An intelligent detection method and system for defects of an aviation part

Abstract

The present application relates to the technical field of defect detection, and provides an intelligent detection method and system for defects of aviation parts, which comprises collecting working surface bright field and dark field images, and simultaneously extracting texture main direction from a grinding process file; a one-dimensional gray sequence is extracted along the normal section line perpendicular to the texture, two-dimensional whole image processing is converted into one-dimensional sequence analysis, and whole image texture removal is avoided to flatten micro-defect features. After the same scale sliding smoothing is performed on the two types of gray sequences, the gray difference value is calculated, the periodic texture fluctuation is filtered out, the signal decoupling of defect features and texture background is realized, the difference value mutation sites are screened and combined into candidate sections, and finally the overlapping area is reserved as the final defect through double image space position matching. Thus, with the aid of the consistency of double modal imaging, the periodic gray fluctuation of the directional grinding texture is avoided to cover the micro-defects, and the accuracy and anti-interference of the surface defect detection are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and more specifically, to an intelligent method and system for detecting defects in aerospace components. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] The working surfaces of aerospace components are mostly finished using precision grinding processes. The surface finish directly determines the fatigue strength, wear resistance, and service life of the components, and is a core indicator for ensuring the safety and reliability of aerospace equipment operation. Therefore, surface defect detection after grinding is an indispensable quality control step in the aerospace component manufacturing process.

[0004] Currently, research and industrial applications in the detection of surface defects in aerospace components generally focus on large-scale surface defects with significant visual characteristics, such as macroscopic cracks, dents, impact damage, and obvious scratches. These defects exhibit significant differences in grayscale and morphology compared to normal substrate surfaces, and possess high distinguishability under conventional optical imaging conditions. Existing detection solutions typically achieve effective detection and identification of these macroscopic defects through common technical processes such as image preprocessing, feature extraction, and target classification and recognition.

[0005] For the working surfaces of parts formed by precision grinding, directional grinding textures extending along a fixed direction are commonly present. These textures exhibit regular, periodic grayscale fluctuations during optical imaging, constituting a major background interference in defect detection. To address this background interference problem, those skilled in the art generally adopt a whole-image detexturing approach. Through algorithms such as frequency domain filtering, global background modeling, and overall texture suppression, uniform texture removal processing is performed on the entire inspection image to reduce the interference of background textures on defect identification.

[0006] However, existing technologies have consistently failed to address the issue that the periodic grayscale fluctuations of directional grinding textures can completely mask micro-defect features. When micro-defects such as micro-pits, micro-cracks, and pits exist on the ground surfaces of aerospace components, the grayscale changes they induce are comparable to or even smaller than the periodic grayscale fluctuations of the grinding texture. The defect features are completely covered by the periodic fluctuations of the texture, making it impossible to form independently distinguishable features in the original image. The generally used whole-image detexturing schemes in this field, while eliminating background textures, easily smooth out the masked micro-defect features along with them; if the detexturing intensity is reduced to retain the defect signal, the interference of periodic textures cannot be completely eliminated, ultimately leading to persistently high rates of missed and false detections of micro-defects.

[0007] In summary, existing surface defect detection technologies for aerospace components have not developed effective solutions for specific scenarios where directional grinding textures conceal micro-defects, making it difficult to meet the high-precision detection requirements of ultra-precision surface micro-defects for high-end aerospace components. Summary of the Invention

[0008] The purpose of this invention is to provide an intelligent detection method and system for defects in aerospace components, thereby improving the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows: Firstly, this application provides an intelligent detection method for defects in aerospace components, including: S101, acquire the bright field image and dark field image of the working surface of the component to be inspected, as well as the main direction parameters of the surface grinding texture corresponding to the working surface of the component; the bright field image is obtained by collecting specular reflected light after the working surface is illuminated by a vertically incident light source, and the dark field image is obtained by collecting scattered light after the working surface is illuminated by an obliquely grazing incident light source; the main direction parameters of the surface grinding texture are extracted from the grinding process document of the component and are used to characterize the extension direction of the grinding texture on the working surface; S102, based on the main direction parameters of the surface grinding texture, multiple texture normal cross-section lines perpendicular to the grinding texture extension direction are defined in the bright field image and the dark field image respectively, and each texture normal cross-section line is arranged in parallel along the grinding texture extension direction; the gray value of the corresponding pixel is extracted sequentially along each texture normal cross-section line to generate the bright field normal gray value sequence and the dark field normal gray value sequence. S103, Perform sliding neighborhood smoothing processing of the same window scale on the bright field normal grayscale sequence and the dark field normal grayscale sequence respectively to obtain the bright field smoothed sequence and the dark field smoothed sequence; Calculate the grayscale difference of the bright field smoothed sequence and the dark field smoothed sequence pixel by pixel to obtain the normal difference sequence corresponding to each texture normal cross-section line. S104. For each texture normal cross-section line, the gray-level difference sequence is compared bit by bit with the gray-level difference change range of adjacent pixels. Pixels with change ranges exceeding the corresponding periodic gray-level fluctuation range are selected and marked as candidate defect sites. Candidate defect sites continuously distributed on the same texture normal cross-section line are merged into candidate defect segments, and the pixel coordinates of each candidate defect segment in the corresponding image are recorded. S105, perform spatial position matching based on pixel coordinates on all candidate defect segments corresponding to the texture normal cross-section lines in the bright field image and dark field image, and retain only the candidate defect segments whose positions overlap in the bright field image and dark field image as the final defect area; output the corresponding position information of the final defect area on the working surface of the part to be inspected.

[0009] Furthermore, multiple texture normal section lines perpendicular to the grinding texture extension direction are defined, specifically including: Select a unified positioning reference point in the corner area of ​​both the bright field and dark field images; Based on the main direction parameters of the surface grinding texture, determine the extension angle and starting position of the first texture normal section line; Along the direction of the grinding texture extension, the remaining texture normal section lines are generated sequentially at equal intervals; In this process, the extension direction of all texture normal cross-section lines is perpendicular to the grinding texture extension direction, and the beginning and end of each texture normal cross-section line extend to the image boundary of the corresponding image, so as to completely cover the entire imaging area of ​​the working surface of the part to be inspected.

