Method for identifying macroscopic morphology defect type of additive manufacturing lattice structure

Abstract

The invention discloses a method for identifying macroscopic morphology defect types of an additive manufacturing lattice structure. The method comprises the following steps: acquiring lattice cell image data of a vertical angle; calibrating the dot matrix cell element image data, constructing a dot matrix cell element identification data set, and identifying single dot matrix cell element image data; segmenting the single dot matrix cell image data, constructing a dot matrix cell top arc region segmentation data set, and segmenting the top arc regions of four rod pieces of the dot matrix cell; processing the top arc areas of the four rod pieces of the segmented dot matrix cell element, and extracting cell element feature information; quantizing the extracted cell feature information, and determining a quantitative index; and identifying and classifying defect types of the dot matrix cell element image data based on the quantitative indexes. According to the method, lattice structure defects are recognized, defect types are quantified, and the printing quality and mechanical properties of the large metal lattice are effectively guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of morphological defect recognition of lattice structures in additive manufacturing, and in particular relates to a method for identifying macroscopic morphological defect types of lattice structures in additive manufacturing. Background Art

[0002] With the development of science and technology, additive manufacturing technology has long been deeply rooted in people's production and life. Among them, the metal lattice structure has structural characteristics such as high porosity, light weight and high strength, as well as functional characteristics such as shock absorption and noise reduction, impact resistance and energy absorption. At present, the additive process for large lattice structures mainly adopts wire arc additive manufacturing (WAAM), which can eliminate the limitation of part size, and compared with metal powder technology, WAAM has lower cost and higher material utilization rate. However, the survey found that the WAAM forming process is highly sensitive to process parameters. The forming process is usually accompanied by gradient changes of multiple energy signals, which leads to significant thermal stress inside the parts, thereby causing the formation of various defects such as cracks and geometric faults. The existence of these defects seriously affects the mechanical properties of the lattice structure, and then has a negative impact on the service life of the lattice and its ideal effect. In the prior art, the internal structure is scanned and detected by industrial CT to obtain a two-dimensional grayscale image of its transverse fault section, and the distribution characteristics of the grayscale values ​​of the pixels in the two-dimensional grayscale image are used to judge whether there are internal defects. In the prior art, there is a method that combines convolutional neural networks and industrial CT to scan multi-layer metal lattice structure material samples, obtain their three-dimensional structures, and extract two-dimensional grayscale images of transverse sections. Defects are identified and located by learning from defect samples in the image. Problems existing in the prior art include: the generalization ability of industrial CT is not ideal, it mostly relies on manual judgment of images, the information dimension is low, the missed detection / false detection rate is high, it is time-consuming and costly, and it only considers the internal defects of the lattice, and does not consider the detection of macroscopic morphological defects of the lattice. Therefore, there is an urgent need for a method for identifying the types of macroscopic morphological defects in additive manufacturing lattice structures. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method for identifying the macroscopic morphological defect types of additive manufacturing lattice structures, which can identify lattice structure defects and quantify defect types, and effectively ensure the printing quality and mechanical properties of large metal lattices.

[0004] To achieve the above object, the present invention provides a method for identifying the type of macroscopic morphology defects of an additively manufactured lattice structure, comprising:

[0005] Acquire the dot matrix cell image data at a vertical angle;

[0006] Calibrate the dot matrix cell image data and construct a dot matrix cell recognition data set to recognize individual dot matrix cell image data;

[0007] Segment the individual dot matrix cell image data and construct a dot matrix cell top arc region segmentation data set to segment the top arc regions of the four rods of the dot matrix cell;

[0008] Process the segmented top arc regions of the four rods of the dot matrix cell to extract cell feature information;

[0009] Quantify the extracted cell feature information to determine quantization indexes;

[0010] Based on the quantization indexes, recognize and classify the defect types of the dot matrix cell image data.

[0011] Optionally, obtaining dot matrix cell image data at a vertical angle includes:

[0012] Measure the printed dot matrix structure, scan it in a horizontal posture along a set S-shaped trajectory by a drone with a camera, and use computer vision technology to intercept video data frame by frame. If an image contains a complete dot matrix cell, obtain the starting arc positions of the four rods of the cell, and use the center of the regular quadrilateral formed by the four vertices as the calibration point. With the center of the image as the origin and a preset radius size, if the calibration point belongs to the interior of the circle, then the cell image is an image taken directly from above, which is the dot matrix cell image data at a vertical angle; if the calibration point does not belong to the interior of the circle, then discard this image; if adjacent frame images contain several different rods of a dot matrix cell but do not each contain a complete cell, then use image stitching technology to stitch the adjacent image frames into dot matrix cell image data at a vertical angle.

