A method and system for online defect detection of a TPMS gradient structure

Abstract

The application is suitable for the field of intelligent detection, and provides an online defect detection method and system for a TPMS gradient structure, which comprises the following steps: collecting a two-dimensional transmission light field image of a TPMS structure with light transmission; pre-processing and feature quantization are performed on the two-dimensional transmission light field image, overall statistical characteristics of the image are calculated to obtain a comprehensive optical feature vector capable of representing internal scattering intensity; the comprehensive optical feature vector is input into a pre-trained deep learning segmentation model to obtain a segmentation atlas with the same size as the two-dimensional transmission light field image; the total projection area of the internal blocked area of the TPMS structure is calculated according to the segmentation atlas, and is taken as a quantitative index of the internal blocking degree; and the index is compared with a preset qualified threshold. The application takes the projection area as the quantitative index, and can accurately evaluate the internal blocking degree of the light transmission TPMS structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent detection, in particular to an online defect detection method and system for TPMS gradient structure. BACKGROUND

[0002] Three-dimensional periodic minimal surface (TPMS) gradient structure, such as Gyroid, Schwarz D and the like, has huge application potential in the fields of aerospace, biological medicine and automobile industry due to its excellent specific strength, energy absorption characteristics and fluid conduction performance.

[0003] In additive manufacturing technologies such as powder bed fusion and light solidification, when manufacturing TPMS gradient structure made of materials such as high molecular polymer, special light-transmitting ceramic and the like which have good light transmission to specific wavelength light, a difficult problem will occur. During printing and post-processing, fine support powder or incompletely solidified resin is easily captured in the complex and winding channels inside the structure, forming internal residues that are difficult to remove. Such defects have a huge impact on components that rely on internal channels to achieve functions. It can completely block the flow or significantly reduce the mass transfer efficiency.

[0004] The existing conventional non-destructive testing methods have no way to deal with such internal blockage. The external machine vision system cannot see through the inside of the component. Although industrial CT can accurately image, its high cost, long detection cycle and radiation characteristics make it impossible to be integrated into the production line as an online and full detection means. Therefore, an online defect detection method and system for TPMS gradient structure are proposed to solve the above problems. SUMMARY

[0005] In view of the deficiencies of the prior art, the purpose of the present application is to provide an online defect detection method and system for TPMS gradient structure to solve the problems in the background art.

[0006] The present application is implemented as follows: an online defect detection method for TPMS gradient structure, the method comprising the following steps:

[0007] Collecting a two-dimensional transmission light field image of a TPMS structure with light transmission, the two-dimensional transmission light field image being obtained by the cooperation of a uniform backlight light source and an industrial camera, the uniform backlight light source being capable of emitting parallel light of a specific wavelength to penetrate the TPMS structure;

[0008] Pretreating and feature quantifying the two-dimensional transmission light field image, and calculating the overall statistical characteristics of the image to obtain a comprehensive optical feature vector capable of representing the internal scattering intensity;

[0009] The comprehensive optical feature vector is input into the pre-trained deep learning segmentation model to obtain a segmentation atlas with the same size as the two-dimensional transmission light field image.

[0010] The total projection area of the blocked area inside the TPMS structure is calculated according to the segmentation atlas, and the total projection area is taken as a quantitative index of the internal blocking degree, and the index is compared with a preset qualified threshold.

[0011] As a further scheme of the present application, the step of collecting the two-dimensional transmission light field image of the TPMS structure with light transmission specifically includes:

[0012] A position detection image capable of representing the position of the TPMS structure is obtained by an industrial camera, and the position detection image is compared with a preset imaging area;

[0013] When the TPMS structure is located in the imaging area, a cooperative instruction is generated and sent to the device corresponding to the uniform backlight light source, so that the device can emit parallel light of two different wavelengths in turn, including a wavelength with strong penetration to the base material and a wavelength with strong scattering to the residual powder;

[0014] An adjustment instruction is generated and sent to the industrial camera to adjust the exposure time and gain of the industrial camera, so as to obtain a group of original images under different imaging parameters;

[0015] The group of original images are fused and registered to generate the two-dimensional transmission light field image.

[0016] As a further scheme of the present application, the step of preprocessing and feature quantization of the two-dimensional transmission light field image specifically includes:

[0017] The obtained two-dimensional transmission light field image is dark-field and flat-field corrected to eliminate the influence of light source unevenness and camera dark noise;

[0018] The image is decomposed into low-frequency and high-frequency components by Fourier transform in the image frequency domain, and the proportion of high-frequency energy in the total energy is calculated to obtain a first feature quantity capable of representing microscopic scattering;

[0019] The texture of the image is analyzed by using a local binary pattern algorithm in the image spatial domain, and the local contrast mean value of the whole image is calculated to obtain a second feature quantity capable of representing macroscopic scattering unevenness;

[0020] The first feature quantity and the second feature quantity are spliced and normalized with the global gray standard deviation of the image to obtain the comprehensive optical feature vector.

