Assembly geometric digital twin modeling and pose calculation method based on deep learning

Abstract

The invention discloses an assembly geometric digital twin modeling and pose calculation method based on deep learning, and belongs to the technical field of precision mechanical system modeling. Comprising the following steps: firstly, selecting a small number of product key contact surfaces for point cloud measurement; secondly, performing redundancy removal, expansion and arrangement on the data by adopting an up-sampling / down-sampling and data enhancement method; then, an assembly data set is established through assembly pose parameter calculation and data enhancement; then training a neural network by using a deep learning technology, and establishing a geometric distribution error and assembly pose mapping model; according to the method, part surface geometric distribution errors and assembly pose errors in an assembly body are considered, a geometric digital twin model with complete and accurate error information is established, and a model basis is provided for product precision prediction, product assembly process optimization, product assembly visualization and product quality improvement; and the method is suitable for precision mechanical products with high precision and large-batch production requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of precision mechanical system modeling, and in particular to a method for assembly geometry digital twin modeling and posture calculation based on deep learning. Background Art

[0002] At present, the existing digital twin modeling methods manually build and assemble ideal three-dimensional CAD models to characterize the geometric features and assembly relationships of virtual models. They do not involve non-uniformly distributed surface geometric distribution errors and non-ideal parts assembly postures, and cannot characterize the geometric information of physical models with high precision, high fidelity, and high efficiency. However, in the precision mechanical systems in actual production, the actual geometric distribution errors of parts directly affect the final assembly accuracy during the matching contact and assembly transfer, and ultimately affect the static and dynamic mechanical properties and operating functions of mechanical products during use. At the same time, for mechanical products with complex geometric structures and assembly processes, the modeling of geometric distribution errors and the calculation of assembly error transfer are both technically difficult, labor-intensive, and repetitive mechanical labor-intensive tasks, which consume manpower and time, and are prone to model gross errors due to manual operation errors. However, through the geometric distribution error modeling method and deep learning method, the actual assembly accuracy of mechanical products can be predicted and characterized with high precision, high efficiency, and intelligence. Summary of the invention

[0003] The purpose of the present invention is to provide a method for modeling and posture calculation of assembly geometry digital twins based on deep learning. By measuring the geometric distribution error data of product samples, part error modeling, assembly accuracy calculation, assembly data set construction, and deep learning neural network training, a mapping model between part geometry distribution error data and assembly accuracy parameters is established, and a mechanical product geometry digital twin model is further established, thereby realizing the prediction of mechanical product assembly accuracy.

[0004] To achieve the above object, the present invention provides an assembly geometry digital twin modeling and pose calculation method based on deep learning, comprising the following steps:

[0005] S1. Point cloud measurement: select a small number of product samples, select the key contact surfaces of the product assembly, and measure the geometric distribution errors to obtain the initial point cloud data;

[0006] S2, data processing, upsampling or downsampling the initial point cloud according to the number and arrangement requirements of the point cloud for modeling and deep learning, obtaining the processed point cloud data, and performing data enhancement on the group of data to increase the amount of data;

[0007] S3, constructing an assembly data set, selecting a pair of surface point cloud data to calculate assembly pose parameters, obtaining the translation and rotation of the three contact points and the surface point cloud pose changes, forming a set of assembly data, selecting different surface point cloud data multiple times to calculate the corresponding assembly pose parameters, and multiple sets of assembly data constitute an assembly data set;

[0008] S4, training the neural network, using the assembly data set to perform deep learning neural network training, the deep learning neural network obtained after training is the geometric distribution error and assembly posture mapping model;

[0009] S5. Production product measurement: geometric distribution error measurement of products is performed during the production process to obtain point cloud data;

[0010] S6, secondary data processing, using the same upsampling or downsampling method as step S2 to process the point cloud data to obtain processed point cloud data;

[0011] S7, assembly posture error mapping, the processed point cloud data is input as input into the geometric distribution error and assembly posture mapping model of step S4 to obtain the predicted value of the assembly posture parameter;

[0012] S8, part geometry digital twin modeling, using the point cloud data of step S5 or the processed point cloud data of step S6 to perform 3D geometric error CAD modeling, and obtain a 3D error model of the part with geometric distribution errors on the surface;

[0013] S9, assembly geometry digital twin modeling, digitally assemble the product assembly with the part three-dimensional error model of step S8 according to the assembly posture parameters of step S7, and obtain a geometric digital twin model of the mechanical product assembly. The model includes the geometric distribution error of the part surface and the assembly posture error of the assembly, and is used to accurately characterize the true geometric state of the actual product.

