A 3D Printing Mesh Structure Multi-Domain Feature Defect Detection Method Based on CT Images

Abstract

The present invention provides a method for detecting multi-domain feature defects of a 3D printed grid structure based on CT images. The 3D printed grid model is converted into RAW data through computer tomography and the 3D printed grid structure is extracted. The geometric features of the 3D printed grid structure in the spatial domain are extracted through spatial domain feature analysis. The distribution of the repeating structure is determined through frequency domain analysis and the change curve of the characteristic frequency domain is drawn. The repeating unit is determined by combining the information in the spatial domain and the change curve of the characteristic frequency domain. The features of the repeating unit are extracted based on the fast point feature histogram feature and registration is performed. The features of the repeating unit in the point cloud after registration are traversed and compared to obtain the result and displayed in the form of a heat map. Visualization enables users to intuitively and comprehensively observe the differences in the model, which helps to quickly locate and analyze potential printing defects, achieving the beneficial effects of high detection accuracy and high efficiency.

Description

Technical Field

[0001] The present invention relates to the field of 3D printing technology, and in particular to a method for detecting multi-domain feature defects of 3D printing grid structures based on CT images. Background Art

[0002] 3D printing is a technology that generates three-dimensional entities by continuously stacking physical layers and adding materials layer by layer based on three-dimensional CAD model data. Simply put, it uses a computer to cut the 3D model of the formed part into a series of "thin slices" of a certain thickness, and the 3D printing device manufactures each "thin slice" from bottom to top, and finally stacks and forms a three-dimensional solid part.

[0003] In the current manufacturing field, metal 3D printing technology has become an innovative manufacturing means, which can shorten the processing cycle, reduce production costs, and the forming is not limited by complexity. However, at present, 3D printing grid models are composed of periodically arranged repetitive structures, with complex interiors and difficult to accurately locate and identify surface defects. Problems such as printing omission and dimensional error often occur during the 3D printing process, seriously reducing the structural performance and even leading to collapse. Therefore, identifying these defects is an important part of structural quality inspection. However, existing detection methods mostly rely on two-dimensional image analysis, not only can only identify the surface part and are prone to misjudgment, but also need to take a large number of images from different angles, resulting in high costs and low efficiency. Summary of the Invention

[0004] The present invention provides a method for detecting multi-domain feature defects of 3D printing grid structures based on CT images, so as to solve the defects in the prior art that rely on two-dimensional image analysis, not only can only identify the surface part and are prone to misjudgment, but also need to take a large number of images from different angles, resulting in high costs and low efficiency.

[0005] The present invention provides a method for detecting multi-domain feature defects of 3D printing grid structures based on CT images, including:

[0006] Converting the 3D printing grid model into data in the original data format through computer tomography, and extracting the 3D printing grid structure by using a voxel histogram;

[0007] Extracting the geometric features of the 3D printing grid structure in the spatial domain through spatial domain feature analysis, and the spatial domain feature analysis includes obtaining repetitive structures by statistically analyzing the standard deviation of the three-dimensional point cloud model dimension of the 3D printing grid structure;

[0008] Determining the distribution of the repetitive structures through frequency domain analysis and drawing a change curve of the characteristic frequency domain, and the frequency domain analysis includes converting the information in the spatial domain into information in the frequency domain through fast Fourier transform;

[0009] Determine the repeating unit by combining the information in the spatial domain and the change curve in the characteristic frequency domain, identify and automatically extract the repeating unit at the end of the period, where the end of the period is at the peaks and valleys;

[0010] Based on the fast point feature histogram feature, extract the features of the repeating unit through a feature matching algorithm and find the corresponding feature point pairs in the point cloud to obtain an initial transformation matrix, which is used for rough registration. Obtain a second transformation matrix from the result after the rough registration, and use the second transformation matrix as the input of the ICP algorithm and perform ICP iteration to register the point cloud containing the repeating unit;

[0011] Obtain the measurement result by traversing and comparing the features of the repeating unit in the point cloud after registration;

[0012] Analyze the defect according to the measurement result and display the analysis result in the form of a heat map.

[0013] According to a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the present invention, the extraction of the 3D printing grid structure using the voxel histogram includes: segmenting the data of the 3D printing grid model by selecting the voxel threshold in the voxel histogram and screening out the voxel points representing the 3D printing grid structure to obtain the 3D printing grid structure.

[0014] According to a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the present invention, the repeating structure is data points distributed centrally or regularly, and the standard deviation is used to identify the periodicity of the planar features of the 3D printing grid structure. Among them, the calculation of the standard deviation of the three-dimensional point cloud model dimension of the 3D printing grid structure specifically includes:

[0015] Obtain three-dimensional point cloud data including x coordinates, y coordinates, and z coordinates;

[0016] Group the point cloud data by x coordinate value, calculate the standard deviation of the corresponding y and z coordinate values for each x coordinate value to obtain the standard deviation change curves of the y plane and the z plane in the x-axis direction;

[0017] Calculate the geometric mean of the standard deviation of the y plane and the standard deviation of the z plane and draw a curve graph changing with the x coordinate value;

[0018] Calculate the geometric mean of the standard deviation of the x plane and the standard deviation of the z plane and draw a curve graph changing with the y coordinate value;

[0019] Calculate the geometric mean of the standard deviation of the x plane and the standard deviation of the y plane and draw a curve graph changing with the z coordinate value.

