Visual three-dimensional reconstruction and mechanical property evaluation method for electric arc additive manufacturing lattice structure

Abstract

The invention discloses a visual three-dimensional reconstruction and mechanical property evaluation method for an electric arc additive manufacturing lattice structure. The method comprises the steps that S1, a two-dimensional image of the electric arc additive manufacturing lattice structure is obtained; s2, mapping the two-dimensional image of the electric arc additive manufacturing lattice structure into a three-dimensional image, and constructing a point cloud model based on the three-dimensional image; s3, performing denoising processing on the point cloud model to obtain a denoised point cloud model; s4, carrying out missing point fitting on the point cloud model to obtain a repaired point cloud model; s5, performing Poisson surface reconstruction on the repaired point cloud model; s6, closing the surface patch model and converting the surface patch model into a solid model; s7, additive manufacturing layered reconstruction is conducted on the point cloud models, the point cloud models are registered, aligned and combined through continuous iteration, the combined model is further processed through S1-S6, and an entity model of a lattice structure is obtained; and S8, performing mechanical simulation on the entity model of the lattice structure, and performing structural mechanical property evaluation according to a mechanical simulation model.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical property evaluation, and particularly relates to a method for visual three-dimensional reconstruction and mechanical property evaluation of an arc additive manufacturing lattice structure. Background Art

[0002] Arc additive manufacturing is a metal additive manufacturing technology that uses an arc as a heat source. It is a method of melting metal wire by arc discharge and then stacking layer by layer to produce metal parts. Compared with traditional metal casting technologies, the printing process does not rely on molds and only needs to control the position of the print head and processing parameters, having the advantages of digitization and intelligence. Compared with traditional machining technologies, the processing procedures are greatly reduced, and the manufacturing of complex three-dimensional structures can be realized. Although there has been certain research on the lattice structure arc additive manufacturing method in the current field, due to influencing factors such as environment, materials, and personnel operation, it is difficult to avoid surface defects such as diameter changes, tilt angle changes, twisting, bulging, and bending. To understand the influence of these surface defects on material properties, mechanical analysis of the printed structure is usually required. Currently, mechanical testing of printed lattice structures usually uses printed physical tests.

[0003] Due to the existence of influencing factors such as environment, materials, and personnel operation, lattice structure parts are difficult to avoid surface defects such as diameter changes, tilt angle changes, twisting, bulging, and bending. Surface defect detection is a key means for additive manufacturing part detection and can currently be divided into destructive testing and non-destructive testing. Destructive testing requires the destruction of the manufactured part, with high precision and strong reliability, but it has high costs, low efficiency, and is not suitable for large-scale detection. And the finite element analysis basically uses an ideal data model, and the prediction of phenomena such as deformation and fracture of the rod is inaccurate. Visual inspection is a non-destructive testing method, which has the advantages of low cost, easy operation, and high efficiency compared with destructive testing. Therefore, the present invention proposes a method for visual three-dimensional reconstruction of a real manufactured part based on visual three-dimensional reconstruction, and then through processes such as point cloud processing, surface reconstruction, and converting the surface patch body into a solid, a data model highly similar to the real manufactured part is generated, providing a reliable basis for detecting the mechanical properties of the lattice structure through finite element analysis. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a method for visual three-dimensional reconstruction and mechanical property evaluation of an arc additive manufacturing lattice structure, which realizes non-destructive testing of the lattice structure through three-dimensional reconstruction and mechanical simulation of the printed part, and has low cost, high efficiency, and easy operation.

[0005] To achieve the above object, the present invention provides a method for visual three-dimensional reconstruction and mechanical property evaluation of an arc additive manufacturing lattice structure, including:

[0006] S1. Obtain two-dimensional images of the arc additive manufacturing lattice structure;

[0007] S2. Map the two-dimensional image of the arc additive manufacturing lattice structure into a three-dimensional image, and construct a point cloud model based on the three-dimensional image;

[0008] S3. Denoise the point cloud model to obtain a denoised point cloud model;

[0009] S4. Fit the missing points of the point cloud model to obtain a repaired point cloud model;

[0010] S5. Perform Poisson surface reconstruction on the repaired point cloud model;

[0011] S6. Close the patch model and convert it into a solid model;

[0012] S7. Perform additive manufacturing layer-by-layer reconstruction on the point cloud model. Through continuous iteration, register and align the point cloud models and merge them to obtain a merged model. Further process the merged model through S1 - S6 to obtain a solid model of the lattice structure;

[0013] S8. Perform mechanical simulation on the solid model of the lattice structure to obtain a mechanical simulation model, and evaluate the structural mechanical properties according to the mechanical simulation model.

