A parameter configuration method for automatic assembly of finite element meshes

Abstract

The invention discloses a parameter configuration method for automatically assembling a finite element mesh, and belongs to the technical field of parameter configuration of a finite element mesh. The invention extracts an image of a first simulation component and an image of a second simulation component, characterizes shape features of the simulation component through the image, characterizes size data of the simulation component through finite element data, uses a matching module to verify whether the constructed first simulation component and the second simulation component match in shape structure, and estimates a matching value. The higher the matching value, the more matched the first simulation component and the second simulation component are in shape structure, thereby having a higher degree of fit and being easier to assemble. Combined with the grid space data and the matching value of the second simulation component, a parameter prediction model is used to predict assembly parameters, thereby avoiding the construction of a mathematical differential equation model and reducing the complexity of obtaining assembly parameters.

Description

Technical Field

[0001] The invention relates to the technical field of parameter configuration of finite element grids, and in particular to a parameter configuration method for automatically assembling finite element grids. Background Art

[0002] In the field of engineering design and simulation, the assembly of computer-aided design (CAD) models is a critical and complex task. The overall CAD model is usually composed of multiple parts, which may have complex geometric shapes and structures, and involve multiple materials and processes. Before engineering analysis and simulation calculations, these CAD models must be assembled into a whole for meshing and simulation calculations. However, the assembly of CAD models is usually complex and cumbersome, and faces many challenges:

[0003] 1. Geometric complexity:

[0004] The geometry of CAD models can be very complex, including curved surfaces, complex curves, and special geometric bodies, making it very difficult to accurately assemble the model, requiring highly precise geometry processing technology and time.

[0005] 2. Component quantity and size differences:

[0006] CAD models are usually composed of multiple parts, each of which may have different sizes, shapes, and materials. The increase in the number of parts makes the assembly process more complicated, especially when there are size differences or interference between parts.

[0007] 3. Engineering constraints and requirements:

[0008] When assembling CAD models, it is usually necessary to consider engineering requirements and design specifications such as clearance, fit, axial alignment, constraints, etc. These constraints increase the complexity of the assembly process and require a high degree of technical ability and expertise.

[0009] 4. Model update and change management:

[0010] As engineering design progresses, CAD models may be frequently changed and updated. Therefore, for assembled models, it is necessary to be able to handle changes quickly and efficiently and ensure that assembly results are updated and accurate in a timely manner.

[0011] In order to meet these challenges, traditional CAD model assembly methods are often time-consuming and error-prone, requiring a lot of manpower and resource investment. Therefore, there is an urgent need for an efficient and reliable finite element mesh model assembly method in this field to improve the efficiency and accuracy of simulation. Summary of the invention

[0012] In view of the above-mentioned deficiencies in the prior art, the present invention provides a parameter configuration method for automatically assembling finite element meshes, which solves the problem of complex model construction and calculation process in the existing methods for predicting mesh pressing force and mesh displacement during assembly of components.

[0013] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: a parameter configuration method for automatically assembling a finite element mesh, comprising the following steps:

[0014] S1, extracting an image of a first simulation component and an image of a second simulation component;

[0015] S2, extracting finite element data of the first simulation component;

[0016] S3, extracting finite element data and grid space data of the second simulation component;

[0017] S4, using a matching module to process the image of the first simulation component, the image of the second simulation component, the finite element data of the first simulation component, and the finite element data of the second simulation component to obtain matching values ​​of the two simulation components;

[0018] S5. When the matching value is greater than the matching threshold, the assembly parameters are predicted using a parameter prediction model according to the grid space data and the matching value of the second simulation component.

[0019] The beneficial effects of the present invention are as follows: in the present invention, the first simulation component is a casting, and the second simulation component is a casting mold. The present invention extracts an image of the first simulation component and an image of the second simulation component, characterizes the shape features of the simulation components through images, characterizes the size data of the simulation components through finite element data, and uses a matching module to verify whether the constructed first simulation component and the second simulation component match in shape and structure, and estimates the matching value. The higher the matching value, the more matched the first simulation component and the second simulation component are in shape and structure, thereby the higher the fit and the easier it is to assemble. Combined with the grid space data and the matching value of the second simulation component, a parameter prediction model is used to predict assembly parameters, thereby avoiding the construction of a mathematical differential equation model and reducing the complexity of obtaining assembly parameters.