[0020] A 3D printing grid structure multi-domain feature defect detection method based on CT images according to the present invention determines the distribution of the repeating structure through frequency domain analysis and draws a change curve of the feature frequency domain. The frequency domain analysis includes converting the information in the spatial domain into information in the frequency domain through fast Fourier transform, specifically including:

[0021] Fix the x coordinate value, and perform feature frequency domain processing on the y coordinate value and z coordinate value corresponding to the x coordinate value through the fast Fourier transform;

[0022] Separate the feature frequency domains of the y plane and the z plane into feature frequency domain curves based on the x coordinate value;

[0023] Perform geometric averaging on the feature frequency domain curves of the y plane and the z plane and draw a feature frequency domain change curve graph that changes with the x coordinate value;

[0024] Perform geometric averaging on the feature frequency domain curves of the x plane and the z plane and draw a feature frequency domain change curve graph that changes with the y coordinate value;

[0025] Perform geometric averaging on the feature frequency domain curves of the x plane and the y plane and draw a feature frequency domain change curve graph that changes with the z coordinate value.

[0026] A 3D printing grid structure multi-domain feature defect detection method based on CT images according to the present invention, where the x coordinate value is fixed, and the y coordinate value and z coordinate value corresponding to the x coordinate value are subjected to feature frequency domain processing through the fast Fourier transform, including:

[0027] Group the data of the 3D printing grid structure in the spatial domain according to the x-axis value. For the fixed x coordinate value, extract the corresponding y and z coordinate values to form a two-dimensional data group. After transformation through the fft function in the numpy library of Python, a complex number array is obtained. Each element in the complex number array corresponds to a frequency component, and the position serial number of the element in the complex number array represents the frequency index. The relationship between the frequency index and the corresponding actual frequency value is , where f is the actual frequency, k is the frequency index, N is the data length, and fs is the sampling frequency.

[0028] A 3D printing grid structure multi-domain feature defect detection method based on CT images according to the present invention determines the wave peaks and wave valleys by setting a discrete data sequence. The wave peak is the local maximum value in the discrete data sequence. By and determine the local maximum value, where i is the data point in the discrete data sequence, is the local maximum value, and after filtering out the noise points by the median filtering method, the wave peak is obtained. The wave trough is the local minimum value in the discrete data sequence, and the local minimum value is determined by and where m is the data point in the discrete data sequence, is the local minimum value, and after filtering out the noise points by the median filtering method, the wave trough is obtained.

[0029] According to a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the present invention, constructing the fast point feature histogram includes: performing weighted processing on the simple point feature histogram of the neighborhood points relative to the selected point, and then combining the simple point feature histogram of the selected point and the weighted simple point feature histogram of the neighborhood points. Specifically, the formula is used:

[0030]

[0031] In the formula: is the selected point, is the fast point feature histogram of the selected point, is the simple point feature histogram of the selected point, is the simple point feature histogram of the neighborhood point, is the selected point and the neighborhood point The weight between them, k represents the number of neighborhood points relative to the selected point.

[0032] According to a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the present invention, by traversing and comparing the features of the repeating units in the point cloud after registration and obtaining the measurement results, including:

[0033] Traversing and comparing the points of the repeating unit with the points at the same position of the adjacent repeating unit to obtain the Euclidean distance and obtain the average value. The average value is used as the difference value of the points of the repeating unit, and the points of the repeating unit are replaced and the operation is repeated.

[0034] According to a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the present invention, the specific method for obtaining the average value is:

[0035]

[0036] where x is the point of the repeating unit, is the point cloud of the repeating unit numbered i, q is The point in, and N is the total number of repeating units.

[0037] A 3D printing grid structure multi-domain feature defect detection method based on CT images according to the present invention, analyzing defects based on the measurement results and displaying the analysis results in the form of a heat map, includes:

[0038] Mapping the index value of the defect to the color range of the heat map;

[0039] Marking the coordinate axes, scales, and color legends on the heat map;

[0040] Adding an interaction function, the interaction function includes displaying the information of the defect when the mouse hovers, and being able to zoom in or out the heat map.

[0041] A 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the present invention converts a 3D printing grid model into data in the original data format through computed tomography, and extracts the 3D printing grid structure by using a voxel histogram. The geometric features of the 3D printing grid structure in the spatial domain are extracted through spatial domain feature analysis. The spatial domain feature analysis includes obtaining repetitive structures by statistically analyzing the standard deviation of the dimensions of the three-dimensional point cloud model. In addition, the distribution of the repetitive structures is determined through frequency domain analysis and the change curve of the characteristic frequency domain is drawn. The frequency domain analysis includes converting the information in the spatial domain into information in the frequency domain through fast Fourier transform. The repetitive unit is determined by combining the information in the spatial domain and the change curve of the characteristic frequency domain. The repetitive unit is identified and automatically extracted at the end of the period. The end of the period is at the peaks and valleys. Based on the features of the fast point feature histogram, the features of the repetitive unit are extracted through a feature matching algorithm and the corresponding feature point pairs in the point cloud are found to obtain an initial transformation matrix. The initial transformation matrix is used for rough registration. A second transformation matrix is obtained through the result after the rough registration. The second transformation matrix is used as the input of the ICP algorithm and the ICP iteration is performed to register the point cloud containing the repetitive unit. Finally, the features of the repetitive units in the point cloud after registration are traversed and compared to obtain the measurement results. Defects are analyzed based on the measurement results and the analysis results are displayed in the form of a heat map. The visualization of the defect results enables users to intuitively observe the differences in the model, helps to quickly locate and analyze potential printing defects, and has the beneficial effects of high detection accuracy and high efficiency. Description of the Drawings