[0010] Furthermore, a unified positioning reference point is selected in the corner regions of both the bright-field and dark-field images, specifically including: The preset corner of the working surface of the component to be inspected is selected as the imaging position in the image as the positioning reference point.

[0011] Furthermore, the grayscale values ​​of the corresponding pixels are extracted sequentially along each texture normal cross-section line to generate a bright-field normal grayscale sequence and a dark-field normal grayscale sequence, specifically including: Starting with one endpoint of the texture normal cross-section line on the boundary between the bright field and dark field images as the starting pixel, the process proceeds sequentially towards adjacent pixels along the extension direction of the texture normal cross-section line. Each pixel reached during this process is taken as the grayscale extraction object. The grayscale value of the grayscale extraction object in the bright field image is read, and the read grayscale values ​​are arranged in the order of the process to obtain the corresponding bright field normal grayscale sequence and dark field normal grayscale sequence.

[0012] Furthermore, the periodic grayscale fluctuation amplitude is determined in the following ways, specifically including: In the bright field image, a local area that has been pre-confirmed as defect-free is selected as the bright field defect-free area, and in the dark field image, a local area that has been pre-confirmed as defect-free is selected as the dark field defect-free area. Along the direction of the texture normal section line, extract the bright-field defect-free grayscale sequence in the bright-field defect-free area, calculate the grayscale difference of each pair of adjacent pixels in the bright-field defect-free grayscale sequence, and take the magnitude of the grayscale difference as the amplitude of the bright-field grayscale difference change; count the amplitude of all bright-field grayscale difference changes obtained in the bright-field defect-free area, and take the amplitude of the bright-field grayscale difference change with the highest frequency as the periodic grayscale fluctuation amplitude corresponding to the bright-field image; Along the direction of the texture normal cross-section line, extract the dark field defect-free grayscale sequence in the dark field defect-free area, calculate the grayscale difference of each pair of adjacent pixels in the dark field defect-free grayscale sequence, and take the magnitude of the grayscale difference as the dark field grayscale difference change amplitude; count all the dark field grayscale difference change amplitudes obtained in the dark field defect-free area, and take the dark field grayscale difference change amplitude with the highest frequency as the periodic grayscale fluctuation amplitude corresponding to the dark field image.

[0013] Furthermore, candidate defect sites that are continuously distributed along the same texture normal section line are merged into candidate defect segments, specifically including: All candidate defect sites are traversed sequentially along the extension direction of the texture normal section line. Candidate defect sites that are adjacent to each other are grouped into the same continuous set, and each continuous set corresponds to a candidate defect segment. For each candidate defect segment, its starting pixel coordinates, ending pixel coordinates, and the range of all pixel coordinates covered by the segment in the corresponding image are recorded.

[0014] Furthermore, spatial location matching based on pixel coordinates is performed on all candidate defect segments corresponding to texture normal cross-sections in the bright-field and dark-field images, specifically including: Using a unified image coordinate system as a reference, the pixel coordinates of each candidate defect segment in the bright field image are mapped and compared one by one with the pixel coordinates of each candidate defect segment in the dark field image. When it is determined that the coordinate coverage of two candidate defect segments overlaps and the overlapping area meets the overlap judgment condition, the two are identified as candidate defect segments with overlapping positions, and the corresponding matching process is completed.

[0015] Secondly, this application also provides an intelligent detection system for defects in aerospace components, comprising: The bright and dark field texture module is used to acquire the bright field image and dark field image of the working surface of the part to be inspected, as well as the main direction parameters of the surface grinding texture corresponding to the working surface of the part. The bright field image is obtained by collecting specular reflected light after the working surface is illuminated by a vertically incident light source, and the dark field image is obtained by collecting scattered light after the working surface is illuminated by an obliquely grazing incident light source. The main direction parameters of the surface grinding texture are extracted from the grinding process file of the part and are used to characterize the extension direction of the grinding texture on the working surface. The normal sequence extraction module is used to define multiple texture normal cross-section lines perpendicular to the grinding texture extension direction in the bright field image and dark field image based on the main direction parameters of the surface grinding texture. Each texture normal cross-section line is arranged in parallel along the grinding texture extension direction. The gray value of the corresponding pixel is extracted sequentially along each texture normal cross-section line to generate the bright field normal gray value sequence and the dark field normal gray value sequence. The smoothing difference module is used to perform sliding neighborhood smoothing processing on the bright field normal grayscale sequence and the dark field normal grayscale sequence respectively, with the same window scale, to obtain the bright field smoothed sequence and the dark field smoothed sequence; the grayscale difference of the bright field smoothed sequence and the dark field smoothed sequence is calculated pixel by pixel to obtain the normal difference sequence corresponding to each texture normal cross-section line. The defect candidate extraction module is used to compare the gray-level difference changes of adjacent pixels one by one for the normal difference sequence corresponding to each texture normal cross-section line, and filter out the pixel sites whose change exceeds the corresponding periodic gray-level fluctuation range, and mark them as candidate defect sites; merge the candidate defect sites continuously distributed on the same texture normal cross-section line into candidate defect segments, and record the pixel coordinate position of each candidate defect segment in the corresponding image. The spatial matching and filtering module is used to perform spatial position matching based on pixel coordinates for all candidate defect segments corresponding to texture normal cross-sections in the bright field image and dark field image. Only candidate defect segments whose positions overlap in the bright field image and dark field image are retained as the final defect area. The module outputs the corresponding position information of the final defect area on the working surface of the part to be inspected.

[0016] Thirdly, this application also provides an electronic device, including: Memory, used to store computer programs; A processor is used to implement the method steps as described in the first aspect when executing a computer program.

[0017] Fourthly, this application also provides a readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the method steps of the first aspect.

