A lattice data compression method, system and medium for additive manufacturing

Abstract

The application discloses a kind of lattice data compression methods, systems and media for additive manufacturing, belong to additive manufacturing technical field, for solving the technical problem of the high efficiency compression method for lattice data needed urgently at present, to simultaneously improve the efficiency and precision of lattice structure modeling.Method includes: based on predefined unit cell type, construct unit cell volume field dataset;Construct neural network model based on auto-encoder architecture;Through the unit cell volume field dataset, the neural network model is trained, and a unit cell volume field prediction model is obtained;Through the unit cell volume field prediction model, the three-dimensional grid structure of target unit cell volume field data is extracted;The extracted three-dimensional grid structure is verified accurately, and the unit cell volume field prediction model is optimized and trained according to the verification result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of additive manufacturing, and in particular to a lattice data compression method, system and medium for additive manufacturing. BACKGROUND

[0002] With the rapid development of additive manufacturing technology (such as 3D printing technology), the lattice structure in it has been widely used in the fields of aerospace, medical treatment, industry, etc. due to its lightweight, energy absorption, surface area optimization and other characteristics. The digital modeling of the lattice structure is a key link in the additive manufacturing process. For cloud-based computer-aided design software, such as CAD design software, its lattice modeling function directly affects the design efficiency and innovation ability of engineers.

[0003] However, the existing cloud CAD lattice modeling function has the following shortcomings: first, the generation efficiency is slow. The traditional method has a large amount of calculation when generating a complex lattice structure, resulting in a long modeling time and failing to meet the rapid design requirement. Second, the data transmission amount is large. The lattice volume field data contains a large amount of three-dimensional discrete point data, which occupies a large amount of resources in the storage and transmission process, especially in the cloud collaboration or remote manufacturing scene, the transmission delay significantly affects the efficiency.

[0004] In the prior art, some solutions modeling algorithm relies on CPU serial calculation, which is difficult to parallel processing, resulting in limited generation speed of complex lattice. The volume field data is directly stored and transmitted in the original format without effective compression, causing waste of storage space and network transmission delay. The generation and transmission take a long time, affecting the design feedback cycle, and designers are difficult to quickly verify the combination of different lattice parameters, which restricts the innovation efficiency. Although some solutions use simplified models or low-precision approximation algorithms, the precision and design flexibility of the lattice structure are sacrificed, and it is difficult to balance efficiency and high fidelity. The traditional general compression algorithm does not exploit the periodicity, topological correlation and other characteristics of the lattice data, resulting in low compression rate and easy loss of key physical information, which cannot meet the strict requirements of additive manufacturing on precision.

[0005] In summary, there is an urgent need for an efficient compression method for lattice data to improve the efficiency and precision of lattice structure modeling. SUMMARY

[0006] The embodiments of the present application provide a lattice data compression method, system and medium for additive manufacturing, which solves the technical problem that there is an urgent need for an efficient compression method for lattice data to improve the efficiency and precision of lattice structure modeling.

[0007] The embodiments of the present application adopt the following technical solutions:

[0008] In one aspect, the embodiment of the present application provides a lattice data compression method for additive manufacturing, which comprises the following steps:

[0009] constructing a neural network model based on an autoencoder architecture;

[0010] training the neural network model based on the unit cell volume field dataset to obtain a unit cell volume field prediction model;

[0011] extracting a three-dimensional grid structure of target unit cell volume field data through the unit cell volume field prediction model;

[0012] verifying the accuracy of the extracted three-dimensional grid structure, and optimizing and training the unit cell volume field prediction model according to the verification result.

[0013] In one possible implementation, the unit cell volume field dataset is constructed based on a predefined unit cell type, specifically comprising the following steps:

[0014] constructing a unit cell basic topology based on the predefined unit cell type;

[0015] calculating a signed distance field (SDF) of each unit cell based on the unit cell basic topology, and converting the SDF into a volume field representing a structure proportion;

[0016] performing boundary reflection filling on the volume field of each unit cell to obtain a complete unit cell volume field;

[0017] randomly sampling data points in the complete unit cell volume field, and unifying the coordinate ranges of the data points to generate the unit cell volume field dataset.

