A strain field correction method and device based on finite element proxy model

Abstract

The present application belongs to the technical field of aircraft structure strength test, and relates to a strain field correction method and device based on a finite element proxy model. The method comprises: step S1, selecting a plurality of designated strain gauges that are close to the strain unit to be predicted; step S2, determining the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted; step S3, obtaining the strain simulation data at each strain gauge based on a pre-built strain prediction model, and determining the change rate of each strain simulation data compared to the strain test data fed back by each strain gauge; step S4, according to the weight of each strain gauge, weighted summing the change rate of each strain gauge as the comprehensive change rate of the strain unit to be predicted; step S5, obtaining the strain simulation data of the strain unit to be predicted based on the strain prediction model, and correcting it by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted. The present application improves the strain field construction speed.

Description

Technical Field

[0001] The present application belongs to the technical field of aircraft structure strength testing, and in particular relates to a strain field correction method and device based on a finite element proxy model. Background Art

[0002] Finite element simulation is a necessary step to evaluate the structural strength of test pieces, simulate the test process, and predict loading damage before the aircraft structural strength test. However, the finite element simulation based on complex mechanical mechanisms has low computational efficiency, consumes a lot of computational time to readjust the model, and cannot meet the requirements of real-time and rapid response at the test site. In addition, the data error of finite element simulation of complex structures is uncontrollable and cannot be predicted before the test. Therefore, combining real-time test data to quickly correct the simulation results is an important technical breakthrough for on-site strength tests. Summary of the invention

[0003] In order to solve at least one of the above-mentioned technical problems, the present application provides a strain field correction method and device based on a finite element proxy model, which uses a finite element proxy model based on machine learning to predict the strain field, replacing the traditional finite element model, providing a more effective means for test data utilization and test site monitoring.

[0004] The first aspect of the present application provides a strain field correction method based on a finite element proxy model, which mainly includes:

[0005] Step S1, selecting a plurality of designated strain gauges that are close to the strain unit to be predicted;

[0006] Step S2, determining the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted;

[0007] Step S3, acquiring strain simulation data at each strain gauge based on a pre-built strain prediction model, and determining a change rate of the strain simulation data at each strain gauge compared to the strain test data fed back by each strain gauge;

[0008] Step S4, performing weighted summation on the change rates of the strain gauges according to the weights of the strain gauges, and using the weighted summation result as the comprehensive change rate of the strain unit to be predicted;

[0009] Step S5, acquiring strain simulation data of the strain unit to be predicted based on the pre-built strain prediction model, and correcting the strain simulation data by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted.

[0010] Preferably, in step S1, the number of the plurality of strain gauges is specified to be four.

[0011] Preferably, in step S2, the weight w of each strain gauge is determined by the following formula: i :

[0012] w i =f i / w;

[0013] Among them, f i =1 / (d i ) 2 ; w = ∑ f i ;d i is the Euclidean distance between the ith strain gauge and the unit centroid.

[0014] Preferably, before step S3, the step further includes constructing a strain prediction model, and constructing the strain prediction model includes:

[0015] Step S31, extracting parameters from the aircraft structure finite element model to construct a training set and a test set, wherein the parameters of the training set and the test set include unit centroid and unit strain data;

[0016] Step S32: training the strain prediction model using the training set, and verifying the trained strain prediction model using the test set.

[0017] Preferably, step S31 further comprises:

[0018] Step S311, eliminating stress concentration units in the finite element model, and retaining units with continuous strain values ​​in each plane of the aircraft structure;

[0019] Step S312: planes in the same span direction in the aircraft structure are arranged in the same group, and different groups and different strain types correspond to different strain prediction models.

[0020] The second aspect of the present application provides a strain field correction device based on a finite element proxy model, which mainly includes:

[0021] A strain gauge selection module is used to select a plurality of designated strain gauges that are close to the strain unit to be predicted;

[0022] A strain gauge weight calculation module, used to determine the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted;

[0023] A strain gauge change rate calculation module is used to obtain strain simulation data at each strain gauge based on a pre-built strain prediction model, and determine the change rate of the strain simulation data at each strain gauge compared to the strain test data fed back by each strain gauge;

[0024] A comprehensive change rate calculation module is used to perform weighted summation on the change rates of each strain gauge according to the weight of each strain gauge, and use the weighted summation result as the comprehensive change rate of the strain unit to be predicted;

[0025] The strain prediction module is used to obtain the strain simulation data of the strain unit to be predicted based on the pre-built strain prediction model, and correct it by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted.

