CN115762675A - Method and system for predicting mechanical property of truss type lattice structure - Google Patents
Method and system for predicting mechanical property of truss type lattice structure Download PDFInfo
- Publication number
- CN115762675A CN115762675A CN202211312327.3A CN202211312327A CN115762675A CN 115762675 A CN115762675 A CN 115762675A CN 202211312327 A CN202211312327 A CN 202211312327A CN 115762675 A CN115762675 A CN 115762675A
- Authority
- CN
- China
- Prior art keywords
- sample
- cell
- graph
- lattice structure
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
The invention relates to a method and a system for predicting mechanical property of a truss type lattice structure, which comprises the following steps: acquiring a diagram of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of cell elements; the graph nodes of the graph are cell nodes of a cell; the sides of the figure are the rods of the cell; inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property predicted value of the target truss type lattice structure; the node information comprises position information of the cell node in the cell; the side information comprises diameter information, cross section area information and/or cross section shape information of the rod piece of the cell element; the invention can represent the cells with different bar member geometrical characteristics under different cell basic shapes as a graph based on a graph theory mode; the mechanical property of the target truss type lattice structure is predicted by adopting the trained graph neural network model, so that the universality and the speed of predicting the mechanical property of the truss type lattice structure are improved.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a method and a system for predicting mechanical property of a truss type lattice structure based on a graph neural network model.
Background
The basic cells of the truss lattice structure are composed of nodes and rod members connecting the nodes, and the more common cell configurations include Body Centered Cubic (BCC), face Centered Cubic (FCC), octahedral (Octahedron), octet, and the like. Through the reasonable design of the cell structure, the truss lattice structure obtains higher mechanical performance indexes such as specific modulus, specific strength, specific energy absorption and the like, and is the basic research content of lattice material structure design. Common methods for predicting the mechanical properties of a truss-type lattice structure include finite element simulation methods and mechanical calculation methods based on simplified physical models. However, the two methods have limitations respectively, and the finite element simulation method has long calculation time and cannot rapidly acquire the mechanical properties of the truss lattice structure. In the mechanical calculation method based on the simplified physical model, the performance prediction based on the structural mechanics and the solid mechanics depends on expert knowledge for carrying out physical modeling on a specific structure, the physical model constructed by different structures is not suitable for other structures, and the universality is low.
Disclosure of Invention
The invention aims to provide a method and a system for predicting mechanical properties of a truss type lattice structure, which improve the universality and the speed of predicting the mechanical properties of the truss type lattice structure.
In order to achieve the purpose, the invention provides the following scheme:
a method for predicting mechanical properties of a truss lattice structure comprises the following steps:
acquiring a diagram of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of the cell elements; the graph nodes of the graph correspond to cell nodes of the cells; the edges of the graph correspond to the bars of the cells;
inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property predicted value of the target truss type lattice structure; the node information comprises location information of the cell node within the cell; the side information comprises diameter information, cross-sectional area information and/or cross-sectional shape information of the rod pieces of the cells; the trained graph neural network model is a model obtained by training by taking sample node information and sample side information of a sample graph of a sample cell as input and taking a real value of the sample mechanical property of the sample truss type lattice structure as a label; the sample truss type lattice structure is obtained by periodically arranging a plurality of sample cells.
Optionally, the trained neural network model is a signal transmission neural network model.
Optionally, the trained neural network model of the graph includes an information transfer module and a reading module;
the information transmission module is used for carrying out iterative update on the node characteristic vector of each graph node according to the node characteristic vector of the adjacent graph node connected with the graph node through the edge to obtain an updated node characteristic vector;
and the reading module is used for splicing all the updated node characteristic vectors and inputting a full-connection layer neural network which takes a mechanical property numerical value as output to obtain a mechanical property predicted value of the target truss type lattice structure.
Optionally, before the node information and the side information of the graph are input into the trained graph neural network model, training the graph neural network model is further included, and the training process is as follows:
acquiring a data set; the data set comprises sample node information and sample side information of a sample graph of each sample cell in a plurality of sample cells with different configurations and a real value of sample mechanical property of the sample truss type lattice structure corresponding to the sample cell;
and training the graph neural network model by using the data set to obtain the trained graph neural network model.