[0014] Optionally, obtaining the two-dimensional image of the arc additive manufacturing lattice structure includes:

[0015] Use a spiral - rising method to surround the manufactured part with a camera, and the horizontal rotation angle of the camera is directed towards the center point of the arc, within a preset vertical angle range, to obtain the two - dimensional image of the arc additive manufacturing lattice structure.

[0016] Optionally, constructing a point cloud model based on the three - dimensional image;

[0017] Extract features from the three - dimensional image, match the corresponding points between different images, combine the image sequences, reconstruct the camera movement and the structure of the manufactured part, and generate a sparse point cloud model and camera pose data;

[0018] Based on the generated sparse point cloud model and camera pose data, use photometric consistency to perform stereo matching on the same point on different images, and use the depth map registration principle to fuse the depth maps to generate a dense point cloud model, completing the construction of the point cloud model.

[0019] Optionally, fitting the missing points of the point cloud model to obtain a repaired point cloud model includes:

[0020] Determine the axis of the cylinder, and the cylinder contains a number of point cloud data;

[0021] Centralize the point cloud data contained in the cylinder to obtain the centralized data points and the covariance matrix;

[0022] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, and determine the direction of the cylinder axis;

[0023] Perform slicing on the cylinder perpendicular to the axis to obtain slice data;

[0024] Perform projection processing on the slice data, delete abnormal points and interpolate to fill in missing points to obtain a repaired point cloud model.

[0025] Optionally, performing Poisson surface reconstruction on the surface reconstruction result in the three-dimensional space includes:

[0026] Implicitly fit an indicator function derived from the object to the surface reconstruction result in the three-dimensional space to obtain a Laplace equation for a smooth object surface estimate;

[0027] Find the indicator function by iteratively solving the Laplace equation through the Laplace matrix.

[0028] Optionally, closing the patch model and converting it into a solid model includes:

[0029] Perform Boolean operations based on the patch model to calculate the intersection line or intersection surface of two geometric bodies;

[0030] Redivide the topological structure of the geometric body according to the intersection line or intersection surface;

[0031] Divide the geometric body along the intersection line into new patches and update the topological relationship of the geometric body;

[0032] Classify the new patches, determine whether they belong to the geometric body, and select the patches to be retained according to the type of Boolean operation;

[0033] Sew the retained patches into a new closed geometric body to complete the solidification of the patch model.

[0034] Optionally, calculating the intersection line or intersection surface of two geometric bodies includes:

[0035] If it is the intersection of plane and plane:

[0036] N1·(P - P1) = 0;

[0037] N2·(P - P2) = 0;

[0038] where N1 and N2 are the normal vectors of the plane, P1 and P2 are the points on the plane, and P is the point on the intersection line;

[0039] If it is the intersection of surface and surface, the intersection line is obtained by solving the nonlinear equation system through the Newton iteration method.

[0040] Optionally, performing additive manufacturing layer-by-layer reconstruction on the solid model includes:

[0041] For each preset printing height, the print head of the printer prints layer by layer according to the preset layer height;

[0042] Using a displacement sensor based on the Z-axis, when the print head completes the printing of a certain layer and rises to the height of the next layer, it is determined that the printing of this layer has been completed.

[0043] Technical effects of the present invention: The present invention discloses a method for visual three-dimensional reconstruction and mechanical property evaluation of an arc additive manufacturing lattice structure. Based on the three-dimensional reconstruction model of the printed lattice structure, mechanical property experiments are carried out to achieve non-destructive testing. Solve the axis of the inclined rod of the lattice structure to realize the fitting of missing points based on slicing. Add structures to the existing point cloud model and calculate the coordinates of the newly added point cloud data using the overlapping part. The prior art needs to use a physical object printed with a lattice structure for mechanical experiments, resulting in the destruction of the printed part. However, usually the lattice structure is large in volume, and the manufacturing is time-consuming and costly, so it is not suitable for large-scale detection. The present invention realizes non-destructive testing of the lattice structure through three-dimensional reconstruction and mechanical simulation of the printed part, and has the advantages of low cost, high efficiency and easy operation. Description of the Drawings

[0044] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0045] Figure 1 is a schematic flow chart of the method for visual three-dimensional reconstruction and mechanical property evaluation of the arc additive manufacturing lattice structure according to the embodiment of the present invention;

[0046] Figure 2 is a schematic diagram of the shooting path according to the embodiment of the present invention;

[0047] Figure 3 is a schematic diagram of the conversion relationship according to the embodiment of the present invention

[0048] Figure 4 is a top view according to the embodiment of the present invention;

[0049] Figure 5 is a schematic diagram of the axis according to the embodiment of the present invention;

[0050] Figure 6 is a schematic diagram of the slice according to the embodiment of the present invention;

[0051] Figure 7 is a schematic diagram of the fitting of missing points according to the embodiment of the present invention. Detailed Embodiments

[0052] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The following will describe the present application in detail with reference to the accompanying drawings and in combination with the embodiments.