[0018] In one possible implementation, after the unit cell volume field dataset is constructed based on the predefined unit cell type, the method further comprises the following steps:

[0019] visualizing and verifying the unit cell volume field data in the unit cell volume field dataset through an existing lattice structure digital display method, and determining whether the unit cell volume field data meet verification standards;

[0020] if not, optimizing and adjusting the unqualified items until the verification standards are met; or if yes, saving the unit cell volume field dataset in a file form in a local device.

[0021] In one possible implementation, the neural network model based on the autoencoder architecture is constructed, specifically comprising the following steps:

[0022] An encoder submodule and a decoder submodule are constructed, an output end of the encoder submodule is connected with an input end of the decoder submodule, to obtain a self-encoder architecture; wherein the encoder submodule is used for mapping input data to a latent space to obtain a latent encoding vector; the decoder submodule is used for mapping the latent encoding vector back to an original data space;

[0023] The self-encoder architecture is embedded into a neural network structure to obtain the neural network model based on the self-encoder architecture;

[0024] The neural network model is deployed in parallel threads of a GPU to realize parallel compression and reconstruction of the unit cell volume field data.

[0025] In a feasible implementation, the neural network model is trained by using the unit cell volume field data set to obtain a unit cell volume field prediction model, and specifically includes:

[0026] The unit cell volume field data set is input into the neural network model, the input unit cell volume field data is converted into a latent encoding vector by using the encoder submodule, and the latent encoding vector is reconstructed into unit cell volume field data by using the decoder submodule to obtain predicted volume field data, thereby forming a complete end-to-end training process;

[0027] In the training process, the weight parameters and bias parameters of the network are updated by using an Adam optimizer;

[0028] The loss value of the output predicted volume field data and the target value is calculated by using a mean square error loss function, and the neural network model is parameter-optimized according to the loss value until the network converges, thereby obtaining the unit cell volume field prediction model.

[0029] In a feasible implementation, the three-dimensional grid structure of the target unit cell volume field data is extracted by using the unit cell volume field prediction model, and specifically includes:

[0030] The pre-trained unit cell volume field prediction model and the model parameters thereof are loaded;

[0031] A target unit cell type is selected, and a volume field threshold value is preset as a judgment criterion for grid extraction;

[0032] The target unit cell volume field data is input into the unit cell volume field prediction model to perform volume field reconstruction, thereby obtaining target unit cell volume field reconstruction data;

[0033] According to the unit cell type and the preset volume field threshold value, the corresponding three-dimensional grid structure is extracted from the target unit cell volume field reconstruction data by using a double-contour extraction algorithm ODC.

[0034] In an implementable embodiment, after the three-dimensional grid structure of the target unit cell volume field data is extracted by the unit cell volume field prediction model, the method further comprises:

[0035] uploading the extracted three-dimensional grid structure to a visualization software for visual display to obtain a three-dimensional unit cell grid structure diagram corresponding to the target unit cell volume field data;

[0036] based on the three-dimensional unit cell grid structure diagram, performing additive manufacturing on the target object.

[0037] In an implementable embodiment, the extracted three-dimensional grid structure is verified for accuracy, and the unit cell volume field prediction model is optimized and trained according to the verification result, specifically comprising:

[0038] in the visualization software, the three-dimensional unit cell grid structure diagram is verified for accuracy and completeness to obtain a construction precision and a construction completeness;

[0039] based on a comparison result of the construction precision and a preset precision threshold and a comparison result of the construction completeness and a preset completeness threshold, the model parameters of the unit cell volume field prediction model are adjusted and optimized and trained.

