A dimensionality reduction method, electronic device, and medium for a three-dimensional mesh model

Abstract

The present invention provides a method for dimensionality reduction of a three-dimensional mesh model, an electronic device, and a medium. The dimensionality reduction method extracts the simulation result data of the three-dimensional mesh model into a matrix, then maps it into an image, adjusts the amount of data participating in the calculation through pixel adjustment, and then screens out the dimensionality reduction region and the non-dimensionality reduction region based on the color gradient change, effectively avoiding the problems of difficult solution and long time consumption caused by huge dimensions during the calculation of grid node data. The dimensionality reduction method realizes the dimensionality reduction of the mesh model in the dimensionality reduction region and the prediction of node information in the non-dimensionality reduction region through a neural network algorithm, realizes high-precision dimensionality reduction of the three-dimensional mesh model, obtains complete grid node information, achieves fast simulation, and realizes real-time diagnosis of the operating state of the physical device.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural machinery, and particularly relates to a method for dimension reduction of a three-dimensional grid model, an electronic device, and a medium. Background Art

[0002] Simulation technology can help designers provide design basis and guidance. Most simulation models are grid models, including fields such as structure, fluid, and electromagnetism. Existing grid model calculations mainly rely on finite element simulation software such as ABAQUS, ANSYS, FLUENT, STARCCM+, etc. for solution calculations. For some complex models, in order to ensure calculation accuracy, it is often necessary to divide dozens or even millions of grids. When solving the grid model, a multi-core parallel simulation method is often used for solution, which not only consumes a large amount of CPU resources but also takes a long time.

[0003] There are three types of grid models, namely one-dimensional, two-dimensional, and three-dimensional. One-dimensional and two-dimensional grid models are simple in model and fast in calculation speed, and do not require dimension reduction. In a conventional simulation calculation, the number of grids of a three-dimensional grid model can usually reach one hundred thousand, or even millions or tens of millions. More grid numbers can often effectively improve the calculation accuracy of the model, but at the same time, it also requires a large amount of computing resources. The conventional calculation method of the grid model depends on finite element analysis software, which takes a lot of time. At the same time, the characteristics of its data itself are less input and huge output dimensions. When using traditional machine learning algorithms for surrogate model dimension reduction, the curse of dimensionality will occur, and it is difficult to construct a surrogate model.

[0004] In the prior art, the method for constructing a digital twin model uses the principal component analysis method to reduce the data dimension. The disadvantages are that, on the one hand, there is no specific physical meaning, and on the other hand, data loss will occur during data restoration, and 100% data restoration cannot be achieved. On the other hand, this type of algorithm is applicable to input data with correlation, and will not be applicable when the input data is completely orthogonal or uncorrelated. The existing model dimension reduction calculation method uses a parallel simulation method to improve the simulation efficiency. The disadvantage is that a large amount of computing resources are required to complete the solution calculation. The fast reanalysis method excludes the grids that do not participate in the calculation in the model first, which can narrow the simulation solution range. The disadvantage is that the initial matrix dimension of the model is huge. If the dimension of the processed data is not restricted, simple data processing still requires a large amount of running memory, and the solution speed of the grid model after narrowing the range still cannot meet the needs of designers for rapid evaluation after scheme change.

[0005] In summary, there is an urgent need for a method for dimension reduction of a three-dimensional grid model that can better respond to design requirement changes and timely evaluate the involved scheme to solve the problems existing in the prior art. Summary of the Invention

[0006] The object of the present invention is to provide a method for dimension reduction of a three-dimensional mesh model, an electronic device and a medium. The specific technical solutions are as follows:

[0007] A method for dimension reduction of a three-dimensional mesh model, the specific steps are as follows:

[0008] Step S1: Simulation of the three-dimensional mesh model. Specifically, finite element analysis is performed on the three-dimensional mesh model to obtain the spatial coordinates and node information of each mesh node. Based on the spatial coordinates and node information, the three-dimensional mesh model is simulated multiple times to obtain simulation result data. The simulation result data is saved as matrix A in the form of a matrix. The simulation result data includes the node maximum value, node spatial coordinate information, node stress value, and node strain value;

[0009] Step S2: Divide the dimension reduction area. Specifically, the nodes of matrix A in step S1 are given colors and a node three-dimensional map is established. The greater the stress value in the simulation result data, the darker the color, and vice versa. The node three-dimensional map is converted into a red-green-blue value table. Based on the color difference change, the area where the color difference changes in the node three-dimensional map is divided into the dimension reduction area. The nodes in the dimension reduction area are reconstructed into a matrix and saved as matrix B. The nodes outside the dimension reduction area are reconstructed into a matrix and saved as matrix C;

[0010] Step S3: Perform dimension reduction processing to obtain a dimension reduction model. Specifically, a training network is established through a neural network algorithm. The simulation boundary conditions are used as input parameters, and the node spatial coordinate information, node stress value, and node strain value of matrix B in step S2 are used as output parameters. Matrix B is processed for dimension reduction to obtain a dimension reduction model.

