A 3D printing method and device based on 3D point cloud repair model

Abstract

The present application is a 3D printing method and device based on a three-dimensional point cloud repair model, including: S1, scanning the original casting to obtain original point cloud data; S2, processing the original point cloud data to obtain three-dimensional printing data; S3, performing SLM printing according to the three-dimensional printing data. Among them, the present application can generate low-resolution original point cloud data into high-precision three-dimensional printing data in step S2. At the same time, the geometric features of the original workpiece are identified through the model, and the casting is divided into voxel data spliced ​​by basic voxel blocks. Only the basic voxel blocks are repaired, and then the repaired basic voxel blocks are spliced ​​into casting voxel data according to the geometric classification results, which can reduce the computing power requirements of the hardware for processing casting voxel data.

Description

Technical Field

[0001] The present application relates to the field of SLM casting technology, and in particular to a 3D printing method and device based on a three-dimensional point cloud repair model. Background Art

[0002] In the prior art, the selective laser melting technology (i.e., SLM) is widely used in 3D printing of metal castings. The SLM is to lay metal powder layer by layer according to the structural characteristics of the casting, melt the metal powder in the selected area by laser beam, and manufacture three-dimensional castings through condensation molding and layer-by-layer stacking. SLM is widely used in aerospace, medical devices and their implants, molds, and automobiles due to its ability to form complex and fine structures in an integrated manner.

[0003] In order to obtain high-resolution point cloud data, scanners used in existing SLM printing devices usually need to be equipped with structured light cameras and lidars. The hardware is expensive and the scanning time is long.

[0004] However, in low-end 3D printing equipment, the scanner can only obtain low-resolution point cloud data, and the printed castings need to cut off the incorrectly printed parts.

[0005] For low-resolution point cloud data, the printing accuracy can be improved by reconstructing the point cloud data into voxel data. Therefore, constructing high-resolution voxel data from existing low-resolution point cloud data is widely cited as an effective method.

[0006] However, voxel conversion also has the disadvantage of requiring high computing power of computing devices. In order to speed up the generation speed and accuracy of voxel data of 3D printed castings, the present application provides a 3D printing method based on a three-dimensional point cloud repair model. Summary of the invention

[0007] In order to overcome the problems existing in the related art, the first aspect of the present application provides a 3D printing method based on a 3D point cloud repair model, comprising:

[0008] S1, scan the original casting to obtain original point cloud data;

[0009] S2, processing the original point cloud data to obtain three-dimensional printing data;

[0010] S3, performing SLM printing according to the three-dimensional printing data;

[0011] The original point cloud data is processed to obtain three-dimensional printing data, specifically including:

[0012] S201, identifying the original point cloud data to obtain a type classification result and a geometric classification result corresponding to the original point cloud data;

[0013] S202, voxelizing the original point cloud data to obtain casting voxel data;

[0014] S203, obtaining a block division strategy according to the geometric classification result;

[0015] The block partitioning strategies include: axisymmetric partitioning, rotational symmetric partitioning, spiral symmetric partitioning and translational symmetric partitioning;

[0016] S204, dividing the casting voxel data into basic voxel blocks according to the block division strategy; the basic voxel block is the smallest unit constituting the casting voxel data;

[0017] S205 , extracting features from the basic voxel block, and repairing the basic voxel block using a point cloud repair model to obtain three-dimensional printing data.

[0018] In one implementation, before S201, the process further includes: converting the format of the original point cloud data.

[0019] In one implementation, S201 specifically includes:

[0020] The original point cloud data is identified based on a point cloud recognition model, and a type classification result and a geometric classification result of the original point cloud data are output. The point cloud recognition model is a neural network model pre-trained by a point cloud data set.

[0021] In one implementation, S202 specifically includes:

[0022] S301, defining voxel size;

[0023] S302, creating a voxel grid;

[0024] S303, filling the voxel grid.

[0025] In one implementation, S205 specifically includes:

[0026] S401, extracting contour features of the basic voxel block through an encoder; the contour features include contour feature codes of the basic voxel block in six viewing directions;

[0027] S402, inputting the contour feature codes in six viewing directions into the point cloud repair model to perform contour correction;

[0028] S403, generating the restored basic voxel block through a decoder;

[0029] S404: Convert the basic voxel block into three-dimensional printing data.

