Machine vision-based forging aluminum blade surface defect detection method and system

Abstract

The application discloses a forging aluminum blade surface defect detection method and system based on machine vision, relates to the field of defect detection, and comprises the following steps: a collecting module is used for adaptively collecting blade surface images according to the posture of a blade curved surface and the distribution of forging texture; an enhancing module is used for performing targeted noise reduction on the collected images to extract surface potential defect correlation information and then performing feature contour strengthening; after the blade image is adaptively collected, the potential defect correlation information is extracted through targeted noise reduction, the feature contour is strengthened, and the suspected defect area is locked in combination with the metal streamline direction and the forging stress distribution characteristics.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, specifically to a method and system for detecting surface defects in forged aluminum blades based on machine vision. Background Technology

[0002] Forged aluminum blades are made of high-quality aluminum alloy, possessing high strength and lightweight characteristics. They have a dense and uniform structure, excellent fatigue resistance and corrosion resistance, and high transmission efficiency. Compared with cast aluminum blades, they have dense grains without porosity defects, superior strength and fatigue resistance, and precise and uniform profiles. Compared with steel blades, they are lighter, have a superior strength-to-weight ratio, and excellent thermal conductivity.

[0003] The invention patent application with application number 202010847852.X discloses a machine vision-based method for detecting defects in wind turbine blades. This method aims to solve the problem that "the visual detection effect depends on the quality of the input image. Since wind turbine blades are large in size and have irregular curved surfaces, the use of machine vision-based detection methods inevitably results in uneven lighting, high similarity between defects and non-defects, and large defect detection errors."

[0004] However, most existing methods for detecting surface defects on metal blades focus on optimizing image quality, acquisition posture, and the comprehensiveness of acquisition, making it difficult to utilize the deeper information contained in the images for defect detection.

[0005] To this end, we propose a machine vision-based method and system for detecting surface defects in forged aluminum blades. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method and system for detecting surface defects of forged aluminum blades based on machine vision, which can effectively solve the problems of the prior art.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a machine vision-based surface defect detection system for forged aluminum blades, comprising: The system comprises the following modules: an acquisition module for adaptively acquiring images of the blade surface based on its curved surface attitude and forging texture distribution; an enhancement module for targeted noise reduction of the acquired images to extract information related to potential surface defects, followed by feature contour enhancement; a localization module for establishing feature matching logic between defects and processes based on the flow lines and forging stress distribution characteristics formed during the forging process of the aluminum blade, to pinpoint the location of suspected defects; an identification module for receiving the location of suspected defects, extracting the geometric shape and texture distribution features of the suspected defect area, and constructing a defect discrimination model based on process correlation to distinguish defect types and attributes; a quantification module for calculating the spatial size, distribution density, and depth of the defect location after the defect type and attribute distinction, and outputting quantitative characterization data of the defect; and a judgment module for receiving the quantitative characterization data of the defect, comparing it with a preset acceptable range, judging whether the blade surface defects meet the usage requirements, and outputting the judgment result. The acquisition module is interconnected with an enhancement module via a wireless network. The enhancement module is interconnected with a positioning module via a wireless network. The positioning module is interconnected with an identification module via a wireless network. The identification module is interconnected with a quantization module via a wireless network. The quantization module is interconnected with a judgment module via a wireless network.

[0008] Furthermore, the acquisition module includes a surface pose sensing unit and a texture adaptation unit; The curved surface attitude sensing unit extracts the three-dimensional coordinates of at least three non-collinear sampling points on the blade surface in real time, calculates the principal curvature value and normal vector direction of the local curved surface of the blade, controls the gimbal structure of the acquisition device to keep the shooting optical axis and the normal vector at a preset angle, and calibrates the focal length through an automatic focusing algorithm based on the average distance from the sampling points to the lens of the acquisition device. The texture adaptation unit extracts the line density of the forging texture. When the texture density is higher than the preset density threshold, the sampling frequency of image acquisition is increased by a preset ratio, and the exposure time is decreased by a preset ratio. When the texture density is lower than the preset density threshold, the sampling frequency is decreased by a preset ratio, and the exposure parameters are adjusted by a preset ratio to make the image grayscale within a preset effective range. Among them, the line density of the forging texture ; In the formula: This represents the total number of texture line pixels within a preset statistical area after directional filtering. This represents the actual physical length corresponding to a single pixel. This is a correction factor for line connectivity; , These represent the width and height of the preset statistical region in the image coordinate system, respectively. To collect the angle between the optical axis and the normal vector of the statistical region.