[0018] The beneficial effects of this invention are as follows: This invention first acquires bright-field and dark-field images of the working surface, and simultaneously extracts the principal direction parameters of the texture from the grinding process file. It then utilizes the difference in reflectivity between defects and textures under two different lighting conditions to establish a basis for dual-modal signal comparison. Subsequently, based on the principal direction of the texture, normal cross-sections perpendicular to the texture extension direction are defined in both types of images, and grayscale values ​​are extracted pixel-by-pixel along these cross-sections to generate corresponding sequences. The two-dimensional whole-image processing is transformed into one-dimensional line sequence analysis, precisely targeting the normal change dimension of texture undulations, avoiding the smoothing of micro-defect features during whole-image detexturing. Next, sliding neighborhood smoothing of the same window scale is performed on the two types of grayscale sequences, and then the grayscale difference is calculated pixel-by-pixel to obtain the normal difference sequence. By smoothing and filtering out the grayscale fluctuations of periodic textures, the non-periodic grayscale abrupt changes caused by defects are retained, achieving signal decoupling between defect features and the texture background. Then, the variation amplitude of adjacent pixels in the difference sequence is compared bit-by-bit, and candidate defect sites exceeding the periodic fluctuation range are selected. Continuous sites are merged into candidate defect segments, and their pixel coordinates are recorded. Finally, the candidate defect segments in the bright and dark field images are spatially matched, and only the areas where the two images overlap are retained as the final defects. By leveraging the consistency of dual-modal imaging, the periodic gray-level fluctuations of directional grinding textures are avoided from masking micro-defects, thus improving the accuracy and anti-interference of the detection results. Attached Figure Description

[0019] Figure 1 A flowchart of an intelligent detection method for defects in aerospace components provided by the present invention; Figure 2 A schematic diagram of an intelligent defect detection system for aerospace components provided by the present invention; Figure 3 This is a schematic diagram of an electronic device provided by the present invention.

[0020] In the diagram: 201, Bright and Dark Field Texture Module; 202, Normal Sequence Extraction Module; 203, Smoothing Difference Module; 204, Defect Candidate Extraction Module; 205, Spatial Matching and Filtering Module; 301, Processor; 302, Memory. Detailed Implementation

[0021] 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 embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] like Figure 1 As shown in the figure, an intelligent detection method for defects in aerospace components proposed in this embodiment of the invention includes: S101, acquire the bright field image and dark field image of the working surface of the component to be inspected, as well as the main direction parameters of the surface grinding texture corresponding to the working surface of the component; the bright field image is obtained by collecting specular reflected light after the working surface is illuminated by a vertically incident light source, and the dark field image is obtained by collecting scattered light after the working surface is illuminated by an obliquely grazing incident light source; the main direction parameters of the surface grinding texture are extracted from the grinding process document of the component and are used to characterize the extension direction of the grinding texture on the working surface.

[0023] Specifically, the bright-field and dark-field images of the working surface of the component to be inspected, as well as the principal direction parameters of the surface grinding texture corresponding to the working surface, are first acquired. The bright-field image is obtained by collecting specular reflection light after the working surface is vertically illuminated by a vertically incident light source. The imaging principle is that a flat grinding texture surface will produce regular specular reflection. The reflected light enters the image acquisition device in the opposite direction along the incident light path, making the overall image present a uniform high grayscale background. However, surface pits, cracks and other defects will destroy the specular reflection conditions, causing the reflected light to deviate from the acquisition path and appear as low grayscale dark areas in the image. This imaging method is highly sensitive to the contour boundaries of surface depression-like morphological defects, and can clearly preserve the morphological details of the defects, providing a high-contrast contour basis for subsequent defect localization. Dark-field images are obtained by collecting surface-scattered light from a grazing incident light source at a small angle close to the working surface. The imaging principle is that the regular and flat grinding texture surface produces very little scattered light, and the dark-field image presents a low grayscale background. However, the rough morphology at the defect causes diffuse reflection of the incident light, and some of the scattered light enters the acquisition device and presents a high grayscale bright area. This imaging method can effectively suppress the background signal of uniform grinding texture, highlight the scattering characteristics of micro-scale defects, and form a feature complement with the bright-field image, which greatly improves the anti-interference ability of subsequent defect identification and reduces the probability of missed detection in a single imaging mode.

[0024] The formula for grayscale imaging of this bright-field image is: (1) In the formula, Bright field image The grayscale value at that location; Bright field image The specular reflectance at the location is close to 1 in the smooth texture area and significantly reduced in the defect area; This is the reference gray value corresponding to the reference light intensity of a vertically incident light source. This represents the random noise term in the acquisition of the bright-field image.

[0025] The formula for grayscale imaging of this dark field image is: (2) In the formula, For dark field images The grayscale value at that location; For dark field images The diffuse scattering rate at the location is close to 0 in the smooth texture area and significantly increases in the defect area; This is the reference gray value corresponding to the reference light intensity of the obliquely grazing incident light source; This represents the random noise term during the acquisition of the dark field image.

[0026] The principal direction parameter of the surface grinding texture is extracted from the grinding process document of the part. It is used to characterize the extension direction of the grinding texture on the working surface. The principle is that during the grinding process, the grinding wheel moves along a fixed feed direction and forms a periodic parallel texture extending along the feed direction on the working surface. The direction of this texture is directly determined by the processing parameters. Extracting it directly from the process document can avoid the direction deviation caused by image algorithm estimation, ensure the accuracy of the direction parameter, and provide an accurate direction reference for the subsequent extraction of grayscale sequence along the texture normal. For example, for the grinding working surface of nickel-based alloy of high-pressure compressor blade of aero-engine, the principal direction of the surface grinding texture extracted from its grinding process document is found to be at an angle of 15° with the blade length direction. The bright field image is obtained by using a coaxially arranged 550nm wavelength vertically incident LED light source with a field array camera to collect specular reflected light, and the dark field image is obtained by using two sets of symmetrically arranged 30° grazing incident strip light sources to collect surface scattered light. Finally, the bright field image, dark field image and principal direction parameter of the surface grinding texture of the corresponding working surface are obtained simultaneously.

[0027] S102, based on the main direction parameters of the surface grinding texture, multiple texture normal cross-section lines perpendicular to the grinding texture extension direction are defined in the bright field image and the dark field image, and each texture normal cross-section line is arranged in parallel along the grinding texture extension direction; the gray value of the corresponding pixel is extracted sequentially along each texture normal cross-section line to generate the bright field normal gray value sequence and the dark field normal gray value sequence.