[0042] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1It is a flowchart of a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by an embodiment of the present invention;

[0044] Figure 2 It is a flowchart of the statistical standard deviation of a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by an embodiment of the present invention;

[0045] Figure 3 It is a flowchart of the frequency domain analysis of a 3D printing grid structure multi-domain feature defect detection method based on CT images provided by an embodiment of the present invention. Detailed implementation manners

[0046] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] The present invention provides a 3D printing grid structure multi-domain feature defect detection method based on CT images, including: converting a 3D printing grid model into data in an original data format through computed tomography, and extracting the 3D printing grid structure by using a voxel histogram; extracting geometric features of the 3D printing grid structure in the spatial domain through spatial domain feature analysis, where the spatial domain feature analysis includes obtaining repeated structures by statistically calculating the standard deviation of the dimensions of a three-dimensional point cloud model, the repeated structures being data points distributed in a concentrated or regular manner, and the standard deviation being used to identify the periodicity of planar features; determining the distribution of the repeated structures through frequency domain analysis and plotting a change curve of the characteristic frequency domain, where the frequency domain analysis includes converting the information in the spatial domain into information in the frequency domain through fast Fourier transform; determining repeated units by combining the information in the spatial domain and the change curve of the characteristic frequency domain, identifying and automatically extracting the repeated units at the end of the period, where the end of the period is at the peaks and valleys; based on the fast point feature histogram feature, extracting the features of the repeated units through a feature matching algorithm and finding corresponding feature point pairs in the point cloud to obtain an initial transformation matrix, the initial transformation matrix being used for rough registration, obtaining a second transformation matrix through the result of the rough registration, taking the second transformation matrix as the input of the ICP algorithm and performing ICP iteration to register the point cloud containing the repeated units; traversing and comparing the features of the repeated units in the point cloud after registration and obtaining a measurement result; analyzing defects based on the measurement result and displaying the analysis result in the form of a heat map.

[0048] Among them, the full name of CT is "Computed Tomography", that is, computed tomography. CT images are computed tomography images. In addition, the extraction of the 3D printing grid structure using the voxel histogram includes: performing data segmentation on the 3D printing grid model by selecting the voxel threshold in the voxel histogram to screen out the voxel points representing the 3D printing grid structure, and obtaining the 3D printing grid structure.

[0049] Among them, the statistical standard deviation of the three-dimensional point cloud model dimension of the 3D printing grid structure specifically includes:

[0050] Obtain three-dimensional point cloud data including x coordinate, y coordinate, and z coordinate;

[0051] Group the point cloud data by x coordinate value, and calculate the standard deviation of the corresponding y and z coordinate values for each x coordinate value to obtain the standard deviation change curves of the y plane and z plane along the x-axis direction;

[0052] Perform geometric mean on the standard deviation of the y plane and the standard deviation of the z plane and draw a curve graph changing with the x coordinate value;

[0053] Perform geometric mean on the standard deviation of the x plane and the standard deviation of the z plane and draw a curve graph changing with the y coordinate value;

[0054] Perform geometric mean on the standard deviation of the x plane and the standard deviation of the y plane and draw a curve graph changing with the z coordinate value.

[0055] Among them, the distribution of the repeating structure is determined by frequency domain analysis and the change curve of the characteristic frequency domain is drawn. The frequency domain analysis includes converting the information in the spatial domain into the information in the frequency domain through fast Fourier transform, specifically including:

[0056] Fix the x coordinate value, and perform characteristic frequency domain processing on the corresponding y coordinate value and z coordinate value through the fast Fourier transform;

[0057] Respectively draw the characteristic frequency domain curves of the y plane and the z plane based on the x coordinate value;

[0058] Perform geometric mean on the characteristic frequency domain curve of the y plane and the characteristic frequency domain curve of the z plane and draw a characteristic frequency domain change curve graph changing with the x coordinate value;

[0059] Perform geometric mean on the characteristic frequency domain curve of the x plane and the characteristic frequency domain curve of the z plane and draw a characteristic frequency domain change curve graph changing with the y coordinate value;

[0060] Perform geometric mean on the characteristic frequency domain curve of the x plane and the characteristic frequency domain curve of the y plane and draw a characteristic frequency domain change curve graph changing with the z coordinate value.

[0061] Among them, for the fixed x - coordinate value, the fast Fourier transform performs characteristic frequency - domain processing on the y - coordinate value and z - coordinate value corresponding to the x - coordinate value, including:

[0062] Group the data of the 3D printing grid structure in the spatial domain according to the x - axis value. For the fixed x - coordinate value, extract the corresponding y and z coordinate values to form a two - dimensional data group. After transformation by the fft function in the numpy library of Python, a complex - number array is obtained. Each element in the complex - number array corresponds to a frequency component, and the position serial number of the element in the complex - number array represents the frequency index. The relationship between the frequency index and the corresponding actual frequency value is , where f is the actual frequency, k is the frequency index, N is the data length, and fs is the sampling frequency.

