Aircraft part three-dimensional measurement point cloud reconstruction CAD method based on autoregression model

Abstract

The invention relates to the technical field of three-dimensional point cloud reconstruction, solves the technical problem that a reconstruction result is inaccurate or inconsistent due to the fact that a traditional reconstruction method depends on manual operation or a rule-based algorithm to generate a CAD command sequence, and particularly relates to an autoregression model-based aircraft part three-dimensional measurement point cloud reconstruction CAD method. Comprising the following steps: constructing a three-dimensional measurement point cloud reconstruction CAD model; training and testing the three-dimensional measurement point cloud reconstruction CAD model; and inputting the three-dimensional measurement point cloud data into the three-dimensional measurement point cloud reconstruction CAD model, and reconstructing a three-dimensional model of the aircraft part according to the predicted CAD command marking sequence. According to the method, the measurement point cloud which is easy to obtain can be used as input, the CAD command sequence of the part is predicted by using the autoregression model, accurate and automatic reconstruction of the aircraft part is realized, and the reconstruction quality of the aircraft part is improved.

Description

Technical Field

[0001] The invention relates to the technical field of three-dimensional point cloud reconstruction, and in particular to a CAD method for three-dimensional measurement point cloud reconstruction of aircraft parts based on an autoregressive model. Background Art

[0002] The reconstruction of point cloud for 3D measurement of aircraft parts is of great significance in the field of aviation manufacturing and maintenance. First of all, through 3D measurement technology, the geometric shape and size information of the surface of aircraft parts can be accurately obtained, which is crucial to ensure the manufacturing accuracy and assembly quality of parts. By converting these measurement data into 3D models, designers can intuitively see the 3D form of parts, which helps to conduct more in-depth analysis and optimize the design. By comparing the actual parts with the design model, errors in the manufacturing process can be found and corrected. Point cloud reconstruction technology can help detect the wear and damage of aircraft parts, improve maintenance efficiency and accuracy, and promptly detect potential defects such as cracks and dents, thereby avoiding safety hazards.

[0003] In summary, the 3D measurement point cloud reconstruction technology for aircraft parts is of great significance in improving manufacturing accuracy, optimizing design and maintenance efficiency, and is an indispensable technology for the aviation industry. However, traditional methods for reconstructing aircraft parts may rely on complex or difficult-to-obtain data sources, such as high-precision 3D scanning equipment or professional measurement services, which may be costly or complicated to operate. In addition, the reconstruction process relies on manual operations or rule-based algorithms to generate CAD command sequences, which may be inefficient and error-prone, and involve multiple tedious manual steps, such as data preprocessing, model repair and optimization, which may lead to inaccurate or inconsistent reconstruction results, and cannot meet application requirements in terms of timeliness. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a CAD method for reconstructing three-dimensional measured point clouds of aircraft parts based on an autoregressive model, which solves the technical problem that traditional reconstruction methods rely on manual operations or rule-based algorithms to generate CAD command sequences, resulting in inaccurate or inconsistent reconstruction results.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts based on an autoregressive model, the method comprising the following steps: S1. Obtain the three-dimensional measurement point cloud data of any aircraft component and perform preprocessing; S2, constructing a CAD model of a 3D measured point cloud reconstructing a 3D model of an aircraft component according to the predicted CAD command tag sequence based on an autoregressive model; S3, obtaining the true CAD values ​​of aircraft parts, and using the true CAD values ​​of aircraft parts to train and test the CAD model reconstructed from the three-dimensional measured point cloud; S4, inputting the three-dimensional measurement point cloud data into the three-dimensional measurement point cloud reconstruction CAD model, and reconstructing the three-dimensional model of the aircraft parts according to the predicted CAD command mark sequence.

[0006] Furthermore, in step S1, the specific process includes the following steps: S11. Use MetraSCAN scanner to collect 3D measurement point cloud data of the surface of aircraft parts; S12, segmenting the three-dimensional measurement point cloud data corresponding to the aircraft parts with complex structures into a plurality of parts point cloud data using a point cloud segmentation algorithm; S13. Use a point cloud completion algorithm to complete the hole area of ​​the point cloud data of each component.