[0014] Preferably, in step S2, for the geometric error morphology of a single surface, a small amount of point cloud measurement data and data upsampling are used to predict the error morphology of the unmeasured area, and a large amount of point cloud measurement data and downsampling are used to delete redundant point clouds while retaining the error morphology characteristics, thereby reducing the data volume and regularizing the point cloud distribution.

[0015] Preferably, in step S2, with respect to the geometric error morphology of a large number of products, the geometric error morphology characteristics of a large number of other product surfaces are predicted through a small amount of product surface point cloud measurement data and data enhancement.

[0016] Preferably, in step S3, selecting a pair of surface point cloud data to calculate assembly posture parameters is to calculate a pair of surface error point cloud contact point conditions and assembly posture parameters.

[0017] Preferably, deep learning technology is used in step S4, and a deep learning neural network is trained with point clouds and corresponding assembly pose parameters to establish a geometric distribution error and assembly pose mapping model. The mapping model adopts an "end-to-end" strategy to quickly calculate the assembly pose parameters.

[0018] Preferably, in step S5, step S6 and step S7, a data processing method and a trained deep learning neural network are used to achieve rapid prediction from product measurement data to product assembly posture parameters.

[0019] Preferably, in steps S8 and S9, the processed surface point cloud data, predicted assembly posture parameters and computer-aided design modeling technology are used to establish a high-precision, high-fidelity, and high-reduction geometric digital twin assembly model of the actual product assembly.

[0020] Therefore, the present invention adopts a method for modeling and calculating the geometric digital twin of an assembly based on deep learning of the above structure, which has the following beneficial effects:

[0021] (1) The present invention can obtain a geometric distribution error and assembly posture mapping model by measuring the surface geometric distribution error, processing point cloud data, calculating assembly posture and training neural networks on a small number of products, and then obtain a mechanical product assembly geometry digital twin modeling by measuring the surface geometric distribution error, processing point cloud data, predicting assembly posture, 3D geometric error modeling and digital assembly on the products in actual production. It can also predict assembly errors and establish a visualized digital model containing geometric distribution errors and assembly posture errors.

[0022] (2) The present invention takes into account the geometric distribution error of the part surface and the assembly posture error in the assembly, and establishes a geometric digital twin model with complete and accurate error information, providing a model basis for product accuracy prediction, product assembly process optimization, product assembly visualization, and product quality improvement.

[0023] (3) The present invention adopts a deep learning method, an end-to-end mapping method, in the geometric digital twin modeling process. The use of a trained deep learning neural network can save complex intermediate modeling and calculation links, simplify manual operations, and reduce the technical threshold requirements for front-line production operators.

[0024] (4) The deep learning network of the present invention can achieve high accuracy and high reliability in assembly posture parameter prediction with only one training. The response time is in seconds and the calculation speed is fast. Except for the part measurement link, all other aspects can be automatically modeled through programming, meeting the requirements for online prediction of product accuracy in mass production.

[0025] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of a modeling process of a method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to the present invention;

[0027] Figure 2 A schematic diagram of circular surface point cloud upsampling of an assembly geometry digital twin modeling and pose calculation method based on deep learning of the present invention;

[0028] Figure 3 This is a schematic diagram of a geometric digital twin model of a circular surface assembly structure of an assembly geometric digital twin modeling and posture calculation method based on deep learning in the present invention. DETAILED DESCRIPTION

[0029] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0030] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0031] Example

[0032] The purpose of the present invention is to provide a deep learning-based geometric digital twin modeling method for mechanical product assemblies for industrial production. The geometric digital twin model established according to this method can concisely, efficiently and accurately predict and characterize the actual assembly errors of mechanical products. Figure 1 The present invention provides a geometric digital twin of a circular surface assembly structure as an example for illustration, and includes the following steps:

[0033] S1. Point cloud measurement:

[0034] Start to build the geometric distribution error and assembly posture mapping model. First, perform assembly analysis on the target assembly. The circular surface assembly structure has two circular surfaces as contact mating surfaces. Plan the distribution of measurement points, measure 11 circles, 25 points per circle, and a total of 275 points. Select the OGP optical imager to measure multiple circular contact surfaces and obtain multiple sets of point cloud XYZ coordinate values.