[0011] Specifically, in step S1, the node information includes the range of force application conditions. The three-dimensional mesh model selects at least two force application conditions within the range of force application conditions for simulation calculation. The force application conditions include the maximum force application condition and the minimum force application condition.

[0012] Specifically, in step S2, the specific steps of assigning colors to the nodes in matrix A are as follows:

[0013] Establish a color interval, divide the interval into N small intervals equally and assign different colors to different intervals. The value of N is 3-50; the node stress values of the nodes are normalized, and the nodes are mapped to the color interval one by one according to the node stress values to obtain a node three-dimensional map.

[0014] Specifically, the range of the color interval is [0, 1].

[0015] Specifically, in step S2, the numerical values of colors are all represented by numbers from 0 to 256. When any one of these numerical values changes by more than K, it is considered that the color difference has changed. The value range of K is from 0 to 50.

[0016] Specifically, in step S3, the neural network algorithm structure includes an input layer, an output layer, a hidden layer, and an export layer. The input layer is used to write input parameters, the output layer is used to present output parameters, the hidden layer is used to calculate the dimensionality reduction model, and the export layer is used to save the dimensionality reduction model.

[0017] Specifically, in step S3, the parameters of the hidden layer are set as follows: the number of hidden layers is set to 5 - 100, the activation function is the sigmoid or tanh function, the validation data and test data are selected as 5 - 30% of the total data volume, and the number of iterations is 100 - 10000.

[0018] In addition, the present invention also provides an electronic device, including:

[0019] A memory for storing a computer program;

[0020] A processor for implementing the dimensionality reduction method as described above when executing the computer program.

[0021] In addition, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the dimensionality reduction method as described above is implemented.

[0022] Applying the technical solution of the present invention has the following beneficial effects:

[0023] The dimensionality reduction method of the present invention can perform data matrix processing on any mesh model, is applicable to mesh models in multiple fields, and reduces the capacity of the simulation result file.

[0024] The dimensionality reduction method of the present invention divides the dimensionality reduction area through image analysis, uses pixel adjustment to maximize the calculation amount, and greatly reduces the data operation amount.

[0025] The dimensionality reduction method of the present invention realizes the prediction of node information in the dimensionality reduction area and non-dimensionality reduction area of the mesh model in the dimensionality reduction area through the neural network algorithm, and at the same time can realize the rapid calculation of simulation results, realizing the rapid simulation technology.

[0026] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The present invention will be further described in detail below with reference to the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which form a part of this application, are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0028] Figure 1 is a flowchart of the steps of the dimensionality reduction method;

[0029] Figure 2 is a schematic diagram of a three-dimensional grid model;

[0030] Figure 3 is a schematic diagram of the coordinate system of the three-dimensional view of the nodes;

[0031] Figure 4 is a schematic diagram of the selection of a color map (schematic with a grayscale map);

[0032] Figure 5 is Figure 3 a schematic diagram of the coordinate system for removing the area without color difference change of

[0033] Figure 6 is a schematic diagram of the neural network algorithm structure. Detailed Embodiment

[0034] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, the present invention can be implemented in many different ways as defined and covered by the claims.

[0035] Embodiment 1:

[0036] To solve the technical problems that the simulation model runs slowly, the result file is large and not easy to save, and it cannot quickly respond to the evaluation requirements of design changes, as Figure 1 shown, this embodiment discloses a dimensionality reduction method for a three-dimensional grid model. The preferred three-dimensional grid model in this embodiment is as Figure 2 shown. It should be noted that the dimensionality reduction model is a black box model, and the specific steps of the dimensionality reduction method are as follows:

[0037] Step S1: Simulation of the three-dimensional grid model. Specifically, finite element analysis is performed on the three-dimensional grid model to obtain the spatial coordinates and node information of each grid node. The node information includes the range of force application conditions. At least two force application conditions are selected from the range of force application conditions for the three-dimensional grid model for simulation calculation. The force application conditions include the maximum force application condition and the minimum force application condition. Based on the spatial coordinates and the force application conditions, the three-dimensional grid model is simulated multiple times to obtain simulation result data. The simulation result data is saved as matrix A in the form of a matrix. The rows of matrix A represent the node numbers. In this embodiment, the preferred order of node numbers is the same as the order of nodes in the finite element software. The columns of matrix A are sequentially placed with the simulation result data. Matrix processing is adopted, which is applicable to multi-domain grid models and reduces the capacity of the simulation result file;

[0038] Further, the simulation result data includes node maximum values, node space coordinate information, node stress values, and node strain values. In this embodiment, the first column of the preferred matrix A stores the node number, the second column, the third column, and the fourth column of the matrix A store the node space coordinate X, Y, and Z information in sequence, the fifth column of the matrix A stores the node stress value, and the sixth column of the matrix A stores the node strain value.