[0007] Further, the three-dimensional measurement point cloud reconstruction CAD model includes a point cloud feature sequence extraction module for extracting a point cloud feature sequence from the three-dimensional measurement point cloud data, and a point cloud feature adapter for further processing and converting the point cloud feature sequence; The method also includes a CAD command tag building module for mapping the parameters of each CAD command to a tag space according to the CAD command sequence of the CAD true value of the aircraft parts, and obtaining the CAD command tag sequence, and a CAD command tag embedding module for extracting the embedding of the CAD command tag sequence and obtaining the CAD command tag embedding sequence, and, The point cloud feature sequence and the CAD command marker embedding sequence are spliced ​​to obtain a position encoding module of the point cloud feature-CAD command marker embedding sequence with position encoding, an autoregressive model of the CAD command marker sequence predicted according to the output of the point cloud feature-CAD command marker embedding sequence, and a CAD command reconstruction module that generates a CAD command sequence according to the CAD command marker sequence and constructs an aircraft parts CAD.

[0008] Furthermore, the CAD command mark embedding module is composed of an embedding layer, the position encoding module is composed of a position encoding layer, and the autoregressive model is composed of a Transformer decoder.

[0009] Furthermore, in step S3, the CAD model reconstructed from the three-dimensional measurement point cloud is trained, and the specific process includes the following steps: S311, inputting the CAD command sequence of the CAD true value of the aircraft parts into the CAD command tag construction module, mapping the parameters of each CAD command to the tag space, obtaining the tag corresponding to each CAD command, the tags of all CAD commands constitute a CAD command tag sequence, and adding fixed start tags and end tags at the beginning and end of the sequence respectively; S312, inputting the CAD command tag sequence into the CAD command tag embedding module to extract the embedding of all tags and obtain the CAD command tag embedding sequence; S313, concatenating the point cloud feature sequence and the CAD command mark embedding sequence and inputting them into a position coding module to obtain a point cloud feature-CAD command mark embedding sequence with position coding; S314, input the point cloud feature with position encoding-CAD command label embedding sequence into the autoregressive model, output the predicted CAD command label sequence, and supervise the predicted CAD command label sequence using the cross entropy loss function until convergence, thereby completing the training of the autoregressive model.

[0010] Furthermore, in step S3, the CAD model reconstructed from the three-dimensional measurement point cloud is tested, and the specific process includes the following steps: S321, the initial CAD command tag sequence only includes the start tag; S322, inputting the CAD command tag sequence into the CAD command tag embedding module to extract the embedding of all tags and obtain the CAD command tag embedding sequence; S323, concatenating the point cloud feature sequence and the CAD command mark embedding sequence and inputting them into a position coding module to obtain a point cloud feature-CAD command mark embedding sequence with position coding; S324, inputting the point cloud feature with position coding-CAD command tag embedding sequence into the trained autoregressive model, predicting the next CAD command tag, and concatenating the CAD command tag sequence with the predicted next CAD command tag; S325. Repeat steps S322-S324 until a CAD command mark sequence of a specified length is obtained.

[0011] Furthermore, in step S4, the specific process includes the following steps: S41, intercepting the mark between the start mark and the end mark from the predicted CAD command mark sequence as a valid CAD command mark sequence; S42, inputting the valid CAD command tag sequence into the CAD command reconstruction module, mapping each tag to a parameter of the CAD command, obtaining the CAD command corresponding to each tag, and all CAD commands constitute a CAD command sequence; S443. According to the CAD command sequence, construct the aircraft parts CAD according to the commands.

[0012] By means of the above technical solution, the present invention provides a CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts based on an autoregressive model, which has at least the following beneficial effects: 1. The present invention uses easily accessible measurement point clouds as input, which solves the problem of difficulty in obtaining data sources and reduces the cost and complexity of data acquisition. By introducing an autoregressive model, it can automatically predict and generate CAD command sequences that meet design requirements, thereby improving the degree of automation and accuracy. In addition, through the automated reconstruction process, it reduces the reliance on manual operations and improves the accuracy and consistency of reconstruction.