[0053] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0054] As Figure 1 shown, this embodiment provides a method for visual three-dimensional reconstruction and mechanical property evaluation of an arc additive manufacturing lattice structure, including:

[0055] S1. Obtain a two-dimensional image of the arc additive manufacturing lattice structure;

[0056] S2. Map the two-dimensional image of the arc additive manufacturing lattice structure into a three-dimensional image, and construct a point cloud model based on the three-dimensional image;

[0057] S3. Denoise the point cloud model to obtain a denoised point cloud model;

[0058] S4. Fit the missing points of the point cloud model to obtain a repaired point cloud model;

[0059] S5. Perform Poisson surface reconstruction on the repaired point cloud model;

[0060] S6. Close the patch model and convert it into a solid model;

[0061] S7. Perform additive manufacturing layer-by-layer reconstruction on the point cloud model, continuously iterate to register and align the point cloud model and merge to obtain a merged model, and further process the merged model through S1-S6 to obtain a solid model of the lattice structure;

[0062] S8. Perform mechanical simulation on the solid model of the lattice structure to obtain a mechanical simulation model, and evaluate the structural mechanical properties according to the mechanical simulation model.

[0063] Furthermore, obtaining a two-dimensional image of the arc additive manufacturing lattice structure includes:

[0064] Adopt a spiral upward method to surround the manufactured part with a camera, and the horizontal rotation angle of the camera faces the center point of the arc, meeting the preset vertical angle range, to obtain a two-dimensional image of the arc additive manufacturing lattice structure.

[0065] Specifically, as Figure 2As shown in the figure, place the manufactured part in an environment with sufficient ambient light, simple structure, single texture, and pure color, and avoid environments with strong reflective materials such as tiles and glass (to reduce the interference of environmental factors on point cloud reconstruction). Use a camera to surround the manufactured part to collect images, ensuring that each visible part can be collected. At the same time, the horizontal rotation angle of the camera is oriented towards the center point of the arc, and the vertical angle is between 20 - 30°, to ensure that each collected picture overlaps with other pictures.

[0066] Since a single cell has a pyramid structure, the pictures are collected in a spiral ascending manner. The spiral equation is as follows:

[0067]

[0068] In the formula, h is the pitch, β is the helix angle, taking a positive sign for a right-handed helix and a negative sign for a left-handed helix.

[0069] Furthermore, construct a point cloud model based on the three-dimensional image;

[0070] Extract features from the three-dimensional image, match the corresponding points between different images, and combine the image sequence to reconstruct the camera motion and the structure of the manufactured part, generating a sparse point cloud model and camera pose data;

[0071] Based on the generated sparse point cloud model and camera pose data, use photometric consistency to perform stereo matching on the same point on different images, and use the depth map registration principle to fuse the depth maps to generate a dense point cloud model, completing the construction of the point cloud model.

[0072] Specifically, establish a sparse point cloud model

[0073] The objects in the three-dimensional world and the two-dimensional images are connected by a camera, and the three-dimensional world can be restored from the two-dimensional images through their mapping relationship. In the homogeneous coordinate system, the relationship from P(x, y, z) in the camera coordinate system to P′(x′, y′) in the image plane coordinate system is:

[0074]

[0075] M is the projection matrix. Among them, α = fk, β = fl, which are called the internal parameters of the camera, simply referred to as internal parameters, f is the camera focal length, and k and l are the unit conversion parameters in the x and y directions of the image plane coordinate system respectively. c(c x , c y ) is the coordinate of the center point of the image plane in the image plane coordinate system.

[0076] Due to the errors in the manufacturing process of the camera imaging element, there is an inclination angle θ in the imaging plane. At this time:

[0077]

[0078] K is the camera internal parameter matrix, simply referred to as the internal parameter matrix.

[0079] If a certain point in the real world is used as the origin to establish the world coordinate system i w j w k w , then there is a conversion relationship between the world coordinate system and the camera coordinate system, as Figure 3 shown.

[0080] The world coordinate system can obtain the camera coordinate system through rotation and translation operations. R and T are the rotation matrix and the translation matrix respectively. Then the mapping relationship between the point P W and P′ in the world coordinate system is:

[0081] P′ = K[R T]P w = MP w ;

[0082] [R T] is the camera external parameter matrix, and M is the projection matrix. Therefore, knowing the projection matrix M, the world coordinates and image plane coordinates of any point can be converted to each other.