[0009] Furthermore, the targeted noise reduction operation for the acquired image in the enhancement module is as follows: ; In the formula: The noise reduction weights at coordinates (i,j) in the image; The texture gradient value at coordinates (i,j); The grayscale difference between coordinates (i,j) and surrounding pixels; They are respectively The x and y coordinates of the inner pixel; The local neighborhood range of coordinates (i,j); This is the balance coefficient between texture gradient and grayscale difference; This is the sensitivity adjustment coefficient; Let (i,j) be the confidence score of the pixel at coordinates (i,j). This is the confidence baseline value; The noise reduction weights of each pixel in the image, after being calculated, follow the following principle in the application stage: when performing local neighborhood smoothing on the pixel at coordinates (i,j), the weights are... As a weight reserved for the grayscale value of the pixel itself, the remaining weights are distributed evenly among the grayscale values ​​of other pixels in the neighborhood. The smoothed pixel grayscale value is calculated by weighted averaging; when When the value is higher than a preset weight threshold, retain more than 80% of the original grayscale information of the pixel. When the value is below a preset weight threshold, only 30% or less of the original grayscale information of that pixel is retained. The information related to potential surface defects is extracted by constructing an anomaly detection index: ; In the formula: These are the weighting coefficients for texture continuity, grayscale distribution, and pixel connectivity. The texture continuity deviation of the coordinate (x,y) region is the integral value of the deviation between the texture direction of this region and the standard direction of the metal streamline of the aluminum blade. The gray-level distribution deviation of the (x,y) region is calculated by taking the Barcol distance between the gray-level histogram of this region and the gray-level histogram of the normal forging texture. The pixel connectivity of the coordinate (x,y) region is calculated based on the degree of clustering of consecutive pixels within that region, using a connected component labeling algorithm. when When it does not match the preset judgment threshold, a judgment is made. The pointed pixel is the defective pixel, and the area where all defective pixels with adjacent defective pixels are located is the suspected defective area; When performing feature contour enhancement on the image, the following applies: ; In the formula: This is to enhance the strength of the defect boundary; For enhancement coefficient; These are the edge response values ​​of the boundary pixels calculated based on a multi-scale edge detection algorithm. This represents the average grayscale difference between the defect boundary and the background area. This is the standard edge response baseline value; This represents the overall average grayscale value of the image. The fitting weights for the broken contour segments; These are the fitting coefficients; , For curvature deviation penalty factor and orientation deviation penalty factor; Here are the curvatures of the contour segment to be fitted and the reference curvature of the target contour; Here are the extension direction angles of the contour segment to be fitted and the reference direction angles of the target contour; This is the set of contour segments to be matched; Let be the curvature of the p-th contour segment in set P, and be the extension direction angle of the p-th contour segment in set P. This is the anomaly detection index for the corresponding region.

[0010] Furthermore, the feature matching logic of the positioning module calculates the correlation and similarity between the suspected defect area and the metal flow lines and forging stress, in order to lock down the area where the suspected defect is located. ; In the formula: Similarity based on relevance; The characteristic direction vector of the suspected defect region and the metal streamline direction vector of the corresponding region; Let be the magnitude of the vector; Weighting for metal streamlines and stress distribution; The standard deviation of stress distribution in the suspected defect area and the standard deviation of standard stress distribution under the corresponding forging process; This is the penalty coefficient for directional deviation; The principal extension direction angle of the suspected defect area and the principal direction angle of the forging stress in the corresponding area; When S is greater than the preset similarity threshold, the region is locked as a suspected defect region.

[0011] Furthermore, the geometric features extracted by the identification module include the equivalent circle diameter, aspect ratio, roundness, and edge concavity / convexity of the defect, and the texture distribution features include the grayscale entropy, texture contrast, and orientation consistency of the defect area. The expression for the defect discrimination model is: ; In the formula: This is the comprehensive discrimination vector; A normalized vector representing geometric morphological features; This is the weight matrix for geometric morphological features; This is a normalized vector of texture distribution features; This is the process correlation factor matrix; This is the texture distribution feature weight matrix; This is a process deviation correction item; Indicates transpose; in, The row vectors correspond to die wear, forging temperature fluctuations, and differences in metal fluidity, while the column vectors correspond to defects such as cracks, porosity, inclusions, and scratches, i.e., defect types. By comparing the numerical values ​​of each element in the comprehensive discrimination vector Y, the type associated with the column vector corresponding to the largest value is determined as the defect type. At the same time, the magnitude of the comprehensive discrimination vector Y is calculated. When it is greater than the preset attribute threshold, the defect is determined to be an irreversible defect in the forging process; otherwise, it is determined to be a reversible defect in the process.

[0012] Furthermore, the logic for calculating the defect location depth in the quantization module is as follows: ; In the formula: For the depth of the defect; The material's grayscale-to-depth conversion factor; This represents the average grayscale difference between the defective area and the surrounding normal area. The optical reflectance coefficient of the blade surface; The incident angle for the acquisition module; This represents the actual density of the defective region; This refers to the standard density of aluminum forgings; This is the density correction factor; This refers to the forging stress fluctuation coefficient in the corresponding region. The spatial dimensions are calculated through pixel calibration and three-dimensional coordinate fitting, and the distribution density is characterized by the number of defects per unit area and the equivalent area ratio.

[0013] Furthermore, the defect quantification data received by the determination module during its operation phase includes spatial size, distribution density, and depth. It is deemed qualified only when the spatial size, distribution density, and depth all conform to their respective preset qualified ranges; If the distance between adjacent defect locations is less than the preset distance, a set of defect quantitative characterization data is recombined based on the maximum values ​​of spatial size, distribution density and depth among the adjacent defect locations, and the defect quantitative characterization data is corrected. Then, the corrected defect quantitative characterization data is added to the pass / fail judgment process. The correction of the reconstructed defect quantification data is as follows: the spatial size, distribution density and depth are multiplied by correction factors to obtain the reconstructed corrected defect quantification data. The correction factor is calculated by multiplying the total number of adjacent defect locations of the defect location with the largest defect space size among adjacent defect locations by 0.1, and then summing the result with 1.