[0021] As a further scheme of the present application: the deep learning segmentation model is a U-Net variant introducing a channel-spatial dual attention mechanism, which enables the model to calibrate the weights of different scattering features in the channel dimension and focus on the local distortion area strongly related to internal blockage in the spatial dimension, for identifying the projection of powder residue in the TPMS background structure.

[0022] As a further scheme of the present application: the step of calculating the total projection area of the internal blockage area of the TPMS structure according to the segmentation map and taking it as a quantitative indicator of the internal blockage degree, and comparing it with a preset qualified threshold, specifically includes:

[0023] Performing connected component analysis on the obtained segmentation map to identify and label each independent suspected blockage area;

[0024] Based on the internal pore space distribution map of the three-dimensional structure, different weight coefficients are assigned to each blockage area under the current two-dimensional projection view, and the internal pore space distribution map is obtained through the digital model of the TPMS structure;

[0025] Calculate the weighted projection area sum of all suspected blockage areas to obtain the quantitative indicator, and the weighted projection area sum is the pixel area of each area multiplied by the weight coefficient of its position and then accumulated;

[0026] Compare the weighted projection area sum with the preset qualified threshold, and automatically generate sorting instructions and traceable quality reports containing component ID, defect image and quantitative indicator according to the comparison result.

[0027] As a further scheme of the present application: the method further includes:

[0028] When the quantitative indicator is within a preset critical interval based on the qualified threshold, a rotation control instruction is generated for controlling the rotation of the rotating bearing device, so that the TPMS structure is rotated by two different angles, and the TPMS structure is located on the rotating bearing device;

[0029] Collect two-dimensional transmission light field images at each angle through an industrial camera, and arrange to obtain a set of multi-view projection images;

[0030] Data fusion is performed on the multi-view projection images, a three-dimensional spatial distribution model of the internal blockage of the TPMS structure is generated through a three-dimensional reconstruction algorithm, and the total volume of the blockage is calculated based on the model;

[0031] Compare the total volume with a preset second qualified threshold, and when the total volume exceeds the second qualified threshold, generate a prompt information indicating that the TPMS structure is unqualified.

[0032] Another object of the present application is to provide an online defect detection system for TPMS gradient structure, which comprises:

[0033] An image acquisition module is configured to acquire a two-dimensional transmission light field image of the TPMS structure with light transmission, which is obtained by cooperation of a uniform backlight light source and an industrial camera, and the uniform backlight light source can emit parallel light of a specific wavelength to penetrate the TPMS structure;

[0034] An image processing module is configured to pre-process and feature quantify the two-dimensional transmission light field image, and calculate overall statistical features of the image to obtain a comprehensive optical feature vector capable of representing internal scattering intensity;

[0035] A model processing module is configured to input the comprehensive optical feature vector into a pre-trained deep learning segmentation model to obtain a segmentation atlas with the same size as the two-dimensional transmission light field image;

[0036] A quantitative calculation module is configured to calculate the total projection area of the internal blockage area of the TPMS structure according to the segmentation atlas, and take the total projection area as a quantitative index of the internal blockage degree, and compare the index with a preset qualified threshold.

[0037] As a further scheme of the present application, the image acquisition module comprises:

[0038] A position detection unit is configured to acquire a position detection image capable of representing the position of the TPMS structure by the industrial camera, and compare the position detection image with a preset imaging area;

[0039] A light source control unit is configured to generate a cooperation instruction and send the cooperation instruction to a device corresponding to the uniform backlight light source when the TPMS structure is located in the imaging area, so that the device can emit parallel light of two different wavelengths in sequence, including a wavelength with strong penetration to the base material and a wavelength with strong scattering to residual powder;

[0040] An image acquisition unit is configured to generate an adjustment instruction and send the adjustment instruction to the industrial camera to adjust the exposure time and gain of the industrial camera, so as to acquire a group of original images under different imaging parameters;

[0041] An image fusion unit is configured to perform image fusion and registration on the group of original images to generate the two-dimensional transmission light field image.

[0042] As a further scheme of the present application, the image processing module comprises:

[0043] An image correction unit is configured to perform dark field and flat field correction on the acquired two-dimensional transmission light field image, so as to eliminate the influence of light source unevenness and camera dark noise;

[0044] The first feature acquisition unit is configured to decompose the image into low-frequency and high-frequency components by Fourier transform in the image frequency domain, and calculate the proportion of high-frequency energy in total energy to obtain a first feature quantity capable of representing microscopic scattering;

[0045] The second feature acquisition unit is configured to analyze the texture of the image by using a local binary pattern algorithm in the image spatial domain, and calculate the local contrast mean of the whole image to obtain a second feature quantity capable of representing macroscopic scattering unevenness.