[0013] 2. The present invention generates a high-precision 3D model based on a large amount of point cloud data, accurately reflects the shape and structural characteristics of the object, and automatically generates a high-precision 3D model, greatly improving the accuracy and efficiency of mapping. It can be conveniently used in subsequent design optimization, process planning, production and processing, etc., to achieve close coordination between design and manufacturing, and greatly shorten the product design and manufacturing cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 It is a flow chart of the CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts in the present invention; Figure 2 This is a network structure diagram of the CAD model reconstructed from the three-dimensional measurement point cloud in the present invention. DETAILED DESCRIPTION

[0015] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0016] Please refer to Figure 1-Figure 2 This embodiment proposes a CAD method for reconstructing three-dimensional measured point clouds of aircraft parts based on an autoregressive model, which can use easily obtained measured point clouds as input, use the autoregressive model to predict the CAD command sequence of the parts, and achieve accurate and automated reconstruction of aircraft parts, thereby improving the reconstruction quality of aircraft parts. Figure 1 As shown, the method comprises the following steps: S1. Obtain the 3D measurement point cloud data of any aircraft component and perform preprocessing. The specific process includes the following steps: S11. Use MetraSCAN scanner to collect 3D measurement point cloud data of the surface of aircraft parts; S12. Use a point cloud segmentation algorithm to segment the three-dimensional measured point cloud data corresponding to the aircraft parts with complex structures into multiple parts point cloud data. The point cloud segmentation algorithm can be a RANSAC algorithm (Random Sample Consensus) or a Region Growing algorithm (Region Growing).

[0017] S13. Use point cloud completion algorithms to complete the hole areas of each component’s point cloud data. The PointNet algorithm can process unknown point clouds through symmetric equations, or the PointNet++ algorithm can learn multi-scale point cloud feature information by hierarchically segmenting multiple local point clouds, or the DGCNN algorithm can efficiently extract point cloud features using edge-related computational operations in graph theory, or the VRCNet algorithm can enhance structural relationships to complete point cloud completion.

[0018] S2, based on the autoregressive model, a 3D measured point cloud is constructed to reconstruct a CAD model of the 3D model of the aircraft parts according to the predicted CAD command tag sequence. Figure 2 As shown in FIG. 1 , the CAD model reconstructed from the 3D measured point cloud includes a point cloud feature sequence extraction module for extracting point cloud feature sequences from the 3D measured point cloud data, and a point cloud feature adapter for further processing and converting the point cloud feature sequences. The function of the point cloud adapter is mainly to further process and convert the point cloud features to meet the needs of subsequent tasks or models. Specifically, the point cloud feature adapter consists of a fully connected layer and layer normalization, and its functions can be summarized as follows: Feature transformation: The fully connected layer can perform linear transformation on the input point cloud features to extract higher-level feature representations. This transformation helps capture complex patterns and relationships in point cloud data and provide more useful feature information for subsequent tasks.

[0019] Feature fusion: In some cases, point cloud feature adapters may be used to fuse features from different sources or different levels. Through the transformation of the fully connected layer, these features can be combined to form a comprehensive feature representation to better describe the point cloud data.

[0020] Layer normalization: Layer normalization is a technique used to accelerate the neural network training process and improve the model stability. It normalizes the input of each layer so that the input of each layer remains within a relatively stable range. This helps to reduce the gradient vanishing or gradient exploding problems of the model during training, thereby improving the convergence speed and performance of the model.

[0021] Feature Adaptation: One of the main functions of the point cloud feature adapter is to perform feature adaptation. Since different tasks or models may have different requirements for features, it is necessary to properly adjust the features through the adapter. The combination of fully connected layers and layer normalization can flexibly adjust the spatial distribution and scale of features to make them more suitable for subsequent tasks or models.

[0022] The three-dimensional measurement point cloud reconstruction CAD model proposed in this embodiment also includes a CAD command tag construction module for mapping the parameters of each CAD command to the tag space according to the CAD command sequence of the true value of the CAD of the aircraft parts, and obtaining the CAD command tag sequence; a CAD command tag embedding module for extracting the embedding of the CAD command tag sequence and obtaining the CAD command tag embedding sequence; a position encoding module for splicing the point cloud feature sequence and the CAD command tag embedding sequence to obtain a point cloud feature-CAD command tag embedding sequence with position encoding; an autoregressive model for outputting a predicted CAD command tag sequence based on the point cloud feature-CAD command tag embedding sequence; and a CAD command reconstruction module for generating a CAD command sequence according to the CAD command tag sequence and constructing the CAD of the aircraft parts.