[0040] On the other hand, the embodiment of the present application also provides a unit cell data compression system for additive manufacturing, which comprises:

[0041] a data set construction module for constructing a unit cell volume field data set based on a predefined unit cell type;

[0042] a model training module for constructing a neural network model based on an autoencoder architecture; training the neural network model by using the unit cell volume field data set to obtain a unit cell volume field prediction model;

[0043] a grid extraction module for extracting a three-dimensional grid structure of target unit cell volume field data by using the unit cell volume field prediction model; verifying the extracted three-dimensional grid structure for accuracy, and optimizing and training the unit cell volume field prediction model according to the verification result.

[0044] Finally, the embodiment of the present application also provides a storage medium, which is a non-volatile computer readable storage medium, and the non-volatile computer readable storage medium stores at least one program, each of the programs comprising instructions, which, when executed by a terminal, cause the terminal to perform the method for compressing unit cell data for additive manufacturing.

[0045] Compared with the prior art, the method, system and medium for compressing unit cell data for additive manufacturing provided by the embodiment of the present application have the following beneficial effects:

[0046] Firstly, compared with the traditional method, the application realizes efficient compression of unit cell volume field data through the autoencoder architecture, realizes compact representation of lattice data, and greatly reduces the data storage and transmission cost. At the same time, the autoencoder output result is combined with the ODC algorithm to realize grid extraction, realizing the organic combination of unit cell volume field reconstruction and grid extraction algorithm. And data preprocessing, model training, reasoning and grid extraction are all included in the GPU acceleration framework, realizing end-to-end parallel efficiency optimization. It can reduce the time-consuming of cloud CAD lattice modeling from minutes to seconds, fully utilize the server computing power, and adapt to large-scale lattice structure processing demand.

[0047] Secondly, the application ensures that the reconstructed unit cell structure is highly consistent with the original structure through the deep learning model, and guarantees the geometric accuracy; the grid extraction based on the ODC algorithm ensures the topological correctness and geometric accuracy of the reconstructed lattice, and meets the accuracy requirements of additive manufacturing.

[0048] Finally, the application supports 34 common unit cell structure types, has wide application range; through the extensible neural network architecture, it can adapt to different types of lattice design requirements, and provide technical support for various additive manufacturing application scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor. In the drawings:

[0050] Figure 1 A lattice data compression method for additive manufacturing is provided for the embodiment of the application;

[0051] Figure 2 A lattice data compression method for additive manufacturing is provided for the embodiment of the application;

[0052] Figure 3 A unit cell volume field prediction model training flow chart is provided for the embodiment of the application;

[0053] Figure 4 A body-centered cubic unit cell structure visualization schematic diagram is provided for the embodiment of the application;

[0054] Figure 5 A structure schematic diagram of a lattice data compression system for additive manufacturing is provided for the embodiment of the application. DETAILED DESCRIPTION

[0055] In order for those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0056] The embodiment of the present application provides a lattice data compression method for additive manufacturing, as shown in Figure 1 The lattice data compression method for additive manufacturing specifically includes steps S101-S105:

[0057] S101, based on a predefined unit cell type, a unit cell volume field dataset is constructed.

[0058] Specifically, based on the predefined unit cell type, a unit cell basic topology is constructed; based on the unit cell basic topology, a signed distance field SDF of each unit cell is calculated, and the signed distance field SDF is converted into a volume field representing a structure proportion; the volume field of each unit cell is filled by boundary reflection to obtain a complete unit cell volume field; data points are randomly sampled in the complete unit cell volume field, and the coordinate ranges of the various data points are unified to generate a unit cell volume field dataset.

[0059] In data set preprocessing, especially in image, signal, time series data and other data with "spatial / sequence continuity" processing, the core significance of boundary reflection padding (also called "mirror padding") is: when supplementing the "virtual area" outside the data boundary, the feature continuity of the original data is maximized to avoid information loss, distortion or artifacts caused by boundary truncation or operation (such as convolution).