[0046] The data normalization unit is configured to splice and normalize the first feature quantity and the second feature quantity with the global gray standard deviation of the image to obtain the comprehensive optical feature vector.

[0047] As a further scheme of the present application, the system further comprises a precision review module, and the precision review module comprises:

[0048] The index judgment unit is configured to generate a rotation control instruction for controlling the rotation of the bearing device when the quantitative index is within a preset critical interval based on the qualified threshold, so as to rotate the TPMS structural member by two different angles, and the TPMS structural member is located on the bearing device.

[0049] The multi-view acquisition unit is configured to acquire two-dimensional transmission light field images under each angle by using an industrial camera, and arrange a group of multi-view projection images.

[0050] The multi-view fusion unit is configured to perform data fusion on the multi-view projection images, generate a three-dimensional spatial distribution model of the blockage in the TPMS structural member by using a three-dimensional reconstruction algorithm, and calculate the total volume of the blockage based on the model.

[0051] The comparison unit is configured to compare the total volume with a preset second qualified threshold, and generate prompt information indicating that the TPMS structural member is unqualified when the total volume exceeds the second qualified threshold.

[0052] Compared with the prior art, the present application has the following advantages:

[0053] The present application is mainly aimed at the internal blockage detection of light-transmitting TPMS gradient structure. Firstly, a stable and reliable transmission imaging system is constructed by combining the uniformly designed backlight source with the industrial camera, which can effectively capture the light field scattering signal caused by the internal powder residue. In the image processing link, the global statistical characteristics of the image are analyzed to construct the optical feature vector which can accurately represent the internal blockage condition. A specially trained deep learning segmentation model is introduced, which can more accurately distinguish between normal structure shadow and real blockage area, significantly improving the accuracy and reliability of the detection. In addition, the present application creatively uses the projection area as a quantitative indicator, which not only ensures the detection efficiency, but also ensures the high relevance of the evaluation results to the actual performance of the component. In terms of cost control, only conventional optical elements are needed; in terms of detection efficiency, the evaluation of a single component can be quickly completed, adapting to the rhythm of the industrial production line. It should be noted that the whole system is easy to integrate into the existing production process, providing a practical technical path for realizing the full number of online detection of TPMS structure. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 Flow chart of a TPMS gradient structure online defect detection method.

[0055] Figure 2 Flow chart of collecting two-dimensional transmission light field images in a TPMS gradient structure online defect detection method.

[0056] Figure 3 Flow chart of pre-processing and feature quantization of two-dimensional transmission light field images in a TPMS gradient structure online defect detection method.

[0057] Figure 4 Flow chart of internal blockage degree quantization in a TPMS gradient structure online defect detection method.

[0058] Figure 5 Flow chart of another detection method in a TPMS gradient structure online defect detection method.

[0059] Figure 6 Structural schematic diagram of a TPMS gradient structure online defect detection system.

[0060] Figure 7 Structural schematic diagram of an image acquisition module in a TPMS gradient structure online defect detection system.

[0061] Figure 8 Structural schematic diagram of an image processing module in a TPMS gradient structure online defect detection system.

[0062] Figure 9It is a structure diagram of a precision review module in an online defect detection system of a TPMS gradient structure. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0064] The specific implementation of the present application will be described in detail below in combination with specific embodiments.

[0065] As shown in the drawings, the embodiment of the present application provides an online defect detection method for a TPMS gradient structure, which comprises the following steps: Figure 1 S100, a two-dimensional transmission light field image of a TPMS structure with light transmission is collected, the two-dimensional transmission light field image is obtained by the cooperation of a uniform backlight light source and an industrial camera, and the uniform backlight light source can emit parallel light of a specific wavelength to penetrate the TPMS structure;

[0066] S200, the two-dimensional transmission light field image is preprocessed and feature-quantized, and the overall statistical feature of the image is calculated to obtain a comprehensive optical feature vector capable of representing the internal scattering intensity;

[0067] S300, the comprehensive optical feature vector is input into a pre-trained deep learning segmentation model to obtain a segmentation atlas with the same size as the two-dimensional transmission light field image;

[0068] S400, the total projection area of the internal blockage area of the TPMS structure is calculated according to the segmentation atlas, and is taken as a quantitative index of the internal blockage degree, and the index is compared with a pre-set qualified threshold.

[0069] It should be noted that the deep learning segmentation model is a U-Net variant with a channel-spatial double attention mechanism, which enables the model to calibrate the weights of different scattering features in the channel dimension and focus on the local distortion area strongly related to the internal blockage in the spatial dimension, and is used to identify the projection of powder residues in the TPMS background structure.