[0013] Optionally, the design process of the S-shaped trajectory includes:

[0014] Measure the overall physical size of the dot matrix structure part, where the dot matrix cells are printed according to the settings of the G code and arranged equidistantly in sequence. According to the size of the dot matrix cells and in combination with the pinhole model, calculate the effective projection range to obtain the length and spacing of the S-shaped trajectory.

[0015] Optionally, recognizing individual dot matrix cell image data includes:

[0016] Obtain several dot matrix cell image data taken at a directly overhead angle by scanning. Based on the dot matrix cell image data, mark them with labels, set the size of the input image, train the dot matrix cell recognition target detection neural network, and obtain the trained dot matrix cell recognition target detection neural network to recognize all individual dot matrix cell image data.

[0017] Optionally, the top arc regions of the four rods of the lattice cell include:

[0018] Divide the top arc regions of the lattice cell rods from the lattice cell image data, set the size of the input image, train the segmentation network for the top arc regions of the lattice cell rods, obtain the trained segmentation network for the top arc regions of the lattice cell rods, and segment the top arc regions of the four rods of the lattice cell.

[0019] Optionally, extracting the cell feature information includes:

[0020] Process the top arc regions of the four rods of the segmented lattice cell, perform binary processing and edge detection on the image, obtain the boundaries of the top arc regions of the four rods, fit each top arc region of the rod into a standard circle, connect the centers of the four circles, and obtain the cell feature information of a quadrilateral.

[0021] Optionally, determining the quantization indexes includes:

[0022] Index 1 is that the quadrilateral is a convex quadrilateral;

[0023] Index 2 is whether the ratio of the area of the quadrilateral to the diameter of the lattice rod is greater than 625;

[0024] Index 3 is whether the shortest side length of the quadrilateral is less than 5 mm;

[0025] Index 4 is whether the longest side length of the quadrilateral is greater than 40 mm;

[0026] Index 5 is whether the logarithm of the ratio of the longest side length to the shortest side length of the quadrilateral is greater than 0.9;

[0027] Index 6 is whether the diagonal of the quadrilateral is less than 5 mm;

[0028] Index 7 is whether the diagonal of the quadrilateral is greater than 60 mm;

[0029] Index 8 is whether the ratio of the two diagonals of the quadrilateral is 1.

[0030] Optionally, identifying the defect types of the lattice cell image data includes:

[0031] Identify the defect types of the lattice cell image data. If all eight indexes do not conform, it is determined as a normal cell; if Index 1 is met and the other indexes do not conform, the defect type is determined as rod offset; if Index 2 or 4 or 7 is met and the other indexes do not conform, the defect type is determined as rod missing; if Index 3 or 5 or 6 is met and the other indexes do not conform, the defect type is determined as rod adhesion; if Index 8 is met and the other indexes do not conform, the defect type is determined as asymmetric lattice.

[0032] Technical effects of the present invention: The present invention discloses a method for identifying the types of macroscopic morphology defects of an additive manufacturing lattice structure. Although some existing technologies and researches can also identify some defects in 3D printing products or manufacturing processes, the research on the macroscopic morphology problems of lattice structures is very scarce. Moreover, the present invention proposes specific quantitative indicators for determining whether a lattice cell is qualified and the types of lattice cell defects, which is more scientific. The present invention combines the S-shaped trajectory scanning technology of an unmanned aerial vehicle to realize a method for accurately photographing vertical lattice images. By steps such as scanning method, lattice cell identification, feature extraction, defect quantification, and defect classification, the purpose of identifying lattice structure defects and quantifying the types of defects is achieved, effectively ensuring the printing quality and mechanical properties of large metal lattices. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0034] Figure 1 is a schematic flowchart of the method for identifying the types of macroscopic morphology defects of an additive manufacturing lattice structure according to an embodiment of the present invention;

[0035] Figure 2 is a flowchart of the module for obtaining vertical lattice cell pictures according to an embodiment of the present invention;

[0036] Figure 3 is a schematic diagram for determining whether a calibration point is within the determination range according to an embodiment of the present invention;

[0037] Figure 4 is a flowchart of the module for extracting cell feature information by using computer vision technology according to an embodiment of the present invention;

[0038] Figure 5 is a flowchart for classifying lattice defect types according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0040] It should be noted that the steps shown in the flowchart of the 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.