[0020] Further, the finite element data of the first simulation component in S2 includes: the number of finite element grids of the inner diameter of the first simulation component, the number of finite element grids of the outer diameter of the first simulation component, and the number of finite element grids of the length of the first simulation component;

[0021] The finite element data of the second simulation component in S3 includes: the number of finite element grids of the inner diameter of the second simulation component, the number of finite element grids of the outer diameter of the second simulation component, and the number of finite element grids of the height of the second simulation component;

[0022] The grid space data in S3 includes: space coordinates and node numbers.

[0023] Further, the matching module in S4 includes: a first image feature extraction unit, a second image feature extraction unit, a first feature splicing unit, a second feature splicing unit, a first multi-scale feature extraction unit, a second multi-scale feature extraction unit and a matching value calculation unit;

[0024] The input end of the first image feature extraction unit is used to input the image of the first simulation component, and the output end thereof is connected to the first input end of the first feature splicing unit; the second input end of the first feature splicing unit is used to input the finite element data of the first simulation component, and the output end thereof is connected to the input end of the first multi-scale feature extraction unit; the input end of the second image feature extraction unit is used to input the image of the second simulation component, and the output end thereof is connected to the first input end of the second feature splicing unit; the second input end of the second feature splicing unit is used to input the finite element data of the second simulation component, and the output end thereof is connected to the input end of the second multi-scale feature extraction unit; the input end of the matching value calculation unit is respectively connected to the output end of the first multi-scale feature extraction unit and the output end of the second multi-scale feature extraction unit, and the output end thereof serves as the output end of the matching module.

[0025] The beneficial effect of the above-mentioned further scheme is: the present invention adopts the second image feature extraction unit to extract the image features of the second simulation component, adopts the first image feature extraction unit to extract the image features of the first simulation component, splices the image features of the second simulation component and the finite element data of the second simulation component, and adopts the second multi-scale feature extraction unit to extract multi-scale features, splices the image features of the first simulation component and the finite element data of the first simulation component, and adopts the first multi-scale feature extraction unit to extract multi-scale features, calculates the matching value of the two multi-scale features through the matching value calculation unit, thereby realizing automatic verification of the matching status of the first simulation component and the second simulation component.

[0026] Further, the first image feature extraction unit and the second image feature extraction unit each include: a first convolution layer, a MaxPool layer, an AvgPool layer, a feature enhancement layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an adder A1 and an upsampling layer;

[0027] The input end of the first convolutional layer serves as the input end of the first image feature extraction unit or the second image feature extraction unit, and its output end is respectively connected to the input end of the MaxPool layer and the input end of the AvgPool layer; the input end of the feature enhancement layer is respectively connected to the output end of the MaxPool layer and the output end of the AvgPool layer, and its output end is respectively connected to the input end of the second convolutional layer and the input end of the third convolutional layer; the output end of the third convolutional layer is connected to the input end of the fourth convolutional layer; the input end of the adder A1 is respectively connected to the output end of the second convolutional layer and the output end of the fourth convolutional layer, and its output end is connected to the input end of the upsampling layer; the output end of the upsampling layer serves as the output end of the first image feature extraction unit or the second image feature extraction unit.

[0028] The beneficial effect of the above-mentioned further scheme is as follows: the present invention adopts the MaxPool layer to extract the significant features of the output of the first convolutional layer, adopts the AvgPool layer to extract the global average features of the output of the first convolutional layer, and enhances the significant features and the average features in the feature enhancement layer, thereby enhancing the adaptive adjustment capability of the first image feature extraction unit or the second image feature extraction unit, and then adopts a two-way convolution structure to construct a residual structure to solve the problem of gradient disappearance and improve the depth of feature extraction.

[0029] Furthermore, the expression of the feature enhancement layer is:

[0030] ,

[0031] Among them, H is the output of the feature enhancement layer, H MaxPool is the output of the MaxPool layer, H AvgPool is the output of the AvgPool layer, FC1 is the first fully connected layer, FC2 is the second fully connected layer, T is the transposition operation, It is Hadamard.