[0070]

[0071] ​In the embodiments of the present application, the present application is mainly aimed at the internal blockage detection of the light-transmitting TPMS gradient structure. First, a stable and reliable transmission imaging system is constructed by combining a carefully designed uniform backlight source with an industrial camera, which can effectively capture the light field scattering signals caused by internal powder residues. In the image processing link, the global statistical characteristics of the image are analyzed to construct an optical feature vector that can accurately represent the internal blockage condition. A specially trained deep learning segmentation model is introduced, which can more accurately distinguish between normal structure shadows and real blockage areas, significantly improving the accuracy and reliability of the detection. In addition, the present application creatively uses the projection area as a quantitative indicator, which not only ensures the detection efficiency, but also ensures the high relevance of the evaluation results to the actual performance of the component. In terms of cost control, only conventional optical elements are required; in terms of detection efficiency, the evaluation of a single component can be quickly completed, adapting to the rhythm of the industrial production line. It should be noted that the entire system is easy to integrate into the existing production process, providing a practical technical path for realizing the full number of online detection of TPMS structure.

[0072] As shown in Figure 2 , as a preferred embodiment of the present application, the step of collecting a two-dimensional transmission light field image of the TPMS structure with light transmission, specifically includes:

[0073] S101, acquiring a position detection image capable of representing the position of the TPMS structure by an industrial camera, and comparing it with a preset imaging area;

[0074] S102, when the TPMS structure is located in the imaging area, generating a coordination instruction and issuing it to the device corresponding to the uniform backlight source, so that it can emit parallel light of two different wavelengths in turn, including a wavelength with strong penetration to the base material and a wavelength with strong scattering to the residual powder;

[0075] S103, generating an adjustment instruction and issuing it to the industrial camera to adjust the exposure time and gain of the industrial camera, for acquiring a group of original images under different imaging parameters;

[0076] S104, image fusion and registration of the group of original images to generate the two-dimensional transmission light field image.

[0077] In this embodiment of the invention, the quality and efficiency of acquiring two-dimensional transmission light field images are ensured through four steps. First, a spatial reference is established by rapidly capturing component position images using an industrial camera and comparing them in real time with a preset ideal imaging area. This establishes a precise coordinate system for the entire detection system, ensuring that subsequent imaging will not result in blind spots due to positional deviations. Then, based on the precise positioning results, a multispectral imaging sequence is intelligently triggered. A uniform backlight source is controlled to emit two complementary parallel lights (e.g., using the near-infrared band to efficiently penetrate the substrate material to obtain structural contour information, while simultaneously using a specific visible blue light band to enhance defect signals by utilizing its strong scattering characteristics of residual powder). This dual-band collaborative illumination strategy enhances the observation capabilities of the entire system. Next, the exposure time and gain parameters of the industrial camera are adjusted to optimize the acquisition settings for the characteristics of different wavelengths of light. For example, shorter exposures can be used for wavelengths with high penetration to avoid oversaturation, while the gain can be appropriately increased for wavelengths with high scattering to improve the signal-to-noise ratio, thereby obtaining a set of high-quality original images containing different dimensions of information. Finally, advanced image fusion algorithms are used to register and synthesize image sequences acquired under multiple bands and parameters, effectively integrating the advantageous features under different imaging conditions to generate a two-dimensional transmission light field image with higher dynamic range and more complete defect information.

[0078] like Figure 3 As shown, in a preferred embodiment of the present invention, the steps of preprocessing and feature quantization of the two-dimensional transmitted light field image specifically include:

[0079] S201 performs dark field and flat field correction on the acquired two-dimensional transmitted light field image to eliminate the effects of light source inhomogeneity and camera dark noise.

[0080] S202, in the image frequency domain, the image is decomposed into low-frequency and high-frequency components by Fourier transform, and the proportion of high-frequency energy in the total energy is calculated to obtain the first characteristic quantity that can represent microscopic scattering.

[0081] S203 uses the local binary mode algorithm to analyze the texture of the image in the spatial domain and calculates the mean local contrast of the entire image to obtain a second feature quantity that can represent macroscopic scattering unevenness.

[0082] S204, the first feature quantity and the second feature quantity are concatenated and normalized with the global grayscale standard deviation of the image to obtain the comprehensive optical feature vector.

[0083] In the embodiment of the present application, firstly, dark field and flat field correction is performed to eliminate the influence of camera dark current noise and uneven light source intensity distribution from the two-dimensional transmission light field image, and ensure that the image gray scale change only reflects the light transmission characteristics of the sample itself. In frequency domain analysis, the image is decomposed into components of different spatial frequencies by Fourier transform, and the proportion of high frequency energy in the total energy is calculated. This characteristic quantity is sensitive to subtle scattering signals in the image, and can effectively capture local optical property changes caused by residual micro-powder; in spatial domain analysis, the local binary pattern algorithm is used to quantify the image texture features, and the local contrast mean value of the entire image is obtained by calculating the gray difference between each pixel point and its neighborhood pixels. This feature can represent the macroscopic scattering unevenness phenomenon caused by resin residue or mass blockage in a larger area. Finally, the frequency domain high frequency energy proportion, the spatial domain local contrast mean value, and the global gray scale standard deviation reflecting the overall light transmission rate change are standardized and spliced to form a comprehensive optical feature vector, which provides feature information covering multiple scales from micro to macro for subsequent deep learning models.