[0083] To recover the three-dimensional coordinate points, usually two or more images are required. The corresponding constraint relationships between multiple images only depend on the internal and external parameter matrices of the camera and have nothing to do with the structure of the scene. In two imaging views at different angles, the homogeneous coordinates of the point P in the O1 image plane coordinate system are p(u, v, 1), and the coordinates in the O2 image plane coordinate system are p′(u′, v′, 1). At this time, the corresponding constraint relationship between the two points can be expressed as:

[0084] p′ T K′ -T EK -1 p = p′ T Fp = 0;

[0085] where F = K′ -T EK -1 is the fundamental matrix, and E = [T×R]] is the essential matrix.

[0086] When establishing a sparse point cloud model of the lattice structure, use feature extraction algorithms such as the SIFT algorithm to extract features from each image, and then use the matching algorithm to match the corresponding points between different images to provide key spatial information for subsequent calculations.

[0087] According to the image sequence and the spatial information obtained from the above processing, based on three-dimensional reconstruction methods such as the incremental SFM algorithm, the camera motion and the 3D structure of the manufactured part are reconstructed to generate a sparse point cloud model and camera pose data.

[0088] 2) Establish a dense point cloud model

[0089] The sparse point cloud has a small data volume and the reconstruction effect is not ideal. Therefore, dense reconstruction is required to improve the point cloud data.

[0090] Given the known camera parameters, using photometric consistency, stereo matching is performed on the same point in different images, and the depth maps are fused using the depth map registration principle, thereby increasing the density of the sparse point cloud and generating a dense point cloud. Among them, depth estimation is based on the parallax principle. The top view is as Figure 4 shown.

[0091] Given the matching feature points and the camera projection matrix, the depth value of each pixel point can be further calculated:

[0092]

[0093] Among them, B is the distance between the optical centers of the two cameras, f is the camera focal length, z is the depth information of point P, and p u -p′ u is the parallax.

[0094] Therefore, according to the camera parameters B and f, the depth map can be restored. Finally, the depth maps of each image are fused to recover the depth information of the scene and complete the dense reconstruction.

[0095] 3) Straight line calibration

[0096] During the manufacturing process, it is difficult to avoid situations such as bent rods. Therefore, it is necessary to perform straight line calibration on the point cloud model to improve the accuracy and precision of the straight line features in the point cloud.

[0097] First, the three-dimensional straight line segments in the model need to be extracted. For a three-dimensional projection plane area, its coplanar points are projected onto a two-dimensional projection plane. After obtaining the projection image, the straight line segments in the image are extracted using an algorithm, and then the two-dimensional straight line segments are back-projected from 2D to 3D to obtain three-dimensional line segments.

[0098] Next, geometric calibration and optimization are performed on the extracted straight lines. Based on all the extracted straight line segments, the distance error between all related points and the straight line is minimized by the least squares method to calibrate the straight line.

[0099] Furthermore, the point cloud data denoising includes:

[0100] The point cloud model usually contains noise, outliers, or other inaccurate parts. To avoid the influence of noise on the subsequent reconstruction quality and improve the accuracy and effect of the reconstruction, it is necessary to perform denoising processing on the point cloud model. Given that the lattice structure consists of thin plates and inner rods, in order to preserve details (especially at the edge part), bilateral filtering is used to perform denoising processing on the model.

[0101] Bilateral filtering takes into account both distance factors and pixel value differences. The pixel value weight is represented as G r , and the spatial distance weight is represented as G s .

[0102]

[0103] Then the entire filter can be represented as BF, and the filtering result is:

[0104]

[0105] where W q is the sum of the weights of each pixel value within the filtering window and is used for weight normalization:

[0106]

[0107] Furthermore, fitting the missing points of the point cloud model, obtaining the repaired point cloud model includes:

[0108] Determine the axis of the cylinder, and the cylinder contains a number of point cloud data;

[0109] Centralize the point cloud data contained in the cylinder to obtain the centered points and covariance matrix;

[0110] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, and determine the direction of the cylinder axis;

[0111] Slice the cylinder perpendicular to the axis to obtain slice data;

[0112] Perform projection processing on the slice data, delete abnormal points and interpolate to fill in the missing points to obtain the repaired point cloud model.

[0113] Specifically, due to limitations such as object self-occlusion and perspective transformation, the obtained point cloud data is incomplete. To avoid affecting the quality of subsequent processing, it is necessary to fit the missing points of the model.