[0041] As Figure 1 shown, in this embodiment, a method for identifying the types of macroscopic morphology defects of an additive manufacturing lattice structure is provided, including:

[0042] Measure the printed dot matrix structure to obtain the image data of the dot matrix cell in the vertical angle;

[0043] Calibrate the dot matrix cell image data and construct a dot matrix cell recognition data set to recognize the image data of a single dot matrix cell;

[0044] Segment the dot matrix cell image data and construct a data set for segmenting the top arc region of the dot matrix cell to segment the top arc regions of the four rods of the dot matrix cell;

[0045] Process the segmented top arc regions of the four rods of the dot matrix cell to extract the cell feature information;

[0046] Quantify the extracted cell feature information to determine the quantization index;

[0047] Based on the quantization index, identify and classify the defect types of the dot matrix cell image data.

[0048] Further, obtaining the image data of the dot matrix cell in the vertical angle includes:

[0049] Measure the printed dot matrix structure, scan it in a horizontal posture along a set S-shaped trajectory by a drone with a camera, and use computer vision technology to intercept video data frame by frame. If an image contains a complete dot matrix cell, obtain the starting arc positions of the four rods of the cell, and use the center of the regular quadrilateral formed by the four vertices as the calibration point. With the center of the image as the origin and a preset radius size, if the calibration point belongs to the interior of the circle, the cell image is an image taken from directly above, which is the image data of the dot matrix cell in the vertical angle; if the calibration point does not belong to the interior of the circle, discard this image; if adjacent frame images contain several different rods of a dot matrix cell but do not each contain a complete cell, use image stitching technology to stitch the adjacent image frames into the image data of the dot matrix cell in the vertical angle.

[0050] Specifically, as Figure 2 For the flowchart of obtaining the vertical dot matrix cell picture, use a drone equipped with a camera to scan the dot matrix structure along an S-shaped trajectory. The specific design of the S-shaped trajectory is as follows: Measure the overall physical size of the dot matrix structure part, where the dot matrix cells are printed according to the G-code setting and arranged equidistantly in sequence, such as ten rows and ten columns. Subsequently, according to the size (x, y, z) of the dot matrix cell and combined with the pinhole model, calculate the effective projection range, and then the length and spacing of the S-shaped trajectory can be obtained. For example, preset the flight height of the drone to be H, and know the focal length f of the drone camera and the size of the image sensor (width W and height H s )

[0051] The drone takes a vertical shot, and the distance between the image plane and the camera is the focal length f. Therefore, the position of the image plane:

[0052] Z = H - f,

[0053] According to the pinhole model, the projection of an object on the image plane can be calculated by the following formula:

[0054]

[0055] where (X, Y) are the coordinates of the lattice cell in the world coordinate system, and (x, y) are the coordinates of the lattice cell on the image plane.

[0056] Subsequently, the width W of the effective projection range is calculated p :

[0057]

[0058] where d is the diagonal length of the image sensor, FOV is the field of view of the camera, and W is the width of the image sensor.

[0059] Based on the drone sensor and flight control, ensure that the fuselage remains as horizontal as possible during the shooting process. For example, for a rotary-wing drone, the rotation speed can be measured by the gyroscope built into the drone to judge its attitude. If the readings of the gyroscope show that the drone is continuously rotating or tilting, it means that it is not in a horizontal attitude; when the drone is in a non-horizontal attitude, the lift is changed by adjusting the rotation speeds of multiple motors to change the rotation speeds of the propellers, so as to achieve the purpose of horizontal flight control. Based on this, the drone finally completes the shooting task in a state of following an S-shaped trajectory and maintaining a horizontal posture throughout the process.

[0060] Furthermore, identifying the image data of a single lattice cell includes:

[0061] Obtain several pieces of image data of the lattice cell taken at the directly overhead angle by scanning. Based on the image data of the lattice cell, mark them with labels, and divide them into a training set, a validation set, and a test set. Set the size of the input image, train the object detection neural network for lattice cell recognition, obtain the trained object detection neural network for lattice cell recognition, and identify all the image data of single lattice cells.