Optionally, the acquiring the data set specifically includes:
acquiring a plurality of sample cells with different configurations;
representing each sample cell based on a graph theory mode to obtain a sample graph of each sample cell; a sample map node of the sample map is a sample cell node of the sample cell; a sample edge of the sample map is a sample rod of the sample cell; the sample node information comprises location information of the sample cell node within the sample cell; the sample side information comprises diameter information, cross-sectional area information, and/or cross-sectional shape information of sample rods of the sample cells;
periodically arranging each sample cell to obtain a plurality of sample truss type lattice structures;
performing mechanical property test on each sample truss type lattice structure under a preset load condition to obtain a numerical value of a mechanical property index; the numerical value of the mechanical property index is a real value of the sample mechanical property of the sample truss type lattice structure; the mechanical property indexes comprise elastic modulus, yield strength, compression strength, tensile strength and bending strength.
Optionally, the acquiring a plurality of sample cells of different configurations specifically includes:
and 4, repeating the steps 2-3 to obtain a plurality of sample cells with different configurations.
Optionally, the preset load condition is a compressive load, a tensile load or a bending load.
Optionally, the basic shape of the predetermined cell element is a polyhedron; the polyhedron can fill a three-dimensional space without a gap by a periodic arrangement.
Optionally, the basic shape of the preset cell element is a cube, a rhombohedron, a triangular prism, a hexagonal prism, a rhombic dodecahedron or a tetradecahedron.
The invention also provides a system for predicting the mechanical property of the truss type lattice structure, which comprises the following steps:
the image acquisition module is used for acquiring an image of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of the cell elements; the graph nodes of the graph correspond to cell nodes of the cells; the sides of the graph correspond to the bars of the cells;
the mechanical property prediction module is used for inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property prediction value of the target truss type lattice structure; the node information comprises location information of the cell node within the cell; the side information comprises diameter information, cross-sectional area information and/or cross-sectional shape information of the rod pieces of the cells; the trained graph neural network model is obtained by training by taking the sample node information and the sample side information of the sample graph of the sample cell as input and taking the real value of the sample mechanical property of the sample truss-type lattice structure as a label; the sample truss type lattice structure is obtained by periodically arranging a plurality of sample cells.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a method and a system for predicting mechanical property of a truss type lattice structure, which comprises the following steps: acquiring a diagram of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of cell elements; the graph node of the graph corresponds to the cell node of the cell; the sides of the figure correspond to the bars of the cell; inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property predicted value of the target truss type lattice structure; the node information comprises the position information of the cell node in the cell; the side information comprises diameter information, cross section area information and/or cross section shape information of the member of the cell element, and the method can represent the cell elements with different geometric characteristics of the member element as a graph based on a graph theory under different basic cell element shapes, thereby improving the universality of mechanical property prediction of a truss type lattice structure; and the mechanical property of the target truss type lattice structure is predicted by adopting the trained graph neural network model, so that the speed of predicting the mechanical property of the truss type lattice structure is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a method for predicting mechanical properties of a lattice structure in a truss type according to embodiment 1 of the present invention;
fig. 2 is a schematic structural diagram of a neural network model provided in embodiment 1 of the present invention;
FIG. 3 is a flow chart of data set acquisition and training of the neural network model provided in embodiment 1 of the present invention;
fig. 4 is a schematic diagram of a default cell basic shape, and sample cell nodes and sample bars within sample cells according to embodiment 1 of the present invention;
fig. 5 is a schematic diagram of a method for converting sample cell nodes and sample bars of sample cells into sample graph nodes and sample edges according to embodiment 1 of the present invention;
fig. 6 is a deformation topography diagram of a sample truss-like lattice structure under quasi-static compressive loading according to embodiment 1 of the present invention;
fig. 7 is a stress-strain curve of a sample truss-type lattice structure under quasi-static compressive loading according to example 1 of the present invention;
fig. 8 is a graph showing a relationship between a compressive strength prediction result and a true compressive strength of a truss-type lattice structure of a test set sample provided in embodiment 1 of the present invention;
fig. 9 is a block diagram of a system for predicting mechanical properties of a lattice structure in a truss type according to embodiment 2 of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The Graph neural network is an important branch of a deep learning model, and can be used for exploring rules from a data structure such as a Graph (Graph) and realizing a Node-level or Graph-level prediction task. The neural network has been applied to chemical property prediction of drug molecules and novel drug development. Nodes and rods inside the lattice structure can be analogous to atoms and chemical bonds of chemical molecules, respectively. Furthermore, any unit cell can be converted into a representation mode of a graph, and the mechanical property prediction of the truss lattice structure is realized through a graph neural network model.