[0063] Among them, by setting a discrete data sequence to find the peaks and valleys. The peak is the local maximum value in the discrete data sequence, and is determined by and , where i is the data point in the discrete data sequence, is the local maximum value. After filtering out noise points by the median filtering method, the peak is obtained. The valley is opposite to the peak. The valley is the local minimum value in the discrete data sequence, and is determined by and , where m is the data point in the discrete data sequence, is the local minimum value. After filtering out noise points by the median filtering method, the valley is obtained.

[0064] Among them, constructing the fast point - feature histogram includes: weighting the simple point - feature histogram of the neighborhood points relative to the selected point, and then combining the simple point - feature histogram of the selected point and the weighted simple point - feature histogram of the neighborhood points, using the formula:

[0065]

[0066] In the formula: is the selected point, is the fast point - feature histogram of the selected point, is the simple point - feature histogram of the selected point, is the simple point - feature histogram of the neighborhood points, is the selected point and the neighborhood point the weight between them, and k represents the number of neighborhood points relative to the selected point.

[0067] Among them, by traversing and comparing the features of the repeating units in the point cloud after registration and obtaining measurement results, it includes: traversing and comparing the points of the repeating unit with the points at the same position of the adjacent repeating unit to obtain the Euclidean distance and obtaining the average value, using the average value as the difference value of the points of the repeating unit, and replacing the points of the repeating unit to repeat the operation.

[0068] Among them, the specific method for obtaining the average value is:

[0069]

[0070] Among them, x is the point of the repeating unit, is the point cloud of the repeating unit numbered i, q is the point in the point cloud, and N is the total number of the repeating units.

[0071] Among them, analyzing the defects according to the measurement results and displaying the analysis results in the form of a heat map includes: mapping the index value of the defect to the color range of the heat map; marking the coordinate axis, scale, and color legend on the heat map; adding an interaction function, and the interaction function includes mouse hovering to display the information of the defect and being able to zoom in or out the heat map.

[0072] In this embodiment, please refer to Figure 1 、 Figure 2 and Figure 3 , which are respectively the flowchart of the 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the embodiment of the present invention, the flowchart of the statistical standard deviation of the 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the embodiment of the present invention, and the flowchart of the frequency domain analysis of the 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the embodiment of the present invention.

[0073] Specifically, in this embodiment, as Figure 1 shown, the 3D printing grid structure multi-domain feature defect detection method based on CT images provided by the embodiment of the present invention includes:

[0074] S1: Convert the 3D printing grid model into data in the original data format through computer tomography, and extract the 3D printing grid structure by using the voxel histogram;

[0075] S2: Extract the geometric features of the 3D printing grid structure in the spatial domain through spatial domain feature analysis;

[0076] S3: Determine the distribution of the repeating structure through frequency domain analysis and draw the change curve of the characteristic frequency domain, and the frequency domain analysis includes converting the information in the spatial domain into the information in the frequency domain through fast Fourier transform;

[0077] S4: Determine the repeating unit by combining the information in the spatial domain and the change curve in the characteristic frequency domain, identify and automatically extract the repeating unit at the end of the period;

[0078] S5: Based on the fast point feature histogram feature, extract the features of the repeating unit through the feature matching algorithm and find the corresponding feature point pairs in the point cloud to obtain the initial transformation matrix, and register the point cloud containing the repeating unit through the ICP algorithm;

[0079] S6: Obtain the measurement result by traversing and comparing the features of the repeating unit in the point cloud after registration;

[0080] S7: Analyze the defect according to the measurement result and display the analysis result in the form of a heat map.

[0081] In this embodiment, the above steps are specifically described. Among them, S1: Convert the 3D printing grid model into data in the original data format through computer tomography, and extract the 3D printing grid structure by using the voxel histogram. In this embodiment, the extraction of the 3D printing grid structure by using the voxel histogram includes: performing data segmentation on the 3D printing grid model by selecting the voxel threshold in the voxel histogram to filter out the voxel points representing the 3D printing grid structure, and obtaining the 3D printing grid structure.

[0082] In this embodiment, specifically, S2: Extract the geometric features of the 3D printing grid structure in the spatial domain through spatial domain feature analysis. Among them, the spatial domain feature analysis includes obtaining the repeating structure by statistically calculating the standard deviation of the dimensions of the three-dimensional point cloud model. The repeating structure is data points distributed concentratedly or regularly. The standard deviation is used to identify the periodicity of the plane feature. In this embodiment, the statistical calculation of the standard deviation of the dimensions of the three-dimensional point cloud model, as Figure 2 shown, specifically includes:

[0083] S10: Obtain three-dimensional point cloud data including the x coordinate, y coordinate, and z coordinate;

[0084] S20: Group the point cloud data according to the x coordinate value, calculate the standard deviation of the corresponding y and z coordinate values for each x coordinate value to obtain the standard deviation change curves of the y plane and z plane along the x-axis direction;

[0085] S30: Perform geometric averaging on the standard deviation of the y plane and the standard deviation of the z plane and draw a curve graph changing with the x coordinate value.