[0114] First, it is necessary to determine the axis of the cylinder. It is known that the point cloud data of each rod contains multiple 3D coordinate points:

[0115] P = {(x1, y1, z1), (x2, y2, z2), (x3, y3, z3),..., (x n , y n , z n )};

[0116] These points constitute a part of the cylinder surface and interior. Centralize these point cloud data to make their mean value become 0:

[0117]

[0118] p′ i is the point after data centralization.

[0119] The covariance matrix Σ can reflect the relationship between each dimension in the point cloud data. By performing eigenvalue decomposition on the covariance matrix ∑, eigenvalues and eigenvectors are obtained. The eigenvalue represents the variance magnitude in the direction of the corresponding eigenvector, and the eigenvector represents the main direction of the data (i.e., the cylinder axis direction).

[0120]

[0121] Among them, v i is the eigenvector, and λ i is the corresponding eigenvalue. The eigenvector corresponding to the minimum value among all eigenvalues is the cylinder axis direction, as Figure 5 shown.

[0122] Then, select appropriate number of slices and thickness, and perform slicing processing perpendicular to the axis of the cylinder to obtain slices E1, E2... En, as Figure 6 shown.

[0123] Perform projection processing on the slice data. Given that the ideal surface contour line should be circular, use the moving least squares method to integrate the projection data. Compare and calculate the original point cloud data with the integrated point cloud data to obtain the residuals and errors of each data point after fitting. Points exceeding a certain error are considered abnormal points and should be deleted, otherwise they are retained. After completion, recalculate using the remaining points and loop the above steps until the errors of all points are within the required range. At the edge of the slice data, there may be insufficient data, and interpolation can be used to fill in the missing data. Finally, splice the fitting results of each slice to obtain the surface reconstruction result in the entire three-dimensional space, as Figure 7 shown.

[0124] Since there may be discontinuous situations such as local protrusions or depressions on the surface when splicing the slice fitting results, the reconstructed model is smoothed.

[0125] Furthermore, performing Poisson surface reconstruction on the surface reconstruction result in the three-dimensional space includes:

[0126] Obtaining a Laplace equation for a smooth object surface estimate by implicitly fitting an indicator function derived from the object to the surface reconstruction result in the three-dimensional space;

[0127] Finding the indicator function by iteratively solving the Laplace equation through the Laplace matrix.

[0128] Specifically, a smooth and continuous three-dimensional surface is recovered from the point cloud data through Poisson surface reconstruction. Poisson surface reconstruction can give a smooth estimate of the object surface by implicitly fitting an indicator function derived from the object. Given a region M and its boundary Indicator function χ M is defined as:

[0129]

[0130] This indicator function is a piecewise function, which defines that the value inside the model is greater than 0, the value outside is less than 0, and the part equal to 0 is the isosurface, that is, the surface of the target model.

[0131] Spatial vector and the indicator function satisfy:

[0132]

[0133] To obtain the least squares solution, the above formula is differentiated on both sides to obtain the Laplace equation:

[0134]

[0135] By iteratively solving this equation through the Laplace matrix, the indicator function can be found, and thus the isosurface can be obtained to complete the surface reconstruction.

[0136] Furthermore, closing the patch model and converting it into a solid model includes:

[0137] Performing Boolean operations based on the patch model to calculate the intersection line or intersection surface of two geometric bodies;

[0138] Redefining the topological structure of the geometric body according to the intersection line or intersection surface;

[0139] Dividing the geometric body along the intersection line into new patches and updating the topological relationship of the geometric body;

[0140] Classifying the new patches to determine whether they belong to the geometric body, and selecting the patches to be retained according to the type of Boolean operation;

[0141] Sewing the retained patches into a new closed geometric body to complete the solidification of the patch model.

[0142] Specifically, since the required CAD model is a closed model, and the patch model may have non-closed, overlapping, and irregular connections, it is necessary to close the patch model and convert it into a solid for processing.

[0143] Use software to optimize the surface of the patch model, including but not limited to patch repair, patch subdivision, smoothing, re-wrapping, etc. The hard requirement for optimization is that the model after converting to a solid can successfully perform finite element mesh generation. Then, convert the hollow closed patch model into a CAD solid model that can be input into finite element analysis. The conversion of the patch model to a solid is completed based on Boolean operations. Boolean operations are based on set theory, and their mathematical representation is as follows:

[0144] ① Union: A ∪ B = {x | x ∈ A or x ∈ B};

[0145] ② Intersection: A ∩ B = {x | x ∈ A and x ∈ B};

[0146] ③ Difference set:

[0147] When performing Boolean operations, first calculate the intersection line or intersection surface of two geometric bodies (such as plane and plane, surface and surface). If finding the intersection of a plane and a plane:

[0148] N1 · (P - P1) = 0;

[0149] N2 · (P - P2) = 0;

[0150] where N1 and N2 are the normal vectors of the planes, P1 and P2 are points on the planes, and P is a point on the intersection line.