[0032] The beneficial effects of the above further scheme are as follows: the present invention uses the first fully connected layer to process the output of the MaxPool layer, and uses the second fully connected layer to process the output of the AvgPool layer, so as to integrate the extracted features and capture global information, and performs a transposition operation on the output of the first fully connected layer, and uses the Hadamard product to perform feature fusion, so as to achieve enhanced fusion of significant features and average features.

[0033] Furthermore, the expressions of the first multi-scale feature extraction unit and the second multi-scale feature extraction unit are:

[0034] ,

[0035] ,

[0036] ,

[0037] ,

[0038] Wherein, O is the output of the first multi-scale feature extraction unit or the second multi-scale feature extraction unit, S is the output of the feature splicing unit, The window size is The average pooling layer, The window size is The average pooling layer, The window size is The average pooling layer, For Hadamard, The convolution kernel is The convolution layer is σ, σ is the S-type function, Concat is the concatenation operation, X1 is the first scale feature, X2 is the second scale feature, and X3 is the third scale feature.

[0039] The beneficial effects of the above further scheme are as follows: the present invention uses average pooling layers of different window sizes to extract different features, and calculates different attentions based on different features, thereby improving the model's ability to capture important information and enhancing the model's representation and generalization capabilities. The present invention calculates attention based on the average features extracted under each window and assigns weights to the average features, thereby assigning different weights to different parts of information. Important information can receive more attention, while secondary information has a relatively reduced impact. The present invention splices the first scale features, the second scale features, and the third scale features to improve the accuracy of feature extraction.

[0040] Furthermore, the expression of the matching value calculation unit is:

[0041] ,

[0042] Among them, P is the matching value, r 1,i is the i-th element in the output O1 of the first multi-scale feature extraction unit, r 2,i is the i-th element in the output O2 of the second multi-scale feature extraction unit, w i is the i-th weight in the matching value calculation unit, b i is the i-th bias in the matching value calculation element, N is the number of elements in the output O1 of the first multi-scale feature extraction unit or the number of elements in the output O2 of the second multi-scale feature extraction unit, and i is the number of the element.

[0043] The beneficial effect of the above further scheme is: the present invention calculates the difference between the output of the first multi-scale feature extraction unit and the output of the second multi-scale feature extraction unit, and assigns weights and biases to each difference, so as to facilitate the optimization of parameters during matching module training.

[0044] Further, the parameter prediction model in S5 includes: a grid pressing force prediction unit and a grid displacement prediction unit;

[0045] The grid pressing force prediction unit is used to predict the grid pressing force during assembly according to the grid space data and the matching value of the second simulation component;

[0046] The grid displacement prediction unit is used to predict the grid displacement during assembly according to the grid space data of the second simulation component.

[0047] Further, the grid pressing force prediction unit includes: a grid pressing force prediction module and a compensation module;

[0048] The expression of the grid compression force prediction module is:

[0049] ,

[0050] ,

[0051] ,

[0052] ,

[0053] Among them, Y s is the predicted initial grid compression force, tanh is the hyperbolic tangent activation function, and w s1 is the first parameter weight in the grid compression force prediction module, b s1 is the first parameter bias in the grid compression force prediction module, w s2 is the second parameter weight in the grid compression force prediction module, b s2 is the second parameter bias in the grid compression force prediction module, w s3 is the third parameter weight in the grid compression force prediction module, b s3 is the third parameter bias in the grid compression force prediction module, c1 is the grid position priority coefficient 1, c2 is the grid position priority coefficient 2, c3 is the grid position priority coefficient 3, y s1 is the first grid compression force component, y s2 is the second grid compression force component, y s3 is the third grid compression force component, w ys1 is the weight of the first grid compression force component, w ys2 is the weight of the second grid compression force component, w ys3is the weight of the third grid compression force component;

[0054] The expression of the compensation module is:

[0055] ,

[0056] Among them, Y c is the mesh compression force after compensation, e is the natural constant, P is the matching value, and P th is the matching threshold, and γ is the compensation coefficient.