[0028] Specifically, based on the principal direction parameters of the surface grinding texture extracted from the grinding process document, multiple texture normal cross-section lines perpendicular to the grinding texture extension direction are delineated in the simultaneously acquired bright-field and dark-field images. These texture normal cross-section lines are arranged parallel to the grinding texture extension direction. The parametric equations for the texture normal cross-section lines are as follows: (3) In the formula, The index of the texture normal section line; The pixel coordinates of the positioning reference point shared by bright field and dark field images; The number of pixels between adjacent cross-section lines along the main direction of the texture; The principal direction angle of the grinding texture; The normal direction angle of the texture; These are the path parameters for the cross-section line. , , These are the path parameters corresponding to the two intersection points of the cross-section line and the image boundary, which define the start and end range of the cross-section line.

[0029] Because grinding creates periodic parallel textures along the feed direction on the working surface, surface defects such as pits and cracks disrupt the texture continuity in this direction, causing abrupt changes in grayscale in the normal direction perpendicular to the texture. Therefore, extracting grayscale information along the normal cross-section line is the most sensitive way to capture the difference between the defect signal and the texture background. To ensure a strict correspondence between the spatial positions of the cross-section lines in the bright-field and dark-field images, a unified positioning reference point needs to be selected in the corner areas of the two images. That is, the imaging position of a preset corner of the working surface of the component to be inspected in the image is selected as the positioning reference point. For example, for the nickel-based alloy grinding working surface of the high-pressure compressor blade of an aero-engine, the corner point of the blade root tenon near the intake edge can be selected as the reference point. This corner point has the same pixel coordinates in the bright-field and dark-field images, thereby establishing a unified geometric reference system and avoiding subsequent matching errors caused by image misalignment.

[0030] After determining the reference point, the extension angle and starting position of the first texture normal cross-section line are determined based on the principal direction parameters of the surface grinding texture. The principal direction parameters of the surface grinding texture are directly given by the grinding process document. For example, if the principal direction of the grinding texture on the working surface of the blade is extracted from the process document and forms a 15° angle with the blade length direction, the extension angle of the normal cross-section line can be deduced to be 15° + 90° = 105°. This approach completely avoids the angular deviation that may occur when estimating the texture direction using image gradients, ensuring that the normal cross-section line is strictly perpendicular to the grinding texture, providing an accurate directional reference for extracting the grayscale sequence along the normal direction. The starting position of the first cross-section line is determined by offsetting the reference point by a set distance along the texture extension direction, such as 10 pixels, to avoid image grayscale anomalies caused by possible boundary chamfers or clamping marks.

[0031] Subsequently, along the grinding texture extension direction, the remaining texture normal cross-section lines are generated sequentially at equal intervals. The equal interval can be set according to the minimum defect size to be detected and the image resolution, for example, set to 20 pixels, so that there are no detection blind spots between adjacent cross-section lines and the computational redundancy caused by over-dense sampling is avoided. The extension direction of all texture normal cross-section lines is perpendicular to the grinding texture extension direction, and the beginning and end of each texture normal cross-section line extend to the image boundary of the corresponding image, so as to completely cover the entire imaging area of ​​the working surface of the part to be inspected, ensuring that no matter where the defect is located on the working surface, it will be penetrated by at least one normal cross-section line, thus eliminating missed detections from a mechanism perspective.

[0032] After the cross-section lines are defined, the gray values ​​of the corresponding pixels are extracted sequentially along each texture normal cross-section line to generate the bright field normal gray value sequence and the dark field normal gray value sequence. Specifically, the implementation is as follows: taking one endpoint of the texture normal cross-section line on the boundary between the bright field image and the dark field image as the starting pixel, usually the intersection point in the upper left is uniformly selected as the starting point, and advancing towards the adjacent pixels one by one along the extension direction of the texture normal cross-section line. The pixel reached by each advancement is taken as the gray value extraction object, and the gray value of the gray value extraction object in the bright field image is read. The gray value of the corresponding pixel in the dark field image is read in the same advancement order. The read gray values ​​are arranged in the advancement order to obtain the corresponding bright field normal gray value sequence and dark field normal gray value sequence. Among them, the expressions of the bright field normal gray value sequence and the dark field normal gray value sequence based on formulas (1) and (2) are as follows: (4) In the formula, For the first The first bright-field normal grayscale sequence corresponding to the cross-section line Bit grayscale value; For the first The first dark field normal grayscale sequence corresponding to the cross-sectional line Bit grayscale value; This refers to the pixel position index within the sequence. =1, 2...L, where L is the total pixel length of a single grayscale sequence; , The first The first section line The coordinates of each pixel are obtained from the parametric equation (3) of the texture normal section line.

[0033] This extraction method, employing a unified starting point and direction of advancement, ensures a one-to-one spatial correspondence between the bright-field and dark-field sequences, providing a precisely aligned one-dimensional signal for subsequent point-by-point gray-level difference calculations. The process of extracting the gray-level sequence along the normal cross-section essentially reduces the two-dimensional image to a one-dimensional fluctuation signal. This signal superimposed the periodic gray-level fluctuations caused by the uniform grinding texture and the local gray-level abrupt changes caused by defects. In the bright-field image, defects disrupt specular reflection, presenting low-gray-level dark areas that form downward pulses in the sequence; in the dark-field image, defects enhance diffuse reflection, presenting high-gray-level bright areas that form upward pulses in the sequence. These two characteristics exhibit inverse changes at the defect location. Representing the sequence using the normal direction not only preserves the complete morphological details of the defects in the vertical direction of the texture but also significantly reduces the computational dimensionality of subsequent smoothing and interpolation processes. This allows for effective suppression of the texture background and enhancement of the defect signal using simple one-dimensional signal processing techniques. Taking the aforementioned blade as an example, if the working area occupies a width of 800 pixels in the image, then the length of the bright field and dark field normal grayscale sequence generated by each cross-section line is 800 grayscale values. Multiple parallel sequences together constitute a complete scanning array for the defect information of the working surface, realizing a reliable conversion from a two-dimensional image to a one-dimensional feature signal.

[0034] S103, Perform sliding neighborhood smoothing processing of the same window scale on the bright field normal grayscale sequence and the dark field normal grayscale sequence respectively to obtain the bright field smoothed sequence and the dark field smoothed sequence; Calculate the grayscale difference of the bright field smoothed sequence and the dark field smoothed sequence pixel by pixel to obtain the normal difference sequence corresponding to each texture normal cross-section line.