[0026] Preferably, the number of the plurality of strain gauges is specified to be four.

[0027] Preferably, in the strain gauge weight calculation module, the weight w of each strain gauge is determined by the following formula: i :

[0028] w i =f i / w;

[0029] Among them, f i =1 / (d i ) 2 ; w = ∑ f i ;d i is the Euclidean distance between the ith strain gauge and the unit centroid.

[0030] Preferably, the strain field correction device further comprises a strain prediction model construction module for constructing the strain prediction model, and the strain prediction model construction module comprises:

[0031] A data extraction unit, used to extract parameters from the aircraft structure finite element model to construct a training set and a test set, wherein the parameters of the training set and the test set include unit centroid and unit strain data;

[0032] The model training unit is used to train the strain prediction model through the training set and verify the trained strain prediction model through the test set.

[0033] Preferably, the data extraction unit comprises:

[0034] Data screening subunit, used to eliminate stress concentration units in the finite element model and retain units with continuous strain values ​​in each plane of the aircraft structure;

[0035] The model grouping subunit is used to set the planes in the same span direction in the aircraft structure in the same group. Different groups and different strain types correspond to different strain prediction models.

[0036] This application ensures the ability to modify the model in real time based on the test site data and to display the strain cloud map in real time, thereby increasing the speed of strain field construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of a preferred embodiment of the strain field correction method based on the finite element proxy model of the present application.

[0038] Figure 2 This is a schematic diagram of strain gauge selection.

[0039] Figure 3 Schematic diagram of the relationship between the centroid of the strain unit to be predicted and the strain simulation data.

[0040] Figure 4 This is a schematic diagram of the grouping of aircraft structural units. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical scheme and advantages of the implementation of this application clearer, the technical scheme in the implementation of this application will be described in more detail in combination with the drawings in the implementation of this application. In the drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described implementation is a part of the implementation of this application, not all of the implementations. The implementation described below with reference to the drawings is exemplary and is intended to be used to explain this application, and cannot be understood as a limitation on this application. Based on the implementation in this application, all other implementations obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The implementation of this application is described in detail below in combination with the drawings.

[0042] The first aspect of the present application provides a strain field correction method based on a finite element proxy model, such as Figure 1 As shown, the method mainly includes:

[0043] Step S1, selecting a plurality of designated strain gauges that are close to the strain unit to be predicted;

[0044] Step S2, determining the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted;

[0045] Step S3, acquiring strain simulation data at each strain gauge based on a pre-built strain prediction model, and determining a change rate of the strain simulation data at each strain gauge compared to the strain test data fed back by each strain gauge;

[0046] Step S4, performing weighted summation on the change rates of the strain gauges according to the weights of the strain gauges, and using the weighted summation result as the comprehensive change rate of the strain unit to be predicted;

[0047] Step S5, acquiring strain simulation data of the strain unit to be predicted based on the pre-built strain prediction model, and correcting the strain simulation data by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted.

[0048] With the rise of artificial intelligence algorithms, there are more and more cases of combining mechanics with artificial intelligence. Machine learning can explore the hidden laws of mechanical data, skip the complex calculations of mechanical mechanisms and achieve rapid response of mechanical properties. Machine learning-based material mechanics proxy models are also emerging in an endless stream, which has proved that machine learning has a large development space in the field of mechanics. To this end, this application also constructs a finite element proxy model based on machine learning, namely a strain prediction model, to replace the traditional finite element model, so as to quickly predict the strain data of various parts of the aircraft structure and avoid the additional constraints of the slow response of the traditional finite element model.

[0049] In the present application, in steps S1 to S3, the strain at the strain gauge attached to the aircraft structure is first predicted based on the strain prediction model, and the change rate can be determined by comparing it with the test data. The change rates of multiple strain gauges around the strain unit to be predicted can comprehensively reflect the change rate at the strain unit to be predicted, that is, the comprehensive change rate of step S4. Finally, in step S5, the strain at the strain unit to be predicted on the aircraft structure is predicted based on the strain prediction model, and it is corrected by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted under the test. In this way, the entire strain field is corrected and predicted, ensuring that the model can be quickly corrected based on the on-site test data, and the strain cloud map can be quickly formed.