[0084] As shown in Figure 4 As a preferred embodiment of the present application, the step of calculating the total projection area of the internal blockage area of the TPMS structural member according to the segmentation map and taking it as a quantitative index of the internal blockage degree, and comparing the index with a preset qualified threshold, specifically includes:

[0085] S401, performing connected component analysis on the obtained segmentation map to identify and label each independent suspected blockage area;

[0086] S402, based on the internal channel space distribution map of the three-dimensional structure, different weight coefficients are given to each blockage area under the current two-dimensional projection viewing angle, and the internal channel space distribution map is obtained through the digital model of the TPMS structural member;

[0087] S403, calculating the weighted projection area sum of all suspected blockage areas to obtain the quantitative index, and the weighted projection area sum is the pixel area of each area multiplied by the weight coefficient of its position and then accumulated;

[0088] S404, comparing the weighted projection area sum with a preset qualified threshold, and automatically generating a sorting instruction and a traceable quality report containing the component ID, defect image and quantitative index according to the comparison result.

[0089] In the embodiment of the present application, by introducing a quantitative evaluation system based on the importance of three-dimensional structure function, the influence degree of internal blockage is accurately evaluated. First, the segmentation atlas output by the deep learning model is morphologically optimized and connected domain analyzed, all independent suspected blockage regions are identified and labeled through pixel neighborhood traversal algorithm, and a complete database containing the outline boundary, centroid coordinates and pixel statistical features of each region is established. Then, based on the original three-dimensional digital model of the TPMS component, the internal channel function distribution map under the current imaging view is obtained through spatial projection transformation, and different weight coefficients are given according to the functional importance of different regions in the overall structure, for example, the core channel region which bears the main fluid conveying function is given the highest weight coefficient 1.0, the secondary branch channel and connection region is given the medium weight coefficient 0.6-0.8, and the non-functional structural support region is given zero weight, so as to ensure that the evaluation index is closely related to the actual use performance. In the quantitative calculation stage, the system calculates the projection area of each independent blockage region, multiplies it by the weight coefficient at the corresponding position, and finally adds up the weighted values of all regions to obtain the comprehensive quantitative index. This calculation process not only considers the geometric size of the blockage, but also takes into account the functional importance of its spatial position. Finally, the weighted index is automatically compared with the qualified threshold value determined through a large number of process tests and fluid simulation, and when the index exceeds the threshold value, the TPMS component is unqualified, and a complete quality report containing the unique identification of the component, multi-angle defect images, detailed quantitative index and spatial distribution information is automatically generated, and the long-term storage and traceability analysis of quality data are realized through the database interface, providing data support for process optimization and quality control.

[0090] As Figure 5 shown, as a preferred embodiment of the present application, the online defect detection method of the TPMS gradient structure also includes:

[0091] S501, when the quantitative index is in a preset critical interval based on the qualified threshold value, a rotation control instruction for controlling the rotating bearing device is generated, which is used to rotate the TPMS structure by two different angles, and the TPMS structure is located on the rotating bearing device;

[0092] S502, a set of multi-view projection images are obtained by collecting two-dimensional transmission light field images at each angle through an industrial camera;

[0093] S503, the multi-view projection images are data fused, a three-dimensional spatial distribution model of the internal blockage of the TPMS structure is generated through a three-dimensional reconstruction algorithm, and the total volume of the blockage is calculated based on the model;

[0094] S504, comparing the total volume with a preset second qualified threshold, when the total volume exceeds the second qualified threshold, generating prompt information representing that the TPMS structure is unqualified.

[0095] In the embodiment of the present application, on the basis of two-dimensional detection, a three-dimensional review mechanism based on critical state triggering is innovatively introduced, and a hierarchical and progressive intelligent detection system is constructed. When the quantitative index is in a preset critical interval near the qualified threshold, the system automatically starts the accurate review process. First, accurate rotation control instructions are generated to drive the rotating bearing equipment to position the TPMS structure to two different imaging angles in turn (the selection of these two angles is strictly calculated, and they are usually different by 60° or 90° to ensure that complementary projection information is obtained). Then, two-dimensional transmission light field images under each angle are collected by the same industrial camera to obtain a set of multi-view projection images containing spatial structure information. The set of multi-view projection images is sent to a special three-dimensional reconstruction algorithm for processing. Through the principle of stereo vision and spatial geometric transformation, two-dimensional projection information is converted into a three-dimensional spatial distribution model, and the spatial distribution form of the blockage in the TPMS structure is accurately reconstructed. Based on this model, the accurate total volume of the blockage is calculated by voxel counting and spatial interpolation algorithm, which is a more accurate indicator of the severity of the blockage than the two-dimensional projection area. Finally, the system compares the calculated total volume with the second qualified threshold determined by fluid dynamics simulation, which takes into account the functional requirements and fluid performance indicators of the structure. When the total volume exceeds the threshold, an unqualified judgment result is generated. The above method not only ensures the detection efficiency under normal circumstances, but also provides a more scientific decision basis for critical states, forming a perfect detection closed loop.