[0023] S3, obtain the true CAD value of the aircraft parts, and use the true CAD value of the aircraft parts to train and test the CAD model reconstructed from the three-dimensional measurement point cloud; in this embodiment, the true CAD value of the aircraft parts represents the standard CAD three-dimensional model of the aircraft parts in the original design, and the acquisition method can be directly obtained from the local standard model library of the original design or the manufacturer. Specifically, the training of the CAD model reconstructed from the three-dimensional measurement point cloud includes the following steps: S311. Input the CAD command sequence of the CAD true value of the aircraft parts into the CAD command tag construction module, map the parameters of each CAD command to the tag space, obtain the tag corresponding to each CAD command, and the tags of all CAD commands constitute a CAD command tag sequence, and add fixed start tags and end tags at the beginning and end of the sequence respectively.

[0024] S312, inputting the CAD command tag sequence into the CAD command tag embedding module to extract the embedding of all tags and obtain the CAD command tag embedding sequence, the CAD command tag embedding module is composed of embedding layers.

[0025] S313, concatenating the point cloud feature sequence and the CAD command mark embedding sequence and inputting them into a position coding module to obtain a point cloud feature-CAD command mark embedding sequence with position coding. The position coding module is composed of a position coding layer.

[0026] S314. Input the point cloud feature with position encoding-CAD command label embedding sequence into the autoregressive model, output the predicted CAD command label sequence, and supervise the predicted CAD command label sequence using the cross entropy loss function until convergence, thereby completing the training of the autoregressive model, which is composed of the decoder of the Transformer.

[0027] Furthermore, the CAD model reconstructed from the 3D measured point cloud is tested. The specific process includes the following steps: S321, the initial CAD command tag sequence only includes the start tag; S322, inputting the CAD command tag sequence into the CAD command tag embedding module to extract the embedding of all tags and obtain the CAD command tag embedding sequence; S323, concatenating the point cloud feature sequence and the CAD command mark embedding sequence and inputting them into a position coding module to obtain a point cloud feature-CAD command mark embedding sequence with position coding; S324, inputting the point cloud feature with position coding-CAD command tag embedding sequence into the trained autoregressive model, predicting the next CAD command tag, and concatenating the CAD command tag sequence with the predicted next CAD command tag; S325. Repeat steps S322-S324 until a CAD command mark sequence of a specified length is obtained.

[0028] S4, inputting the 3D measured point cloud data into the 3D measured point cloud reconstruction CAD model, and reconstructing the 3D model of the aircraft parts according to the predicted CAD command mark sequence. The specific process includes the following steps: S41, intercepting the mark between the start mark and the end mark from the predicted CAD command mark sequence as a valid CAD command mark sequence; S42, inputting the valid CAD command tag sequence into the CAD command reconstruction module, mapping each tag to a parameter of the CAD command, obtaining the CAD command corresponding to each tag, and all CAD commands constitute a CAD command sequence; S443. According to the CAD command sequence, construct the aircraft parts CAD according to the commands.

[0029] The present invention solves the problem of difficulty in obtaining data sources by using easily accessible measurement point clouds as input, and reduces the cost and complexity of data acquisition. By introducing an autoregressive model, it can automatically predict and generate CAD command sequences that meet design requirements, thereby improving the degree of automation and accuracy. In addition, through the automated reconstruction process, it reduces the reliance on manual operations and improves the accuracy and consistency of reconstruction.

[0030] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, so the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0031] The above implementation methods have been described in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts based on an autoregressive model, characterized in that: The method comprises the following steps: S1. Obtain the three-dimensional measurement point cloud data of any aircraft component and perform preprocessing; S2, constructing a CAD model of a 3D measured point cloud reconstructing a 3D model of an aircraft component according to the predicted CAD command tag sequence based on an autoregressive model; S3, obtaining the true CAD values ​​of aircraft parts, and using the true CAD values ​​of aircraft parts to train and test the CAD model reconstructed from the three-dimensional measured point cloud; S4, inputting the three-dimensional measurement point cloud data into the three-dimensional measurement point cloud reconstruction CAD model, and reconstructing the three-dimensional model of the aircraft parts according to the predicted CAD command mark sequence.