[0057] Furthermore, the expression of the grid displacement prediction unit is:

[0058] ,

[0059] ,

[0060] ,

[0061] ,

[0062] Among them, Y r is the grid displacement, c1 is the grid position priority coefficient 1, c2 is the grid position priority coefficient 2, c3 is the grid position priority coefficient 3, tanh is the hyperbolic tangent activation function, w r1 is the first parameter weight in the grid displacement prediction unit, b r1 is the first parameter bias in the grid displacement prediction unit, w r2 is the second parameter weight in the grid displacement prediction unit, b r2 is the second parameter bias in the grid displacement prediction unit, w r3 is the third parameter weight in the grid displacement prediction unit, b r3 is the third parameter bias in the grid displacement prediction unit, y r1 is the first grid displacement component, y r2 is the second grid displacement component, y r3 is the third grid displacement component, w yr1 is the weight of the first grid displacement component, w yr2 is the weight of the second grid displacement component, w yr3 is the weight of the third grid displacement component.

[0063] The beneficial effect of the above further scheme is: when constructing the prediction expression of grid compression force and grid displacement, the present invention uses and improves the BP neural network, divides the input into three parts, and each part of the input adopts a hyperbolic tangent activation function. Each hyperbolic tangent activation function is used to predict the impact of each grid space data, thereby combining the three components to further adjust the output. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flowchart of a parameter configuration method for automatically assembling a finite element mesh;

[0065] Figure 2 A schematic diagram of the connection relationship between the matching module and the parameter prediction model;

[0066] Figure 3 It is a schematic diagram of the structure of the first image feature extraction unit and the second image feature extraction unit. DETAILED DESCRIPTION

[0067] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0068] like Figure 1 As shown, a parameter configuration method for automatically assembling a finite element mesh comprises the following steps:

[0069] S1, extracting an image of a first simulation component and an image of a second simulation component;

[0070] S2, extracting finite element data of the first simulation component;

[0071] S3, extracting finite element data and grid space data of the second simulation component;

[0072] S4, using a matching module to process the image of the first simulation component, the image of the second simulation component, the finite element data of the first simulation component, and the finite element data of the second simulation component to obtain matching values ​​of the two simulation components;

[0073] S5. When the matching value is greater than the matching threshold, the assembly parameters are predicted using a parameter prediction model according to the grid space data and the matching value of the second simulation component.

[0074] In this embodiment, the first simulation component is a casting, and the second simulation component is a casting mold, which are used to test the matching condition of the casting and the casting mold when they are assembled.

[0075] The finite element data of the first simulation component in S2 includes: the number of finite element grids of the inner diameter of the first simulation component, the number of finite element grids of the outer diameter of the first simulation component, and the number of finite element grids of the length of the first simulation component;

[0076] The finite element data of the second simulation component in S3 includes: the number of finite element grids of the inner diameter of the second simulation component, the number of finite element grids of the outer diameter of the second simulation component, and the number of finite element grids of the height of the second simulation component;

[0077] The grid space data in S3 includes: space coordinates and node numbers.

[0078] The present invention extracts the number of finite element meshes at various locations, and reflects the size of the simulation component through the number of finite element meshes at various locations, thereby facilitating the acquisition of the matching status of the first simulation component and the second simulation component.

[0079] like Figure 2 As shown, the matching module in S4 includes: a first image feature extraction unit, a second image feature extraction unit, a first feature splicing unit, a second feature splicing unit, a first multi-scale feature extraction unit, a second multi-scale feature extraction unit and a matching value calculation unit;

[0080] The input end of the first image feature extraction unit is used to input the image of the first simulation component, and the output end thereof is connected to the first input end of the first feature splicing unit; the second input end of the first feature splicing unit is used to input the finite element data of the first simulation component, and the output end thereof is connected to the input end of the first multi-scale feature extraction unit; the input end of the second image feature extraction unit is used to input the image of the second simulation component, and the output end thereof is connected to the first input end of the second feature splicing unit; the second input end of the second feature splicing unit is used to input the finite element data of the second simulation component, and the output end thereof is connected to the input end of the second multi-scale feature extraction unit; the input end of the matching value calculation unit is respectively connected to the output end of the first multi-scale feature extraction unit and the output end of the second multi-scale feature extraction unit, and the output end thereof serves as the output end of the matching module.