[0062] Specifically, computer vision technology is used to process the video stream data captured by the drone, and the video is intercepted frame by frame. First, it is analyzed whether a single-frame picture contains a complete lattice cell. The specific judgment process is as follows: First, the lattice cell is identified through the lattice cell recognition algorithm, and then, based on the object tracking algorithm of feature points, it is tracked in several consecutive frames to find the frame in which the center point of the cell is closest to the center point of the image, and the cell is intercepted for the image, combining the flight data of the drone with the sequential record number (the Xth row and the Yth column). Based on the above process, a picture containing a complete lattice cell is obtained, the starting arc positions of the four rods of the cell are obtained, and the center of the square formed by these four vertices is used as the calibration point. The pixel size of the image is converted into the physical size, and a circle is drawn within a certain range with the center of the image as the origin (such as a radius of 35 mm). If the calibration point belongs to the interior of the circle, the cell is considered to be an image taken directly from above, that is, an image without projective transformation; if the calibration point does not belong to the interior of the circle, the picture is discarded. As Figure 3 Schematic diagram for judging whether the calibration point is within the determination range.

[0063]

[0064] In the formula, O2 represents the calibration point, and S O1 represents a circle with a certain range centered on the center of the image as the radius.

[0065] Secondly, if several different rods of a lattice cell are included in adjacent frame pictures, but each does not contain a complete cell, image stitching technology is used to stitch these adjacent image frames into a complete lattice cell in the vertical angle. The detailed process of image stitching technology is as follows: First is feature detection, for example, detecting the feature points of the four rods of the lattice cell, and using the SIFT feature detection algorithm to extract key points and descriptors from each cell image. Subsequently, feature matching is performed, and by comparing the descriptors (such as using the lattice cell image in the vertical angle), the matching point pairs between different images are found. Common methods include brute-force matching and FLANN (Fast Library for Approximate Nearest Neighbors). In addition, the distance between the matching points needs to be calculated, generally obtained through the Euclidean distance:

[0066]

[0067] In the formula, x i and y i represent the coordinates of different matching points.

[0068] Subsequently, a ratio test (such as Lowe's ratio test) is used to screen the matching points to improve the accuracy of matching. Through the image transformation structure, image mapping is achieved. Then, the APAP algorithm is adopted to align the feature points of the lattice cell. The graph cut method is used to automatically select the appropriate cell images for stitching. Finally, the lattice cell image fusion is realized according to the multi-band blending strategy:

[0069] I fused (x,y) = α·I1(x,y)+(1 - α)·I2(x,y),

[0070] where I1 and I2 are the transformed image and the target image, and α is the fusion weight. The specific weight setting will vary according to the characteristics of the image.

[0071] Obtain a dataset of lattice cell images taken at a good vertical angle. To meet the requirements of the training dataset and improve the learning and generalization ability of the model, data augmentation methods such as changing brightness, contrast, saturation, and rotating pictures are used to expand the samples in the original dataset. Finally, multiple lattice cell images are obtained. In addition, labels are assigned to the images, and the training set, validation set, and test set are divided according to 7:2:1, and finally a complete dataset is obtained. The input image size is 500*500*3, or the input picture is resized to the same size. The small-sample lattice cell recognition object detection algorithm is used as the backbone network for training, such as Meta-DETR, UniT, etc., to obtain the optimal network hyperparameters. The trained lattice cell recognition object detection neural network is used to recognize all lattice cells from the image. The input is the image, and the output is the position of the lattice cells in the picture.

[0072] y = F(x),

[0073] where F represents the backbone neural network, such as the Meta-DETR model, x is the input image, and y is the detection result.

[0074] The process of converting pixel size to physical size: To ensure that the length of a certain area in the lattice cell images taken by cameras with different resolutions and pixels remains consistent physically, the physical size calibration method is adopted to unify the standard. First, the physical size needs to be calibrated. When taking pictures, a known-size object (such as a standard ruler) is used as the calibration object. Ensure that the calibrated lattice cells are clearly visible in the image and are on the same plane as the shooting object. Subsequently, calculate the relationship between pixels and physical size:

[0075] Physical length = Pixel length × Physical length per pixel.

[0076] It is necessary to further obtain the relevant feature information of the arc region at the top of the lattice cell rod to facilitate the quantification of the defective lattice. First, divide the obtained lattice cell image into the arc region at the top of the lattice cell rod, and also divide the training set, validation set, and test set according to 7:2:1. Finally, a complete data set is obtained. The input image size is 500*500*3, or the input picture can be resized to the same size. Train with the zero-shot segmentation algorithm as the backbone network, such as the FastSAM, ViTDet models, etc., to obtain the optimal network hyperparameters. The trained neural network is used to segment the arc region at the top of the lattice cell rod from the image. The input is the image, and the output is the arc region at the top of the lattice cell rod in the figure. For example, the arc region at the top of the lattice cell rod is assigned a value of 1, and the rest of the region is assigned a value of 0 to obtain the segmentation mask.