[0151] If finding the intersection of a surface and a surface, then solve the non-linear equations through the Newton iteration method to obtain the intersection line.

[0152] Then, re-divide the topological structure of the geometric body according to the intersection line or intersection surface. Divide the geometric body along the intersection line into new patches and update the topological relationship of the geometric body (such as the connection relationship of vertices, lines, and faces). Classify the new patches to determine whether they belong to the geometric body, and select the patches to be retained according to the type of Boolean operation. Finally, stitch the retained patches into a new closed geometric body to complete the solidification.

[0153] Furthermore, the additive manufacturing layer-by-layer reconstruction of the solid model includes:

[0154] Set a preset printing height for each layer, and the printer nozzle prints layer by layer according to the preset layer height;

[0155] Use a displacement sensor based on the Z-axis. When the nozzle completes the printing of a certain layer and lifts to the height of the next layer, it is determined that this layer has been completed.

[0156] Specifically, since the printing height of each layer is preset, and during the additive manufacturing process, the printer's nozzle will print layer by layer according to the preset layer height, a Z-axis-based displacement sensor is used. When the nozzle completes the printing of a certain layer and lifts to the height of the next layer, it can be determined that the layer has been printed. At this time, the machine can be stopped and the material can be taken out for shooting.

[0157] Take a picture of the latest printed material and establish a point cloud model. Since the newly obtained point cloud model and the previously processed point cloud model are not in the same coordinate system, they need to be aligned through registration.

[0158] Suppose that λ layers have been printed, the (λ - 1)-th obtained point cloud model is the source point cloud P, and the λ-th (i.e., the latest obtained) point cloud model is the target point cloud Q. There is an overlapping part between the two sets of point cloud models, that is, the point cloud data of the lattice structure from the 1st layer to the (λ - 1)-th layer. Therefore, the feature point pairs of this overlapping part are used to register and align the point clouds P and Q. Point cloud registration is to iteratively minimize the distance between point clouds and solve the optimal rotation matrix R and translation vector t. It is expressed as:

[0159]

[0160] where p i , q i are the points in the source point cloud P and the target point cloud Q respectively, w i represents the weight of each point, and d represents the dimension. For three-dimensional point clouds, d = 3. That is, by constructing a least squares problem, it is transformed into an optimization problem of the objective function.

[0161] ① Calculate the translation vector t

[0162] Assume that R is a constant, and let The objective function can be obtained:

[0163]

[0164] Let the centroids of p and q be:

[0165]

[0166]

[0167] Then it can be obtained:

[0168] μ p -Rμ q = t;

[0169] Substitute the above formula into the original E(R, t), and it can be obtained:

[0170]

[0171] where p′ i and q′ i are the centroid - removed coordinates calculated for each point cloud according to its centroid coordinates μ p and μ q respectively. When the above - mentioned objective function is minimized, the translation vector t can be obtained.

[0172] ② Calculate the rotation matrix R

[0173]

[0174] As can be seen from the above formula, except for the second term, the other parts of the objective function are independent of the parameter R, so they can be ignored during the calculation. Therefore, the objective function can be simplified as:

[0175]

[0176] By solving the above formula, the rotation matrix R can be obtained.

[0177] By continuous iteration, the two point - cloud models can be registered and aligned.

[0178] After alignment, the two point clouds are merged into one point - cloud model, and uniform sampling is performed in the overlapping area to reduce the point density. The merged point - cloud model can repeat steps (1)-(6) to achieve further processing of the model.

[0179] Furthermore, in S8, perform mechanical simulation on the solid model of the lattice structure to obtain a mechanical simulation model. The structural mechanical property evaluation based on the mechanical simulation model includes:

[0180] For the lattice sandwich structure, the focus is on the rods rather than the upper and lower flat plates. The overly complex flat - plate entity surface will affect the finite - element analysis and also increase the storage size, affecting the efficiency. And since the shape of the flat plate is known to be rectangular, the upper and lower flat plates are manually built using CAD instead of using the surfaces described by the obtained solid model to obtain a complete CAD model for facilitating subsequent structural mechanical property evaluation.

[0181] When performing mechanical simulation on the lattice structure, the following mechanical simulation algorithms are usually adopted:

[0182] Element birth and death: Element birth and death is a non - linear element that allows simulation of the material failure process. Element birth and death has an adaptive feature. It can fail under specific conditions and lose stiffness after failure. It can be used in damage mechanics, fracture mechanics, fatigue analysis, etc. to predict the life and reliability of the structure under actual working conditions.