[0050] In step S1 of the present application, multiple strain gauges with the closest spatial distance are selected according to the location of the strain unit to be predicted. The closer the point is, the greater the influence on the predicted strain unit is, and therefore the greater the weight assigned to it in step S2 is.

[0051] In some optional implementations, in step S1, the number of the plurality of strain gauges is specified to be four. In alternative implementations, more strain gauges may be selected, such as six or eight.

[0052] Step S2 is a step of assigning weights. As mentioned above, the weight value is positively correlated with the distance. For example, in some optional implementations, in step S2, the weight w of each strain gauge is determined by the following formula: i :

[0053] w i =f i / w;

[0054] Among them, f i =1 / (d i ) 2 ; w = ∑ f i ;d i is the Euclidean distance between the ith strain gauge and the unit centroid.

[0055] Take four strain gauges as an example, Figure 2As shown, A is the strain unit to be predicted, B1-B16 are strain gauges, and according to the Euclidean distance, the four strain gauges B5, B6, B9, and B10 are determined to be closest to it, and weights are assigned to these four strain gauges according to their distances.

[0056] Step S3 is used to predict the simulation values ​​of these strain gauges, that is, the strain simulation data, with the help of the strain prediction model, and then calculate the rate of change relative to the strain test data. Therefore, it is necessary to first generate a strain prediction model by pre-training, and the strain prediction model is used to replace the existing finite element simulation model to improve the calculation speed.

[0057] In some optional implementations, constructing the strain prediction model includes:

[0058] Step S31, extracting parameters from the aircraft structure finite element model to construct a training set and a test set, wherein the parameters of the training set and the test set include unit centroid and unit strain data;

[0059] Step S32: training the strain prediction model using the training set, and verifying the trained strain prediction model using the test set.

[0060] In some optional implementations, step S31 further includes:

[0061] Step S311, eliminating stress concentration units in the finite element model, and retaining units with continuous strain values ​​in each plane of the aircraft structure;

[0062] Step S312: planes in the same span direction in the aircraft structure are arranged in the same group, and different groups and different strain types correspond to different strain prediction models.

[0063] In the above embodiment, it is necessary to extract training data and test data based on the finite element model in step S31, and then train the model in step S32. Two points are mainly considered when extracting training data and test data. One is the data screening mentioned in step S311, and the other is the data division mentioned in step S312. Data screening is used to prevent abnormal data from affecting the training results of the model, and data division is used to train different strain prediction models for different types of data to improve the prediction accuracy of the model.

[0064] Specifically, in step S311, the finite element data is screened, and the unit quality of the finite element model is checked in combination with the strain cloud map. Irregular units due to structural reasons such as triangular units and slender quadrilateral units are checked to determine whether there are stress concentration units. If so, these stress concentration units are eliminated to ensure that the strain values ​​of each plane of the aircraft structure are continuous. The tools in Hpyermesh can be used to search for irregular units. In step S312, the unit grouping, unit-related node information, node coordinates, unit strain type, and unit strain simulation data are extracted from the finite element model file and the calculation result file, such as Figure 3 As shown, the unit-related node information and node coordinates are used to calculate the centroid (x, y, z) of each unit. The centroid (x, y, z) and strain simulation data R of each unit are used as the input of the model during model training to train the strain prediction model, which can be simply understood as the training mapping function f: R=f(x, y, z). Unit grouping and unit strain type are used to divide different strain prediction models, that is, to form multiple training mapping functions f1, f2...fn. In this way, when the model is used, the strain simulation data R of the specified type of the strain unit to be predicted can be output according to the centroid (x, y, z) of the strain unit to be predicted and the specified mapping function f. The unit grouping is explained here. For general test structures with staggered structures such as wall panels, reinforcements, and long girders, the planes in the same span direction are set in the same group. For common reinforced wall panel structures, the wall panels and all reinforcements can be set as two groups, such as Figure 4 As shown, the entire wall panel at the bottom can be divided into the first group C, and the three ribs above the wall panel can be divided into the second group D, and then two strain prediction models are used to train these two groups respectively.