[0081] The present invention adopts a second image feature extraction unit to extract image features of the second simulation component, adopts a first image feature extraction unit to extract image features of the first simulation component, splices the image features of the second simulation component and the finite element data of the second simulation component, adopts a second multi-scale feature extraction unit to extract multi-scale features, splices the image features of the first simulation component and the finite element data of the first simulation component, adopts the first multi-scale feature extraction unit to extract multi-scale features, calculates the matching value of the two multi-scale features through a matching value calculation unit, thereby realizing automatic verification of the matching status of the first simulation component and the second simulation component.

[0082] like Figure 3As shown, the first image feature extraction unit and the second image feature extraction unit both include: a first convolution layer, a MaxPool layer, an AvgPool layer, a feature enhancement layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an adder A1 and an upsampling layer;

[0083] The input end of the first convolutional layer serves as the input end of the first image feature extraction unit or the second image feature extraction unit, and its output end is respectively connected to the input end of the MaxPool layer and the input end of the AvgPool layer; the input end of the feature enhancement layer is respectively connected to the output end of the MaxPool layer and the output end of the AvgPool layer, and its output end is respectively connected to the input end of the second convolutional layer and the input end of the third convolutional layer; the output end of the third convolutional layer is connected to the input end of the fourth convolutional layer; the input end of the adder A1 is respectively connected to the output end of the second convolutional layer and the output end of the fourth convolutional layer, and its output end is connected to the input end of the upsampling layer; the output end of the upsampling layer serves as the output end of the first image feature extraction unit or the second image feature extraction unit.

[0084] The present invention adopts the MaxPool layer to extract the significant features of the output of the first convolutional layer, adopts the AvgPool layer to extract the global average features of the output of the first convolutional layer, and enhances the significant features and the average features in the feature enhancement layer, thereby enhancing the adaptive adjustment capability of the first image feature extraction unit or the second image feature extraction unit, and then adopts a two-way convolution structure to construct a residual structure to solve the problem of gradient disappearance and improve the depth of feature extraction.

[0085] The expression of the feature enhancement layer is:

[0086] ,

[0087] Among them, H is the output of the feature enhancement layer, H MaxPool is the output of the MaxPool layer, H AvgPool is the output of the AvgPool layer, FC1 is the first fully connected layer, FC2 is the second fully connected layer, T is the transposition operation, It is Hadamard.

[0088] The present invention uses the first fully connected layer to process the output of the MaxPool layer, and uses the second fully connected layer to process the output of the AvgPool layer, so as to integrate the extracted features and capture global information. The output of the first fully connected layer is transposed, and the Hadamard product is used for feature fusion to achieve enhanced fusion of significant features and average features.

[0089] The expressions of the first multi-scale feature extraction unit and the second multi-scale feature extraction unit are:

[0090] ,

[0091] ,

[0092] ,

[0093] ,

[0094] Wherein, O is the output of the first multi-scale feature extraction unit or the second multi-scale feature extraction unit, S is the output of the feature splicing unit, The window size is The average pooling layer, The window size is The average pooling layer, The window size is The average pooling layer, For Hadamard, The convolution kernel is The convolution layer is σ, σ is the S-type function, Concat is the concatenation operation, X1 is the first scale feature, X2 is the second scale feature, and X3 is the third scale feature.

[0095] The present invention uses average pooling layers of different window sizes to extract different features, and calculates different attentions based on different features, thereby improving the model's ability to capture important information and enhancing the model's representation and generalization capabilities. The present invention calculates attention based on the average features extracted under each window, assigns weights to the average features, and assigns different weights to different parts of information, so that important information can get more attention, while secondary information has a relatively reduced impact. The present invention splices the first scale features, the second scale features, and the third scale features to improve the accuracy of feature extraction.

[0096] The expression of the matching value calculation unit is:

[0097] ,

[0098] Among them, P is the matching value, r 1,i is the i-th element in the output O1 of the first multi-scale feature extraction unit, r 2,i is the i-th element in the output O2 of the second multi-scale feature extraction unit, w i is the i-th weight in the matching value calculation unit, b i is the i-th bias in the matching value calculation element, N is the number of elements in the output O1 of the first multi-scale feature extraction unit or the number of elements in the output O2 of the second multi-scale feature extraction unit, and i is the number of the element.