2. The CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts according to claim 1 is characterized in that: In step S1, the specific process includes the following steps: S11. Use MetraSCAN scanner to collect 3D measurement point cloud data of the surface of aircraft parts; S12, segmenting the three-dimensional measurement point cloud data corresponding to the aircraft parts with complex structures into a plurality of parts point cloud data using a point cloud segmentation algorithm; S13. Use a point cloud completion algorithm to complete the hole area of ​​the point cloud data of each component.

3. The CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts according to claim 1, characterized in that: The 3D measurement point cloud reconstruction CAD model includes a point cloud feature sequence extraction module for extracting a point cloud feature sequence from the 3D measurement point cloud data, and a point cloud feature adapter for further processing and converting the point cloud feature sequence; The method also includes a CAD command tag building module for mapping the parameters of each CAD command to a tag space according to the CAD command sequence of the CAD true value of the aircraft parts, and obtaining the CAD command tag sequence, and a CAD command tag embedding module for extracting the embedding of the CAD command tag sequence and obtaining the CAD command tag embedding sequence, and, The point cloud feature sequence and the CAD command marker embedding sequence are spliced ​​to obtain a position encoding module of the point cloud feature-CAD command marker embedding sequence with position encoding, an autoregressive model of the CAD command marker sequence predicted according to the output of the point cloud feature-CAD command marker embedding sequence, and a CAD command reconstruction module that generates a CAD command sequence according to the CAD command marker sequence and constructs an aircraft parts CAD.

4. The CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts according to claim 3 is characterized in that: The CAD command mark embedding module is composed of an embedding layer, the position encoding module is composed of a position encoding layer, and the autoregressive model is composed of a Transformer decoder.

5. The CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts according to claim 1, characterized in that: In step S3, the CAD model reconstructed from the three-dimensional measurement point cloud is trained. The specific process includes the following steps: S311, inputting the CAD command sequence of the CAD true value of the aircraft parts into the CAD command tag construction module, mapping the parameters of each CAD command to the tag space, obtaining the tag corresponding to each CAD command, the tags of all CAD commands constitute a CAD command tag sequence, and adding fixed start tags and end tags at the beginning and end of the sequence respectively; S312, inputting the CAD command tag sequence into the CAD command tag embedding module to extract the embedding of all tags and obtain the CAD command tag embedding sequence; S313, concatenating the point cloud feature sequence and the CAD command mark embedding sequence and inputting them into a position coding module to obtain a point cloud feature-CAD command mark embedding sequence with position coding; S314, input the point cloud feature with position encoding-CAD command label embedding sequence into the autoregressive model, output the predicted CAD command label sequence, and supervise the predicted CAD command label sequence using the cross entropy loss function until convergence, thereby completing the training of the autoregressive model.

6. The CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts according to claim 1, characterized in that: In step S3, the CAD model reconstructed from the three-dimensional measurement point cloud is tested, and the specific process includes the following steps: S321, the initial CAD command tag sequence only includes the start tag; S322, inputting the CAD command tag sequence into the CAD command tag embedding module to extract the embedding of all tags and obtain the CAD command tag embedding sequence; S323, concatenating the point cloud feature sequence and the CAD command mark embedding sequence and inputting them into a position coding module to obtain a point cloud feature-CAD command mark embedding sequence with position coding; S324, inputting the point cloud feature with position coding-CAD command tag embedding sequence into the trained autoregressive model, predicting the next CAD command tag, and concatenating the CAD command tag sequence with the predicted next CAD command tag; S325. Repeat steps S322-S324 until a CAD command mark sequence of a specified length is obtained.

7. The CAD method for reconstructing three-dimensional measurement point clouds of aircraft parts according to claim 1, characterized in that: In step S4, the specific process includes the following steps: S41, intercepting the mark between the start mark and the end mark from the predicted CAD command mark sequence as a valid CAD command mark sequence; S42, inputting the valid CAD command tag sequence into the CAD command reconstruction module, mapping each tag to a parameter of the CAD command, obtaining the CAD command corresponding to each tag, and all CAD commands constitute a CAD command sequence; S443. According to the CAD command sequence, construct the aircraft parts CAD according to the commands.

Images