[0086] In addition, the curve graph that changes with the y - coordinate value and the curve graph that changes with the z - coordinate value are similar to the above - mentioned S10~S30. Specifically, the point cloud data is grouped according to the y - coordinate value, and the standard deviations of the x - and z - coordinate values corresponding to each y - coordinate value are calculated to obtain the standard deviation change curves of the x - plane and the z - plane along the y - axis direction. By taking the geometric mean of the standard deviation of the x - plane and the standard deviation of the z - plane and plotting the curve graph that changes with the y - coordinate value; the point cloud data is grouped according to the z - coordinate value, and the standard deviations of the x - and y - coordinate values corresponding to each z - coordinate value are calculated to obtain the standard deviation change curves of the x - plane and the y - plane along the z - axis direction. By taking the geometric mean of the standard deviation of the x - plane and the standard deviation of the y - plane and plotting the curve graph that changes with the z - coordinate value.

[0087] In this embodiment, specifically, S3: Determine the distribution of the repetitive structure through frequency - domain analysis and plot the change curve of the characteristic frequency domain. The frequency - domain analysis includes converting the information in the spatial domain into the information in the frequency domain through fast Fourier transform, as Figure 3 shown, specifically including:

[0088] S100: Fix the x - coordinate value, and perform characteristic frequency - domain processing on the y - coordinate value and the z - coordinate value corresponding to the x - coordinate value through the fast Fourier transform;

[0089] S200: Plot the characteristic frequency domains of the y - plane and the z - plane respectively as characteristic frequency - domain curves based on the x - coordinate value;

[0090] S300: Perform geometric mean on the characteristic frequency - domain curve of the y - plane and the characteristic frequency - domain curve of the z - plane and plot the characteristic frequency - domain change curve graph that changes with the x - coordinate value.

[0091] In addition, the characteristic frequency - domain change curve graph that changes with the y - coordinate value and the characteristic frequency - domain change curve graph that changes with the z - coordinate value are similar to the above - mentioned S100~S300. Specifically, fix the y - coordinate value, perform characteristic frequency - domain processing on the x - coordinate and the z - coordinate corresponding to the y - coordinate value through the fast Fourier transform, plot the characteristic frequency domains of the x - plane and the z - plane respectively as characteristic frequency - domain curves based on the y - coordinate value, perform geometric mean on the characteristic frequency - domain curve of the x - plane and the characteristic frequency - domain curve of the z - plane and plot the characteristic frequency - domain change curve graph that changes with the y - coordinate value; fix the z - coordinate value, perform characteristic frequency - domain processing on the x - coordinate and the y - coordinate corresponding to the z - coordinate value through the fast Fourier transform, plot the characteristic frequency domains of the x - plane and the y - plane respectively as characteristic frequency - domain curves based on the z - coordinate value, perform geometric mean on the characteristic frequency - domain curve of the x - plane and the characteristic frequency - domain curve of the y - plane and plot the characteristic frequency - domain change curve graph that changes with the z - coordinate value.

[0092] In this embodiment, for the fixed x coordinate value, the fast Fourier transform is used to perform feature frequency domain processing on the y coordinate value and the z coordinate value corresponding to the x coordinate value, including:

[0093] Group the data of the 3D printing grid structure in the spatial domain according to the x-axis value. For the fixed x coordinate value, extract the corresponding y and z coordinate values to form a two-dimensional data group. After transformation by the fft function in the numpy library of Python, a complex number array is obtained. Each element in the complex number array corresponds to a frequency component, and the position serial number of the element in the complex number array represents the frequency index. The relationship between the frequency index and the corresponding actual frequency value is , where f is the actual frequency, k is the frequency index, N is the data length, and fs is the sampling frequency. Similarly, for the fixed y coordinate value, the fast Fourier transform is used to perform feature frequency domain processing on the x coordinate and the z coordinate corresponding to the y coordinate value, and for the fixed z coordinate value, the fast Fourier transform is used to perform feature frequency domain processing on the x coordinate and the y coordinate corresponding to the z coordinate value. Their processing methods are similar to the method of performing feature frequency domain processing on the y coordinate value and the z coordinate value corresponding to the x coordinate value by the fast Fourier transform for the fixed x coordinate value, and will not be elaborated here.

[0094] In this embodiment, S4: Combine the information in the spatial domain and the change curve in the feature frequency domain to determine the repeating unit, and identify and automatically extract the repeating unit at the end of the period. Among them, the end of the period is at the peak and trough. Specifically, find the peak and trough by setting a discrete data sequence. The peak is the local maximum value in the discrete data sequence, and is determined by and , where i is the data point in the discrete data sequence, is the local maximum value. After filtering out the noise points by the median filtering method, the peak is obtained. The trough is opposite to the peak. The trough is the local minimum value in the discrete data sequence, and is determined by and , where m is the data point in the discrete data sequence, is the local minimum value. After filtering out the noise points by the median filtering method, the trough is obtained.