[0065] After the training set and the test set are preprocessed in step S31, they are divided according to a certain ratio, and then in step S32, the strain prediction model is trained and predicted, and the proxy effects of various machine learning algorithms on the structural model are evaluated from the perspectives of numerical errors and cloud map effects.

[0066] First, common machine learning models, such as support vector machines, artificial neural networks, random forests, and gradient boosting trees, were trained using training data sets, and three test sets were used for verification. The MSE and R2 evaluation indicators of the predicted strain and the actual strain were calculated. The MSE evaluation indicator is used to reflect the error between the predicted value and the actual simulated strain data, and the R2 evaluation indicator is used to reflect the deviation between the predicted value and the actual simulated strain data. Secondly, the strain values ​​of all units of the finite element were predicted through machine learning, and the corresponding scatter cloud graph was generated to check the cloud graph fitting effect of the machine learning model. According to the evaluation indicators and cloud graph fitting effect, the strain prediction model after training was determined.

[0067] In step S3, in addition to outputting strain simulation data through the strain prediction model, strain test data can also be obtained through strain gauges. The strain test data is mainly returned by the strain gauge sensor during the strength test. First, clarify the layout information of the structural strain gauges and record the coordinates of each strain gauge corresponding to the finite element model. Then, organize the standard format data according to the strain gauge number, strain gauge coordinates and strain gauge test data, and finally calculate the change rate of each strain gauge:

[0068] a i =(y 2i -y 1i ) / y 1i ;

[0069] Among them, a i is the rate of change of the ith strain gauge, y 2i is the strain test data of the i-th strain gauge, y 1i is the strain simulation data of the i-th strain gauge.

[0070] In step S4, the comprehensive change rate a of the strain unit to be predicted is the weighted change rate of multiple strain gauges closest to its centroid, and the formula is as follows:

[0071] ;

[0072] Wherein, n is the number of strain gauges selected in step S1.

[0073] Finally, in step S5, according to the strain simulation data y1 of the strain unit to be predicted and the comprehensive change rate a calculated in step S4, the predicted strain y2 of the strain unit to be predicted under the test conditions is calculated, and the formula is as follows:

[0074] y2=y1+ay1.

[0075] Through the above method, the present application calculates the change rate based on the strain simulation data of the strain gauge and the actual collected test data, and then performs numerical correction on the strain unit to be predicted in the whole field, thereby achieving rapid response to real-time test data and accelerating the component speed of the strain field.

[0076] The second aspect of the present application provides a strain field correction device based on a finite element proxy model corresponding to the above method, which mainly includes:

[0077] A strain gauge selection module is used to select a plurality of designated strain gauges that are close to the strain unit to be predicted;

[0078] A strain gauge weight calculation module, used to determine the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted;

[0079] A strain gauge change rate calculation module is used to obtain strain simulation data at each strain gauge based on a pre-built strain prediction model, and determine the change rate of the strain simulation data at each strain gauge compared to the strain test data fed back by each strain gauge;

[0080] A comprehensive change rate calculation module is used to perform weighted summation on the change rates of each strain gauge according to the weight of each strain gauge, and use the weighted summation result as the comprehensive change rate of the strain unit to be predicted;

[0081] The strain prediction module is used to obtain the strain simulation data of the strain unit to be predicted based on the pre-built strain prediction model, and correct it by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted.

[0082] In some optional embodiments, the number of the plurality of strain gauges is specified to be four.

[0083] In some optional embodiments, in the strain gauge weight calculation module, the weight w of each strain gauge is determined by the following formula: i :

[0084] w i =f i / w;

[0085] Among them, f i =1 / (d i ) 2 ; w = ∑ f i ;d i is the Euclidean distance between the ith strain gauge and the unit centroid.

[0086] In some optional implementations, the strain field correction device further includes a strain prediction model construction module for constructing the strain prediction model, and the strain prediction model construction module includes:

[0087] A data extraction unit, used to extract parameters from the aircraft structure finite element model to construct a training set and a test set, wherein the parameters of the training set and the test set include unit centroid and unit strain data;

[0088] The model training unit is used to train the strain prediction model through the training set and verify the trained strain prediction model through the test set.