[0096] As shown in Figure 6 The embodiment of the present application also provides an online defect detection system for a TPMS gradient structure, which comprises:

[0097] An image acquisition module 100 is configured to acquire a two-dimensional transmission light field image of a TPMS structure with light transmission. The two-dimensional transmission light field image is obtained by cooperation of a uniform backlight and an industrial camera. The uniform backlight can emit parallel light of a specific wavelength to penetrate the TPMS structure.

[0098] An image processing module 200 is configured to pre-process and feature quantify the two-dimensional transmission light field image, and calculate the overall statistical characteristics of the image to obtain a comprehensive optical feature vector representing the internal scattering intensity.

[0099] A model processing module 300 is configured to input the comprehensive optical feature vector into a pre-trained deep learning segmentation model to obtain a segmentation atlas with the same size as the two-dimensional transmission light field image.

[0100] The quantification calculation module 400 is configured to calculate the total projection area of the blocked area inside the TPMS structure according to the segmentation map, and take the total projection area as a quantitative index of the internal blocking degree, and compare the index with a preset qualified threshold.

[0101] In the embodiment of the present application, the image acquisition module 100 uses a uniform backlight and an industrial camera to form a transmission imaging system to obtain a two-dimensional transmission light field image of the TPMS structure; the image processing module 200 performs dark field and flat field correction on the collected image and quantifies the features to extract the frequency domain high-frequency energy and the spatial texture features to construct a comprehensive optical feature vector; the model processing module 300 inputs the feature vector into a pre-trained deep learning segmentation model to output a pixel-level segmentation map accurately identifying the blocked area; and the quantification calculation module 400 performs connected component analysis and weighted calculation on the segmentation map to obtain a quantitative index of the internal blocking and compare the index with a qualified threshold to realize automatic quality judgment and sorting decision.

[0102] As shown in Figure 7 , as a preferred embodiment of the present application, the image acquisition module 100 comprises:

[0103] The position detection unit 101 is configured to acquire a position detection image capable of representing the position of the TPMS structure through the industrial camera, and compare the position detection image with a preset imaging area;

[0104] The light source control unit 102 is configured to generate a coordination instruction and send the coordination instruction to the device corresponding to the uniform backlight when the TPMS structure is located in the imaging area, so that the device can emit parallel light of two different wavelengths in turn, one of which has strong penetration to the base material and the other of which has strong scattering to the residual powder;

[0105] The image acquisition unit 103 is configured to generate an adjustment instruction and send the adjustment instruction to the industrial camera to adjust the exposure time and gain of the industrial camera, so as to acquire a group of original images under different imaging parameters;

[0106] The image fusion unit 104 is configured to perform image fusion and registration on the group of original images to generate the two-dimensional transmission light field image.

[0107] As shown in Figure 8 , as a preferred embodiment of the present application, the image processing module 200 comprises:

[0108] The image correction unit 201 is configured to perform dark field and flat field correction on the acquired two-dimensional transmission light field image, so as to eliminate the influence of light source unevenness and camera dark noise;

[0109] The first feature acquisition unit 202 is configured to decompose an image into low-frequency and high-frequency components in a frequency domain of the image by Fourier transform, and calculate a proportion of high-frequency energy in total energy to obtain a first feature quantity capable of representing microscopic scattering;

[0110] The second feature acquisition unit 203 is configured to analyze a texture of the image in a spatial domain of the image by using a local binary pattern algorithm, and calculate a local contrast mean value of the whole image to obtain a second feature quantity capable of representing macroscopic scattering unevenness.

[0111] The data normalization unit 204 is configured to splice and normalize the first feature quantity and the second feature quantity with a global gray scale standard deviation of the image to obtain the comprehensive optical feature vector.

[0112] As shown in the figure, as a preferred embodiment of the present application, the online defect detection system of the TPMS gradient structure further comprises a precision review module 500, and the precision review module 500 comprises: Figure 9 The index judgment unit 501 is configured to generate a rotation control instruction for controlling a rotating bearing device when the quantitative index is in a preset critical interval based on a qualified threshold, so as to rotate the TPMS structure by two different angles, and the TPMS structure is located on the rotating bearing device.