[0039] It should be noted that the node numbering sequence may also be arranged from top to bottom or from left to right, and the node numbering does not affect the dimensionality reduction effect of this embodiment.

[0040] Step S2: Dividing the dimension reduction area, specifically, assigning colors to the nodes of the matrix A in step S1 and establishing a three-dimensional graph of the nodes, the greater the stress value in the simulation result data, the darker the color, and vice versa, the lighter the color, converting the three-dimensional graph of the nodes into a red-green-blue numerical table (RGB numerical table), dividing the area in the three-dimensional graph of the nodes where the color difference changes as a dimension reduction area based on the color difference change, reconstructing the matrix of the nodes in the dimension reduction area and saving it as a matrix B, and reconstructing the matrix of the nodes outside the dimension reduction area and saving it as a matrix C;

[0041] In this embodiment, the specific steps of assigning colors to nodes in matrix A are preferably as follows:

[0042] A color interval is established, and its range is [0, 1]. Each node is assigned an initial color. The interval is divided into seven small intervals and defined as red, orange, yellow, green, cyan, blue, and purple intervals respectively. The node stress values ​​of the nodes are normalized. The nodes and color intervals are mapped one by one according to the node stress values ​​to obtain a three-dimensional diagram of the nodes. Figure 3 The three-dimensional graph of the nodes shown in Figure 3 In the figure, grayscale contrast is used as color distinction. The darker the grayscale, the node is in the purple interval or close to the purple interval, and the lighter the grayscale, the node is in the red interval or close to the red interval.

[0043] Furthermore, if Figure 4 As shown in the figure, the middle cube area is the node reconstruction area. It is perpendicular to the YOZ plane. Two 1920×1080 pixel node reconstruction color images are selected from the t1 and t2 directions respectively. The node reconstruction color images are converted into RGB value tables, and the spatial areas where the color difference exists are eliminated. The following is obtained: Figure 5 The three-dimensional graph of nodes with no color difference change areas removed (using grayscale contrast to represent color difference) divides the dimensionality reduction area through image analysis, and uses pixels to adjust the maximum amount of calculation, which greatly reduces the amount of data calculation;

[0044] Further, in this embodiment, preferably, the values of red - green - blue are all represented by numbers from 0 to 256. The color difference change is considered to occur when any one of the red - green - blue values changes by more than K. The value range of K is from 0 to 50, and preferably, the value of K in this embodiment is 10.

[0045] Further, in this embodiment, preferably, the simulation result data stored in matrix C is replaced by variable X. Specifically, the node stress values and node strain values stored in matrix C are respectively assigned to X1 and X2.

[0046] Step S3: Perform dimensionality reduction processing to obtain a dimensionality reduction model. Specifically, establish a training network through a neural network algorithm, use the simulation boundary conditions as input parameters, and use the node space coordinate information, node stress values, and node strain values of matrix B in step S2 as output parameters. Perform dimensionality reduction processing on matrix B to obtain a dimensionality reduction model. The neural network algorithm is used to achieve high - precision dimensionality reduction of the grid model in the dimensionality reduction area and prediction of node information in the non - dimensionality reduction area, and at the same time, fast calculation of simulation results can be realized;

[0047] As Figure 6 shown, the neural network algorithm structure includes an input layer, an output layer, a hidden layer, and an export layer. The input layer is used to write input parameters, the output layer is used to display output parameters, the hidden layer is used to calculate the dimensionality reduction model, and the export layer is used to save the dimensionality reduction model.

[0048] Further, preferably, the parameters of the hidden layer in this embodiment are set as follows: the number of hidden layers is set to 25, the activation function is the sigmoid function, 15% of the total data volume is selected as verification data and test data, and the number of iteration rounds is set to 1000. After the iteration in this embodiment, the correlation coefficient between the dimensionality reduction model and the original model is not less than 99%, realizing the dimensionality reduction processing of the three - dimensional grid model.

[0049] Specifically, the correlation coefficient = Pearson correlation coefficient * 100%. The expression of the Pearson correlation coefficient is as follows:

[0050]

[0051] where cov(x,y) is the covariance, σ is the standard deviation, x represents the original model, and y represents the dimensionality reduction model.