[0014] On the other hand, machine vision-based methods for detecting surface defects in forged aluminum blades include: By extracting the three-dimensional coordinates of non-collinear sampling points on the blade surface, calibrating the optical axis and focal length of the acquisition device to adapt to the curved surface posture, and adjusting the sampling frequency and exposure parameters according to the forging texture line density, images of the blade surface are acquired. Pixel weights are calculated using a targeted noise reduction algorithm and weighted smoothing is performed to extract potential defect association information. Based on edge response and contour fitting logic, defect feature contours are enhanced to improve defect recognition. The correlation similarity between suspected defect areas and process factors is calculated based on the metal flow line direction and forging stress distribution characteristics, locking in suspected defect areas that meet feature matching logic. The geometric shape and texture distribution features of suspected defect areas are extracted, and the defect type and process reversible / irreversible attributes are distinguished using a process-related defect discrimination model. The spatial size, distribution density, and depth of identified defects are calculated to generate quantitative characterization data reflecting the defect state. The defect quantitative characterization data is compared with a preset qualified range. After reorganization and correction of adjacent defects, the defect quantitative characterization data is added. When all indicators meet the requirements, the defect is judged as qualified, and the final judgment result is output.

[0015] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention provides a machine vision-based method and system for detecting surface defects in forged aluminum blades. During execution, after adaptively acquiring blade images, the method and system extract potential defect association information through targeted noise reduction and enhance feature contours. Combining the metal flow line direction and forging stress distribution characteristics, it locks suspected defect areas, accurately extracts geometric morphology and texture distribution features to distinguish defect types and attributes, and simultaneously calculates associated parameters such as defect spatial size, distribution density, and depth. Based on a preset acceptable range, it completes compliance judgment. Adjacent defects can be recombined to correct quantitative data, effectively adapting to different forging process conditions and texture density variations. This significantly improves the accuracy and efficiency of defect detection, reduces noise interference and false positives / missed negatives, clearly presents key defect information, provides reliable data support for quality control of forged aluminum blades, and helps optimize production processes and improve product qualification rates. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0017] Figure 1 This is a schematic diagram of a machine vision-based surface defect detection system for forged aluminum blades. Figure 2 This is a flowchart illustrating a machine vision-based method for detecting surface defects in forged aluminum blades. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be further described below with reference to embodiments.