[0113] The multi-view acquisition unit 502 is configured to acquire a two-dimensional transmission light field image at each angle by an industrial camera, and arrange a plurality of multi-view projection images.

[0114] The multi-view fusion unit 503 is configured to perform data fusion on the multi-view projection images, generate a three-dimensional spatial distribution model of the blockage in the TPMS structure by a three-dimensional reconstruction algorithm, and calculate a total volume of the blockage based on the model.

[0115] The comparison unit 504 compares the total volume with a preset second qualified threshold, and generates prompt information indicating that the TPMS structure is unqualified when the total volume exceeds the second qualified threshold.

[0116] The above only describes the preferred embodiments of the present application in detail, and does not limit the present application, and any modification, equivalent replacement and improvement within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0117]

[0118] ​It should be understood that, even though a series of steps are shown in the flowcharts of the embodiments of the present application, the steps are not necessarily performed in the order shown by the arrows. Unless otherwise specified herein, the steps are not necessarily performed in the order shown by the arrows. The steps can be performed in other orders. Moreover, at least some of the steps in the embodiments can include a plurality of sub-steps or a plurality of stages, which are not necessarily performed at the same time, but can be performed at different times, and the order of the sub-steps or stages is not necessarily sequential, but can be performed alternately or alternately with at least some of the other steps or sub-steps or stages of other steps.

[0119] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0120] Other embodiments of the present disclosure will be apparent to those skilled in the art with the disclosure herein. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or conventional techniques in the art not disclosed by the present disclosure. The specification and embodiments are only considered as exemplary, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. An online defect detection method for TPMS gradient structural components, characterized in that, The method includes the following steps: A two-dimensional transmitted light field image of a TPMS structural component with light transmittance is acquired. The two-dimensional transmitted light field image is obtained by the coordinated action of a uniform backlight source and an industrial camera. The uniform backlight source can emit parallel light of a specific wavelength that penetrates the TPMS structural component. The two-dimensional transmitted light field image is preprocessed and its features are quantized. The overall statistical features of the image are calculated to obtain a comprehensive optical feature vector that can characterize the internal scattering intensity. By inputting the comprehensive optical feature vector into the pre-trained deep learning segmentation model, a segmentation map of the same size as the two-dimensional transmitted light field image is obtained. The total projected area of ​​the internal blockage region of the TPMS structural component is calculated based on the segmentation map, and it is used as a quantitative indicator of the degree of internal blockage. This indicator is then compared with a preset qualified threshold. The steps of preprocessing and feature quantization of the two-dimensional transmitted light field image specifically include: Dark field and flat field corrections are performed on the acquired two-dimensional transmitted light field image to eliminate the effects of light source inhomogeneity and camera dark noise. In the image frequency domain, the image is decomposed into low-frequency and high-frequency components by Fourier transform, and the proportion of high-frequency energy in the total energy is calculated to obtain the first characteristic quantity that can represent microscopic scattering. The texture of an image is analyzed in the spatial domain using the local binary mode algorithm, and the mean local contrast of the entire image is calculated to obtain a second feature quantity that can represent macroscopic scattering unevenness. The first feature value and the second feature value are concatenated and normalized with the global grayscale standard deviation of the image to obtain the comprehensive optical feature vector.

2. The online defect detection method for TPMS gradient structural components according to claim 1, characterized in that, The step of acquiring a two-dimensional transmitted light field image of a TPMS structural component with light transmittance specifically includes: The position detection image that can characterize the position of the TPMS structural component is acquired by an industrial camera and compared with a preset imaging area; When the TPMS structure is located within the imaging area, a coordination command is generated and sent to the device corresponding to the uniform backlight source, enabling it to emit two different wavelengths of parallel light in sequence, including a wavelength that has strong penetration into the substrate material and a wavelength that has strong scattering of residual powder. The system generates adjustment commands and sends them to the industrial camera to adjust the exposure time and gain of the industrial camera in order to acquire a set of original images with different imaging parameters. The two-dimensional transmission light field image is generated by image fusion and registration of the set of original images.

3. The online defect detection method for TPMS gradient structural components according to claim 1, characterized in that, The deep learning segmentation model is a variant of U-Net that introduces a channel-space dual attention mechanism. This mechanism enables the model to calibrate the weights of different scattering features in the channel dimension and focus on local distortion regions strongly correlated with internal blockage in the spatial dimension, in order to identify the projection of powder residue in the TPMS background structure.