[0077] y = G(x),

[0078] where G represents the backbone neural network, such as the FastSAM model, x is the input image, and y is the segmentation result.

[0079] Furthermore, the segmentation of the arc regions at the tops of the four rods of the lattice cell includes:

[0080] Divide the lattice cell image data into the arc region at the top of the lattice cell rod, divide the training set, validation set, and test set, set the size of the input image, train the segmentation network for the arc region at the top of the lattice cell rod, obtain the trained segmentation network for the arc region at the top of the lattice cell rod, and segment the arc regions at the tops of the four rods of the lattice cell.

[0081] Furthermore, the extraction of cell feature information includes:

[0082] Process the segmented arc regions at the tops of the four rods of the lattice cell, perform binary processing and edge detection on the image, obtain the boundaries of the arc regions at the tops of the four rods, fit each arc region at the top of the rod into a standard circle, connect the centers of the four circles, and obtain the cell feature information of a quadrilateral.

[0083] Specifically, as Figure 4 is the flowchart for processing the arc region at the top of the rod using computer vision technology. The detailed processing process is as follows:

[0084] According to the segmentation mask, perform binary processing on the image to facilitate obtaining the boundaries of the arc regions at the tops of the four rods. The RGB values of each point in the image matrix are different. Binary processing is to set a threshold so that the RGB value of each point in the matrix becomes (0, 0, 0) [black] or (255, 255, 255) [white].

[0085] Edge detection is performed on the binary image. Edge detection aims to identify regions in the image where the brightness changes significantly, i.e., edges. Here, the edges correspond to the boundary contours of the circular arc regions at the top of the rods. The main steps include five processes: smoothing, gradient calculation, non-maximum suppression, double-threshold processing, and edge connection. Among them, Gaussian filtering is often used for smoothing, and the Gaussian function is:

[0086]

[0087] where σ is the standard deviation, which controls the degree of smoothing.

[0088] The Sobel operator is often used for gradient calculation, and it has convolution kernels in two directions:

[0089]

[0090] Calculate the gradient magnitude and direction:

[0091]

[0092] Each circular arc region at the top of the rod is fitted into a standard circle through the Hough circle detection algorithm. The Hough circle detection algorithm actually maps the points in the image space to the parameter space to identify the existence of a circle. Each edge point can be regarded as a candidate center of a circle, and the algorithm determines the final circle by accumulating the parameters of these candidate circles. For a circle, its parameters can be represented by three values: the center coordinates (a, b) and the radius r. Therefore, the parameter space of the Hough circle detection is three-dimensional, denoted as (a, b, r). Subsequently, a voting mechanism needs to be used to select the best approximate circle. For each edge point (x0, y0), the Hough circle detection algorithm calculates all possible combinations of the center and radius. For each possible center (a, b), it is calculated through the following formula:

[0093]

[0094] Add a vote for each (a, b, r) combination in the parameter space, indicating that this combination is a possible circle. A three-dimensional accumulator is used to record the vote count for each (a, b, r) combination. The higher the vote count, the greater the possibility of this circle.

[0095] For each edge point, update the accumulator:

[0096] accumulator[a][b][r]+=1,

[0097] By setting a threshold, find the point with the highest vote count in the accumulator, i.e., (a*, b*, r*) is the parameter of the detected best circle:

[0098] argmax_abr=(a* , b * , r * ),

[0099] Finally, four circles are obtained. Connect the centers of each circle to get a quadrilateral.

[0100] Furthermore, the determined quantization indexes include:

[0101] Index 1: The quadrilateral is a convex quadrilateral;

[0102] Index 2: Whether the ratio of the area of the quadrilateral to the diameter of the lattice rod is greater than 625;

[0103] Index 3: Whether the shortest side length of the quadrilateral is less than 5 mm;

[0104] Index 4: Whether the longest side length of the quadrilateral is greater than 40 mm;

[0105] Index 5: Whether the logarithm of the ratio of the longest side length to the shortest side length of the quadrilateral is greater than 0.9;

[0106] Index 6: Whether the diagonal of the quadrilateral is less than 5 mm;

[0107] Index 7: Whether the diagonal of the quadrilateral is greater than 60 mm;

[0108] Index 8: Whether the ratio of the two diagonals of the quadrilateral is 1.