The invention aims to provide a method and a system for predicting mechanical properties of a truss type lattice structure, which improve the universality and the speed of predicting the mechanical properties of the truss type lattice structure.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
Example 1
The embodiment provides a method for predicting mechanical properties of a lattice structure, which is shown in fig. 1 and includes:
s101: acquiring a diagram of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of the cell elements; the graph nodes of the graph correspond to cell nodes of the cells; the sides of the figure correspond to the rods of the cell.
S102: inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property predicted value of the target truss type lattice structure; the node information comprises location information of the cell node within the cell; the side information comprises diameter information, cross-sectional area information and/or cross-sectional shape information of the rod pieces of the cells; the trained graph neural network model is a model obtained by training by taking sample node information and sample side information of a sample graph of a sample cell as input and taking a real value of the sample mechanical property of the sample truss type lattice structure as a label; the sample truss type lattice structure is obtained by periodically arranging a plurality of sample cells. In this embodiment, the trained Neural Network model is a signaling Neural Network Model (MPNN), and the structure of the trained Neural Network model is shown in fig. 2.
The trained graph neural network model comprises an information transmission module and a reading module;
the information transmission module is used for carrying out iterative update on the node characteristic vector of each graph node according to the node characteristic vector of the adjacent graph node connected with the graph node through the edge to obtain an updated node characteristic vector;
and the reading module is used for splicing all the updated node characteristic vectors and inputting a full-connection layer neural network which takes a mechanical property numerical value as output to obtain a mechanical property predicted value of the target truss type lattice structure.
Specifically, the information transfer module is responsible for performing T-round information transfer between nodes of the graph, and finally integrating the structural information in the graph and mapping the structural information into a multidimensional vector. And the reading module is responsible for mapping the multidimensional vector to a mechanical property space and outputting a mechanical property predicted value.
Any graph node v of the graph of the input neural network i All depend on non-directional edges (i.e. edges of the graph) to connect with adjacent graph nodes and are given a random initialization node feature vector h i 0 The node feature vector is also referred to as the state of the node. After entering the information transfer module, in each round of information transfer, any graph node updates its own node state based on the current node state (i.e. current node feature vector) of the neighboring graph node, and the node state update formula is as follows:
wherein the content of the first and second substances,
in order that the graph node with the number i is adjacent to the graph node v in the process of the t +1 th iteration information transfer j Integration of the received information, A ij Is a weight matrix learned by a neural network,
for nodes v of adjacent graphs j Current node feature vector, N (v) of the t-th iterative information transfer process i ) Finger diagram node v i Of the neighbor graph node v j J is a neighborhoodThe number of graph nodes. The GRU is a Gated Recurrent Unit (GRU) belonging to a recurrent neural network, and can be based on integrated information
Node v of the graph i Node state of
Is updated to
After T rounds of iterative information transmission, all graph nodes obtain final states
And the final state h of each graph node is read by the reading module T Further integrating and mapping to output the mechanical property value predicted by the model
It is not determined by the node state of a single graph node, but by the structure information on the whole graph hierarchy. The information processing of the readout module can be represented by the following formula:
where f (-), m (-), n (-), denote fully connected neural network layers, an element-by-element multiplication symbol, and V denotes a set of graph nodes.
In this embodiment, the method further includes a training process of the graph neural network model, including the following steps:
acquiring a data set; the data set comprises sample node information and sample side information of a sample graph of each sample cell in a plurality of sample cells with different configurations and a sample mechanical property real value of the sample truss type lattice structure corresponding to the sample cell.