[0020] Example 1: This embodiment describes a machine vision-based surface defect detection system for forged aluminum blades, such as... Figure 1 As shown, it includes: The acquisition module is used to adaptively acquire images of the blade surface based on the blade's curved surface posture and forging texture distribution. The acquisition module includes a surface pose sensing unit and a texture adaptation unit; The curved surface attitude sensing unit extracts the three-dimensional coordinates of at least three non-collinear sampling points on the blade surface in real time, calculates the principal curvature value and normal vector direction of the local curved surface of the blade, controls the gimbal structure of the acquisition device to keep the shooting optical axis and the normal vector at a preset angle, and calibrates the focal length through an autofocus algorithm based on the average distance from the sampling points to the lens of the acquisition device. The texture adaptation unit extracts the line density of the forging texture. When the texture density is higher than the preset density threshold, the sampling frequency of image acquisition is increased by a preset ratio, and the exposure time is decreased by a preset ratio. When the texture density is lower than the preset density threshold, the sampling frequency is decreased by a preset ratio, and the exposure parameters are adjusted by a preset ratio to keep the image grayscale within the preset effective range. Among them, the line density of the forging texture ; In the formula: This represents the total number of texture line pixels within a preset statistical area after directional filtering. This represents the actual physical length corresponding to a single pixel. This is a correction factor for line connectivity; , These represent the width and height of the preset statistical region in the image coordinate system, respectively. To collect the angle between the optical axis and the normal vector of the statistical region; The above formula combines core parameters such as the total number of texture line pixels after filtering within the preset statistical area and the actual physical length of a single pixel, and introduces the line connectivity correction coefficient and the cosine value of the angle between the acquisition optical axis and the normal vector of the statistical area. At the same time, the correction coefficient is dynamically adjusted according to the line connectivity to ensure the accuracy of density calculation under different connectivity states. ∈[0.8,1.0], the better the connectivity of adjacent line pixels of the forging texture within the preset statistical area, that is, no break and gap less than the preset threshold, the larger the value; the worse the connectivity, that is, there is a break and gap greater than the preset threshold, the smaller the value. The enhancement module is used to perform targeted noise reduction on the acquired images to extract information related to potential surface defects, and then perform feature contour enhancement. The targeted noise reduction operation for the acquired images in the enhancement module is as follows: ; In the formula: The noise reduction weights at coordinates (i,j) in the image; The texture gradient value at coordinates (i,j); The grayscale difference between coordinates (i,j) and surrounding pixels; They are respectively The x and y coordinates of the inner pixel; The local neighborhood range of coordinates (i,j); This is the balance coefficient between texture gradient and grayscale difference; This is the sensitivity adjustment coefficient; Let (i,j) be the confidence score of the pixel at coordinates (i,j). This is the confidence baseline value; The above formula integrates the texture gradient value at the image coordinates with the grayscale difference of the surrounding pixels. It is then normalized by summing the local neighborhood and combined with a balance coefficient between the texture gradient and the grayscale difference. At the same time, it uses a sigmoid function term based on pixel confidence to consider both the texture characteristics itself and the differences between the pixel and the background, as well as the pixel confidence. This achieves a reasonable allocation of targeted noise reduction weights, allowing the noise reduction process to accurately preserve effective texture information and suppress noise. in, >0, and the value is larger when the forging texture is denser and smaller when the texture is sparser; >0, and the value is larger when the difference between defects and background is lower, and smaller when the difference is higher; The value is larger when the pixel grayscale conforms to the distribution pattern of the forging texture, and smaller when it deviates from the distribution pattern; The noise reduction weights of each pixel in the image, after being calculated, follow the following principle in the application stage: when performing local neighborhood smoothing on the pixel at coordinates (i,j), the weights are... As a weight reserved for the grayscale value of the pixel itself, the remaining weights are distributed evenly among the grayscale values ​​of other pixels in the neighborhood. The smoothed pixel grayscale value is calculated by weighted averaging; when When the value is higher than a preset weight threshold, retain more than 80% of the original grayscale information of the pixel. When the value is below a preset weight threshold, only 30% or less of the original grayscale information of that pixel is retained. Information related to potential surface defects is extracted by constructing an anomaly detection index: ; In the formula: These are the weighting coefficients for texture continuity, grayscale distribution, and pixel connectivity. The texture continuity deviation of the coordinate (x,y) region is the integral value of the deviation between the texture direction of this region and the standard direction of the metal streamline of the aluminum blade. The gray-level distribution deviation of the (x,y) region is calculated by taking the Barcol distance between the gray-level histogram of this region and the gray-level histogram of the normal forging texture. The pixel connectivity of the coordinate (x,y) region is calculated based on the degree of clustering of consecutive pixels within that region, using a connected component labeling algorithm. The above formula comprehensively considers three dimensions: texture continuity, grayscale distribution, and pixel connectivity. By assigning different importance to each dimension through custom weighting coefficients, the deviation of each dimension is quantified by texture direction deviation integral value, grayscale histogram Bach distance, and connected component clustering degree. Multi-dimensional information is integrated into a unified anomaly judgment index, thereby comprehensively capturing the feature differences of defective pixels and achieving accurate screening of potential defective areas. when When it does not match the preset judgment threshold, a judgment is made. The pointed pixel is the defective pixel, and the area where all defective pixels with adjacent defective pixels are located is the suspected defective area; The sum is 1, and the values ​​of each weight coefficient are defined by the user on the system side; When performing feature contour enhancement on an image, the following rules apply: ; In the formula: This is to enhance the strength of the defect boundary; For enhancement coefficient; These are the edge response values ​​of the boundary pixels calculated based on a multi-scale edge detection algorithm. This represents the average grayscale difference between the defect boundary and the background area. This is the standard edge response baseline value; This represents the overall average grayscale value of the image. The fitting weights for the broken contour segments; These are the fitting coefficients; , For curvature deviation penalty factor and orientation deviation penalty factor; Here are the curvatures of the contour segment to be fitted and the reference curvature of the target contour; Here are the extension direction angles of the contour segment to be fitted and the reference direction angles of the target contour; This is the set of contour segments to be matched; Let be the curvature of the p-th contour segment in set P, and be the extension direction angle of the p-th contour segment in set P. This is the anomaly detection index for the corresponding region; In the above formula, the enhancement strength calculation combines parameters such as edge response value and grayscale difference between defect and background. By comparing with the standard benchmark value, the enhancement coefficient is dynamically adjusted to ensure that defect boundaries with different discrimination can be effectively strengthened. The fitting weights introduce curvature deviation and direction deviation penalty factors, and combine the fracture status of the contour segment and the anomaly judgment index to perform targeted fitting on the fracture contour segment. This not only strengthens the defect boundary, but also accurately repairs the fracture contour, thereby improving the integrity and accuracy of the defect contour. in, >0, and the larger the value is when the difference in grayscale between the defect boundary and the background is smaller and the edge response is weaker, the smaller the value is when the distinction between the defect boundary and the background is higher and the edge response is stronger; The value is greater than 0, and the value is larger when the gap between the contour segments is larger and the integrity is worse, and smaller when the gap between the contour segments is smaller and the integrity is better. >0, and the value is larger when the curvature deviation between the contour segment to be fitted and the target contour is larger, and the value is smaller when the curvature deviation between the two is smaller; >0, and the value is larger when the deviation of the extension direction angle between the contour segment to be fitted and the target contour is larger, and the value is smaller when the deviation of the direction angle between the two is smaller; The positioning module is used to establish feature matching logic that associates defects with processes based on the direction of metal flow lines and