[0089] In some optional implementations, the data extraction unit includes:

[0090] Data screening subunit, used to eliminate stress concentration units in the finite element model and retain units with continuous strain values ​​in each plane of the aircraft structure;

[0091] The model grouping subunit is used to set the planes in the same span direction in the aircraft structure in the same group. Different groups and different strain types correspond to different strain prediction models.

[0092] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A strain field correction method based on a finite element proxy model, characterized in that: include: Step S1, selecting a plurality of designated strain gauges that are close to the strain unit to be predicted; Step S2, determining the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted; Step S3, acquiring strain simulation data at each strain gauge based on a pre-built strain prediction model, and determining a change rate of the strain simulation data at each strain gauge compared to the strain test data fed back by each strain gauge; Step S4, performing weighted summation on the change rates of the strain gauges according to the weights of the strain gauges, and using the weighted summation result as the comprehensive change rate of the strain unit to be predicted; Step S5, acquiring strain simulation data of the strain unit to be predicted based on the pre-built strain prediction model, and correcting the strain simulation data by the comprehensive change rate to obtain the predicted strain of the strain unit to be predicted; In step S2, the weight w of each strain gauge is determined by the following formula: i : w i =f i / w; Among them, f i =1 / (d i ) 2 ; w = ∑f i ;d i is the Euclidean distance between the ith strain gauge and the unit centroid; Wherein, before step S3, it further includes constructing a strain prediction model, and constructing the strain prediction model includes: Step S31, extracting parameters from the aircraft structure finite element model to construct a training set and a test set, wherein the parameters of the training set and the test set include unit centroid and unit strain data; Step S32: training the strain prediction model using the training set, and verifying the trained strain prediction model using the test set; Step S31 further includes: Step S311, eliminating stress concentration units in the finite element model, and retaining units with continuous strain values ​​in each plane of the aircraft structure; Step S312: planes in the same span direction in the aircraft structure are arranged in the same group, and different groups and different strain types correspond to different strain prediction models.

2. The strain field correction method based on the finite element proxy model according to claim 1, characterized in that: In step S1 , the number of the plurality of strain gauges is specified to be four.

3. A strain field correction device based on a finite element proxy model, characterized in that: include: A strain gauge selection module is used to select a plurality of designated strain gauges that are close to the strain unit to be predicted; A strain gauge weight calculation module, used to determine the weight of each strain gauge according to the distance between each strain gauge and the strain unit to be predicted; A strain gauge change rate calculation module is used to obtain strain simulation data at each strain gauge based on a pre-built strain prediction model, and determine the change rate of the strain simulation data at each strain gauge compared to the strain test data fed back by each strain gauge; A comprehensive change rate calculation module is used to perform weighted summation on the change rates of each strain gauge according to the weight of each strain gauge, and use the weighted summation result as the comprehensive change rate of the strain unit to be predicted; A strain prediction module, used for acquiring strain simulation data of a strain unit to be predicted based on a pre-built strain prediction model, and correcting the strain simulation data by the comprehensive change rate to obtain a predicted strain of the strain unit to be predicted; Among them, in the strain gauge weight calculation module, the weight w of each strain gauge is determined by the following formula: i : w i =f i / w; Among them, f i =1 / (d i ) 2 ; w = ∑f i ;d i is the Euclidean distance between the ith strain gauge and the unit centroid; The strain field correction device further comprises a strain prediction model construction module for constructing the strain prediction model, and the strain prediction model construction module comprises: A data extraction unit, used to extract parameters from the aircraft structure finite element model to construct a training set and a test set, wherein the parameters of the training set and the test set include unit centroid and unit strain data; A model training unit, used to train the strain prediction model through a training set and verify the trained strain prediction model through a test set; The data extraction unit comprises: Data screening subunit, used to eliminate stress concentration units in the finite element model and retain units with continuous strain values ​​in each plane of the aircraft structure; The model grouping subunit is used to set the planes in the same span direction in the aircraft structure in the same group. Different groups and different strain types correspond to different strain prediction models.

4. The strain field correction device based on the finite element proxy model according to claim 3, characterized in that: Specify the number of multiple strain gauges to be four.

Images