[0030] In one implementation, before S205, the method further includes:

[0031] S501, obtaining a casting voxel training set;

[0032] S502, constructing a point cloud restoration model and defining a fusion loss function; the fusion loss function includes a contour loss function and a semantic loss function;

[0033] S503, configuring training parameters;

[0034] S504: Input the casting voxel training set into the point cloud repair model for gradient training.

[0035] In one implementation, the fusion loss function is:

[0036]

[0037] in, is the contour loss function, is the semantic loss function, is the fusion loss function, To adjust the weight;

[0038] In one embodiment, the contour loss function is:

[0039]

[0040] in, is the Euclidean distance, is the corrected contour feature vector of the i-th view direction, is the true label of the contour in the i-th view direction, .

[0041] In one implementation, the semantic loss function is:

[0042]

[0043] is the probability that the three-dimensional printing data belongs to the current classification result, and c is the total number of classification results.

[0044] A second aspect of the present application provides a 3D printing device based on a three-dimensional point cloud repair model, comprising: a scanning unit, a point cloud processing unit, and a casting printing unit;

[0045] The scanning unit is used to scan the original casting to obtain original point cloud data; the point cloud processing unit is used to process the original point cloud data to obtain three-dimensional printing data; and the casting printing unit is used to perform SLM printing according to the three-dimensional printing data.

[0046] The technical solution provided by this application may have the following beneficial effects:

[0047] After identifying the geometric features of the original workpiece through a point cloud recognition model, the present application divides the casting into voxel data composed of basic voxel blocks, only repairs the basic voxel blocks, and then splices the repaired basic voxel blocks into casting voxel data based on the geometric classification results, which can reduce the computing power requirements of the hardware for processing the casting voxel data.

[0048] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and other objects, features and advantages of the present application will become more apparent through a more detailed description of exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0050] Figure 1 A schematic diagram of a 3D printing method according to an embodiment of the present application;

[0051] Figure 2 for Figure 1 A schematic flow chart of step S2 in the 3D printing method shown;

[0052] Figure 3 A three-dimensional schematic diagram of a 3D printed casting shown in an embodiment of the present application;

[0053] Figure 4 A three-dimensional schematic diagram of a basic voxel block shown in an embodiment of the present application;

[0054] Figure 5 A network structure diagram of a point cloud repair model shown in an embodiment of the present application. DETAILED DESCRIPTION

[0055] The preferred embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0056] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0057] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0058] Embodiment 1

[0059] Before SLM printing, it is necessary to first obtain the point cloud data of the casting. The higher the resolution of the point cloud data, the higher the accuracy during printing.

[0060] However, due to the limitations of the accuracy of the scanning device and the shooting environment during scanning, the acquired point cloud data does not meet the requirements of high resolution.

[0061] Point clouds collected by LiDAR, especially those from outdoor scenes, suffer from different kinds of quality issues such as noise, outliers, and missing points. So our goal is to complete the missing points and remove outliers.

[0062] In order to improve the accuracy of SLM printing, the present application embodiment provides a 3D printing method based on a three-dimensional point cloud repair model, such as Figure 1 As shown, including:

[0063] S1, scan the original casting to obtain original point cloud data;

[0064] S2, processing the original point cloud data to obtain three-dimensional printing data;

[0065] S3, performing SLM printing according to the three-dimensional printing data;

[0066] Among them, Figure 2 The steps shown in S2 specifically include:

[0067] S201, identifying the original point cloud data to obtain a type classification result and a geometric classification result corresponding to the original point cloud data;

[0068] Furthermore, before S201, the method further includes: converting the format of the original point cloud data.

[0069] Specifically, the original point cloud data is recognized based on a point cloud recognition model to obtain classification features of the original point cloud data; the point cloud recognition model is a CLIP model, which is a comparative language-image pre-training model.

[0070] In the embodiment of the present application, the CLIP model is a neural network model pre-trained for identifying point cloud data. In the classification task of point cloud data, the CLIP model can use the category name as a prompt to compare the point cloud data vector with the text vector to achieve classification.

[0071] During training, the CLIP model combines the point cloud data and text data embedding models, maps the text data and point cloud data into the same vector space, and learns the correlation between the point cloud feature vector and the text feature vector, thereby achieving type classification results and geometric classification results for the original point cloud data.