[0035] Specifically, in the sliding neighborhood smoothing process, for each grayscale sequence along the texture normal cross-section, a fixed-length sliding window is used to advance along the sequence. The center of the window is aligned with each pixel position sequentially, and the arithmetic mean of all the grayscale values ​​covered by the window is taken as the smoothed grayscale value of the center pixel. This process is repeated throughout the entire sequence to form a bright-field smoothed sequence and a dark-field smoothed sequence. The formula for the sliding neighborhood smoothing process is as follows: (5) In the formula, For the first The bright-field smoothing sequence corresponding to the section line. Bit value; For the first The dark field smoothing sequence corresponding to the cross-section line. Bit value; The pixel length (window scale) of the sliding window is used, with the same value for both bright and dark fields; This is a floor function; This represents the relative positional offset within the window. , These are the original bright field and dark field normal grayscale sequences, respectively.

[0036] Here, the bright-field normal grayscale sequence and the dark-field normal grayscale sequence use the same window scale to ensure consistent spatial frequency attenuation characteristics and avoid introducing spurious differences due to varying smoothness. The window scale is set based on the texture period along the main direction of the surface grinding texture, preferably covering the number of pixels corresponding to at least one complete texture period. This way, the periodic grayscale fluctuations caused by uniform grinding texture are largely offset by the averaging operation, resulting in a smoother background. The local grayscale abrupt changes caused by defects, being non-periodic and exhibiting inverse variation characteristics in both bright and dark field images, are weakened in magnitude during smoothing, but their trend direction is still well preserved. This smoothing operation is essentially equivalent to low-pass filtering the one-dimensional sequence, effectively suppressing high-frequency noise in the textured background and highlighting the weak defect signals that were originally submerged in regular fluctuations, thus reducing the risk of false alarms in subsequent processing.

[0037] Next, pixel-by-pixel grayscale difference calculation is performed. The grayscale values ​​at the same pixel location are subtracted from the grayscale values ​​of the bright-field smoothed sequence and the dark-field smoothed sequence to form a normal difference sequence. The expression for the difference sequence is as follows: (6) In the formula, For the first The normal difference sequence corresponding to the cross-section line. Bit value; For the first The bright-field smoothing sequence corresponding to the section line. Bit value; For the first The dark field smoothing sequence corresponding to the cross-section line. Bit value.

[0038] This differential operation fully utilizes the complementary characteristics of bright-field and dark-field imaging: in a smooth, flawless ground texture, bright-field specular reflection is strong, resulting in a high grayscale value after smoothing; dark-field scattering is extremely weak, resulting in a low grayscale value after smoothing. Subtracting the two yields a stable positive baseline value. When defects such as pits or cracks exist, specular reflection is disrupted under bright-field conditions, causing a decrease in grayscale; diffuse reflection is enhanced under dark-field conditions, causing an increase in grayscale. This results in a decrease in the bright-field smoothing value and an increase in the dark-field smoothing value at that location. The difference between the two deviates significantly from the baseline, forming a significant negative pulse. This method allows even minute defects to form clearly discernible anomalous features, providing a highly sensitive basis for the selection of candidate defect sites based on the magnitude of grayscale difference changes in step S104.

[0039] S104. For each texture normal cross-section line, the gray-level difference sequence is compared bit by bit with the gray-level difference change range of adjacent pixels. Pixel sites whose change range exceeds the corresponding periodic gray-level fluctuation range are selected and marked as candidate defect sites. Candidate defect sites continuously distributed on the same texture normal cross-section line are merged into candidate defect segments, and the pixel coordinates of each candidate defect segment in the corresponding image are recorded.

[0040] Specifically, this step involves using the periodic fluctuations of the grinding texture itself as an adaptive criterion in the differentially enhanced normal difference sequence to accurately separate the gray-level anomaly signal caused by defects from the normal texture fluctuation signal. Simultaneously, discrete anomaly sites are integrated into continuous defect segments, providing structured candidate regions for subsequent light and dark field spatial matching. This effectively reduces subsequent computational load and eliminates false alarms caused by single-point noise. For each texture normal cross-section line corresponding to the normal difference sequence, the gray-level difference variation amplitude of adjacent pixels is compared bit by bit. Pixel sites with variation amplitudes exceeding the corresponding periodic gray-level fluctuation amplitude are selected and marked as candidate defect sites. The principle behind this determination method is that the fluctuation of the normal difference sequence in the defect-free area is entirely dominated by the alternation of light and dark in the regular grinding texture. The grayscale change amplitude of its adjacent pixels is stable within a limited range. However, defects such as pits and cracks will break the periodic structure of the texture, causing grayscale jumps that are much greater than those of normal texture fluctuations. Therefore, using the periodic fluctuation amplitude of the texture itself as the threshold benchmark, it can adaptively adapt to the surface of parts with different processing techniques and materials without the need for manual preset of fixed thresholds, which greatly improves the versatility and scene adaptability of the detection method.

[0041] The periodic grayscale fluctuation amplitude is obtained by statistically analyzing the intrinsic texture fluctuation characteristics of the defect-free region. Since the normal difference sequence is obtained by subtracting the bright-field smoothing sequence and the dark-field smoothing sequence pixel by pixel, its background fluctuation is composed of the superposition of the periodic texture fluctuations of the bright-field and dark-field. Therefore, it is necessary to obtain the intrinsic texture fluctuation amplitude under the two imaging modes separately. Specifically, this includes selecting a pre-confirmed defect-free local area in the bright-field image as the bright-field defect-free region, and selecting a pre-confirmed defect-free local area in the dark-field image as the dark-field defect-free region. The defect-free region is usually a rectangular area at the center of the working surface that has been pre-confirmed to be defect-free. Its coverage area must include at least 20 complete texture cycles to ensure that the statistical sample is sufficiently representative and to avoid local sporadic noise interfering with the accuracy of the baseline value. Along the direction of the texture normal cross-section line, the bright-field defect-free grayscale sequence is extracted in the bright-field defect-free region. The grayscale difference of each pair of adjacent pixels in the bright-field defect-free grayscale sequence is calculated, and the magnitude of the grayscale difference is taken as the bright-field grayscale difference change amplitude. The expression for the grayscale change amplitude of adjacent pixels in the defect-free region is as follows: (7) In the formula, For the bright field defect-free grayscale sequence The magnitude of grayscale change between adjacent pixels; For the dark field defect-free grayscale sequence, the first... The magnitude of grayscale change between adjacent pixels; This is a defect-free grayscale sequence extracted from the defect-free region of the bright field; This is a defect-free grayscale sequence extracted from the defect-free region in the dark field.