[0183] Finite element analysis: The continuum is discretized into a finite number of simple geometric shapes (i.e., finite elements), and a mathematical model is established on these elements. Visualization results are obtained based on the applied boundary conditions and loads. Through finite element analysis, nonlinear analysis, dynamic analysis, buckling analysis, etc. can be carried out.

[0184] Since simulation software such as ABAQUS does not have a fixed dimension system, it is necessary to determine the dimension system used before starting. All input data must ensure unit consistency. The dimension system of the simulation model is shown in Table 1.

[0185] Table 1

[0186] Density Mass Time Length Load Stress Energy <![CDATA[tonne / mm 3 > <![CDATA[tonne(10 3 kg)]]> s mm N <![CDATA[MPa (N / mm 2 )]]> <![CDATA[mJ(10 -3 J)]]>

[0187] To more realistically simulate the entire experimental process, the two flat plates at the upper and lower ends of the lattice structure model simulate the upper and lower indenter of the universal testing machine, and reference points are set at the centers of the two flat plates respectively to apply constraint conditions. For quantitative simulation analysis, the upper and lower flat plates are set as rigid, and the lattice structure is set as flexible. Among them, the rigid flat plate simulating the support plate applies a fully fixed constraint at the reference point, and the rigid flat plate used to simulate the loading plate only releases the degree of freedom perpendicular to the flat plate direction at the reference point, and a quantitative displacement can be added to simulate the actual situation. The contact condition is set as self-contact inside the lattice structure model, the normal direction is hard contact, and an appropriate tangential friction coefficient is set.

[0188] The basic idea of the finite element method is to discretize the continuum, use simplified geometric elements to approximate the continuum, and then perform comprehensive solutions according to the deformation compatibility conditions. Therefore, the selection of mesh element type and size is crucial, and the division quality plays a decisive role in the simulation results. If the size is too large, the accuracy is low; if the size is too small and the number of meshes is large, the calculation time increases and the efficiency decreases. Considering the influence of mesh sensitivity, adaptive size adjustment can be used, with finer meshes in high-gradient regions and coarser meshes in flat regions.

[0189] In addition, parameters such as material type, density, Poisson's ratio, etc. also need to be set.

[0190] After all parameter settings are completed, the following mechanical simulation experiments are carried out according to the experimental requirements:

[0191] ① Compression: Usually, the quasi-static compression experiment is carried out at a constant strain rate and temperature, and the stress-strain curve, deformation behavior, internal stress distribution nephogram, energy absorption capacity curve, etc. of the model can be output.

[0192] ② Tensile: Through the stress-strain curve, the elastic, yield strength, ultimate strength, fracture and other characteristics of the model can be output, and the stress distribution in the entire structure can also be output to identify the stress concentration areas of the material. And under different excitation functions, the peak amplitude, velocity, and acceleration of the structure can be obtained.

[0193] ③ Hydrostatic pressure: Used to analyze the stress distribution and deformation of materials under hydrostatic pressure.

[0194] ④ Vibration: Through modal analysis, the frequencies and corresponding modal shapes of each vibration mode can be obtained, and the natural frequencies of the structure can be output for analyzing the frequencies that cause resonance phenomena.

[0195] ⑤ Explosion: Through explosion shock experiments, study the damage mechanism and failure modes of this structure under strong explosion loads. Compare the failure modes of the lattice structure under different working conditions, such as face-core debonding failure, rod fracture, crushing, local cracks, ply tearing, etc., to evaluate the anti-explosion performance of the structure.

[0196] ⑥ Impact: Impact according to the specified peak acceleration, pulse duration, pulse repetition frequency and number of collisions to study the crashworthiness and energy absorption performance of the lattice structure.

[0197] The present invention discloses a method for visual three-dimensional reconstruction and mechanical property evaluation of an arc additive manufacturing lattice structure. Based on the three-dimensional reconstruction model of the printed lattice structure, mechanical property experiments are carried out to achieve non-destructive testing. Solve the axis of the inclined rods of the lattice structure to realize the fitting of missing points based on slicing. Add a structure to the existing point cloud model and calculate the coordinates of the newly added point cloud data using the overlapping part. The prior art needs to use the printed physical object of the lattice structure for mechanical experiments, resulting in the destruction of the printed parts. However, usually the lattice structure is large in volume, and the manufacturing takes a long time and is costly, so it is not suitable for large-scale detection. The present invention realizes non-destructive testing of the lattice structure through three-dimensional reconstruction and mechanical simulation of the printed parts, and has the advantages of low cost, high efficiency and easy operation.