[0109] Specifically, the judgment basis for whether the lattice is qualified and the type of defect includes eight lattice defect quantization indexes:

[0110] Index 1: Whether the quadrilateral is a convex quadrilateral;

[0111] Index 2: Whether the ratio of the area of the quadrilateral to the diameter of the lattice rod is greater than 625;

[0112]

[0113] Index 3: Whether the shortest side length of the quadrilateral is less than;

[0114] l min < 5 mm,

[0115] Index 4: Whether the longest side length of the quadrilateral is greater than;

[0116] l max > 40 mm,

[0117] Index 5: Whether the logarithm of the ratio of the longest side length to the shortest side length of the quadrilateral is greater than 0.9;

[0118] v l = log(l min / lmax ) > 0.7,

[0119] Index 6: Whether the diagonal of the quadrilateral is less than;

[0120] diag min < 5 mm,

[0121] Index 7: Whether the diagonal of the quadrilateral is greater than;

[0122] diag max > 60 mm,

[0123] Index 8: Whether the ratio of the two diagonals of the quadrilateral is 1;

[0124] v diag = diag1 / diag2 ≠ 1,

[0125] where diag1 and diag2 represent the two diagonals respectively.

[0126] Furthermore, identifying the defect type of the lattice cell image data includes:

[0127] Identifying the defect type of the lattice cell image data. If all eight indicators do not conform, it is determined as a normal cell; if indicator 1 is met and the other indicators do not conform, the defect type is determined as rod offset; if indicator 2 or 4 or 7 is met and the other indicators do not conform, the defect type is determined as rod missing; if indicator 3 or 5 or 6 is met and the other indicators do not conform, the defect type is determined as rod bonding; if indicator 8 is met and the other indicators do not conform, the defect type is determined as asymmetric lattice.

[0128] Specifically, combining the defect quantification indicators, classify the defect types of the lattice cells. Such as Figure 5 is the flowchart for classifying the lattice defect types. The specific definitions of the lattice defect types are as follows:

[0129] Rod offset: The lamination direction of some or certain rods in the lattice cell deviates significantly from the preset direction;

[0130] Rod missing: Some or certain rods in the lattice cell are broken;

[0131] Rod bonding: The arc regions at the tops of some rods in the lattice cell are bonded together;

[0132] Asymmetric lattice: The four rods of the lattice cell present an asymmetric shape.

[0133] The specific classification basis for the lattice defect types is as follows: If none of the above eight indicators are met, it is determined as a normal cell, and the cell state is assigned: s(i) = 0; If indicator 1 is met and the other indicators are not met, the defect type is determined as rod offset, and the cell state is assigned: s(i) = 1; If indicator 2 or 4 or 7 is met and the other indicators are not met, the defect type is determined as rod missing, and the cell state is assigned: s(i) = 2; If indicator 3 or 5 or 6 is met and the other indicators are not met, the defect type is determined as rod bonding, and the cell state is assigned: s(i) = 3; If indicator 8 is met and the other indicators are not met, the defect type is determined as an asymmetric lattice, and the cell state is assigned: s(i) = 4.

[0134] The present invention discloses a method for identifying the macro-morphology defect types of an additive manufacturing lattice structure. Although some existing technologies and researches can also identify some defects in 3D printing manufacturing products or manufacturing processes, the research on the macro-morphology problems of lattice structures is very scarce. Moreover, the present invention proposes specific quantitative indicators for determining whether a lattice cell is qualified and the lattice cell defect types, which is more scientific; The present invention combines the S-shaped trajectory scanning technology of an unmanned aerial vehicle to realize a method for accurately photographing vertical lattice images. Through steps such as scanning method, lattice cell identification, defect quantification, feature extraction, (feature extraction first and then defect quantification) defect classification, etc., the purpose of identifying lattice structure defects and quantifying defect types is achieved, effectively ensuring the printing quality and mechanical properties of large-scale metal lattices.