[0052] Further, to better illustrate the advantages and purposes of this embodiment, this embodiment also discloses a fast simulation method based on the above - mentioned dimensionality reduction model. The specific steps are as follows:

[0053] Empty the data in the fifth and sixth columns of matrices A and B, and retain the node numbers and coordinate information. Input the boundary conditions into the dimensionality reduction model, and the model will quickly output the node spatial coordinate information, node stress values, and node strain values in matrix B as output parameters. Assign the node stress values and node strain values in matrix B to the corresponding nodes in matrix A according to the node spatial coordinate information and node numbers in matrix B (corresponding based on node spatial coordinate information and node numbers). For other unassigned nodes in matrix A, set the node information to the node stress values and node strain values. The three-dimensional mesh model solution can be quickly completed according to the output results of the dimensionality reduction model, and the calculation accuracy of the model depends on the correlation coefficient of the third-step dimensionality reduction model. By ensuring that the model correlation coefficient is above 99%, ensure the calculation accuracy of the three-dimensional mesh model.

[0054] In addition, this embodiment also discloses an electronic device, including:

[0055] A memory for storing a computer program;

[0056] A processor for implementing the dimensionality reduction method as described above when executing the computer program.

[0057] In addition, this embodiment also discloses a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the dimensionality reduction method as described above.

[0058] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0059] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0060] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for dimensionality reduction of a three-dimensional grid model, characterized in that, the specific steps are as follows: Step S1: Simulation of the three-dimensional grid model. Specifically, finite element analysis is performed on the three-dimensional grid model to obtain the spatial coordinates and node information of each grid node. Based on the spatial coordinates and node information, the three-dimensional grid model is simulated multiple times to obtain simulation result data. The simulation result data is saved as matrix A in the form of a matrix. The simulation result data includes the node maximum value, node spatial coordinate information, node stress value, and node strain value; Step S2: Divide the dimensionality reduction area. Specifically, the nodes of matrix A in Step S1 are assigned colors and a node three-dimensional map is established. Based on the color difference change, the area where the color difference changes in the node three-dimensional map is divided into the dimensionality reduction area. The nodes in the dimensionality reduction area are reconstructed into a matrix and saved as matrix B, and the nodes outside the dimensionality reduction area are reconstructed into a matrix and saved as matrix C; Step S3: Dimensionality reduction processing to obtain a dimensionality reduction model. Specifically, a training network is established through a neural network algorithm. The simulation boundary conditions are used as input parameters, and the node spatial coordinate information, node stress value, and node strain value of matrix B in Step S2 are used as output parameters. Matrix B is subjected to dimensionality reduction processing to obtain a dimensionality reduction model; In Step S1, the node information includes the range of force application conditions. The three-dimensional grid model selects at least two force application conditions within the range of force application conditions for simulation calculation. The force application conditions include the maximum force application condition and the minimum force application condition.

2. The dimensionality reduction method according to claim 1, characterized in that, in Step S2, the specific steps for assigning colors to the nodes in matrix A are as follows: Establish a color interval, assign an initial color to each node, divide the interval into N small intervals equally and assign different colors to different intervals. The value of N is 3 - 50; the node stress value of the node is normalized, and the node is mapped to the color interval one by one according to the node stress value to obtain a node three-dimensional map.

3. The dimensionality reduction method according to claim 2, characterized in that, the range of the color interval is [0, 1].

4. The dimensionality reduction method according to claim 1, characterized in that, in Step S2, the color value is represented by 0 to 256; when any value changes by more than K, it is considered that the color difference has changed. The value range of K is 0 to 50.

5. The dimensionality reduction method according to claim 1, characterized in that, in Step S3, the neural network algorithm structure includes an input layer, an output layer, a hidden layer, and an export layer. The input layer is used to write input parameters, the output layer is used to display output parameters, the hidden layer is used to calculate the dimensionality reduction model, and the export layer is used to save the dimensionality reduction model.

6. The dimensionality reduction method according to claim 5, characterized in that, in Step S3, the parameters of the hidden layer are set as follows: the number of hidden layers is set to 5 - 100, the activation function is the sigmoid or tanh function, the validation data and test data are selected as 5 - 30% of the total data volume, and the number of iteration rounds is 100 - 10000.

7. An electronic device, characterized in that, comprising: A memory for storing a computer program; A processor for implementing the dimensionality reduction method according to any one of claims 1 to 6 when executing the computer program.

8. A computer-readable storage medium, characterized in that, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the dimensionality reduction method according to any one of claims 1 to 6.

Images