[0099] The present invention calculates the difference between the output of the first multi-scale feature extraction unit and the output of the second multi-scale feature extraction unit, and assigns weights and biases to each difference, so as to facilitate the optimization of parameters during matching module training.

[0100] The parameter prediction model in S5 includes: a grid pressing force prediction unit and a grid displacement prediction unit;

[0101] The grid pressing force prediction unit is used to predict the grid pressing force during assembly according to the grid space data and the matching value of the second simulation component;

[0102] The grid displacement prediction unit is used to predict the grid displacement during assembly according to the grid space data of the second simulation component.

[0103] The grid pressing force prediction unit comprises: a grid pressing force prediction module and a compensation module;

[0104] The expression of the grid compression force prediction module is:

[0105] ,

[0106] ,

[0107] ,

[0108] ,

[0109] Among them, Y s is the predicted initial grid compression force, tanh is the hyperbolic tangent activation function, and w s1 is the first parameter weight in the grid compression force prediction module, b s1 is the first parameter bias in the grid compression force prediction module, w s2 is the second parameter weight in the grid compression force prediction module, b s2 is the second parameter bias in the grid compression force prediction module, w s3 is the third parameter weight in the grid compression force prediction module, b s3 is the third parameter bias in the grid compression force prediction module, c1 is the grid position priority coefficient 1, c2 is the grid position priority coefficient 2, c3 is the grid position priority coefficient 3, y s1 is the first grid compression force component, y s2 is the second grid compression force component, y s3 is the third grid compression force component, w ys1 is the weight of the first grid compression force component, w ys2 is the weight of the second grid compression force component, w ys3 is the weight of the third grid compression force component;

[0110] The expression of the compensation module is:

[0111] ,

[0112] Among them, Y c is the mesh compression force after compensation, e is the natural constant, P is the matching value, and P th is the matching threshold, and γ is the compensation coefficient.

[0113] In this embodiment, when the first simulation component is a casting and the second simulation component is a mold, the grid pressing force is related to the degree of matching between the casting and the mold. When the matching value is larger, the casting and the mold are more matched, and the required grid pressing force is smaller. When the matching value is smaller, the casting and the mold are more difficult to match, and the required grid pressing force is larger. Therefore, after predicting the initial grid pressing force, the grid pressing force is compensated according to the matching value, and the compensation coefficient is set to flexibly adjust the compensation ratio.

[0114] The expression of the grid displacement prediction unit is:

[0115] ,

[0116] ,

[0117] ,

[0118] ,

[0119] Among them, Y r is the grid displacement, c1 is the grid position priority coefficient 1, c2 is the grid position priority coefficient 2, c3 is the grid position priority coefficient 3, tanh is the hyperbolic tangent activation function, w r1 is the first parameter weight in the grid displacement prediction unit, b r1 is the first parameter bias in the grid displacement prediction unit, w r2 is the second parameter weight in the grid displacement prediction unit, b r2 is the second parameter bias in the grid displacement prediction unit, w r3 is the third parameter weight in the grid displacement prediction unit, b r3 is the third parameter bias in the grid displacement prediction unit, y r1 is the first grid displacement component, y r2 is the second grid displacement component, y r3 is the third grid displacement component, w yr1 is the weight of the first grid displacement component, w yr2 is the weight of the second grid displacement component, w yr3 is the weight of the third grid displacement component.

[0120] When constructing the prediction expressions of grid compression force and grid displacement, the present invention uses and improves the BP neural network, divides the input into three parts, and uses a hyperbolic tangent activation function for each part of the input. Each hyperbolic tangent activation function is used to predict the impact of each grid space data, thereby combining the three components to further adjust the output.

[0121] In this embodiment, the matching module, the grid pressing force prediction unit and the grid displacement prediction unit of the present invention need to be trained, and the training method can use the gradient descent method. For the matching module, the image of the first simulation component, the image of the second simulation component, the finite element data of the first simulation component and the finite element data of the second simulation component are constructed as samples, the target matching value is used as a label, and a plurality of samples and labels are constructed as a training set, and the training set is used to train the matching module; for the grid pressing force prediction unit, the grid space data and matching value of the second simulation component are constructed as samples, and the target grid pressing force is used as a label, so as to train the grid pressing force prediction unit; for the grid displacement prediction unit, the grid space data of the second simulation component is used as a sample, and the target grid displacement is used as a label, so as to train the grid displacement prediction unit.