And training the graph neural network model by using the data set to obtain the trained graph neural network model.
The acquiring the data set specifically includes:
acquiring a plurality of sample cells with different configurations;
representing each of the sample cells based on graph theory to obtain a sample map of each of the sample cells; a sample map node of the sample map is a sample cell node of the sample cell; a sample edge of the sample map is a sample bar of the sample cell; the sample node information comprises location information of the sample cell node within the sample cell; the sample side information comprises diameter information, cross-sectional area information, and/or cross-sectional shape information of sample rods of the sample cells;
periodically arranging each sample cell to obtain a plurality of sample truss type lattice structures;
performing mechanical property test on each sample truss type lattice structure under a preset load condition to obtain a numerical value of a mechanical property index; the numerical value of the mechanical property index is a real value of the sample mechanical property of the sample truss type lattice structure; the mechanical property indexes comprise elastic modulus, yield strength, compression strength, tensile strength and bending strength.
The acquiring a plurality of sample cells of different configurations includes:
And 3, taking the selected multiple sample reference points as sample cell nodes of the sample cells, and taking all sample rods in a sample rod set as sample rods of the sample cells to obtain the sample cells.
And 4, repeating the steps 2-3 to obtain a plurality of sample cells with different configurations.
The following describes the data set acquisition and training process of the neural network model, as shown in fig. 3:
s201: selecting a basic cell shape of a sample cell, establishing a space coordinate system, and placing the basic cell shape in a three-dimensional coordinate system, specifically: a polyhedron capable of filling a three-dimensional space without a gap by periodic arrangement is selected as a basic shape of a cell of a lattice structure (i.e., a basic shape of the cell), a spatial coordinate system is established, and the basic shape of the cell is placed in the three-dimensional coordinate system. In this embodiment, a cube is selected as the basic shape of the cell, and the side length of the cube is set to be 5mm. The cube is placed in the coordinate system in the manner shown in fig. 4.
S202: a plurality of sample reference points are selected within and/or at the boundary of the basic cell shape, and a set of sample reference points is formed. In this embodiment, 27 sample reference points are selected to form a sample reference point set Q0, which includes the body center, the face center, the vertex, and the midpoint of each edge of the cube. Reference point q of each sample i A number i is given, the spatial position of the sample reference point is identified by using the sphere in fig. 4, the spatial coordinates of the 27 sample reference points in this embodiment adopt the side length of a cube as a scale, for example, the coordinates of the sample reference point on the x-axis are (0.5, 0), and the spatial coordinates and the numbers of the 27 sample reference points selected in this embodiment are as shown in table 1.
TABLE 1 spatial coordinates and numbering of sample reference points
S203: selecting a plurality of pairs of sample reference points in the sample reference point set pairwise, allowing the same sample reference point to exist in the selected plurality of pairs of sample reference points at the same time, and constructing a sample rod piece by taking each pair of sample reference points as end points to obtain a sample rod piece set. Specifically, the method comprises the following steps:
from the set Q 0 Mid-pair selection of sample reference points (q) i ,q j ) I is not equal to j, and a sample rod l is established by taking two coordinate points as end points ij According to the method, a plurality of pairs of sample reference points are selected, a plurality of mutually non-overlapping and non-intersecting sample rod pieces corresponding to the sample reference points are established, and a sample rod piece set L is formed, so that all sample rod pieces in the sample rod piece set L can effectively bear loads in all directions through a truss type lattice structure formed by rigid connection, and the sample rod piece set L exists as a sample cell configuration. Repeating the steps to obtain a sample cell library formed by M sample cells with different configurations. Fig. 4 shows one of the sample cells generated according to the step S203, in which the gray spheres represent the sample cell nodes selected as sample cells, the black spheres represent the sample reference points of the sample cell nodes not selected as sample cells, and the gray columns represent the sample rods constituting the sample cells. In the present embodiment, the cross-sectional shape of the sample rod is set to be circular for all sample cells in the sample cell library, and the diameter of the sample rod is 0.5mm.