[0035] S2. Data processing:

[0036] In order to meet the requirements of the number and arrangement of point clouds of the deep learning neural network training data set, the initial measured point cloud of each circular surface is upsampled by interpolation to obtain the processed point cloud data. Each group of point cloud data has 30 circles, 74 points per circle, and a total of 2220 points. Figure 2 In order to improve the accuracy and reliability of the learning neural network, the processed point cloud data set is enhanced, and the number of point cloud groups in the point cloud data set is expanded according to artificially formulated rules and mathematical calculation methods to obtain more groups of circular surface point clouds.

[0037] S3. Build assembly dataset:

[0038] According to the surface matching relationship in the assembly, two circular surface point clouds are extracted from the processed point cloud data set for virtual assembly. The lower part is fixed, and the differential surface contact algorithm and assembly error transfer method are used to calculate the assembly posture parameter group T of the upper part after assembly. i (d x , d y , d z ,θ x ,θ y ,θ z ), including the translation of the upper part along the X, Y, and Z axes and the rotation around the X, Y, and Z axes in the assembly. In addition to the difference surface method, the contact algorithm can also choose the minimum potential energy method, the minimum neighbor distance method, etc. The assembly error transfer method can choose the secondary transformation matrix method, the Jacobi spinor method, etc. The two sets of circular surface point clouds P involved in the virtual assembly in the above calculation 1i and P 2i And the corresponding assembly pose parameter T of the upper part i By repeatedly selecting different circular surface point clouds and calculating assembly pose parameters, multiple assembly data groups can be obtained, which is an assembly data set.

[0039] S4. Training neural network:

[0040] Assemble all circular surface point clouds P in the dataset 1i and P 2i And all corresponding assembly pose parameters T i The coordinate pose parameter d of a part inx ,d y ,d z ,θ x ,θ y Or θ z One of the selected single assembly pose parameter training set is used to train the deep learning neural network and establish a single assembly pose parameter mapping model. According to the requirements of product assembly accuracy, multiple single assembly pose parameter mapping models from point cloud to any part with any degree of freedom can be obtained by training the neural network independently multiple times. Multiple single assembly pose parameter mapping models used in actual production are integrated into a geometric distribution error and assembly pose mapping model.

[0041] S5. Production product measurement:

[0042] In actual production, a geometric digital twin model of a circular surface assembly is constructed. First, the geometric distribution errors of all key surfaces of the product are measured to obtain the initial point cloud data. The measurement method is the same as that of S1 to obtain the initial point cloud set of the product.

[0043] S6. Secondary data processing:

[0044] The same upsampling or downsampling method as S2 is used to process the initial point cloud data to obtain the processed point cloud data of the two key surfaces of the product.

[0045] S7. Assembly posture error mapping:

[0046] According to the production process requirements, the processed point cloud data of the key surfaces are respectively imported into the trained deep learning neural network to obtain the predicted values ​​of the assembly pose parameters of the corresponding parts as output. And output in the form of homogeneous transformation matrix:

[0047]

[0048] The translation amount d x =0,d y =0,d z =2.1655, the sine of the rotation amount sinθ x =0.0167, sinθ y =-0.0468, sinθ z =0.

[0049] S8. Part geometry digital twin modeling:

[0050] According to different needs, different modeling techniques are used to perform geometric distribution error CAD (Computer-Aided Design) software modeling on the processed point cloud data of S6. Including but not limited to the following methods: the surface model of the part can be established using the NURBS (NonUniform Rational B-Spline) surface model, and stored in the IGES (Initial Graphics Exchange Specification) format as a model file that can be read by the CAD modeling software system, and further combined with the three-dimensional ideal model of the part to establish a three-dimensional error model of the part containing geometric distribution errors. The point cloud can also be triangulated to establish a triangular mesh part model based on the error point cloud, and stored in the STL (Standard Triangle Language) format as a model file that can be read by the CAD modeling software system. This embodiment uses the NURBS modeling method to establish a part geometry digital twin model.

[0051] S9. Assembly geometry digital twin modeling:

[0052] In the CAD modeling software system, according to the multi-degree-of-freedom assembly posture parameters of each part of S7, the three-dimensional error model of each part of S8 is digitally assembled and posture-aligned to obtain the geometric digital twin model of the assembly, such as Figure 3 As shown in the figure. This model includes the geometric distribution error of the part surface and the assembly posture error of the assembly, which can accurately characterize the real geometric state of the actual product. It can also further obtain the assembly accuracy of the mechanical product, such as coaxiality, parallelism, verticality, etc., through the assembly posture parameter calculation method or the CAD modeling software geometric feature data extraction method.

