose of a vector.

[0050] In the formula, This reflects the sharpness of the defect patches; taking the absolute value and then squaring it is to amplify the influence of the sharpness. outer product of direction vectors Construct a A symmetric matrix, the elements of which are products of the components of the direction vectors, is obtained by multiplying the symmetric matrix at each point by its sharpness weight. After summing, if the orientations of most high-curvature points in the defect patch are nearly consistent, these symmetric matrices will accumulate in the same direction during summation, resulting in a larger element value of the final symmetric matrix. If the orientations are random, the symmetric matrices in different directions will cancel each other out during summation, causing the final symmetric matrix to be close to a zero matrix. Taking the square root of the sum of the squares of all elements of the final symmetric matrix yields the value used to evaluate the directional consistency of sharp features. The larger the value, the more the high-curvature points tend to be arranged along a dominant direction, and the closer they are to the microcrack morphology. The smaller the value, the more random the orientation of the sharp points is, and the less obvious the structural features are.

[0051] S4. Multiply the shadow depth index of the defect patch by the stress concentration to obtain the hazard index of the defect patch, and complete the detection and quality assessment of the defect patch based on the hazard index.

[0052] It should be noted that the ultimate hazard of a defect patch is determined by both its depth and stress concentration. This invention combines the shadow depth index and stress concentration of the defect patch to form a hazard index for the defect patch.

[0053] Specifically, for each defect patch The hazard index satisfies the expression:

[0054]

[0055] in, It is the first A defective patch The hazard index; It is the first A defective patch The shadow depth index; It is the first A defective patch Stress concentration.

[0056] A defect patch has a very high hazard index only when it has both significant depth and sharp, organized micro-geometry. The multiplicative relationship can effectively identify the most dangerous defect patch types.

[0057] Furthermore, the system collects and analyzes the hazard index of all detected defect patches using the same process and materials within a stable production cycle over a period of time. These historical hazard index data are statistically sorted, and a high percentile, such as the 95th percentile, is selected as the alarm threshold. In subsequent real-time detection, the system compares the hazard index of newly discovered defect patches with this threshold. When the hazard index is less than or equal to the threshold, the defect patch is determined to be within an acceptable range. When the hazard index is greater than the threshold, the defect patch is determined to trigger an alarm.

[0058] This invention also discloses an electron beam welding quality inspection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an electron beam welding quality inspection method according to the present invention.

[0059] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for inspecting the quality of electron beam welding, characterized in that, include: Electron beam weld images under fixed-direction light source illumination are acquired, and the images are preprocessed and binarized to segment and identify potential defect patches. For each defect patch, connected component aggregation is performed and its extended neighborhood is extracted; the average gray value of the normal weld area surrounding the defect patch is obtained based on the extended neighborhood; The local three-dimensional surface of the defect patch is reconstructed based on the extended neighborhood; Setting up a standard calibration object and determining the light source direction unit vector based on the image of the standard calibration object includes: placing a standard calibration object on the detection focal plane of the weld to be tested under a fixed camera and light source layout identical to the actual detection conditions; acquiring a clear image of the standard calibration object; locating the outline of the standard calibration object in the two-dimensional image using image algorithms; calculating the coordinates of the geometric center point of the standard calibration object; searching for the region with the highest brightness value within the area defined by the outline of the standard calibration object, and calculating the centroid of the region with the highest brightness value; normalizing the vector from the geometric center point to the centroid to obtain the light source direction unit vector; and obtaining the shadow depth index of the defect patch based on the light source direction unit vector, the geometric centroid of the defect patch, and the average gray value of the normal weld area surrounding the defect patch, including: ;in, It is a defect plaque index; It is the first One defective patch; It is a defective patch. The shadow depth index; It is the pixel index within the defect patch; It is the first The first defective patch 1 pixel; It is a pixel. grayscale value; It is the first The average gray value of the normal weld area surrounding the defective patch; From the first The first defective patch The pixel points to the first The vector of the centroid of a defective patch; It is the unit vector of the light source direction; It is the dot product of two vectors; It is a symbolic function; It is the absolute value symbol; Calculate the Hessian matrix of each pixel on the local three-dimensional surface of the defect patch, and extract the maximum principal curvature and its direction vector of each pixel; obtain the stress concentration of the defect patch based on the maximum principal curvature and its direction vector of each pixel, including: ;in, It is a defective patch. Stress concentration; It is a pixel. Maximum principal curvature; It is a pixel. The unit direction vector; It is the Frobenius norm of the matrix; Indicates vector transpose; The hazard index of the defect patch is obtained by multiplying the shadow depth index of the defect patch by the stress concentration, and the detection and quality assessment of the defect patch are completed based on the hazard index.

2. The electron beam welding quality inspection method according to claim 1, characterized in that, The step of performing connected component aggregation and extracting the extended neighborhood for each defect patch includes: Define a structuring element; apply this structuring element to the binary mask of each individual defect patch and perform an expansion operation to form an expanded region that includes the original defect patch.

3. The electron beam welding quality inspection method according to claim 1, characterized in that, The local three-dimensional surface of the defect patch reconstructed based on the extended neighborhood includes: All pixels and their corresponding gray values ​​within the extended neighborhood of the defect patch are extracted; the pixel coordinates are used as the bottom coordinates in three-dimensional space, and the gray values ​​are used as the height values, so that each defect patch is reconstructed into a local three-dimensional surface.

4. The electron beam welding quality inspection method according to claim 1, characterized in that, The calculation of the Hessian matrix of each pixel on the local three-dimensional surface of the defect patch, and the extraction of the maximum principal curvature and its direction vector of each pixel, includes: For each defect patch corresponding to the local three-dimensional surface, calculate the Hessian matrix of each pixel on the surface, and obtain the maximum principal curvature of each pixel and its corresponding unit direction vector through eigenvalue decomposition.

5. The electron beam welding quality inspection method according to claim 1, characterized in that, The hazard index of the defective plaques satisfies the expression: ； in, It is the first A defective patch The harm index.

6. The electron beam welding quality inspection method according to claim 1, characterized in that, Based on the aforementioned hazard index, the detection and quality assessment of defective patches are completed, including: Collect and analyze the hazard index of all detected defect patches of the same process and materials within a stable production cycle over a period of time; set a threshold for the hazard index; trigger an alarm in response to the hazard index of a defect patch exceeding the threshold.

7. An electron beam welding quality inspection system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement an electron beam welding quality inspection method according to any one of claims 1-6.

Images