[0135] 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 identifying the types of macroscopic morphology defects of an additive manufacturing lattice structure, characterized in that Including: Obtaining dot cell image data of the vertical angle; Calibrating the dot cell image data and constructing a dot cell recognition data set to recognize individual dot cell image data; Segmenting the individual dot cell image data and constructing a dot cell top arc region segmentation data set to segment the top arc regions of the four rods of the dot cell; Processing the segmented top arc regions of the four rods of the dot cell to extract cell feature information; Quantifying the extracted cell feature information to determine quantization indicators; Based on the quantization indicators, identifying and classifying the defect types of the dot cell image data.

2. The method for identifying the macroscopic morphology defect types of the additive manufacturing lattice structure according to claim 1, wherein Obtaining dot cell image data of the vertical angle includes: Measuring the printed dot structure, scanning it in a horizontal attitude along a set S-shaped trajectory by a drone with a camera, and using computer vision technology to intercept video data frame by frame. If an image contains a complete dot cell, obtain the starting arc positions of the four rods of the cell, and use the center of the regular quadrilateral formed by the four vertices as the calibration point. With the center of the image as the origin and a preset radius size, if the calibration point belongs to the interior of the circle, then the cell image is an image taken directly from above, which is the dot cell image data of the vertical angle; if the calibration point does not belong to the interior of the circle, then discard this image; if adjacent frame images contain several different rods of a dot cell but do not each contain a complete cell, then use image stitching technology to stitch the adjacent image frames into dot cell image data of the vertical angle.

3. The method for identifying the macro-morphology defect types of the additive manufacturing lattice structure according to claim 1, wherein The design process of the S-shaped trajectory includes: Measuring the overall physical size of the dot structure member, where the dot cells are printed according to the settings of the G code and arranged equidistantly in sequence. According to the size of the dot cells and combined with the needle eye model, calculate the effective projection range to obtain the length and spacing of the S-shaped trajectory.

4. The method for identifying the types of macroscopic morphology defects of the additive manufacturing lattice structure according to claim 2, wherein Identifying individual dot cell image data includes: Obtaining several dot cell image data taken from the directly above angle through scanning. Based on the dot cell image data, mark them with labels, set the size of the input image, train the dot cell recognition target detection neural network, obtain the trained dot cell recognition target detection neural network, and identify all individual dot cell image data.

5. The method for identifying the macroscopic morphology defect types of the additive manufacturing lattice structure according to claim 1, wherein, Segmenting the top arc regions of the four rods of the dot cell includes: Dividing the dot cell image data into the top arc regions of the dot cell rods, setting the size of the input image, training the dot cell rod top arc region segmentation network, obtaining the trained dot cell rod top arc region segmentation network, and segmenting the top arc regions of the four rods of the dot cell.

6. The method for identifying the type of macroscopic morphology defects of the additive manufacturing lattice structure according to claim 1, wherein, Extracting cell feature information includes: Processing the segmented top arc regions of the four rods of the dot cell, performing binary processing and edge detection on the image to obtain the boundaries of the four rod top arc regions, fitting each rod top arc region into a standard circle, and connecting the centers of the four circles to obtain the cell feature information of a quadrilateral.

7. The method for identifying the macroscopic morphology defect types of the additive manufacturing lattice structure according to claim 1, wherein, Determining quantization indicators includes: Indicator 1 is that the quadrilateral is a convex quadrilateral; Indicator 2 is whether the ratio of the area of the quadrilateral to the diameter of the dot rod is greater than 625; Indicator 3 is whether the shortest side length of the quadrilateral is less than 5 mm; Indicator 4 is whether the longest side length of the quadrilateral is greater than 40 mm; Indicator 5 is whether the logarithm of the ratio of the longest side length to the shortest side length of the quadrilateral is greater than 0.9; Indicator 6 is whether the diagonal of the quadrilateral is less than 5 mm; Indicator 7 is whether the diagonal of the quadrilateral is greater than 60 mm; Indicator 8 is whether the ratio of the two diagonals of the quadrilateral is 1.

8. The method for identifying the macroscopic morphology defect types of the additive manufacturing lattice structure according to claim 7, characterized in that, Identifying the defect type of the lattice cell image data includes: Identifying the defect type of the lattice cell image data. If all eight indicators do not conform, it is determined as a normal cell; if indicator 1 is met and other indicators do not conform, the defect type is determined as rod offset; if indicator 2 or 4 or 7 is met and other indicators do not conform, the defect type is determined as rod missing; if indicator 3 or 5 or 6 is met and other indicators do not conform, the defect type is determined as rod bonding; if indicator 8 is met and other indicators do not conform, the defect type is determined as asymmetric lattice.

Images