[0060] Further, through the existing lattice structure digital display method, the unit cell volume field data in the unit cell volume field dataset is visualized and verified, and whether it meets the verification standard is determined. If not, the unqualified items are optimized and adjusted until the verification standard is met; if so, the unit cell volume field dataset is saved in the local in the form of a file.

[0061] As a feasible implementation manner, Figure 2 The specific flow chart of the lattice data compression method for additive manufacturing provided by the embodiment of the present application is shown in Figure 2 The present application realizes efficient generation and compression of lattice data by constructing a unit cell volume field automatic encoder based on deep learning combined with GPU acceleration technology. The core process includes:

[0062] a) Unit cell volume field dataset generation: Construct the unit cell volume field dataset by constructing the unit cell topology, volume field calculation, boundary filling, random sampling, etc. data processing operations.

[0063] b) Unit cell volume field training: Design a deep neural network based on the autoencoder architecture to achieve efficient compression and high-precision reconstruction of the unit cell volume field data.

[0064] c) Mesh extraction: Based on the ODC algorithm, extract the three-dimensional mesh from the reconstructed volume field.

[0065] d) GPU acceleration: Use GPU parallel computing capabilities to optimize model training, inference process, and accelerate mesh extraction efficiency.

[0066] The process of generating the unit cell volume field dataset includes:

[0067] 1. Unit cell topology construction: Generate the basic topology structure based on the predefined unit cell type. The unit cell type includes at least 34 common unit cell structure types, such as Diamond, Fluorite, Octet, Truncated Cube, Truncated Octahedron, Kelvin Cell, IsoTruss, Re-entrant, Weaire-Phelan, Triangular Honeycomb, Triangular Honeycomb Rotated, Hexagonal Honeycomb, Re-entrant Honeycomb, Square Honeycomb Rotated, Square Honeycomb, Face Centered Cubic Foam, Body Centered Cubic Foam, Simple Cubic Foam, Gyroid, Schwarz, Schwarz Diamond, Lidinoid, SplitP, Neovius, etc.

[0068] 2. Volume field calculation: Based on the unit cell basic topology structure, calculate the signed distance field SDF of each unit cell, and convert the signed distance field SDF to the volume field representing the structure proportion.

[0069] 3. Boundary filling: Reasonably fill the unit cell boundary to ensure the continuity and integrity of the data.

[0070] 4. Random sampling: Randomly sample points from the complete unit cell to generate the training dataset.

[0071] 5. Coordinate normalization: Normalize the coordinate range to eliminate the training bias caused by the scale difference.

[0072] 6. Visualization verification: The unit cell volume field data in the unit cell volume field data set is visually verified through the existing lattice structure digital display method, and it is judged whether it meets the verification standard.

[0073] S102, construct a neural network model based on an autoencoder architecture.

[0074] Specifically, an encoder submodule and a decoder submodule are constructed, the output end of the encoder submodule is connected with the input end of the decoder submodule, and an autoencoder architecture is obtained; wherein the encoder submodule is used to map the input data to an implicit space to obtain an implicit encoding vector; and the decoder submodule is used to map the implicit encoding vector back to the original data space.

[0075] Further, the autoencoder architecture is embedded in a neural network structure to obtain a neural network model based on the autoencoder architecture. The neural network model is deployed in the parallel threads of the GPU to realize parallel compression and reconstruction of the unit cell volume field data.

[0076] S103, train the neural network model through the unit cell volume field data set to obtain a unit cell volume field prediction model.

[0077] Specifically, the unit cell volume field data set is input into the neural network model, the input unit cell volume field data is converted into an implicit encoding vector through the encoder submodule, and the implicit encoding vector is reconstructed into unit cell volume field data through the decoder submodule to obtain predicted volume field data, forming a complete end-to-end training process.

[0078] During the training process, the weight parameters and bias parameters of the network are updated by the Adam optimizer. The loss value of the output predicted volume field data and the target value is calculated by the mean square error loss function, and the neural network model is optimized according to the loss value until the network converges, and the unit cell volume field prediction model is obtained.