[0042] The principle behind this calculation is that the grayscale difference between adjacent pixels directly reflects the rate of grayscale change of the texture in the normal direction. The difference in periodic textures will exhibit stable and repeating values, while the difference distribution of random noise is discrete. By frequency statistics, the intrinsic fluctuation characteristics of the texture can be effectively separated. By statistically analyzing all the bright-field grayscale difference changes obtained in the defect-free bright-field area, the bright-field grayscale difference change amplitude with the highest frequency is taken as the periodic grayscale fluctuation amplitude corresponding to the bright-field image. Using this method as the benchmark value can eliminate the interference of occasional anomalies such as local dust and minor scratches on the benchmark value, ensuring that the threshold accurately reflects the typical fluctuation level of normal textures and avoiding the threshold being raised by outliers, which could lead to the missed detection of minor defects. Similarly, along the direction of the texture normal cross-section line, a dark-field defect-free grayscale sequence is extracted within the dark-field defect-free region. The grayscale difference between each pair of adjacent pixels in this sequence is calculated, and the magnitude of this difference is taken as the dark-field grayscale difference variation amplitude. All dark-field grayscale difference variation amplitudes obtained within the dark-field defect-free region are statistically analyzed, and the amplitude of the most frequent grayscale difference variation is taken as the periodic grayscale fluctuation amplitude corresponding to the dark-field image. The beneficial effect of calculating the periodic grayscale fluctuation amplitudes of bright-field and dark-field images separately is that the imaging mechanisms of the two imaging modes are different. Bright-field images have high background grayscale and strong texture contrast, while dark-field images have low background grayscale and weak texture signals. The absolute amplitudes of texture fluctuations differ between the two. Separate statistical analysis ensures the accuracy of each threshold and avoids the problem of missed detections in bright-field images or false alarms in dark-field images caused by adapting a single threshold to both types of images.

[0043] For example, regarding the nickel-based alloy grinding surface of the aforementioned high-pressure compressor blades for aero-engines, the most frequent grayscale difference between adjacent pixels in the bright-field defect-free area is 3 grayscale levels, meaning the periodic grayscale fluctuation amplitude in the bright-field is 3. In the dark-field defect-free area, the most frequent grayscale difference between adjacent pixels is 2 grayscale levels, meaning the periodic grayscale fluctuation amplitude in the dark-field is 2. The sum of these two values ​​yields a normal fluctuation baseline amplitude of 5 grayscale levels for the difference sequence. In actual judgment, this baseline amplitude can be multiplied by a safety factor of 1.5 to obtain the final judgment threshold, which can tolerate slight texture non-uniformity and image acquisition noise, thereby further reducing the false alarm rate while ensuring detection sensitivity.

[0044] After screening candidate defect sites, continuously distributed candidate defect sites along the same texture normal cross-section are merged into candidate defect segments, and the pixel coordinates of each candidate defect segment in the corresponding image are recorded. The principle behind this merging operation is that true defects typically appear as a continuous gray-scale anomaly region on the texture normal cross-section, rather than isolated single pixels. Adjacent-to-adjacent merging can directly eliminate isolated noise-generating pseudo-defects, while integrating discrete pixel sites into segments with spatial ranges, facilitating subsequent cross-image position matching and defect size quantification calculations. Specifically, all candidate defect sites are traversed sequentially along the extension direction of the texture normal cross-section, and adjacent candidate defect sites are grouped into the same continuous set. Each continuous set corresponds to a candidate defect segment. For each candidate defect segment, its starting pixel coordinates, ending pixel coordinates, and the entire pixel coordinate range covered by the segment in the corresponding image are recorded. Recording the complete coordinate range provides accurate geometric basis for subsequent spatial overlap matching and can be directly used to calculate the width of the defect in the normal direction, providing quantitative data support for subsequent defect level assessment.

[0045] S105, perform spatial position matching based on pixel coordinates on all candidate defect segments corresponding to the texture normal cross-section lines in the bright field image and dark field image, and retain only the candidate defect segments whose positions overlap in the bright field image and dark field image as the final defect area; output the corresponding position information of the final defect area on the working surface of the part to be inspected.

[0046] Specifically, this step utilizes the synchronous response characteristics of bright-field and dark-field imaging to real defects, and eliminates spurious defect signals under single-mode conditions through spatial position matching, further improving the accuracy and reliability of the detection results. A unified image coordinate system is used as a reference, established based on the positioning reference points at the corners of the working surface selected in step S102. Because the bright-field and dark-field images are acquired synchronously and the pixel coordinates of the reference points are completely consistent, their pixel spaces have a one-to-one mapping relationship. Coordinate comparison can be performed directly without additional registration, fundamentally avoiding matching errors caused by image misalignment. This ensures matching accuracy while eliminating complex image registration calculations, effectively improving detection efficiency. The pixel coordinates of each candidate defect segment in the bright field image are mapped and compared one by one with the pixel coordinates of each candidate defect segment in the dark field image. Here, a method of matching each cross-section line is adopted. That is, the candidate defect segments with the same index on the texture normal cross-section line in the bright field and dark field are first matched and compared, and then the spatial continuity of adjacent segments that cross the cross-section line is checked. This matching logic is based on the physical characteristic that real defects will pass through the corresponding cross-section lines of the bright and dark fields at the same time, which can significantly reduce the matching search range and reduce the computational complexity. When the coordinate coverage of two candidate defect segments overlaps and the overlapping area meets the overlap determination criteria, the two are identified as candidate defect segments with overlapping positions, and the corresponding matching process is completed. The overlap determination criteria are set so that the ratio of the length of the overlapping area to the length of both the bright-field and dark-field candidate defect segments is not less than a set threshold. This threshold can be set to 60% according to the detection accuracy requirements. The principle is that real defects are highly consistent in position in both bright-field and dark-field images, with only minor differences in imaging diffusion. False defects such as dust and imaging noise only appear in a single mode or have a large positional deviation. The proportional threshold screening can tolerate normal imaging deviations while effectively filtering out single-mode false defects, significantly reducing the false alarm rate. Only candidate defect segments with overlapping positions in both bright-field and dark-field images are retained as the final defect area. This dual-verification mechanism fully leverages the complementary advantages of bright-field and dark-field imaging features. Only areas detected simultaneously in both imaging modes and with matching spatial positions are determined as real defects, thus eliminating problems such as texture interference and noise misjudgment that are prone to occur in a single imaging mode. The system outputs the corresponding position information of the final defect area on the working surface of the component to be inspected. The output includes the pixel coordinate range, normal width, and length along the texture direction of the defect area, which can be directly mapped to the actual physical size of the component, providing a quantitative basis for subsequent defect classification and process optimization.