S204: the structure information of each sample cell is represented by a graph, the sample cell internal sample cell nodes are sample graph nodes of the sample graph, and the sample cell internal rods are sample edges of the sample graph, specifically:
for each sample cell in the sample cell library, a Graph (Graph) representation is established for the structure information of each sample cell, a sample Graph of each sample cell is obtained, all sample cell nodes are abstracted to be sample Graph nodes of the sample Graph, and sample rods in the sample cells are abstracted to be sample edges of the sample Graph. Spatial position information of the sample cell nodes is stored in the sample graph nodes of the sample graph, and information of the sample rods is stored in the sample edges of the sample graph.
Specifically, the sample graph is represented by G = (V, E), and is composed of a sample graph node set V and a sample edge set E. The information of the sample graph node set V and the sample edge set E is stored by the node feature matrix N and the edge feature tensor E respectively. Fig. 5 (b) is a sample graph G obtained by converting the sample cell shown in fig. 5 (a) into a sample graph including 6 sample graph nodes, where the node feature matrix N can be expressed as:
the feature vector representation of the node space position can adopt a space coordinate form or a one-hot coding form. Suppose the ith sample graph node v i With a set of sample reference points Q 0 Middle numbered k sample reference point q k Correspondingly, the feature vector in the form of spatial coordinates is represented as
Wherein
For a sample graph node v i Corresponding sample reference point q k Coordinate position in a spatial coordinate system. The feature vector in the one-hot coding form is selected in this embodiment to represent the spatial position of the sample graph node. For the ith sample graph node V in the sample graph node set V i ，v i Only the kth element in (1) and the remaining elements are 0. The cell shown in figure 5 (a) comprises 6 sample map nodes and a set of sample reference points Q 0 The corresponding relationship of the middle sample reference points is shown in table 2:
TABLE 2 correspondence between node number and coordinate point number
According to the above table 2, a node feature matrix N can be obtained, where the obtained N is a sparse matrix of 6 × 27, and the triplet i, j, k can be expressed as:
where i is the row number, j is the column number, and k is the value of a non-zero element.
In this embodiment, since G is an undirected graph, the edge feature tensor E has symmetry and can be expressed as:
wherein the content of the first and second substances,
storing a single sample edge e for the edge feature vectors of sample rods with sample graph node i and sample graph node j as endpoints ij E, i.e. the diameter, cross-sectional area and/or cross-sectional shape of the sample rod within the sample cell. In this embodiment, the only sample side information included in the sample side of the sample graph G is the diameter of the rod with the cylindrical cross section, so the side feature vector e ij Degenerated into a scalar e representing the diameter value (in mm) ij The edge feature tensor E becomes a sparse and symmetric feature matrix as follows:
wherein, when the element in E is 0, the representative and edge feature vector E ij The corresponding sample rod is absent; e is 0.5, represents the edge feature vector E ij The corresponding sample rod is present and has a diameter of 0.5mm.
S205: acquiring a sample truss type lattice structure mechanical property value corresponding to each sample cell, and establishing a data set, specifically:
and periodically arranging each sample cell to obtain a sample truss type lattice structure with a specific shape specification corresponding to each sample cell, and obtaining a true value of the mechanical property of the sample truss type lattice structure under the same load condition.
The specification of the sample truss type lattice structure refers to the size of a single cell model and the number of periodical repetitions in a three-dimensional space, and the number of repetitions of cells in three array directions which are not overlapped with each other is required to be enough, so that the mechanical property of the sample truss type lattice structure is not significantly changed along with the further increase of the number of repetitions of the cells.
The sample graph representing the structural information of the specific cell and the corresponding actual value of the mechanical property of the sample form a data point. The mechanical property test is carried out on the sample truss type lattice structure under the three load conditions of compression, tension and bending, and the indexes obtained by the mechanical property test are respectively as follows: and under the condition of compressive load, the value of the elastic modulus and/or the compressive strength is used as the actual mechanical property value of the sample truss-type lattice structure, under the condition of tensile load, the value of the elastic modulus and/or the tensile strength is used as the actual mechanical property value of the sample truss-type lattice structure, and under the condition of bending load, the value of the bending strength is used as the actual mechanical property value of the sample truss-type lattice structure. The data points corresponding to all sample cells in the sample cell library comprise a data set. In this embodiment, a finite element simulation method is used to establish a finite element model of a lattice structure (with dimensions of 20mm × 20mm × 20 mm) with a repetition number of 4 × 4 × 4 in x, y, and z directions for 2000 sample cells in a sample cell library, and simulate the deformation process of the lattice structure under quasi-static compressive loading (as shown in fig. 6). In this embodiment, a first stress peak value, i.e. a value of quasi-static compressive strength, is extracted from an engineering stress-strain curve (as shown in fig. 7) and is used as a true value of a sample mechanical property of a sample truss-type lattice structure.