[0122] The present invention extracts an image of a first simulation component and an image of a second simulation component, characterizes shape features of the simulation component through images, characterizes dimensional data of the simulation component through finite element data, uses a matching module to verify whether the constructed first simulation component and the second simulation component match in shape and structure, and estimates a matching value. The higher the matching value, the more matched the first simulation component and the second simulation component are in shape and structure, thereby having a higher degree of fit and being easier to assemble. Combined with the grid space data and the matching value of the second simulation component, a parameter prediction model is used to predict assembly parameters, thereby avoiding the construction of a mathematical differential equation model and reducing the complexity of obtaining assembly parameters.

[0123] The above are only 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 in the protection scope of the present invention.

Claims

1. A parameter configuration method for automatically assembling a finite element mesh, characterized in that: The following steps are involved: S1, extracting an image of a first simulation component and an image of a second simulation component; S2, extracting finite element data of the first simulation component; S3, extracting finite element data and grid space data of the second simulation component; S4, using a matching module to process the image of the first simulation component, the image of the second simulation component, the finite element data of the first simulation component, and the finite element data of the second simulation component to obtain matching values ​​of the two simulation components; S5. When the matching value is greater than the matching threshold, the assembly parameters are predicted using a parameter prediction model according to the grid space data and the matching value of the second simulation component; The finite element data of the first simulation component in S2 includes: the number of finite element grids of the inner diameter of the first simulation component, the number of finite element grids of the outer diameter of the first simulation component, and the number of finite element grids of the length of the first simulation component; The finite element data of the second simulation component in S3 includes: the number of finite element grids of the inner diameter of the second simulation component, the number of finite element grids of the outer diameter of the second simulation component, and the number of finite element grids of the height of the second simulation component; The grid space data in S3 includes: space coordinates and node numbers; The matching module in S4 includes: a first image feature extraction unit, a second image feature extraction unit, a first feature splicing unit, a second feature splicing unit, a first multi-scale feature extraction unit, a second multi-scale feature extraction unit and a matching value calculation unit; The input end of the first image feature extraction unit is used to input the image of the first simulation component, and the output end thereof is connected to the first input end of the first feature splicing unit; the second input end of the first feature splicing unit is used to input the finite element data of the first simulation component, and the output end thereof is connected to the input end of the first multi-scale feature extraction unit; the input end of the second image feature extraction unit is used to input the image of the second simulation component, and the output end thereof is connected to the first input end of the second feature splicing unit; the second input end of the second feature splicing unit is used to input the finite element data of the second simulation component, and the output end thereof is connected to the input end of the second multi-scale feature extraction unit; the input end of the matching value calculation unit is respectively connected to the output end of the first multi-scale feature extraction unit and the output end of the second multi-scale feature extraction unit, and the output end thereof serves as the output end of the matching module.

2. The parameter configuration method for automatically assembling a finite element mesh according to claim 1, characterized in that: The first image feature extraction unit and the second image feature extraction unit both include: a first convolution layer, a MaxPool layer, an AvgPool layer, a feature enhancement layer, a second convolution layer, a third convolution layer, a fourth convolution layer, an adder A1 and an upsampling layer; The input end of the first convolutional layer serves as the input end of the first image feature extraction unit or the second image feature extraction unit, and its output end is respectively connected to the input end of the MaxPool layer and the input end of the AvgPool layer; the input end of the feature enhancement layer is respectively connected to the output end of the MaxPool layer and the output end of the AvgPool layer, and its output end is respectively connected to the input end of the second convolutional layer and the input end of the third convolutional layer; the output end of the third convolutional layer is connected to the input end of the fourth convolutional layer; the input end of the adder A1 is respectively connected to the output end of the second convolutional layer and the output end of the fourth convolutional layer, and its output end is connected to the input end of the upsampling layer; the output end of the upsampling layer serves as the output end of the first image feature extraction unit or the second image feature extraction unit.