[0079] As a feasible implementation manner, Figure 3 A unit cell volume field prediction model training flowchart is provided for the embodiment of the application, as shown in Figure 3 The neural network based on the autoencoder architecture is trained in combination with GPU acceleration to realize efficient compression and high-precision reconstruction of the unit cell volume field data. The module converts the original unit cell volume field data into 128-dimensional implicit encoding through the encoder, and reconstructs the implicit encoding into the original volume field data through the decoder, forming a complete end-to-end training process.

[0080] The autoencoder architecture includes an encoder and a decoder submodule, the encoder maps the input data to an implicit space, and the mathematical expression of the encoder function is: ; wherein P is the point cloud data, V is the volume field data, and z is the implicit encoding.

[0081] The decoder maps the implicit encoding back to the original data space, and the mathematical expression is: ; wherein P is the point cloud data, and z is the implicit encoding, to reconstruct the volume field data.

[0082] Optimizer: The Adam optimizer is used to update the weights and biases of the network during the training process.

[0083] Loss function: The Mean Squared Error (MSE) function is used to calculate the squared difference between the output and the target value.

[0084] Grid extraction module: A high-fidelity three-dimensional grid is extracted from the implicit function Occupancy Function, and GPU parallelization is used to achieve fast calculation.

[0085] S104, through the unit cell volume field prediction model, extracts the three-dimensional grid structure of the target unit cell volume field data.

[0086] Specifically, the pre-trained unit cell volume field prediction model and its model parameters are loaded. The target unit cell type is selected, and the volume field threshold is preset as the judgment standard for grid extraction.

[0087] Further, the target unit cell volume field data is input into the unit cell volume field prediction model for volume field reconstruction to obtain the target unit cell volume field reconstruction data.

[0088] Further, through the double contour extraction algorithm ODC, the corresponding three-dimensional grid structure is extracted from the target unit cell volume field reconstruction data according to the unit cell type and the preset volume field threshold.

[0089] Further, the extracted three-dimensional grid structure is uploaded to the visualization software for visualization display to obtain a three-dimensional unit cell grid structure diagram corresponding to the target unit cell volume field data. Based on the three-dimensional unit cell grid structure diagram, the target object is subjected to additive manufacturing.

[0090] As a feasible implementation, 34 different types of unit cell structure data sets are adopted, each containing a large number of three-dimensional point coordinates and their corresponding volume field values, to construct a comprehensive training data set covering various lattice morphologies. The batch size is set to 2, the optimizer is Adam, the initial learning rate is 0.0005, and the training round is 100000 times to ensure model convergence and reconstruction accuracy.

[0091] Compression effect: successfully compress 3N-dimensional (N is the number of points, 100,000 points) original volume field data to 128-dimensional implicit encoding vector, the original volume field data file size is 106251 KB, the compressed file size is 3510 KB, significantly reducing the data storage and transmission cost.

[0092] Load the pre-trained auto-encoder model, including the encoder and decoder network parameters. Select the cell type and parameters: select the target cell type from 34 cell structures, set the volume field threshold to 0.5 as the grid extraction criterion. Use the decoder network to reconstruct the 128-dimensional implicit encoding vector into the original dimension volume field data. Use the ODC algorithm to extract the three-dimensional grid from the reconstructed volume field.

[0093] Finally, the extracted three-dimensional grid is visualized to verify the geometric accuracy and integrity of the cell structure, and the body-centered cubic cell structure is as shown in Figure 4 .

[0094] S105, verify the accuracy of the extracted three-dimensional grid structure, and optimize the training of the cell volume field prediction model according to the verification result.

[0095] Specifically, in the visualization software, the three-dimensional cell grid structure diagram is verified for accuracy and integrity, and the construction precision and construction integrity are obtained.

[0096] Further, based on the comparison result of the construction precision and the preset precision threshold, and the comparison result of the construction integrity and the preset integrity threshold, the model parameters of the cell volume field prediction model are adjusted and optimized.

[0097] In addition, the embodiment of the present application also provides a lattice data compression system for additive manufacturing, as shown in Figure 5 , the lattice data compression system for additive manufacturing 500 specifically comprises:

[0098] The data set construction module 510 is used to construct a cell volume field data set based on a predefined cell type.

[0099] The model training module 520 is used to construct a neural network model based on an auto-encoder architecture; the neural network model is trained based on the cell volume field data set to obtain a cell volume field prediction model.

[0100] The grid extraction module 530 is used to extract a three-dimensional grid structure of the target cell volume field data through the cell volume field prediction model; the accuracy of the extracted three-dimensional grid structure is verified, and the cell volume field prediction model is optimized and trained according to the verification result.

[0101] Finally, the embodiment of the present application also provides a storage medium, the storage medium is a non-volatile computer readable storage medium, the non-volatile computer readable storage medium stores at least one program, each program includes instructions, the instructions make the terminal execute when being executed by the terminal:

[0102] constructing a unit cell volume field dataset based on a predefined unit cell type;

[0103] constructing a neural network model based on an autoencoder architecture;

[0104] training the neural network model through the unit cell volume field dataset to obtain a unit cell volume field prediction model;

[0105] extracting a three-dimensional grid structure of target unit cell volume field data through the unit cell volume field prediction model;

[0106] verifying the accuracy of the extracted three-dimensional grid structure, and optimizing and training the unit cell volume field prediction model according to the verification result.

[0107] Each of the embodiments of the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment mainly explains the difference from other embodiments. Especially, for the device, equipment and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the related parts can be referred to the part of the method embodiment.

[0108] The above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0109] The above only describes the embodiments of the present application and is not used to limit the present application. The embodiments of the present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present application shall be included in the protection scope of the present application.

Claims

1. A lattice data compression method for additive manufacturing, characterized by, The method comprises: constructing a unit cell volume field dataset based on a predefined unit cell type; constructing a neural network model based on an autoencoder architecture, specifically comprising: constructing an encoder submodule and a decoder submodule, the output end of the encoder submodule being connected to the input end of the decoder submodule to obtain an autoencoder architecture; wherein the encoder submodule is used to map input data to a latent space to obtain a latent code vector; the decoder submodule is used to map the latent code vector back to the original data space; embedding the autoencoder architecture into a neural network structure to obtain the neural network model based on the autoencoder architecture; deploying the neural network model in the parallel threads of a GPU to realize parallel compression and reconstruction of the unit cell volume field data; training the neural network model through the unit cell volume field dataset to obtain a unit cell volume field prediction model; extracting the three-dimensional grid structure of the target unit cell volume field data through the unit cell volume field prediction model; uploading the extracted three-dimensional grid structure to a visualization software for visualization display to obtain a three-dimensional unit cell grid structure diagram corresponding to the target unit cell volume field data; based on the three-dimensional unit cell grid structure diagram, performing additive manufacturing on the target object; verifying the accuracy of the extracted three-dimensional grid structure and optimizing the training of the unit cell volume field prediction model according to the verification result.

2. A lattice data compression method for additive manufacturing according to claim 1, characterized in that, Based on the predefined unit cell type, a unit cell volume field dataset is constructed, specifically comprising: based on the predefined unit cell type, a unit cell basic topology structure is constructed; based on the unit cell basic topology structure, the signed distance field (SDF) of each unit cell is calculated, and the signed distance field (SDF) is converted into a volume field representing the structure proportion; boundary reflection filling is performed on the volume field of each unit cell to obtain a complete unit cell volume field; randomly sampling data points in the complete unit cell volume field and unifying the coordinate ranges of the data points to generate the unit cell volume field dataset.