[0072] For example, if the point cloud data of a gear is input into the CLIP model, the CLIP model can output the classification result of the point cloud data as "gear" and the geometric classification result of the gear. The geometric classification result is the symmetrical structure of the gear.

[0073] S202, voxelizing the original point cloud data to obtain casting voxel data;

[0074] Specifically, the voxelization process is to convert point cloud data into voxel data. Point cloud is a discrete point distributed in the range space. By scanning the casting with laser radar, the shape of the casting is displayed in the three-dimensional space in the form of point cloud.

[0075] However, if you want to print high-precision castings, the accuracy and discreteness of point cloud data do not meet the requirements. Therefore, it is usually necessary to convert point cloud data into voxel data.

[0076] In the embodiment of the present application, the voxelization of the point cloud is to convert the point cloud into voxels. Voxels are the smallest units in three-dimensional space. When converting the point cloud, the three-dimensional space is represented by a voxel grid. If a corresponding point cloud exists in a voxel grid, the voxel grid is filled and the casting modeling is gradually performed.

[0077] Therefore, the size of the voxel grid also determines the accuracy of the casting. The smaller the voxel grid, the more detailed the shape contour of the casting, and the clearer the modeling of the casting.

[0078] Specifically, S202 includes the following steps:

[0079] S301, defining voxel size;

[0080] In the embodiment of the present application, the size of the voxel is first determined, which will determine the resolution of the voxel grid.

[0081] S302, creating a voxel grid;

[0082] Create a regular voxel grid within the spatial extent of the point cloud data. Each voxel can be considered as a small cube with a side length equal to the previously defined voxel size.

[0083] S303, filling the voxel grid.

[0084] Specifically, assign points to voxels: assign each point in the point cloud to a corresponding voxel. This is usually done based on the point's position coordinates.

[0085] Therefore, when the resolution of the voxel grid is constant, the accuracy of the point cloud data directly determines the accuracy of the casting voxel data.

[0086] In order to obtain high-resolution point cloud data, scanners used in existing SLM printing devices usually need to be equipped with structured light cameras and lidars. The hardware is expensive and the scanning time is long.

[0087] However, in low-end 3D printing equipment, the scanner can only obtain low-resolution point cloud data, and the printed castings need to cut off the incorrectly printed parts.

[0088] Therefore, in order to directly print out precise metal castings, after obtaining the point cloud data of the original, the embodiment of the present application first converts the data into a voxel data format, and then repairs the voxel data.

[0089] S203, obtaining a block division strategy according to the geometric classification result;

[0090] Specifically, the block division strategies include: axial symmetry division, rotational symmetry division, spiral symmetry division and translational symmetry division.

[0091] In the embodiment of the present application, the geometric features of the voxel data are first determined, and then the corresponding block division strategy is searched.

[0092] For example, Figure 3 and Figure 4 The cast flange ring shown, Figure 3 is the three-dimensional image of the original workpiece, Figure 4 Schematic diagram of the basic voxel block of the flange ring.

[0093] In the embodiment of the present application, outputting high-resolution voxel data or point cloud data requires high computing power of the hardware device.

[0094] The embodiment of the present application divides the voxel data into blocks through step S203. After the geometric features of the original workpiece are identified through the CLIP model, the casting voxel data is divided into voxel data spliced ​​by basic voxel blocks, and only the basic voxel blocks are repaired. Then, the repaired basic voxel blocks are spliced ​​into casting voxel data according to the geometric classification results.

[0095] The embodiments of the present application can reduce the computing power requirements of the hardware for processing voxel data.

[0096] S204, dividing the casting voxel data into basic voxel blocks according to the block division strategy; the basic voxel block is the smallest unit constituting the casting voxel data.

[0097] Specifically, from Figure 3 and Figure 4 It can be seen that the geometric feature of the flange ring is rotational symmetry, that is, on the horizontal cross section of the flange ring, the basic voxel block coincides with other voxel blocks after rotating around the center point on the horizontal plane.

[0098] In the embodiment of the present application, the flange ring is divided according to the rotational symmetry division strategy to obtain Figure 3 The basic voxel blocks shown.