3. The parameter configuration method for automatically assembling a finite element mesh according to claim 2, characterized in that: The expression of the feature enhancement layer is: ， Among them, H is the output of the feature enhancement layer, H MaxPool is the output of the MaxPool layer, H AvgPool is the output of the AvgPool layer, FC1 is the first fully connected layer, FC2 is the second fully connected layer, T is the transposition operation, It is Hadamard.

4. The parameter configuration method for automatically assembling a finite element mesh according to claim 1, characterized in that: The expressions of the first multi-scale feature extraction unit and the second multi-scale feature extraction unit are: ， ， ， ， Wherein, O is the output of the first multi-scale feature extraction unit or the second multi-scale feature extraction unit, S is the output of the feature splicing unit, The window size is The average pooling layer, The window size is The average pooling layer, The window size is The average pooling layer, For Hadamard, The convolution kernel is The convolution layer is σ, σ is the S-type function, Concat is the concatenation operation, X1 is the first scale feature, X2 is the second scale feature, and X3 is the third scale feature.

5. The parameter configuration method for automatically assembling a finite element mesh according to claim 1, characterized in that: The expression of the matching value calculation unit is: ， Among them, P is the matching value, r 1,i is the i-th element in the output O1 of the first multi-scale feature extraction unit, r 2,i is the i-th element in the output O2 of the second multi-scale feature extraction unit, w i is the i-th weight in the matching value calculation unit, b i is the i-th bias in the matching value calculation element, N is the number of elements in the output O1 of the first multi-scale feature extraction unit or the number of elements in the output O2 of the second multi-scale feature extraction unit, and i is the number of the element.

6. The parameter configuration method for automatically assembling a finite element mesh according to claim 1, characterized in that: The parameter prediction model in S5 includes: a grid pressing force prediction unit and a grid displacement prediction unit; The grid pressing force prediction unit is used to predict the grid pressing force during assembly according to the grid space data and the matching value of the second simulation component; The grid displacement prediction unit is used to predict the grid displacement during assembly according to the grid space data of the second simulation component.

7. The parameter configuration method for automatically assembling a finite element mesh according to claim 6, characterized in that: The grid pressing force prediction unit comprises: a grid pressing force prediction module and a compensation module; The expression of the grid compression force prediction module is: ， ， ， ， Among them, Y s is the predicted initial grid compression force, tanh is the hyperbolic tangent activation function, and w s1 is the first parameter weight in the grid compression force prediction module, b s1 is the first parameter bias in the grid compression force prediction module, w s2 is the second parameter weight in the grid compression force prediction module, b s2 is the second parameter bias in the grid compression force prediction module, w s3 is the third parameter weight in the grid compression force prediction module, b s3 is the third parameter bias in the grid compression force prediction module, c1 is the grid position priority coefficient 1, c2 is the grid position priority coefficient 2, c3 is the grid position priority coefficient 3, y s1 is the first grid compression force component, y s2 is the second grid compression force component, y s3 is the third grid compression force component, w ys1 is the weight of the first grid compression force component, w ys2 is the weight of the second grid compression force component, w ys3 is the weight of the third grid compression force component; The expression of the compensation module is: ， Among them, Y c is the mesh compression force after compensation, e is the natural constant, P is the matching value, and P th is the matching threshold, and γ is the compensation coefficient.

8. The parameter configuration method for automatically assembling a finite element mesh according to claim 6, characterized in that: The expression of the grid displacement prediction unit is: ， ， ， ， Among them, Y r is the grid displacement, c1 is the grid position priority coefficient 1, c2 is the grid position priority coefficient 2, c3 is the grid position priority coefficient 3, tanh is the hyperbolic tangent activation function, w r1 is the first parameter weight in the grid displacement prediction unit, b r1 is the first parameter bias in the grid displacement prediction unit, w r2 is the second parameter weight in the grid displacement prediction unit, b r2 is the second parameter bias in the grid displacement prediction unit, w r3 is the third parameter weight in the grid displacement prediction unit, b r3 is the third parameter bias in the grid displacement prediction unit, y r1 is the first grid displacement component, y r2 is the second grid displacement component, y r3 is the third grid displacement component, w yr1 is the weight of the first grid displacement component, w yr2 is the weight of the second grid displacement component, w yr3 is the weight of the third grid displacement component.

Images