3. A lattice data compression method for additive manufacturing according to claim 2, characterized in that, After constructing the unit cell volume field dataset based on the predefined unit cell type, the method further comprises: visualizing the unit cell volume field data in the unit cell volume field dataset through the existing lattice structure digital display method, and determining whether it meets the verification standard; if not, the unqualified items are optimized and adjusted until the verification standard is met; if so, the unit cell volume field dataset is saved in the local in the form of a file.

4. The lattice data compression method for additive manufacturing of claim 1, wherein, The neural network model is trained through the unit cell volume field dataset to obtain a unit cell volume field prediction model, specifically comprising: inputting the unit cell volume field dataset into the neural network model, converting the input unit cell volume field data into a latent code vector through the encoder submodule, and reconstructing the latent code vector into unit cell volume field data through the decoder submodule to obtain prediction volume field data, forming a complete end-to-end training process; in the training process, the weight parameters and bias parameters of the network are updated through the Adam optimizer; The loss value of the output prediction volume field data and the target value is calculated through a mean square error loss function, and the neural network model is optimized in parameters according to the loss value until the network converges, so as to obtain the unit cell volume field prediction model.

5. The lattice data compression method for additive manufacturing of claim 1, wherein, Through the unit cell volume field prediction model, the three-dimensional grid structure of the target unit cell volume field data is extracted, specifically including: loading the pre-trained unit cell volume field prediction model and its model parameters; selecting a target unit cell type and pre-setting a volume field threshold as the judgment standard for grid extraction; inputting the target unit cell volume field data into the unit cell volume field prediction model to perform volume field reconstruction and obtain target unit cell volume field reconstruction data; extracting the corresponding three-dimensional grid structure in the target unit cell volume field reconstruction data according to the unit cell type and the pre-set volume field threshold through the double contour extraction algorithm ODC.

6. The lattice data compression method for additive manufacturing of claim 1, wherein, The extracted three-dimensional grid structure is verified for accuracy, and the unit cell volume field prediction model is optimized and trained according to the verification result, specifically including: in the visualization software, the three-dimensional unit cell grid structure diagram is verified for accuracy and completeness to obtain construction accuracy and construction completeness; based on the comparison results of the construction accuracy and the pre-set accuracy threshold and the comparison results of the construction completeness and the pre-set completeness threshold, the model parameters of the unit cell volume field prediction model are adjusted and optimized and trained.

7. A lattice data compression system for additive manufacturing, characterized by, The system comprises: a data set construction module for constructing a unit cell volume field data set based on a pre-defined unit cell type; a model training module for constructing a neural network model based on an autoencoder architecture, specifically including: constructing an encoder submodule and a decoder submodule, the output end of the encoder submodule being connected to the input end of the decoder submodule to obtain an autoencoder architecture; wherein the encoder submodule is used to map input data to a hidden space to obtain a hidden code vector; the decoder submodule is used to map the hidden code vector back to the original data space; the autoencoder architecture is embedded in a neural network structure to obtain the neural network model based on the autoencoder architecture; the neural network model is deployed in the parallel threads of a GPU to realize the parallel compression and reconstruction of unit cell volume field data; the neural network model is trained through the unit cell volume field data set to obtain a unit cell volume field prediction model; a grid extraction module for extracting the three-dimensional grid structure of the target unit cell volume field data through the unit cell volume field prediction model; uploading the extracted three-dimensional grid structure to a visualization software for visualization to obtain a three-dimensional unit cell grid structure diagram corresponding to the target unit cell volume field data; based on the three-dimensional unit cell grid structure diagram, a target object is manufactured by additive manufacturing; the extracted three-dimensional grid structure is verified for accuracy, and the unit cell volume field prediction model is optimized and trained according to the verification result.

8. A storage medium, characterized by The storage medium is a nonvolatile computer readable storage medium, and the nonvolatile computer readable storage medium stores at least one program, and each program includes instructions, which, when executed by a terminal, causes the terminal to perform the lattice data compression method for additive manufacturing according to any one of claims 1-6.

Images