[0099] S205, extracting features from the basic voxel block, and repairing the basic voxel block using a point cloud repair model to obtain three-dimensional printing data;

[0100] Specifically, the point cloud repair module is a neural network model pre-trained by a training set.

[0101] Furthermore, S205 includes:

[0102] S401 . Extract contour features of the basic voxel block through an encoder; the contour features include contour feature codes of the basic voxel block in six viewing directions.

[0103] S402, inputting the contour feature codes in six viewing directions into the point cloud repair model to perform contour correction;

[0104] S403, generating the restored basic voxel block through a decoder;

[0105] S404: Convert the basic voxel block into three-dimensional printing data.

[0106] like Figure 5As shown, the cube in the figure represents the basic voxel block before input restoration, and (H, W, C)x6 represents six views of the basic voxel block, namely the front view, top view, back view, bottom view, left view and right view. It represents the contour feature vector extracted from each view, and the output represents the voxel data of the repaired casting.

[0107] In the embodiment of the present application, the projection images (H, W, C) of the basic voxel block in six viewing directions are first obtained. Figure 5 In the figure, (H, W, C)x6 represents 6 three-channel images.

[0108] Then the 6 projection images are input into the encoder for contour feature extraction to obtain the contour feature vector .exist Figure 5 middle, to Represents the contour feature vectors on the 6 projection images.

[0109] The decoder performs contour feature vector Repair, and then the repaired contour feature vector Generate the restored basic voxel blocks. Finally, splice the basic voxel blocks according to the geometric classification results to produce 3D printing data (i.e., the restored casting voxel data).

[0110] In the training phase, it is also necessary to configure training parameters and define loss functions. Furthermore, the training steps of the point cloud restoration model include:

[0111] S501, obtaining a casting voxel training set;

[0112] S502, constructing a point cloud restoration model and defining a fusion loss function; the fusion loss function includes a contour loss function and a semantic loss function;

[0113] S503, configuring training parameters;

[0114] S504: Input the casting voxel training set into the point cloud repair model for gradient training.

[0115] The fusion loss function is:

[0116]

[0117] in, is the contour loss function, is the semantic loss function, is the fusion loss function, To adjust the weight;

[0118] The contour loss function is:

[0119]

[0120] in, is the Euclidean distance, After the correction i The contour feature vector in each view direction, is the true label of the contour in the i-th view direction, . N represents an integer.

[0121] In the embodiment of the present application, by extracting projection images of six viewing directions of voxel data, extracting contour features at the input encoder, and generating new voxel data through the decoder and hidden layer, contour restoration is performed on the basic voxel block in six viewing directions.

[0122] Specifically, the semantic loss function is:

[0123]

[0124] is the probability that the three-dimensional printing data belongs to the current geometric classification result, and c is the total number of classification results.

[0125] The probability of the 3D printing data generated by the encoder being located in each geometric classification result within the range of 1 to c classification results is calculated respectively. If it is determined that the geometric classification result of the 3D printing data is inconsistent with the type classification result of the original point cloud data, it is judged that the restored voxel data has excessive deformation.

[0126] In the embodiment of the present application, the loss function of the training point cloud restoration model is a fusion function, including contour loss and semantic loss. The contour loss is used to judge the distance between the shape restoration of the basic voxel block and the true label to optimize the point cloud restoration model, and the gradient feedback of the semantic loss function is used to limit the shape restoration process of the basic voxel block.

[0127] Embodiment 2

[0128] A 3D printing device based on a three-dimensional point cloud repair model comprises: a scanning unit, a point cloud processing unit and a casting printing unit;

[0129] The scanning unit is used to scan the original casting to obtain original point cloud data;

[0130] The point cloud processing unit is used to process the original point cloud data to obtain three-dimensional printing data;

[0131] The casting printing unit is used to perform SLM printing according to the three-dimensional printing data.

[0132] Regarding the device in the above embodiment, the specific way in which each module performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0133] The scheme of the present application has been described in detail above with reference to the accompanying drawings. In the above embodiments, the description of each embodiment has its own emphasis. For the part that is not described in detail in a certain embodiment, refer to the relevant description of other embodiments. Those skilled in the art should also know that the actions and modules involved in the description are not necessarily required for the present application. In addition, it can be understood that the steps in the method of the embodiment of the present application can be adjusted in order, merged and deleted according to actual needs, and the modules in the device of the embodiment of the present application can be merged, divided and deleted according to actual needs.