[0053] Therefore, the present invention adopts the above-mentioned assembly geometry digital twin modeling and posture calculation method based on deep learning, applies the model to actual product batch production, predicts the assembly posture parameters of mechanical product assemblies; at the same time, a geometric digital twin model of product parts containing microscopic geometric errors is established based on point cloud data; finally, a mechanical product assembly geometry digital twin model containing geometric shape errors and assembly errors is established based on the assembly posture parameters and the geometric digital twin model of parts. The present invention is suitable for precision mechanical products with high-precision and mass production requirements.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for modeling and calculating the pose of an assembly geometry digital twin based on deep learning, characterized in that: The following steps are involved: S1. Point cloud measurement: select a small number of product samples, select the key contact surfaces of the product assembly, and measure the geometric distribution errors to obtain the initial point cloud data; S2, data processing, upsampling or downsampling the initial point cloud according to the number and arrangement requirements of the point cloud for modeling and deep learning, obtaining the processed point cloud data, and performing data enhancement on the group of data to increase the amount of data; S3, constructing an assembly data set, selecting a pair of surface point cloud data to calculate assembly pose parameters, obtaining the translation and rotation of the three contact points and the surface point cloud pose changes, forming a set of assembly data, selecting different surface point cloud data multiple times to calculate the corresponding assembly pose parameters, and multiple sets of assembly data constitute an assembly data set; S4, training the neural network, using the assembly data set to perform deep learning neural network training, the deep learning neural network obtained after training is the geometric distribution error and assembly posture mapping model; S5. Production product measurement: geometric distribution error measurement of products is performed during the production process to obtain point cloud data; S6, secondary data processing, using the same upsampling or downsampling method as step S2 to process the point cloud data to obtain processed point cloud data; S7, assembly posture error mapping, the processed point cloud data is input as input into the geometric distribution error and assembly posture mapping model of step S4 to obtain the predicted value of the assembly posture parameter; S8, part geometry digital twin modeling, using the point cloud data of step S5 or the processed point cloud data of step S6 to perform 3D geometric error CAD modeling, and obtain a 3D error model of the part with geometric distribution errors on the surface; S9, assembly geometry digital twin modeling, digitally assemble the product assembly with the part three-dimensional error model of step S8 according to the assembly posture parameters of step S7, and obtain a geometric digital twin model of the mechanical product assembly. The model includes the geometric distribution error of the part surface and the assembly posture error of the assembly, and is used to accurately characterize the true geometric state of the actual product.

2. The method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to claim 1, characterized in that: In step S2, for the geometric error morphology of a single surface, a small amount of point cloud measurement data and data upsampling are used to predict the error morphology of the unmeasured area. A large amount of point cloud measurement data and downsampling are used to delete redundant point clouds while retaining the error morphology characteristics, thereby reducing the data volume and regularizing the point cloud distribution.

3. The method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to claim 1, characterized in that: In step S2, for the geometric error morphology of a large number of products, the geometric error morphology characteristics of a large number of other product surfaces are predicted through a small amount of product surface point cloud measurement data and data enhancement.

4. The method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to claim 1, characterized in that: In step S3, a pair of surface point cloud data is selected to calculate the assembly posture parameter to calculate the contact point situation of a pair of surface error point clouds and the assembly posture parameter group T (d x , d y , d z ,θ x ,θ y ,θ z ).

5. The method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to claim 1, characterized in that: In step S4, deep learning technology is used to train a deep learning neural network with point clouds and corresponding assembly pose parameters to establish a geometric distribution error and assembly pose mapping model. The mapping model adopts an "end-to-end" strategy to quickly calculate the assembly pose parameters.

6. The method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to claim 1, characterized in that: In step S5, step S6 and step S7, a data processing method and a trained deep learning neural network are used to achieve rapid prediction from product measurement data to product assembly posture parameters.

7. The method for modeling and calculating the geometric digital twin of an assembly based on deep learning according to claim 1, characterized in that: In step S8 and step S9, the processed surface point cloud data, predicted assembly posture parameters and computer-aided design modeling technology are used to establish a high-precision, high-fidelity, and high-reduction geometric digital twin assembly model of the actual product assembly.

Images