4. The online defect detection method for TPMS gradient structural components according to claim 1, characterized in that, The step of calculating the total projected area of ​​the internal blockage region of the TPMS structural component based on the segmentation map, using it as a quantitative indicator of the degree of internal blockage, and comparing this indicator with a preset qualified threshold specifically includes: Connectivity analysis was performed on the obtained segmentation map to identify and mark each individual suspected blockage region; Based on the three-dimensional structure of the internal channel spatial distribution map, different weight coefficients are assigned to each blockage area under the current two-dimensional projection view. The internal channel spatial distribution map is obtained through the digital model of the TPMS structural component. The quantitative index is obtained by calculating the weighted projected area of ​​all suspected congestion areas. The weighted projected area is the sum of the pixel area of ​​each area multiplied by the weight coefficient of its location. The weighted total projected area is compared with a preset qualified threshold, and sorting instructions and a traceable quality report containing component ID, defect image and quantitative indicators are automatically generated based on the comparison results.

5. The online defect detection method for TPMS gradient structural components according to claim 1, characterized in that, The method further includes: When the quantitative index is within a preset critical range based on the qualified threshold, a rotation control command is generated to control the rotating bearing device, which causes the TPMS structural component to rotate at two different angles. The TPMS structural component is located on the rotating bearing device. Two-dimensional transmitted light field images at each angle are captured by an industrial camera and then processed to obtain a set of multi-view projection images. Multi-view projection images are fused together, and a three-dimensional spatial distribution model of the blockage inside the TPMS structure is generated through a three-dimensional reconstruction algorithm. The total volume of the blockage is then calculated based on this model. The total volume is compared with a preset second qualified threshold. If the total volume exceeds the second qualified threshold, a prompt message indicating that the TPMS structural component is unqualified is generated.

6. An online defect detection system for TPMS gradient structural components, characterized in that, The system includes: The image acquisition module is used to acquire a two-dimensional transmitted light field image of a TPMS structural component with light transmittance. The two-dimensional transmitted light field image is obtained by the coordinated action of a uniform backlight source and an industrial camera. The uniform backlight source can emit parallel light of a specific wavelength that penetrates the TPMS structural component. The image processing module is used to preprocess and feature quantize the two-dimensional transmitted light field image, and calculate the overall statistical features of the image to obtain a comprehensive optical feature vector that can characterize the internal scattering intensity. The model processing module is used to input the comprehensive optical feature vector into the pre-trained deep learning segmentation model to obtain a segmentation map of the same size as the two-dimensional transmitted light field image. The quantitative calculation module is used to calculate the total projected area of ​​the internal blockage area of ​​the TPMS structural component based on the segmentation map, and use it as a quantitative indicator of the degree of internal blockage. The indicator is then compared with a preset qualified threshold. The image processing module includes: The image correction unit is used to perform dark field and flat field correction on the acquired two-dimensional transmitted light field image to eliminate the effects of light source inhomogeneity and camera dark noise. The first feature acquisition unit is used to decompose the image into low-frequency and high-frequency components in the image frequency domain by Fourier transform, and calculate the proportion of high-frequency energy in the total energy to obtain the first feature quantity that can represent microscopic scattering. The second feature acquisition unit is used to analyze the texture of the image in the image spatial domain using the local binary mode algorithm, and calculate the mean local contrast of the entire image to obtain a second feature quantity that can represent macroscopic scattering unevenness. The data normalization unit is used to concatenate and normalize the first feature quantity and the second feature quantity with the global grayscale standard deviation of the image to obtain the comprehensive optical feature vector.

7. The online defect detection system for TPMS gradient structural components according to claim 6, characterized in that, The image acquisition module includes: The position detection unit is used to acquire position detection images that can characterize the position of TPMS structural components through an industrial camera and compare them with a preset imaging area; The light source control unit is used to generate a coordination command and send it to the device corresponding to the uniform backlight source when the TPMS structure is located in the imaging area, so that it can emit two different wavelengths of parallel light in sequence, including a wavelength with strong penetration to the substrate material and a wavelength with strong scattering of residual powder. The image acquisition unit is used to generate adjustment instructions and send them to the industrial camera to adjust the exposure time and gain of the industrial camera, and to acquire a set of original images under different imaging parameters. The image fusion unit is used to perform image fusion and registration on the set of original images to generate the two-dimensional transmission light field image.

8. The online defect detection system for TPMS gradient structural components according to claim 6, characterized in that, The system also includes a precise verification module, which includes: The indicator judgment unit is used to generate a rotation control command for controlling the rotating bearing device when the quantitative indicator is within a preset critical range based on the qualified threshold. The command is used to rotate the TPMS structural component at two different angles. The TPMS structural component is located on the rotating bearing device. The multi-view acquisition unit is used to acquire two-dimensional transmitted light field images from each angle using an industrial camera, and then process them to obtain a set of multi-view projection images. The multi-view fusion unit is used to fuse multi-view projected images, generate a three-dimensional spatial distribution model of the blockage inside the TPMS structure through a three-dimensional reconstruction algorithm, and calculate the total volume of the blockage based on this model. The comparison unit compares the total volume with a preset second qualified threshold. If the total volume exceeds the second qualified threshold, a prompt message indicating that the TPMS structural component is unqualified is generated.

Images