The above numerical simulation is performed on the 2000 selected sample cells respectively to obtain the actual value of the mechanical property of the sample of each sample cell, so as to obtain a data set containing 2000 data points. Setting the data proportion of the training set to be 0 < m < 1, dividing the data set into a training set and a testing set according to the proportion of m (1-m) for training the graph neural network model, and preferably selecting the division proportion m of the training set data from the range of 0.5-0.95. In this embodiment, the data sets are represented as 4: a scale of 1 is divided into a training set and a test set.
S206: training the graph neural network model by adopting a data set, specifically:
and carrying out standardization operation on the divided training set data, then inputting the standardized training set data into the graph neural network model for training to obtain the trained graph neural network model, and predicting the mechanical property of the trained graph neural network model based on the sample graph storing the configuration structure information of the specific cell. And predicting the mechanical property of the sample truss type lattice structure corresponding to the sample cell in the test set by adopting the trained graph neural network model, wherein the predicted deviation rate of the mechanical property is less than 15%. The deviation ratio calculation formula is as follows:
wherein E is rr To predict the deviation ratio, Z is the test set data volume, y i To test the actual mechanical properties of the samples in the ith sample cell in the set,
to test the predicted value of the mechanical property of the sample in the ith sample cell in the set, Z =400 in this embodiment.
FIG. 8 shows the comparison of the quasi-static compressive strength prediction results (sample mechanical property predicted values) and the true strength (sample mechanical property true values) of all sample truss-type lattice structures in the test set by the trained neural network model, and it can be obtained from FIG. 8 that all sample mechanical property predicted values are near the baseline, R is near the baseline 2 The value reached 0.96 and the deviation rate for predicting the intensity was 8.67%. This illustrates the present embodimentThe used graph neural network model can accurately predict the mechanical properties of the sample truss type lattice structure with different cell configurations.
Compared with the prior art, the invention has the following advantages:
firstly, the invention abstracts the cell nodes and the member bars in the lattice structure cell into the graph nodes and the edges of the graph, respectively, and is a simple, efficient and universal structure information storage method, and the universality of the method is embodied in that the cell with macroscopically different basic cell shapes and microscopically different geometric characteristics of the member bars can be stored into the graph in the same form.
Secondly, the graph neural network model used by the method can predict the mechanical properties of the truss lattice structure with different cell configurations with higher precision and less time. The mechanical property prediction speed of the trained graph neural network model on the new target truss type lattice structure is far higher than that of numerical simulation.
Example 2
The embodiment provides a mechanical property prediction system of a lattice structure of a truss type, referring to fig. 9, the system includes:
the image acquisition module T1 is used for acquiring an image of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of the cell elements; the graph nodes of the graph correspond to cell nodes of the cells; the sides of the figure correspond to the rods of the cell.
The mechanical property prediction module T2 is used for inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property prediction value of the target truss type lattice structure; the node information comprises location information of the cell node within the cell; the side information comprises diameter information, cross-sectional area information and/or cross-sectional shape information of the rod pieces of the cells; the trained graph neural network model is obtained by training by taking the sample node information and the sample side information of the sample graph of the sample cell as input and taking the real value of the sample mechanical property of the sample truss-type lattice structure as a label; the sample truss type lattice structure is obtained by periodically arranging a plurality of sample cells.
For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the foregoing, the description is not to be taken in a limiting sense.