[0095] In this embodiment, S5: Based on the Fast Point Feature Histogram (FPFH) features, the features of the repeating unit are extracted through a feature matching algorithm, and the corresponding feature point pairs in the point cloud are found to obtain an initial transformation matrix. The point cloud containing the repeating unit is registered by the Iterative Closest Point (ICP) algorithm. Specifically, the initial transformation matrix is used for rough registration. According to the result of the rough registration, a second transformation matrix is calculated. The second transformation matrix is used as the input of the ICP algorithm, and the ICP iteration is performed to register the point cloud containing the repeating unit. In this embodiment, the Chinese name of the ICP algorithm is the Iterative Closest Point algorithm. Constructing the Fast Point Feature Histogram includes: weighting the neighborhood points of the selected point to obtain a simple point feature histogram, and then combining the simple point feature histogram of the selected point and the weighted simple point feature histogram, using the formula:

[0096]

[0097] In the formula: is the selected point, is the Fast Point Feature Histogram of the selected point, is the simple point feature histogram of the selected point, is the selected point and the neighborhood point The weight between them, k represents the number of neighborhood points of the selected point.

[0098] In this embodiment, specifically, S6: By traversing and comparing the features of the repeating unit in the registered point cloud to obtain a measurement result, including: traversing and comparing the points of the repeating unit with the points at the same position of the adjacent repeating unit to obtain the Euclidean distance and obtain the average value. The average value is used as the difference value of the points of the repeating unit. Replace the points of the repeating unit and repeat the operation.

[0099] Among them, the specific method for obtaining the average value is:

[0100]

[0101] Among them, x is the point of the repeating unit, is the point cloud of the repeating unit numbered i, q is The points in the point cloud, N is the total number of the repeating units.

[0102] In this embodiment, specifically, S7: Analyze the defects according to the measurement result and display the analysis result in the form of a heat map, including: mapping the index value of the defect to the color range of the heat map; marking the coordinate axis, scale and color legend on the heat map; adding an interactive function, and the interactive function includes mouse hovering to display the information of the defect and being able to zoom in or out the heat map.

[0103] Specifically, in this embodiment, through the best processing implementation method, the 3D printing mesh model is converted into RAW format data through CT scanning, and the 3D printing mesh structure part is accurately extracted by using the voxel histogram. The voxel histogram of the 3D printing mesh model data can better understand the structure of the 3D printing mesh model by statistically analyzing the voxel quantity and attribute distribution. Among them, different voxel value distributions show different structures. For example, a certain part of the 3D printing mesh structure is stronger than other parts, which is reflected on the histogram as being greater than a certain voxel threshold. By selecting different voxel thresholds for three-dimensional model segmentation, the voxel points representing the 3D printing mesh structure are screened out, and finally, the incomplete structures around the 3D printing mesh model are removed, and the best 3D printing mesh model can be obtained.

[0104] The 3D printing mesh model is composed of several repeating structures arranged periodically. In this embodiment, the repeating structure is revealed by statistically analyzing the standard deviation of each dimension of the three-dimensional point cloud model. The standard deviation describes the degree to which data points deviate from the mean value. The larger the value, the more dispersed the data. After fixing the coordinate axis values, the standard deviation of each plane, such as the yz, xz, and xy planes, is calculated to analyze the point cloud distribution. The repeating structure is manifested as the concentration or regular distribution of data points within the region, and the standard deviation calculation can accurately identify the periodicity of the plane features.

[0105] In this embodiment, the spatial domain feature analysis includes the following steps: First, load the three-dimensional point cloud data containing x, y, and z coordinates. Second, group the three-dimensional point cloud data by x value, calculate the standard deviation of the corresponding y and z coordinates for each x value, and obtain the standard deviation change curve of each plane along the x-axis direction. Finally, to comprehensively measure the variability of each plane (y and z dimensions), calculate the geometric mean of the y and z standard deviations and draw a curve showing its change with the x value. Similarly, the characteristic curves of the corresponding planes in the other two dimensions (y and z) can be obtained.

[0106] Frequency domain analysis is introduced to assist in determining the distribution of the repeating structure and accurately extracting it. The fast Fourier transform (FFT) is an efficient algorithm for calculating the discrete Fourier transform (DFT). The magnitude of the Fourier transform represents the intensity of each frequency present in the function. The frequencies of periodic functions are recorded as local maxima in the magnitude of their Fourier transform. In this embodiment, first, fix each x-axis value and perform fast Fourier transform (FFT) analysis on the y and z coordinates. After ignoring the zero-frequency component, plot the main frequency amplitudes of each plane as a curve based on the x value to analyze the characteristic changes. Then, perform geometric averaging and visual analysis on the frequency domain amplitude curves of the planes (y and z value curves). Similarly for the y-axis and z-axis, characteristic frequency domain change curves of each dimension in the three-dimensional space are obtained.

[0107] In this embodiment, since the 3D printing grid material is composed of multiple repeating small structures and has obvious spatial and dimensional periodicity, the characteristic curve can be divided block by block. By combining the spatial domain and frequency domain characteristic curves, the minimum repeating unit can be determined and identified and extracted at the end of the period, usually at the peaks and valleys.

[0108] For finding the peaks and valleys, assuming there is a function or a discrete data sequence, the peak condition can be understood as a local maximum. In this scenario, the data are several discrete points, and the peak is the local maximum in the discrete data sequence. By and the local maximum is determined, where i is the data point in the discrete data sequence, is the local maximum, and the peak is obtained by filtering out the noise points through the median filtering method. The valley is opposite to the peak. The valley is the local minimum in the discrete data sequence. By and the local minimum is determined, where m is the data point in the discrete data sequence, is the local minimum, and the valley is obtained by filtering out the noise points through the median filtering method.