[0198] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for visual three-dimensional reconstruction and mechanical property evaluation of a lattice structure by arc additive manufacturing, characterized in that Including: S1. Obtain a two-dimensional image of the arc additive manufacturing lattice structure; S2. Map the two-dimensional image of the arc additive manufacturing lattice structure into a three-dimensional image, and construct a point cloud model based on the three-dimensional image; S3. Denoise the point cloud model to obtain a denoised point cloud model; S4. Fit the missing points of the point cloud model to obtain a repaired point cloud model; S5. Perform Poisson surface reconstruction on the repaired point cloud model; S6. Close the patch model and convert it into a solid model; S7. Perform additive manufacturing hierarchical reconstruction on the point cloud model. Through continuous iteration, register and align the point cloud model and merge it to obtain a merged model. Further process the merged model through S1 - S6 to obtain a solid model of the lattice structure; S8. Perform mechanical simulation on the solid model of the lattice structure to obtain a mechanical simulation model, and evaluate the structural mechanical properties according to the mechanical simulation model.

2. The visual three-dimensional reconstruction and mechanical property evaluation method of the arc additive manufacturing lattice structure according to claim 1, characterized in that Obtaining a two-dimensional image of the arc additive manufacturing lattice structure includes: Using a spiral upward method to surround the manufactured part with a camera, and the horizontal rotation angle of the camera is oriented towards the center point of the arc, meeting the preset vertical angle range, to obtain a two-dimensional image of the arc additive manufacturing lattice structure.

3. The visual three-dimensional reconstruction and mechanical property evaluation method of the arc additive manufacturing lattice structure according to claim 1, characterized in that, Construct a point cloud model based on the three-dimensional image; Extract features from the three-dimensional image, match the corresponding points between different images, and combine the image sequence to reconstruct the camera motion and the structure of the manufactured part, generating a sparse point cloud model and camera pose data; Based on the generated sparse point cloud model and camera pose data, use photometric consistency to perform stereo matching on the same point on different images, and use the depth map registration principle to fuse the depth maps to generate a dense point cloud model, completing the construction of the point cloud model.

4. The method for visual three-dimensional reconstruction and mechanical property evaluation of the arc additive manufacturing lattice structure according to claim 1, characterized in that, Fitting the missing points of the point cloud model to obtain a repaired point cloud model includes: Determine the cylindrical axis, and the cylinder contains a number of point cloud data; Centralize the point cloud data contained in the cylinder to obtain the centralized data points and the covariance matrix; Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, and determine the direction of the cylindrical axis; Perform slicing processing perpendicular to the axis of the cylinder to obtain slice data; Perform projection processing on the slice data, delete abnormal points and interpolate to fill in the missing points to obtain a repaired point cloud model.

5. The method for visual three-dimensional reconstruction and mechanical property evaluation of the arc additive manufacturing lattice structure according to claim 1, characterized in that, Performing Poisson surface reconstruction on the surface reconstruction result in the three-dimensional space includes: Implicitly fit an indicator function derived from the object to the surface reconstruction result in the three-dimensional space to obtain a Laplace equation for a smooth object surface estimate; Iteratively solve the Laplace equation through the Laplace matrix to find the indicator function.

6. The visual three-dimensional reconstruction and mechanical property evaluation method of the arc additive manufacturing lattice structure according to claim 1, characterized in that Closing the patch model and converting it into a solid model includes: Perform Boolean operations based on the patch model to calculate the intersection line or intersection surface of two geometric bodies; Redivide the topological structure of the geometric body according to the intersection line or intersection surface; Divide the geometric body along the intersection line into new patches and update the topological relationship of the geometric body; Classify the new patches to determine whether they belong to the geometric body, and select the patches to be retained according to the type of Boolean operation; Stitch the retained patches into a new closed geometric body to complete the solidification of the patch model.

7. The method for visual three-dimensional reconstruction and mechanical property evaluation of the arc additive manufacturing lattice structure according to claim 6, wherein Calculating the intersection line or intersection surface of two geometric bodies includes: If it is the intersection of plane and plane: N1·(P - P1) = 0; N2·(P - P2) = 0; where N1 and N2 are the normal vectors of the planes, P1 and P2 are points on the planes, and P is a point on the intersection line; If the intersection of surfaces is required, the intersection line is obtained by solving the non - linear equations through the Newton - Raphson method.

8. The method for visual three-dimensional reconstruction and mechanical property evaluation of the arc additive manufacturing lattice structure according to claim 1, characterized in that The additive manufacturing layer - by - layer reconstruction of the solid model includes: Presetting the printing height for each layer, and the nozzle of the 3D printer prints layer by layer according to the preset layer height; Using a displacement sensor based on the Z - axis, when the nozzle completes the printing of a certain layer and rises to the height of the next layer, it is determined that the current layer has been printed.

Images