Claims (10)
1. A method for predicting mechanical properties of a truss type lattice structure is characterized by comprising the following steps:
acquiring a diagram of a cell of a target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of the cell elements; the graph nodes of the graph correspond to cell nodes of the cells; the edges of the graph correspond to the bars of the cells;
inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property predicted value of the target truss type lattice structure; the node information comprises location information of the cell node within the cell; the side information comprises diameter information, cross-sectional area information and/or cross-sectional shape information of the rod pieces of the cells; the trained graph neural network model is obtained by training by taking the sample node information and the sample side information of the sample graph of the sample cell as input and taking the real value of the sample mechanical property of the sample truss-type lattice structure as a label; the sample truss type lattice structure is obtained by periodically arranging a plurality of sample cells.
2. The method as claimed in claim 1, wherein the trained neural network model is a signal transmission neural network model.
3. The method for predicting the mechanical property of the lattice structure of the truss type as claimed in claim 2, wherein the trained neural network model comprises an information transmission module and a reading module;
the information transmission module is used for carrying out iterative update on the node characteristic vector of each graph node according to the node characteristic vector of the adjacent graph node connected with the graph node through the edge to obtain an updated node characteristic vector; and the reading module is used for splicing all the updated node characteristic vectors and inputting a full-connection layer neural network which takes a mechanical property numerical value as output to obtain a mechanical property predicted value of the target truss type lattice structure.
4. The method for predicting mechanical properties of a lattice-tied truss structure according to claim 1, wherein before inputting the node information and the side information of the graph into the trained graph neural network model, the method further comprises acquiring a data set and training the graph neural network model, and the method comprises the following steps:
acquiring a data set; the data set comprises sample node information and sample edge information of a sample graph of each sample cell in a plurality of sample cells with different configurations and a sample mechanical property real value of the sample truss type lattice structure corresponding to the sample cell;
and training the graph neural network model by using the data set to obtain the trained graph neural network model.
5. The method for predicting mechanical properties of a lattice structure of a truss type according to claim 4, wherein the acquiring a data set specifically includes:
acquiring a plurality of sample cells with different configurations;
representing each of the sample cells based on graph theory to obtain a sample map of each of the sample cells; a sample map node of the sample map is a sample cell node of the sample cell; a sample edge of the sample map is a sample bar of the sample cell; the sample node information comprises location information of the sample cell node within the sample cell; the sample side information comprises diameter information, cross-sectional area information, and/or cross-sectional shape information of sample rods of the sample cells;
periodically arranging each sample cell to obtain a plurality of sample truss type lattice structures;
performing mechanical property test on each sample truss type lattice structure under a preset load condition to obtain a numerical value of a mechanical property index; the numerical value of the mechanical property index is a real value of the sample mechanical property of the sample truss type lattice structure; the mechanical property indexes comprise elastic modulus, yield strength, compression strength, tensile strength and bending strength.
6. The method of claim 5, wherein the obtaining a plurality of sample cells of different configurations comprises:
step 1, selecting a plurality of sample reference points in the basic shape of a preset cell and/or on the boundary of the basic shape of the preset cell to obtain a sample reference point set; the boundary of the basic shape of the preset cell comprises the edges of the basic shape of the preset cell and a surface formed by the edges;
step 2, selecting a plurality of pairs of sample reference points in the sample reference point set, allowing the same sample reference point to exist in the selected plurality of pairs of sample reference points at the same time, and constructing a sample rod piece by taking each pair of sample reference points as end points to obtain a sample rod piece set; all of the sample rods in the set of sample rods are non-overlapping and non-intersecting with each other;
step 3, using the selected multiple sample reference points as sample cell nodes of the sample cells, and using all the sample rods in the sample rod set as sample rods of the sample cells to obtain the sample cells;
and 4, repeating the steps 2-3 to obtain a plurality of sample cells with different configurations.
7. The method for predicting mechanical properties of the lattice structure of the truss type according to claim 5, wherein the predetermined load condition is a compressive load, a tensile load or a bending load.
8. The method of claim 6, wherein the basic shape of the predetermined cell is a polyhedron; the polyhedron can fill a three-dimensional space without a gap by a periodic arrangement.
9. The method according to claim 8, wherein the basic shape of the predetermined cell is a cube, rhombohedron, triangular prism, hexagonal prism, rhombohedron or tetrakaidecahedron.