[0047] like Figure 2 As shown, based on the same inventive concept, this embodiment provides an intelligent detection system for defects in aerospace components, including: The bright and dark field texture module 201 is used to acquire the bright field image and dark field image of the working surface of the part to be inspected, as well as the main direction parameters of the surface grinding texture corresponding to the working surface of the part. The bright field image is obtained by collecting specular reflected light after the working surface is illuminated by a vertically incident light source, and the dark field image is obtained by collecting scattered light after the working surface is illuminated by an obliquely grazing incident light source. The main direction parameters of the surface grinding texture are extracted from the grinding process file of the part and are used to characterize the extension direction of the grinding texture on the working surface. The normal sequence extraction module 202 is used to define multiple texture normal cross-section lines perpendicular to the grinding texture extension direction in the bright field image and the dark field image based on the main direction parameters of the surface grinding texture. Each texture normal cross-section line is arranged in parallel along the grinding texture extension direction. The gray value of the corresponding pixel is extracted sequentially along each texture normal cross-section line to generate a bright field normal gray value sequence and a dark field normal gray value sequence. The smoothing difference module 203 is used to perform sliding neighborhood smoothing processing on the bright field normal grayscale sequence and the dark field normal grayscale sequence respectively, to obtain the bright field smoothed sequence and the dark field smoothed sequence; and to calculate the grayscale difference of the bright field smoothed sequence and the dark field smoothed sequence pixel by pixel to obtain the normal difference sequence corresponding to each texture normal cross-section line. The defect candidate extraction module 204 is used to compare the gray-level difference change range of adjacent pixels bit by bit for the normal difference sequence corresponding to each texture normal cross-section line, filter out the pixel sites whose change range exceeds the corresponding periodic gray-level fluctuation range, and mark them as candidate defect sites; merge the candidate defect sites continuously distributed on the same texture normal cross-section line into candidate defect segments, and record the pixel coordinate position of each candidate defect segment in the corresponding image. The spatial matching and filtering module 205 is used to perform spatial position matching based on pixel coordinates on all candidate defect segments corresponding to the texture normal cross-section lines in the bright field image and the dark field image, and retain only the candidate defect segments whose positions overlap in the bright field image and the dark field image as the final defect area; and output the corresponding position information of the final defect area on the working surface of the part to be inspected.

[0048] like Figure 3 As shown, based on the same inventive concept, this embodiment provides an electronic device, including: Memory 302 is used to store computer programs; Processor 301 is used to implement the method steps as described in the first aspect when executing a computer program.

[0049] Based on the same inventive concept, this embodiment provides a readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the method steps as described in the first aspect.

[0050] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0051] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent detection of defects in aerospace components, characterized in that, include: S101, acquire the bright field image and dark field image of the working surface of the component to be inspected, as well as the main direction parameters of the surface grinding texture corresponding to the working surface of the component; The bright-field image is obtained by collecting specular reflected light after the working surface is illuminated by a vertically incident light source, and the dark-field image is obtained by collecting scattered light after the working surface is illuminated by an obliquely grazing incident light source; the main direction parameter of the surface grinding texture is extracted from the part grinding process document and is used to characterize the grinding texture extension direction of the working surface. S102, based on the main direction parameters of the surface grinding texture, in the bright field image and the dark field image, multiple texture normal cross-section lines perpendicular to the grinding texture extension direction are defined respectively, and each texture normal cross-section line is arranged parallel to the grinding texture extension direction; the gray value of the corresponding pixel is extracted sequentially along each texture normal cross-section line to generate a bright field normal gray value sequence and a dark field normal gray value sequence; S103, Perform sliding neighborhood smoothing processing of the same window scale on the bright field normal grayscale sequence and the dark field normal grayscale sequence respectively to obtain a bright field smoothed sequence and a dark field smoothed sequence; Calculate the grayscale difference of the bright field smoothed sequence and the dark field smoothed sequence pixel by pixel to obtain the normal difference sequence corresponding to each texture normal cross-section line. S104, for each texture normal cross-section line corresponding to the normal difference sequence, compare the gray level difference change range of adjacent pixels bit by bit, and filter out the pixel sites whose change range exceeds the corresponding periodic gray level fluctuation range, and mark them as candidate defect sites. Candidate defect sites that are continuously distributed along the same texture normal section line are merged into candidate defect segments, and the pixel coordinates of each candidate defect segment in the corresponding image are recorded. S105, perform spatial position matching based on pixel coordinates on all candidate defect segments corresponding to the texture normal cross-section lines in the bright field image and the dark field image, and retain only the candidate defect segments whose positions overlap in the bright field image and the dark field image as the final defect area; Output the location information of the final defect area on the working surface of the component to be inspected.

2. The intelligent detection method for defects in aerospace components according to claim 1, characterized in that, The definition of multiple texture normal cross-section lines perpendicular to the grinding texture extension direction specifically includes: A unified positioning reference point is selected in the corner area of ​​the bright field image and the dark field image; Based on the main direction parameters of the surface grinding texture, determine the extension angle and starting position of the first texture normal section line; Along the direction of the grinding texture extension, the remaining texture normal section lines are generated sequentially at equal intervals; In this process, the extension direction of all texture normal cross-section lines is perpendicular to the grinding texture extension direction, and the beginning and end of each texture normal cross-section line extend to the image boundary of the corresponding image, so as to completely cover the entire imaging area of ​​the working surface of the part to be inspected.