[0109] The periodic curve part may not be completely consistent due to inconsistent point distribution and noise points, but the basic trend remains unchanged. The segmentation points appear according to the period. By setting a certain interval between the peaks and valleys, automatic extraction can be completed. The same method is used for the y-axis and z-axis dimensions to completely extract the repeating unit.

[0110] In this embodiment, an improved ICP registration method based on the Fast Point Feature Histogram (FPFH) is adopted. FPFH is a simplified improvement of the Point Feature Histogram (PFH). For each selected point, FPFH only calculates the relationship between itself and adjacent points, and these are called the Simplified Point Feature Histogram (SPFH). At the same time, the neighborhood points of each point are re-determined, and the final histogram (FPFH) is weighted using the adjacent SPFH. The calculation formula is as follows:

[0111]

[0112] In the formula: is the selected point, is the simple point feature histogram of the selected point, is the selected point and the neighborhood point The weight between them, k represents the number of neighborhood points of the selected point. Based on the FPFH feature descriptor, the corresponding feature point pairs in the two point clouds are found by using the hybrid KD tree search parameters, and these points are used to generate the initial transformation matrix to complete the rough registration. Then, the second transformation matrix is ​​calculated according to the result after the rough registration, and the second transformation matrix is ​​used as the input of the ICP algorithm. The ICP iteration process is performed until the termination condition is met, thereby completing the high-precision registration.

[0113] In this embodiment, in order to reduce the influence of small structures with large errors, a method is adopted in which the average distance is calculated by sequentially calculating the difference between the currently measured repeating unit and other repeating units in the 3D printed mesh model.

[0114] Traverse and compare a point of a repeating unit with the points of the same position of other repeating units, calculate the Euclidean distance and take the average value as the difference value of the point. Repeat this operation to obtain the difference value of each position point of the repeating unit. For example, taking the calculation process of the repeating unit numbered 0 as an example, the final average distance calculation formula is: , where x is the point of the repeating unit, is the point cloud of the repeating unit numbered i, and q is Points in the point cloud, N is the total number of the repeating units, where each point belongs to the repeating unit point cloud numbered 0, is the point cloud of the repeating unit numbered i, and N is the total number of repeating structures in the 3D printed mesh model. Repeating this operation can complete the traversal calculation of the entire 3D printed mesh model. This traversal comparison method can quickly and accurately measure each repeating unit to provide a basis for subsequent difference visualization and defect analysis.

[0115] In this embodiment, point cloud processing technology is used to describe different features based on color information in the 3D printed mesh three-dimensional model to assist in the detection of defect differences in the 3D printed mesh three-dimensional model. The entire 3D printed mesh three-dimensional model is visualized in the form of a heat map, and point cloud segmentation is performed based on the color intensity threshold. As the threshold increases, the number of point clouds identified and retained gradually decreases. The higher the threshold, the farther the point distance is, the greater the difference is, and the shape of the defect will also be intuitively visualized.

[0116] The preprocessing method based on CT scanning combined with voxel histogram and threshold segmentation in the present invention can efficiently extract the three-dimensional geometric structure of the model. This method can quickly and automatically generate accurate three-dimensional point cloud data, greatly improving the efficiency of data processing. In addition, the multi-domain feature analysis method combines the dual features of the spatial domain and the frequency domain, especially characterizing the local changes in the spatial domain through the standard deviation, which can more sensitively capture tiny geometric changes. The frequency domain analysis of Fourier transform increases the recognition ability of periodic and repetitive structures, making the system more suitable for complex grid structures and can automatically extract and analyze the basic units of the model.

[0117] In addition, the present invention can automatically detect structural differences and defects by combining multi-domain features with a registration algorithm. The improved application of the registration algorithm greatly improves the accuracy of difference detection, and the traversal calculation improves the detection efficiency. The visualization of the difference detection results enables users to intuitively observe the differences in the models, which helps to quickly locate and analyze potential printing defects.

[0118] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0119] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the above technical solutions, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images, characterized in that: include: Converting the 3D printed mesh model into data in a raw data format by computer tomography, and extracting the 3D printed mesh structure by using a voxel histogram; Extracting geometric features of the 3D printed grid structure in the spatial domain by spatial domain feature analysis, wherein the spatial domain feature analysis includes obtaining a repetitive structure by counting the standard deviation of the dimension of the three-dimensional point cloud model of the 3D printed grid structure, wherein the repetitive structure is concentrated or regularly distributed data points, and analyzing the point cloud distribution by calculating the standard deviation on each plane of the three-dimensional point cloud model, wherein the standard deviation is used to identify the periodicity of the planar features of the 3D printed grid structure; Determine the distribution of the repetitive structure by frequency domain analysis and draw a change curve of the characteristic frequency domain, wherein the frequency domain analysis includes converting the information of the space domain into the information of the frequency domain by fast Fourier transform; Determine the repeating unit by combining the information in the spatial domain and the change curve in the characteristic frequency domain, and identify and automatically extract the repeating unit at the end of a cycle, where the end of the cycle is a wave peak and a wave trough; Based on the fast point feature histogram feature, the feature of the repeating unit is extracted by a feature matching algorithm and the corresponding feature point pairs in the point cloud are found to obtain an initial transformation matrix, the initial transformation matrix is ​​used for coarse registration, a second transformation matrix is ​​obtained by the result after the coarse registration, the second transformation matrix is ​​used as the input of the ICP algorithm and the ICP iteration is performed to register the point cloud containing the repeating unit; The method comprises: performing traversal comparison on the features of the repeating unit in the point cloud after traversal comparison and registration to obtain a measurement result, including: performing traversal comparison on the points of the repeating unit and the points at the same position of the adjacent repeating unit to obtain the Euclidean distance and obtain the average value, using the average value as the difference value of the points of the repeating unit, and repeating the operation by replacing the points of the repeating unit; The defects are analyzed based on the measurement results and the analysis results are displayed in the form of a heat map.

2. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: The extracting of the 3D printing grid structure by using a voxel histogram includes: segmenting the data of the 3D printing grid model by selecting a voxel threshold in the voxel histogram and screening out voxel points representing the 3D printing grid structure to obtain the 3D printing grid structure.

3. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: The statistical standard deviation of the dimension of the three-dimensional point cloud model of the 3D printed grid structure specifically includes: Get three-dimensional point cloud data including x-coordinate, y-coordinate and z-coordinate; The point cloud data is grouped according to the x-coordinate value, and the standard deviation of the y- and z-coordinate values ​​corresponding to each x-coordinate value is calculated to obtain a standard deviation variation curve of the y-plane and the z-plane along the x-axis direction; By taking the geometric mean of the y-plane standard deviation and the z-plane standard deviation and plotting a curve graph that changes with the x-coordinate value; By taking the geometric mean of the x-plane standard deviation and the z-plane standard deviation and plotting a curve graph that changes with the y-coordinate value; By taking the geometric mean of the x-plane standard deviation and the y-plane standard deviation and plotting it against the z-coordinate value.

4. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: The distribution of the repetitive structure is determined by frequency domain analysis and a characteristic frequency domain change curve is plotted. The frequency domain analysis includes converting the spatial domain information into frequency domain information by fast Fourier transform, specifically including: The x-coordinate value is fixed, and the y-coordinate value and the z-coordinate value corresponding to the x-coordinate value are processed in the characteristic frequency domain by the fast Fourier transform; The characteristic frequency domains of the y plane and the z plane are respectively plotted into characteristic frequency domain curves based on the x coordinate value; Performing geometric averaging on the characteristic frequency domain curve of the y plane and the characteristic frequency domain curve of the z plane and drawing a characteristic frequency domain variation curve graph as the x coordinate value changes; Performing geometric averaging on the characteristic frequency domain curve of the x plane and the characteristic frequency domain curve of the z plane and drawing a characteristic frequency domain variation curve graph as the y coordinate value changes; The characteristic frequency domain curve of the x-plane and the characteristic frequency domain curve of the y-plane are geometrically averaged and a characteristic frequency domain variation curve graph varying with the z-coordinate value is plotted.

5. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 4, characterized in that: The fixed x-coordinate value, and performing characteristic frequency domain processing on the y-coordinate value and the z-coordinate value corresponding to the x-coordinate value through the fast Fourier transform, include: The data of the 3D printed grid structure in the spatial domain are grouped according to the x-axis value. For a fixed x-coordinate value, the corresponding y and z-coordinate values ​​are extracted to form a two-dimensional data group. A complex array is obtained after transformation by the fft function in the Python numpy library. Each element in the complex array corresponds to a frequency component. The position number of the element in the complex array represents the frequency index. The relationship between the frequency index and the corresponding actual frequency value is: , where f is the actual frequency, k is the frequency index, N is the data length, and fs is the sampling frequency.

6. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: The peak and the trough are determined by setting a discrete data sequence, wherein the peak is a local maximum value in the discrete data sequence. and Determine the local maximum, where i is a data point in the discrete data sequence, is the local maximum value, the peak is obtained by filtering out the noise points through the median filtering method, and the trough is the local minimum value in the discrete data sequence. and Determine the local minimum, where m is a data point in the discrete data sequence, is the local minimum value, and the trough is obtained by filtering out noise points using the median filtering method.

7. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: Constructing the fast point feature histogram includes: weighting the simple point feature histogram of the neighborhood points relative to the selected point, and then combining the simple point feature histogram of the selected point with the weighted simple point feature histogram of the neighborhood points, specifically using the formula: Where: is the selected point, is the fast point feature histogram of the selected point, is the simple point feature histogram of the selected point, is the simple point feature histogram of the neighborhood points, For the selected point and neighboring points The weight between them, k represents the number of neighborhood points relative to the selected point.

8. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: The specific method of obtaining the average value is: Wherein, x is the point of the repeating unit, is the point cloud of the repeating unit numbered i, q is , N is the total number of repeating units.

9. The method for detecting multi-domain characteristic defects of 3D printed mesh structures based on CT images according to claim 1, characterized in that: Analyzing defects according to the measurement results and displaying the analysis results in the form of a heat map includes: Mapping the indicator value of the defect to the color range of the heat map; Annotating axes, scales and color legends on the heat map; Add interactive functions, including displaying the defect information by hovering the mouse, and being able to zoom in or out the heat map.

Images