the distribution characteristics of forging stress formed during the forging process of aluminum blades, so as to locate the area where suspected defects are located. The feature matching logic of the localization module calculates the correlation and similarity between the suspected defect area and the metal flow lines and forging stress, in order to locate the area where the suspected defect is located. ; In the formula: Similarity based on relevance; The characteristic direction vector of the suspected defect region and the metal streamline direction vector of the corresponding region; Let be the magnitude of the vector; Weighting for metal streamlines and stress distribution; The standard deviation of stress distribution in the suspected defect area and the standard deviation of standard stress distribution under the corresponding forging process; This is the penalty coefficient for directional deviation; The principal extension direction angle of the suspected defect area and the principal direction angle of the forging stress in the corresponding area; The above formula starts from two core process dimensions: metal flow line matching and forging stress distribution matching. It calculates the directional matching degree through vector dot product, combines the stress distribution standard deviation ratio and directional deviation penalty term, and reasonably allocates the weight coefficients of the two dimensions to quantify the correlation between suspected defect areas and process factors. It can accurately locate suspected defect areas related to the forging process. When S is greater than the preset similarity threshold, the region is locked as a suspected defect region; Where a > 0, the value is larger when the direction of the metal flow lines of the forged aluminum blade has a stronger dominant effect on the formation and distribution of defects, and the value is smaller when the influence of the metal flow lines on defects is weaker than other process factors; b > 0, the value is larger when the correlation between forging stress fluctuation and defect generation is more significant, and the value is smaller when the influence of forging stress on defect formation is relatively weak. >0, the larger the value is when the deviation of the extension direction of the suspected defect area from the main direction of forging stress has a more critical impact on the defect type determination; the smaller the value is when the directional deviation contributes less to the distinguishability of the defect determination. The identification module is used to receive the area where the suspected defect is located, extract the geometric shape and texture distribution features of the area where the suspected defect is located, and construct a defect discrimination model based on process correlation to distinguish the defect type and attribute. The geometric features extracted by the identification module include the equivalent circle diameter, aspect ratio, roundness, and edge concavity / convexity of the defect; the texture distribution features include the grayscale entropy, texture contrast, and orientation consistency of the defect area. The expression for the defect discrimination model is: ; In the formula: This is the comprehensive discrimination vector; A normalized vector representing geometric morphological features; This is the weight matrix for geometric morphological features; This is a normalized vector of texture distribution features; This is the process correlation factor matrix; This is the texture distribution feature weight matrix; This is a process deviation correction item; Indicates transpose; The above formula integrates two types of features: geometric shape and texture distribution. Through normalization and a custom weight matrix, it highlights the contribution of different features to defect judgment. At the same time, it introduces a process correlation factor matrix and a process deviation correction term, which directly associates process factors such as mold wear and temperature fluctuations with defect types such as cracks and porosity. It can also dynamically adjust the correction term according to the actual working condition deviation, so as to achieve accurate differentiation of defect types and attributes. Each element in the equation represents the normalized values ​​of the equivalent circle diameter, aspect ratio, roundness, and edge concavity / convexity. Based on the different sensitivity of different defect types to various geometric features, the results were determined through statistical analysis and verification experiments of pre-set aluminum blade defect samples. Each element in the table represents a grayscale entropy normalization value, a texture contrast normalization value, and a direction consistency normalization value. in, The row vectors correspond to die wear, forging temperature fluctuations, and differences in metal fluidity, while the column vectors correspond to defects such as cracks, porosity, inclusions, and scratches, i.e., defect types. The contribution of each texture distribution feature to the differentiation of defect types is determined by statistical analysis of the sample data of texture distribution features corresponding to different preset defect types, combined with the contribution of each texture distribution feature to the differentiation of defect types. The value range is preset to [0, 0.2]. The larger the deviation of the die wear, forging temperature fluctuation, and metal flow rate from the standard working conditions in the actual forging process, the larger the value; the smaller the deviation, the smaller the value. By comparing the numerical values ​​of each element in the comprehensive discrimination vector Y, the type associated with the column vector corresponding to the largest value is determined as the defect type. At the same time, the magnitude of the comprehensive discrimination vector Y is calculated. When it is greater than the preset attribute threshold, the defect is determined to be an irreversible defect in the forging process; otherwise, it is determined to be a reversible defect in the process. The quantization module is used to calculate the spatial size, distribution density, and depth of the defect location after the defect type and attribute have been distinguished, and output the quantitative characterization data of the defect. The logic for calculating the defect location and depth in the quantization module is as follows: ; In the formula: For the depth of the defect; The material's grayscale-to-depth conversion factor; This represents the average grayscale difference between the defective area and the surrounding normal area. The optical reflectance coefficient of the blade surface; The incident angle for the acquisition module; This represents the actual density of the defective region; This refers to the standard density of aluminum forgings; This is the density correction factor; This refers to the forging stress fluctuation coefficient in the corresponding region. The above formula integrates the material grayscale-depth conversion coefficient, the grayscale difference between the defect and the normal area, and the optical reflection coefficient, while also considering the influence of the shooting incident angle. It also introduces the ratio of the actual density of the defect to the standard density and the forging stress fluctuation coefficient. The calculation results are dynamically adjusted through density correction terms and stress fluctuation correction terms to accurately quantify the defect depth, so that the depth calculation conforms to the optical properties of the material and improves the accuracy of depth quantification. Spatial dimensions are calculated through pixel calibration and three-dimensional coordinate fitting, and distribution density is characterized by the number of defects per unit area and the equivalent area ratio. in, >0, and follows the following rules: the higher the forging process complexity and the greater the blade surface roughness, the larger the value; the simpler the forging process and the smoother the blade surface, the smaller the value. >0, and follows the rule that: the greater the difference between the actual density of the defect area and the standard density of the aluminum forging, the larger the value; the smaller the difference between the two densities, the smaller the value. >0, and follows the rule that: the more violent the stress fluctuation in this area during the forging process, the larger the value; and the more stable the stress distribution, the smaller the value. The judgment module is used to receive defect quantification data, compare it with the preset qualified range, judge whether the surface defects of the blade meet the usage requirements, and output the judgment result. The defect quantification data received during the operation of the judgment module includes spatial size, distribution density, and depth. It is deemed qualified only when the spatial size, distribution density, and depth all conform to their respective preset qualified ranges; If the distance between adjacent defect locations is less than the preset distance, a set of defect quantitative characterization data is recombined based on the maximum values ​​of spatial size, distribution density and depth among the adjacent defect locations, and the defect quantitative characterization data is corrected. Then, the corrected defect quantitative characterization data is added to the pass / fail judgment process. The correction of the reconstructed defect quantification data is as follows: the spatial size, distribution density and depth are multiplied by correction factors to obtain the reconstructed corrected defect quantification data. The correction factor is calculated as follows: the total number of adjacent defect locations of the defect location with the largest defect space size among adjacent defect locations is multiplied by 0.1, and then summed with 1. The acquisition module interacts with the enhancement module via a wireless network. The enhancement module interacts with the positioning module via a wireless network. The positioning module interacts with the identification module via a wireless network. The identification module interacts with the quantization module via a wireless network. The quantization module interacts with the judgment module via a wireless network.

[0021] In this embodiment, the acquisition module adapts to the blade surface attitude and forging texture distribution to acquire images of the blade surface. The enhancement module then performs targeted noise reduction on the acquired images to extract information related to potential surface defects, followed by feature contour enhancement. The positioning module further establishes feature matching logic related to defects and processes based on the metal flow lines and forging stress distribution characteristics formed during the forging process of the aluminum blade to locate the suspected defect area. The identification module then receives the suspected defect area, extracts the geometric shape and texture distribution features of the suspected defect area, and constructs a defect discrimination model based on process correlation to distinguish defect types and attributes. The quantification module calculates the spatial size, distribution density, and depth of the defect location after the defect type and attribute are distinguished, and outputs the quantitative characterization data of the defect. Finally, the judgment module receives the quantitative characterization data of the defect, compares it with the preset qualified range, judges whether the blade surface defects meet the usage requirements, and outputs the judgment result.

[0022] In the above embodiments, the system can accurately adapt to the blade surface and forging texture, effectively filter out interference, enhance defect features, lock and distinguish defect types by combining forging process-related logic, accurately quantify defect parameters, and intelligently process adjacent defects. Based on the qualification standards, it can significantly improve detection efficiency and accuracy, reduce the risk of misjudgment and missed judgment, and provide reliable support for blade quality control.

[0023] Example 2: At the implementation level, based on Example 1, this example refers to... Figure 2 The machine vision-based surface defect detection system for forged aluminum blades in Example 1 will be further described in detail below: A machine vision-based method for detecting surface defects in forged aluminum blades includes: By extracting the three-dimensional coordinates of non-collinear sampling points on the blade surface, calibrating the optical axis and focal length of the acquisition device to adapt to the curved surface posture, and adjusting the sampling frequency and exposure parameters according to the line density of the forging texture, images of the blade surface are acquired. The pixel weights are calculated and weighted smoothing is performed by a targeted noise reduction algorithm to extract potential defect association information. Based on edge response and contour fitting logic, the defect feature contour is enhanced to improve defect recognition. Based on the flow path of metal and the distribution characteristics of forging stress, the correlation similarity between suspected defect areas and process factors is calculated, and suspected defect areas that meet the feature matching logic are identified. Extract the geometric shape and texture distribution features of suspected defect areas, and distinguish the defect type and process reversibility / irreversibility attributes through a process-related defect discrimination model; The spatial size, distribution density, and depth of the identified defects are calculated to generate quantitative characterization data reflecting the defect status. The defect quantitative characterization data is compared with the preset qualified range. After adjacent defects are reorganized and corrected, the defect quantitative characterization data is added. When all indicators meet the requirements, the result is judged as qualified, and the final judgment result is output.

[0024] In summary, during the execution of the system and method in the above embodiments, after adaptively acquiring blade images, potential defect association information is extracted through targeted noise reduction and feature contour enhancement. Combined with the metal flow line direction and forging stress distribution characteristics, suspected defect areas are locked. Geometric shape and texture distribution features are accurately extracted to distinguish defect types and attributes. Corresponding parameters such as defect spatial size, distribution density, and depth are calculated simultaneously. Compliance judgment is completed based on preset qualified intervals. Adjacent defects can be recombined to correct quantitative data, effectively adapting to different forging process conditions and texture density changes. This significantly improves the accuracy and efficiency of defect detection, reduces noise interference and false or missed judgments, clearly presents key defect information, provides reliable data support for quality control of forged aluminum blades, and helps optimize production processes and improve product qualification rates.

[0025] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine vision-based surface defect detection system for forged aluminum blades, characterized in that, include: The acquisition module is used to adaptively acquire images of the blade surface based on the blade's curved surface posture and forging texture distribution. The enhancement module is used to perform targeted noise reduction on the acquired images to extract information related to potential surface defects, and then perform feature contour enhancement. The targeted noise reduction operation for the acquired images in the enhancement module is as follows: ； In the formula: The noise reduction weights at coordinates (i,j) in the image; The texture gradient value at coordinates (i,j); The grayscale difference between coordinates (i,j) and surrounding pixels; They are respectively The x and y coordinates of the inner pixel; The local neighborhood range of coordinates (i,j); This is the balance coefficient between texture gradient and grayscale difference; This is the sensitivity adjustment coefficient; Let (i,j) be the confidence score of the pixel at coordinates (i,j). This is the confidence baseline value; The noise reduction weights of each pixel in the image, after being calculated, follow the following principle in the application stage: when performing local neighborhood smoothing on the pixel at coordinates (i,j), the weights are... As a weight reserved for the grayscale value of the pixel itself, the remaining weights are distributed evenly among the grayscale values ​​of other pixels in the neighborhood. The smoothed pixel grayscale value is calculated by weighted averaging; when When the value is higher than a preset weight threshold, retain more than 80% of the original grayscale information of the pixel. When the value is below a preset weight threshold, only 30% or less of the original grayscale information of that pixel is retained. The information related to potential surface defects is extracted by constructing an anomaly detection index: ； In the formula: These are the weighting coefficients for texture continuity, grayscale distribution, and pixel connectivity. The texture continuity deviation of the coordinate (x,y) region is the integral value of the deviation between the texture direction of this region and the standard direction of the metal streamline of the aluminum blade. The gray-level distribution deviation of the (x,y) region is calculated by taking the Barcol distance between the gray-level histogram of this region and the gray-level histogram of the normal forging texture. The pixel connectivity of the coordinate (x,y) region is calculated based on the degree of clustering of consecutive pixels within that region, using a connected component labeling algorithm. when When it does not match the preset judgment threshold, a judgment is made. The pointed pixel is the defective pixel, and the area where all defective pixels with adjacent defective pixels are located is the suspected defective area; When performing feature contour enhancement on the image, the following applies: ； In the formula: This is to enhance the strength of the defect boundary; For enhancement coefficient; These are the edge response values ​​of the boundary pixels calculated based on a multi-scale edge detection algorithm. This represents the average grayscale difference between the defect boundary and the background area. This is the standard edge response baseline value; This represents the overall average grayscale value of the image. The fitting weights for the broken contour segments; These are the fitting coefficients; , For curvature deviation penalty factor and orientation deviation penalty factor; Here are the curvatures of the contour segment to be fitted and the reference curvature of the target contour; Here are the extension direction angles of the contour segment to be fitted and the reference direction angles of the target contour; This is the set of contour segments to be matched; Let be the curvature of the p-th contour segment in set P, and be the extension direction angle of the p-th contour segment in set P. This is the anomaly detection index for the corresponding region; The positioning module is used to establish feature matching logic that associates defects with processes based on the direction of metal flow lines and the distribution characteristics of forging stress formed during the forging process of aluminum blades, so as to locate the area where suspected defects are located. The identification module is used to receive the area where the suspected defect is located, extract the geometric shape and texture distribution features of the area where the suspected defect is located, and construct a defect discrimination model based on process correlation to distinguish the defect type and attribute. The quantization module is used to calculate the spatial size, distribution density, and depth of the defect location after the defect type and attribute have been distinguished, and output the quantitative characterization data of the defect. The judgment module is used to receive defect quantification data, compare it with the preset qualified range, judge whether the surface defects of the blade meet the usage requirements, and output the judgment result.

2. The machine vision-based surface defect detection system for forged aluminum blades according to claim 1, characterized in that, The acquisition module includes a curved surface posture perception unit and a texture adaptation unit; The curved surface attitude sensing unit extracts the three-dimensional coordinates of at least three non-collinear sampling points on the blade surface in real time, calculates the principal curvature value and normal vector direction of the local curved surface of the blade, controls the gimbal structure of the acquisition device to keep the shooting optical axis and the normal vector at a preset angle, and calibrates the focal length through an automatic focusing algorithm based on the average distance from the sampling points to the lens of the acquisition device. The texture adaptation unit extracts the line density of the forging texture. When the texture density is higher than the preset density threshold, the sampling frequency of image acquisition is increased by a preset ratio, and the exposure time is decreased by a preset ratio. When the texture density is lower than the preset density threshold, the sampling frequency is decreased by a preset ratio, and the exposure parameters are adjusted by a preset ratio to make the image grayscale within a preset effective range. Among them, the line density of the forging texture ; In the formula: This represents the total number of texture line pixels within a preset statistical area after directional filtering. This represents the actual physical length corresponding to a single pixel. This is a correction factor for line connectivity; , These represent the width and height of the preset statistical region in the image coordinate system, respectively. To collect the angle between the optical axis and the normal vector of the statistical region.