10. A system for predicting mechanical properties of a lattice-truss structure, comprising:
the image acquisition module is used for acquiring an image of a cell of the target truss type lattice structure; the target truss type lattice structure is obtained by periodically arranging a plurality of the cell elements; the graph nodes of the graph correspond to cell nodes of the cells; the sides of the graph correspond to the bars of the cells;
the mechanical property prediction module is used for inputting the node information and the side information of the graph into a trained graph neural network model to obtain a mechanical property prediction value of the target truss type lattice structure; the node information comprises location information of the cell node within the cell; the side information comprises diameter information, cross-sectional area information and/or cross-sectional shape information of the rod pieces of the cells; the trained graph neural network model is obtained by training by taking the sample node information and the sample side information of the sample graph of the sample cell as input and taking the real value of the sample mechanical property of the sample truss-type lattice structure as a label; the sample truss type lattice structure is obtained by periodically arranging a plurality of sample cells.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211312327.3A CN115762675A (en) | 2022-10-25 | 2022-10-25 | Method and system for predicting mechanical property of truss type lattice structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211312327.3A CN115762675A (en) | 2022-10-25 | 2022-10-25 | Method and system for predicting mechanical property of truss type lattice structure |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115762675A true CN115762675A (en) | 2023-03-07 |
Family
ID=85353296
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211312327.3A Pending CN115762675A (en) | 2022-10-25 | 2022-10-25 | Method and system for predicting mechanical property of truss type lattice structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115762675A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116127655A (en) * | 2023-04-17 | 2023-05-16 | 之江实验室 | Method and device for manufacturing buffer assembly, storage medium and electronic equipment |
-
2022
- 2022-10-25 CN CN202211312327.3A patent/CN115762675A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116127655A (en) * | 2023-04-17 | 2023-05-16 | 之江实验室 | Method and device for manufacturing buffer assembly, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Oishi et al. | Computational mechanics enhanced by deep learning | |
| Lefik et al. | Artificial neural network as an incremental non-linear constitutive model for a finite element code | |
| Zheng et al. | Deep learning in mechanical metamaterials: from prediction and generation to inverse design | |
| CN115762675A (en) | Method and system for predicting mechanical property of truss type lattice structure | |
| CN114861890B (en) | Method and device for constructing neural network, computing equipment and storage medium | |
| CN114418105B (en) | Method and device for processing quantum application problem based on quantum circuit | |
| CN112560966A (en) | Polarimetric SAR image classification method, medium and equipment based on scattergram convolution network | |
| Oishi et al. | A surface-to-surface contact search method enhanced by deep learning | |
| Mason et al. | Statistical topology of cellular networks in two and three dimensions | |
| Gan | Computational modeling of tensegrity structures | |
| CN112949944B (en) | Intelligent groundwater level prediction method and system based on space-time characteristics | |
| Li et al. | An integrated surrogate model-driven and improved termite life cycle optimizer for damage identification in dams | |
| CN116206690B (en) | Antibacterial peptide generation and identification method and system | |
| CN107203587B (en) | Method for constructing direction adaptability characteristic of geomagnetic map | |
| CN110457841B (en) | Method for representing mechanical property of 3D printing diamond lattice structure | |
| CN111505706A (en) | Microseism P wave first arrival pickup method and device based on depth T-Net network | |
| CN116011682A (en) | Meteorological data prediction method and device, storage medium and electronic device | |
| CN112000673B (en) | Method and device for inquiring transaction elements by quantum circuit | |
| CN113935458A (en) | Air pollution multi-site combined prediction method based on convolution self-coding deep learning | |
| KR102572429B1 (en) | Method, apparatus and storage for storing a program for multi-demensional matrix multiplication | |
| CN115600449B (en) | Simulation prediction method of large-scale particle system based on slender flexible object composition | |
| CN115267883B (en) | Earthquake response prediction model training and predicting method, system, equipment and medium | |
| Wang et al. | Learning the intuitive physics of non-rigid object deformations | |
| CN115937340A (en) | Image reconstruction algorithm of capacitance tomography system based on lstm | |
| CN117173552B (en) | Underwater target detection method, system, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |