A method and device for constructing a dynamic finite element surrogate model based on random forest

Abstract

This application belongs to the technical field of proxy model construction, and particularly relates to a method and device for constructing a dynamic finite element proxy model based on random forest. The method includes: Step S1, obtaining finite element simulation data of an aircraft structure, including physical quantities of each grid unit and node coordinates of four nodes surrounding the grid unit; Step S2, determining physical quantities of each node according to physical quantities of multiple grid units adjacent to the node; Step S3, correcting physical quantities of each node based on the test process; Step S4, using the node coordinates and the test process as expected inputs, and the corrected physical quantity of the node as the expected output, training a plurality of preset random forest models with different specifications, and selecting the optimal random forest model as the dynamic finite element proxy model using a preset evaluation index. This application realizes the mutual mapping of physical quantities and finite element simulation results throughout the test process, and improves the efficiency of aircraft structural strength tests.

Description

Technical Field

[0001] This application belongs to the technical field of proxy model construction, and particularly relates to a method and device for constructing a dynamic finite element proxy model based on a random forest. Background Art

[0002] Finite element simulation is a necessary step for evaluating the structural strength of test pieces, simulating the test process, and predicting loading failure before an aircraft structural strength test. However, it is difficult to map the finite element calculation results based on full-field simulation to the physical test results based on discrete strain gauges. This is because the arrangement of discrete strain gauges is difficult to completely cover the entire test components of the aircraft structure. Discrete strain gauges generally only specify the X, Y, and Z position coordinates in space. When mapping specific strain gauges to finite element model units, factors such as spatial coordinates, pasting positions, detection directions, and model quality need to be considered, and manual chip selection processing is required, which is a large workload for the full-aircraft structural strength test. The mapping process between discrete strain gauges and finite element simulation model units consumes a large amount of time and labor costs, resulting in the simulation process not fully serving the test verification process.

[0003] On the other hand, the finite element model cannot meet the requirement of viewing dynamic simulation results according to the physical test process. When using the finite element simulation method, the finite element model does not reflect the test condition process and strain change process intuitively enough, and can only reflect the simulation results at the end of the test singly, lacking the mapping ability of the strain throughout the entire test process. Summary of the Invention

[0004] To solve the above problems, this application provides a method and device for constructing a dynamic finite element proxy model based on a random forest, which uses a finite element proxy model to replace finite element simulation, and quickly obtains the simulation calculation results at specific positions and specific loading times through the finite element proxy model, avoiding the complex process of manually searching for mapping units, and providing a more effective means for test data utilization, test site monitoring, and virtual-real data consistency evaluation.

[0005] The first aspect of this application provides a method for constructing a dynamic finite element proxy model based on a random forest, which mainly includes:

[0006] Step S1: Obtain the finite element simulation data of the aircraft structure, including the physical quantities of each grid unit and the node coordinates of the four nodes surrounding the grid unit;

[0007] Step S2: Determine the physical quantities of each node according to the physical quantities of multiple grid units adjacent to the node;

[0008] Step S3: Modify the physical quantities of each node based on the test process;

[0009] Step S4: Use the node coordinates and the test process as the expected inputs, and the corrected physical quantity of the node as the expected output to train a plurality of preset random forest models with different specifications, and select the optimal random forest model as the dynamic finite element surrogate model using the preset evaluation metrics.

[0010] Preferably, in step S2, the physical quantity of the node is set to the average value of the physical quantities of the grid cells adjacent to the node.

[0011] Preferably, step S3 further includes:

[0012] When the finite element simulation data of the aircraft structure obtained is linear simulation data, directly use the test process to correct the physical quantity of each node;

[0013] When the finite element simulation data of the aircraft structure obtained is non-linear simulation data, first calculate the correction amount according to the non-linear relationship, and then use the correction amount to correct the physical quantity of each node.

[0014] Preferably, in step S4, the evaluation metrics include the mean square error metric or the R-squared metric.

[0015] Preferably, the physical quantity includes strain.

[0016] The second aspect of the present application provides a device for constructing a dynamic finite element surrogate model based on a random forest, mainly including:

[0017] A finite element simulation data acquisition module for acquiring the finite element simulation data of the aircraft structure, including the physical quantity of each grid cell and the node coordinates of the four nodes surrounding the grid cell;

[0018] A node physical quantity determination module for determining the physical quantity of each node according to the physical quantities of a plurality of grid cells adjacent to the node;

[0019] A physical quantity correction module for correcting the physical quantity of each node based on the test process;

[0020] A random forest model selection module for using the node coordinates and the test process as the expected inputs, and the corrected physical quantity of the node as the expected output to train a plurality of preset random forest models with different specifications, and selecting the optimal random forest model as the dynamic finite element surrogate model using the preset evaluation metrics.

[0021] Preferably, the physical quantity of the node is set to the average value of the physical quantities of the grid cells adjacent to the node.

[0022] Preferably, the physical quantity correction module includes:

[0023] A correction unit based on linear simulation data is used to directly correct the physical quantities of each node using the test process when the finite element simulation data of the aircraft structure obtained is linear simulation data;

[0024] A correction unit based on non - linear simulation data is used to first calculate the correction amount according to the non - linear relationship and then use the correction amount to correct the physical quantities of each node when the finite element simulation data of the aircraft structure obtained is non - linear simulation data.

[0025] Preferably, the evaluation index includes the mean square error index or the R - square index.

[0026] Preferably, the physical quantity includes strain.

[0027] This application realizes the mutual mapping between the physical quantities in the whole test process and the finite element simulation results, improving the efficiency of the aircraft structure strength test. Description of the Drawings

[0028] Figure 1 is a flowchart of a preferred embodiment of the method for constructing a dynamic finite element surrogate model based on random forest in this application.

[0029] Figure 2 is a schematic diagram for determining the physical quantity of a node. Detailed Embodiments

[0030] To make the purpose, technical solutions, and advantages of the implementation of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings in the embodiments of this application. In the drawings, the same or similar reference numerals denote the same or similar elements or elements with the same or similar functions from beginning to end. The described embodiments are some but not all of the embodiments of this application. The embodiments described below by referring to the drawings are exemplary and are intended to explain this application and should not be construed as limiting this application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the drawings.

[0031] The first aspect of this application provides a method for constructing a dynamic finite element surrogate model based on random forest, as Figure 1 shown, mainly including:

[0032] Step S1: Obtain the finite element simulation data of the aircraft structure, including the physical quantities of each mesh element and the node coordinates of the four nodes surrounding the mesh element;

[0033] Step S2: Determine the physical quantity of each node according to the physical quantities of multiple mesh elements adjacent to the node;

[0034] Step S3. Modify the physical quantities of each node based on the test process;

[0035] Step S4. Use the node coordinates and the test process as the expected inputs, and the modified physical quantity of the node as the expected output to train a plurality of preset random forest models with different specifications, and select the optimal random forest model as the dynamic finite element surrogate model using the preset evaluation index.

[0036] First, an explanation of the finite element surrogate model is given. In recent years, with the rise of artificial intelligence algorithms, machine learning has a strong fitting ability for data and has realized the construction of surrogate models in many fields to quickly respond to the data characteristics of the pre-training set. Among them, the integrated model of decision trees is particularly prominent, proving that machine learning has the possibility of becoming a fast-response surrogate model for finite elements. The finite element surrogate model refers to using technologies such as data processing and surrogate models to fit important information such as finite element positions and test results through the constructed surrogate model. That is, training a finite element surrogate model, so that the finite element position can be directly input, and the stress and strain at this position can be calculated using the trained finite element surrogate model. The so-called dynamic finite element surrogate model means that during the training of the surrogate model, not only the finite element position but also the time information is input, and the obtained surrogate model can output different results according to different test processes, so as to be able to synchronously respond to the physical test process at different stages of test verification and provide support for test command decision-making.

[0037] Since the purpose of this application is to quickly obtain simulation results for mapping with physical test data, in steps S1 - S3 of this application, the data used for training the finite element surrogate model is preprocessed, and then these data are divided into a training set and a test set. In step S4, the finite element surrogate model is trained and tested. In step S1, first, prepare the aircraft structure finite element model and the simulation result file corresponding to the physical test, that is, the finite element simulation data, and check the matching between the simulation result file and the finite element model. Then, use HyperWorks software to remove the unexpected stress concentration elements according to the finite element results, and extract the information of the elements and the physical quantities of the simulation results in the remaining finite element model. In some alternative embodiments, the physical quantity includes strain. Refer to Figure 2 Each rectangular box represents a mesh unit. Each mesh unit has strain information, and each rectangular box where each mesh unit is located has four corner points, corresponding to four nodes.

[0038] After that, in step S2, convert the strain of the mesh unit into the strain of the node, as Figure 2The shown node A has its strain calculated based on the strains of the adjacent mesh elements a, b, c, and d. In some alternative embodiments, in step S2, the physical quantity of the node is set to the average of the physical quantities of the mesh elements adjacent to the node. That is, the average of the strains of mesh elements a, b, c, and d is used as the strain of node A.

[0039] In step S3, it is necessary to superimpose time information on the physical quantity of the node. That is, under different test processes, the strain values of the same node may be different. In some alternative embodiments, step S3 further includes:

[0040] When the obtained finite element simulation data of the aircraft structure is linear simulation data, directly use the test process to correct the physical quantities of each node;

[0041] When the obtained finite element simulation data of the aircraft structure is non - linear simulation data, first calculate the correction amount according to the non - linear relationship, and then use the correction amount to correct the physical quantities of each node.

[0042] In this embodiment, for the strength test process T, if the simulation result is a linear simulation result, the test process T can be set to (0, 5%, 10%... 100%) for example. Then, multiply the strain R1 (the strain value at the end of the simulation) obtained in step S2 by the test process T to indirectly calculate the dynamic simulation data. That is, the strain R corresponding to a certain test process T is set as: R = T×R1. If the simulation result is a non - linear simulation result, the test process T is set corresponding to the process of the non - linear simulation result. For example, when the test process T is 20%, theoretically, first find the ordinate value when the abscissa is 20% in the non - linear relationship, then compare this ordinate value with the ordinate value when the abscissa is 100% in the non - linear relationship to calculate the correction amount, and then use this correction amount to correct the physical quantity. In this embodiment, the non - linear relationship is a non - linear function formed with the abscissa as the process and the ordinate as the physical quantity.

[0043] Through the above method, a training data set for machine learning is constructed in the form of [X, Y, Z, T|R], where X, Y, Z, and T respectively represent the X, Y, Z coordinates of the node and the test process, and R is the corrected strain value. Finally, in step S4, the node coordinates and the test process are the expected inputs, and the corrected strain value is the expected output, so as to train multiple random forest models with different specifications to select the optimal random forest model.

[0044] When training a random forest model, the training dataset is usually divided into a training set and a test set, with ratios of 7:3, 5:5, and 3:7 respectively, to simulate three different scenarios where the number of finite element units is large, moderate, and small. In this embodiment, setting multiple specifications of the random forest model mainly means assigning different parameter values to the hyperparameters of the random forest model. For example, the number of decision trees n_estimators, the learning rate learning_rate, the number of features considered when splitting each node (here the node refers to the model node, not the finite element simulation node) max_features, the maximum depth of the decision tree max_depth, the minimum number of samples required for node splitting min_samples_split, the minimum number of samples required for leaf nodes min_samples_leaf, the criterion for measuring the quality of node splitting criterion, etc. Use the training set to train multiple specifications of the random forest model, use the corresponding test set for verification, and calculate the mean squared error MSE or the R-squared index R of the predicted strain and the simulated strain (expected output). 2 The mean squared error MSE is used to reflect the error between the predicted value and the actual simulated strain data, and the R-squared index R 2 is used to reflect the deviation between the predicted value and the actual simulated strain data, so as to be able to select the optimal specification of the random forest model.

[0045] Finally, check the fitting effect. For the optimal random forest model trained on all training datasets, input the coordinates of all nodes of the finite element model, predict the strain values, and form a contour map under the same chromatogram. Compare the finite element simulation contour map and the surrogate model fitting contour map. If the fitting effect is acceptable, the construction of the surrogate model is completed. If the fitting effect is poor, reset the random forest specifications for verification.

[0046] Through the above method, this application focuses on the position of the physical quantities in the strength test and the dynamic process of the test, ensuring that the simulation calculation results at specific positions and specific loading times can be quickly obtained through the finite element surrogate model at the strength test site, avoiding the complex process of manually searching for mapping units, and providing a more effective means for test data utilization, test site monitoring, and virtual-real data consistency evaluation.

[0047] The second aspect of this application provides a device for constructing a dynamic finite element surrogate model based on a random forest corresponding to the above method, mainly including:

[0048] A finite element simulation data acquisition module, used to acquire finite element simulation data of the aircraft structure, including the physical quantities of each grid unit and the node coordinates of the four nodes surrounding the grid unit;

[0049] A node physical quantity determination module, used to determine the physical quantity of each node according to the physical quantities of multiple grid units adjacent to the node;

[0050] A physical quantity correction module for correcting the physical quantities of each node based on the test process;

[0051] A random forest model selection module for training a plurality of preset random forest models with different specifications by using the node coordinates and the test process as the expected inputs and the corrected physical quantity of the node as the expected output, and selecting the optimal random forest model as the dynamic finite element surrogate model by using a preset evaluation index.

[0052] In some alternative embodiments, the physical quantity of the node is set to the average value of the physical quantities of each grid unit adjacent to the node.

[0053] In some alternative embodiments, the physical quantity correction module includes:

[0054] A correction unit based on linear simulation data for directly correcting the physical quantities of each node by using the test process when the finite element simulation data of the aircraft structure obtained is linear simulation data;

[0055] A correction unit based on non-linear simulation data for first calculating a correction amount according to a non-linear relationship and then correcting the physical quantities of each node by using the correction amount when the finite element simulation data of the aircraft structure obtained is non-linear simulation data.

[0056] In some alternative embodiments, the evaluation index includes a mean square error index or an R-squared index.

[0057] In some alternative embodiments, the physical quantity includes strain.

[0058] The above is only the specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any change or replacement that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for constructing a dynamic finite element proxy model based on random forest, characterized in that: include: Step S1, obtaining finite element simulation data of the aircraft structure, including physical quantities of each grid unit and node coordinates of four nodes surrounding the grid unit; Step S2, determining the physical quantity of each node according to the physical quantities of a plurality of grid units adjacent to the node; Step S3, correcting the physical quantity of each node based on the test progress; Step S4: taking the node coordinates and the test process as the expected input and the corrected physical quantity of the node as the expected output, training a plurality of preset random forest models of different specifications, and using the preset evaluation index to select the optimal random forest model as the dynamic finite element proxy model; Wherein, the physical quantity includes strain; Step S3 further comprises: When the finite element simulation data of the aircraft structure obtained is linear simulation data, the physical quantity of each node is directly corrected using the test process; When the acquired finite element simulation data of the aircraft structure is nonlinear simulation data, a correction amount is first calculated according to the nonlinear relationship, and then the correction amount is used to correct the physical quantity of each node.

2. The method for constructing a dynamic finite element proxy model based on random forest according to claim 1, characterized in that: In step S2, the physical quantity of the node is set to the average value of the physical quantities of each grid unit adjacent to the node.

3. The method for constructing a dynamic finite element proxy model based on random forest according to claim 1, characterized in that: In step S4, the evaluation index includes a mean square error index or an R-square index.

4. A dynamic finite element proxy model construction device based on random forest, characterized in that: include: A finite element simulation data acquisition module, used to acquire finite element simulation data of the aircraft structure, including physical quantities of each grid unit and node coordinates of four nodes surrounding the grid unit; A node physical quantity determination module, used to determine the physical quantity of each node according to the physical quantities of a plurality of grid units adjacent to the node; A physical quantity correction module is used to correct the physical quantity of each node based on the test process; The random forest model selection module is used to train multiple preset random forest models of different specifications with node coordinates and test progress as expected inputs and the corrected physical quantity of the node as expected outputs, and select the optimal random forest model as the dynamic finite element proxy model using preset evaluation indicators; Wherein, the physical quantity includes strain; The physical quantity correction module comprises: A correction unit based on linear simulation data, used to directly use the test process to correct the physical quantity of each node when the finite element simulation data of the aircraft structure obtained is linear simulation data; The correction unit based on nonlinear simulation data is used to first calculate the correction amount according to the nonlinear relationship when the acquired finite element simulation data of the aircraft structure is nonlinear simulation data, and then use the correction amount to correct the physical quantity of each node.

5. The random forest-based dynamic finite element proxy model construction device according to claim 4, characterized in that: The physical quantity of the node is set to the average value of the physical quantities of each grid unit adjacent to the node.

6. The random forest-based dynamic finite element proxy model construction device according to claim 4, characterized in that: The evaluation index includes a mean square error index or an R-square index.

Images