3. The intelligent detection method for defects in aerospace components according to claim 2, characterized in that, The step of selecting a unified positioning reference point in the corner regions of the bright field image and the dark field image specifically includes: The preset corner of the working surface of the component to be tested is selected as the imaging position in the image as the positioning reference point.

4. The intelligent detection method for defects in aerospace components according to claim 1, characterized in that, The step of sequentially extracting the grayscale values ​​of corresponding pixels along each of the texture normal cross-section lines to generate a bright field normal grayscale sequence and a dark field normal grayscale sequence specifically includes: Starting with one endpoint of the texture normal cross-section line on the boundary between the bright field image and the dark field image as the starting pixel, the process proceeds sequentially towards adjacent pixels along the extension direction of the texture normal cross-section line. Each pixel reached during the process is taken as the grayscale extraction object. The grayscale value of the grayscale extraction object in the bright field image is read, and the read grayscale values ​​are arranged sequentially according to the order of the process to obtain the corresponding bright field normal grayscale sequence and dark field normal grayscale sequence.

5. The intelligent detection method for defects in aerospace components according to claim 1, characterized in that, The periodic grayscale fluctuation amplitude is determined in the following ways, specifically including: In the bright-field image, a local area that has been pre-confirmed as defect-free is selected as the bright-field defect-free area, and in the dark-field image, a local area that has been pre-confirmed as defect-free is selected as the dark-field defect-free area. Along the direction of the texture normal section line, a bright-field defect-free grayscale sequence is extracted within the bright-field defect-free region. The grayscale difference between each pair of adjacent pixels in the bright-field defect-free grayscale sequence is calculated, and the magnitude of the grayscale difference is taken as the bright-field grayscale difference variation amplitude. All bright-field grayscale difference variation amplitudes obtained within the bright-field defect-free region are statistically analyzed, and the bright-field grayscale difference variation amplitude with the highest frequency is taken as the periodic grayscale fluctuation amplitude corresponding to the bright-field image. Along the direction of the texture normal cross-section line, extract the dark field defect-free grayscale sequence within the dark field defect-free region, calculate the grayscale difference of each pair of adjacent pixels in the dark field defect-free grayscale sequence, and take the magnitude of the grayscale difference as the dark field grayscale difference variation amplitude; count all the dark field grayscale difference variation amplitudes obtained within the dark field defect-free region, and take the dark field grayscale difference variation amplitude with the highest frequency as the periodic grayscale fluctuation amplitude corresponding to the dark field image.

6. The intelligent detection method for defects in aerospace components according to claim 1, characterized in that, The step of merging candidate defect sites that are continuously distributed along the same texture normal cross-section line into a candidate defect segment specifically includes: All candidate defect sites are traversed sequentially along the extension direction of the texture normal section line. Candidate defect sites that are adjacent to each other are grouped into the same continuous set, and each continuous set corresponds to a candidate defect segment. For each candidate defect segment, its starting pixel coordinates, ending pixel coordinates, and the range of all pixel coordinates covered by the segment in the corresponding image are recorded.

7. The intelligent detection method for defects in aerospace components according to claim 1, characterized in that, The step of performing pixel coordinate-based spatial location matching on all candidate defect segments corresponding to the texture normal cross-sections in the bright field image and the dark field image specifically includes: Using a unified image coordinate system as a reference, the pixel coordinates of each candidate defect segment in the bright field image are mapped and compared one by one with the pixel coordinates of each candidate defect segment in the dark field image; when it is determined that the coordinate coverage of two candidate defect segments overlaps and the overlapping area meets the overlap determination condition, the two are identified as candidate defect segments with overlapping positions, and the corresponding matching process is completed.

8. An intelligent detection system for defects in aerospace components, using the intelligent detection method for defects in aerospace components as described in claim 1, characterized in that, include: The bright and dark field texture module is used to acquire bright field images and dark field images of the working surface of the component to be inspected, as well as the principal direction parameters of the surface grinding texture corresponding to the working surface of the component. The bright field image is obtained by collecting specular reflected light after the working surface is illuminated by a vertically incident light source, and the dark field image is obtained by collecting scattered light after the working surface is illuminated by an obliquely grazing incident light source. The principal direction parameters of the surface grinding texture are extracted from the grinding process file of the component and are used to characterize the extension direction of the grinding texture on the working surface. The normal sequence extraction module is used to define multiple texture normal cross-section lines perpendicular to the grinding texture extension direction in the bright field image and the dark field image, based on the main direction parameters of the surface grinding texture. Each texture normal cross-section line is arranged parallel to the grinding texture extension direction. The gray value of the corresponding pixel is extracted sequentially along each texture normal cross-section line to generate a bright field normal gray value sequence and a dark field normal gray value sequence. The smoothing difference module is used to perform sliding neighborhood smoothing processing of the same window scale on the bright field normal grayscale sequence and the dark field normal grayscale sequence respectively to obtain a bright field smoothed sequence and a dark field smoothed sequence; and to calculate the grayscale difference of the bright field smoothed sequence and the dark field smoothed sequence pixel by pixel to obtain the normal difference sequence corresponding to each texture normal cross-section line. The defect candidate extraction module is used to compare the grayscale difference change range of adjacent pixels bit by bit for the normal difference sequence corresponding to each texture normal cross-section line, and filter out pixel sites whose change range exceeds the corresponding periodic grayscale fluctuation range, and mark them as candidate defect sites. Candidate defect sites that are continuously distributed along the same texture normal section line are merged into candidate defect segments, and the pixel coordinates of each candidate defect segment in the corresponding image are recorded. The spatial matching and filtering module is used to perform spatial position matching based on pixel coordinates on all candidate defect segments corresponding to the texture normal cross-section lines in the bright field image and the dark field image, and retain only the candidate defect segments whose positions overlap in the bright field image and the dark field image as the final defect area. Output the location information of the final defect area on the working surface of the component to be inspected.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the intelligent detection method for defects in aerospace components as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements an intelligent detection method for defects in aerospace components as described in any one of claims 1 to 7.

Images