3. The machine vision-based surface defect detection system for forged aluminum blades according to claim 1, characterized in that, The feature matching logic of the positioning module calculates the correlation and similarity between the suspected defect area and the metal flow lines and forging stress, in order to lock the location of the suspected defect area. ； In the formula: Similarity based on relevance; The characteristic direction vector of the suspected defect region and the metal streamline direction vector of the corresponding region; Let be the magnitude of the vector; Weighting for metal streamlines and stress distribution; The standard deviation of stress distribution in the suspected defect area and the standard deviation of standard stress distribution under the corresponding forging process; This is the penalty coefficient for directional deviation; The principal extension direction angle of the suspected defect area and the principal direction angle of the forging stress in the corresponding area; When S is greater than the preset similarity threshold, the region is locked as a suspected defect region.

4. The machine vision-based surface defect detection system for forged aluminum blades according to claim 1, characterized in that, The geometric features extracted by the identification module include the equivalent circle diameter, aspect ratio, roundness, and edge concavity / convexity of the defect; the texture distribution features include the grayscale entropy, texture contrast, and orientation consistency of the defect area. The expression for the defect discrimination model is: ； In the formula: This is the comprehensive discrimination vector; Normalized vectors for geometric morphological features; This is the weight matrix for geometric morphological features; This is a normalized vector of texture distribution features; This is the process correlation factor matrix; This is the texture distribution feature weight matrix; This is a process deviation correction item; Indicates transpose; in, The row vectors correspond to die wear, forging temperature fluctuations, and differences in metal fluidity, while the column vectors correspond to defects such as cracks, porosity, inclusions, and scratches, i.e., defect types. By comparing the numerical values ​​of each element in the comprehensive discrimination vector Y, the type associated with the column vector corresponding to the largest value is determined as the defect type. At the same time, the magnitude of the comprehensive discrimination vector Y is calculated. When it is greater than the preset attribute threshold, the defect is determined to be an irreversible defect in the forging process; otherwise, it is determined to be a reversible defect in the process.

5. The machine vision-based surface defect detection system for forged aluminum blades according to claim 1, characterized in that, The logic for calculating the defect location depth in the quantization module is as follows: ； In the formula: For the depth of the defect; The material's grayscale-to-depth conversion factor; This represents the average grayscale difference between the defective area and the surrounding normal area. The optical reflectance coefficient of the blade surface; The incident angle for the acquisition module; This represents the actual density of the defective region; This refers to the standard density of aluminum forgings; This is the density correction factor; This refers to the forging stress fluctuation coefficient in the corresponding region. The spatial dimensions are calculated through pixel calibration and three-dimensional coordinate fitting, and the distribution density is characterized by the number of defects per unit area and the equivalent area ratio.

6. The machine vision-based surface defect detection system for forged aluminum blades according to claim 1, characterized in that, The defect quantification data received by the judgment module during the operation phase includes spatial size, distribution density, and depth. It is deemed qualified only when the spatial size, distribution density, and depth all conform to their respective preset qualified ranges; If the distance between adjacent defect locations is less than the preset distance, a set of defect quantitative characterization data is recombined based on the maximum values ​​of spatial size, distribution density and depth among the adjacent defect locations, and the defect quantitative characterization data is corrected. Then, the corrected defect quantitative characterization data is added to the pass / fail judgment process. The correction of the reconstructed defect quantification data is as follows: the spatial size, distribution density and depth are multiplied by correction factors to obtain the reconstructed corrected defect quantification data. The correction factor is calculated by multiplying the total number of adjacent defect locations of the defect location with the largest defect space size among adjacent defect locations by 0.1, and then summing the result with 1.

7. The machine vision-based surface defect detection system for forged aluminum blades according to claim 1, characterized in that, The acquisition module is interconnected with an enhancement module via a wireless network. The enhancement module is interconnected with a positioning module via a wireless network. The positioning module is interconnected with an identification module via a wireless network. The identification module is interconnected with a quantization module via a wireless network. The quantization module is interconnected with a judgment module via a wireless network.

8. A method for detecting surface defects in forged aluminum blades based on machine vision, wherein the method is an implementation method of the machine vision-based surface defect detection system for forged aluminum blades as described in any one of claims 1-7, characterized in that, include: By extracting the three-dimensional coordinates of non-collinear sampling points on the blade surface, calibrating the optical axis and focal length of the acquisition device to adapt to the curved surface posture, and adjusting the sampling frequency and exposure parameters according to the line density of the forging texture, images of the blade surface are acquired. The pixel weights are calculated and weighted smoothing is performed by a targeted noise reduction algorithm to extract potential defect association information. Based on edge response and contour fitting logic, the defect feature contour is enhanced to improve defect recognition. Based on the flow path of metal and the distribution characteristics of forging stress, the correlation similarity between suspected defect areas and process factors is calculated, and suspected defect areas that meet the feature matching logic are identified. Extract the geometric shape and texture distribution features of suspected defect areas, and distinguish the defect type and process reversibility / irreversibility attributes through a process-related defect discrimination model; The spatial size, distribution density, and depth of the identified defects are calculated to generate quantitative characterization data reflecting the defect status. The defect quantitative characterization data is compared with the preset qualified range. After adjacent defects are reorganized and corrected, the defect quantitative characterization data is added. When all indicators meet the requirements, the result is judged as qualified, and the final judgment result is output.

Images