[0134] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0135] Alternatively, the present application can also be implemented as a non-temporary machine-readable storage medium (or computer-readable storage medium, or machine-readable storage medium) on which executable code (or computer program, or computer instruction code) is stored. When the executable code (or computer program, or computer instruction code) is executed by a processor of an electronic device (or electronic device, server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0136] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the application herein may be implemented as electronic hardware, computer software, or combinations of both.

[0137] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system and method according to multiple embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a part of a module, a program segment or a code, and the part of the module, the program segment or the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or the flow chart, and the combination of the square boxes in the block diagram and / or the flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0138] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A 3D printing method based on 3D point cloud repair model, It is characterized in that include: S1, scan the original casting to obtain original point cloud data; S2, processing the original point cloud data to obtain three-dimensional printing data; S3, a casting printing unit is used to perform SLM printing according to the three-dimensional printing data; Among them, S2 specifically includes: S201, identifying the original point cloud data to obtain a type classification result and a geometric classification result corresponding to the original point cloud data; S202, voxelizing the original point cloud data to obtain casting voxel data; S203, obtaining a block division strategy according to the geometric classification result; The block partitioning strategies include: axisymmetric partitioning, rotational symmetric partitioning, spiral symmetric partitioning and translational symmetric partitioning; S204, dividing the casting voxel data into basic voxel blocks according to the block division strategy; the basic voxel block is the smallest unit constituting the casting voxel data; S205 , extracting features from the basic voxel block, and repairing the basic voxel block using a point cloud repair model to obtain three-dimensional printing data.

2. A 3D printing method based on a 3D point cloud repair model according to claim 1, It is characterized in that Prior to S201, it also included: The format of the original point cloud data is converted.

3. A 3D printing method based on a 3D point cloud repair model according to claim 1, It is characterized in that S201 specifically includes: The original point cloud data is identified based on a point cloud recognition model, and a type classification result and a geometric classification result of the original point cloud data are output. The point cloud recognition model is a neural network model pre-trained by a point cloud data set.

4. The 3D printing method based on the three-dimensional point cloud repair model according to claim 1, It is characterized in that S202 specifically includes: S301, defining voxel size; S302, creating a voxel grid; S303, filling the voxel grid.

5. The 3D printing method based on the 3D point cloud repair model according to claim 1, It is characterized in that S205 specifically includes: S401, extracting contour features of the basic voxel block through an encoder; the contour features include contour feature codes of the basic voxel block in six viewing directions; S402, inputting the contour feature codes in six viewing directions into the point cloud repair model to perform contour correction; S403, generating the restored basic voxel block through a decoder; S404: Convert the basic voxel block into three-dimensional printing data.

6. A 3D printing method based on a 3D point cloud repair model according to claim 5, It is characterized in that Prior to S205, it also included: S501, obtaining a casting voxel training set; S502, constructing a point cloud restoration model and defining a fusion loss function; the fusion loss function includes a contour loss function and a semantic loss function; S503, configuring training parameters; S504: Input the casting voxel training set into the point cloud repair model for gradient training.

7. A 3D printing method based on a 3D point cloud repair model according to claim 6, It is characterized in that The fusion loss function is: in, is the contour loss function, is the semantic loss function, is the fusion loss function, To adjust the weight.

8. A 3D printing method based on a 3D point cloud repair model according to claim 7, It is characterized in that The contour loss function is: in, is the Euclidean distance, is the corrected contour feature vector of the i-th view direction, is the true label of the contour in the i-th view direction, .

9. The 3D printing method based on the three-dimensional point cloud repair model according to claim 7, It is characterized in that The semantic loss function is: is the probability that the three-dimensional printing data belongs to the current geometric classification result, and c is the total number of classification results.

10. A 3D printing device based on a 3D point cloud repair model. It is characterized in that Used to perform the 3D printing method according to any one of claims 1 to 9, comprising: a scanning unit, a point cloud processing unit and a casting printing unit; The scanning unit is used to scan the original casting to obtain original point cloud data; The point cloud processing unit is used to process the original point cloud data to obtain three-dimensional printing data; The casting printing unit is used to perform SLM printing according to the three-dimensional printing data.

Images