High-stability structure deformation performance inversion method

Abstract

The invention discloses a high-stability structure deformation performance inversion method. The method comprises the following steps: dividing an overall structure body into a plurality of finite structure units; obtaining a structural unit end face parameter set; constructing a structural attitude mapping model and obtaining an attitude angle set of grid cells; constructing a modal response prediction model and obtaining a modal coordinate vector prediction value; and carrying out inversion to obtain an overall strain field of the structure. According to the method, full-field deformation inversion can be realized in real time and with high precision through pose reconstruction and modal coordinate inversion technologies in the absence of comprehensive sensor layout, so that the measurement cost is remarkably reduced, and the prediction precision and the interference resistance are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of structural deformation performance inversion, and particularly relates to a high-stability structural deformation performance inversion method. BACKGROUND

[0002] In modern engineering, rigid-flexible coupled structures are widely used in aerospace, construction, bridges and transportation fields. These structures usually exhibit complex deformation characteristics under dynamic loads, and are easily disturbed by external interference in extreme environments, making it difficult to accurately evaluate their deformation through traditional simulation methods. In particular, at the far end of the structure, it is difficult to lay sensors, making it impossible to realize real-time monitoring of the full-size deformation field of the structure, resulting in great challenges for existing monitoring methods.

[0003] Traditional deformation inversion methods, such as modal superposition method, inverse finite element method and neural network method, can predict the deformation of the structure to some extent, but usually rely on accurate simulation models and sufficient sensor data. Especially under dynamic loading and complex environmental conditions, existing methods still face challenges in real-time, accuracy and stability. Specifically, existing methods are difficult to eliminate errors caused by simulation models, and it is difficult to realize high-precision, real-time full-field deformation inversion during dynamic assembly.

[0004] Therefore, how to realize real-time, high-precision deformation inversion of the structure, especially for the monitoring of the far-end deformation, by relying on effective technical means in the absence of comprehensive sensor layout, has become a technical problem to be solved. SUMMARY

[0005] The technical problem solved by the application is to overcome the shortcomings of the prior art and provide a high-stability structural deformation performance inversion method, which effectively solves the problem of excessive dependence on simulation models in traditional methods, and can realize real-time, high-precision full-field deformation inversion in the absence of comprehensive sensor layout through pose reconstruction and modal coordinate inversion technology, significantly reducing the measurement cost and improving the prediction accuracy and anti-interference performance.

[0006] The technical solution of the application is:

[0007] A high-stability structural deformation performance inversion method comprises:

[0008] 1) dividing the overall structure into a plurality of finite structural units, further dividing each structural unit into N e subsections Z e,i , respectively, and performing mesh division on each subsection to obtain M i grid subunits G i,j ; wherein i=1, 2, …, N e ; j=1, 2, …, Mi ;

[0009] 2) define end face directivity parameter θ e,i denotes the first end interface of sub-section Z e,i normal deflection angle relative to the initial static state after loading; define intra-cell directivity parameter θ i,j denotes the local pose of grid sub-cell G i,j relative to its belonging sub-section Z e,i first end interface normal offset angle;

[0010] 3) convert end face directivity parameter θ e,i and intra-cell directivity parameter θ i,j to a unified global reference coordinate system O G , and perform standardization processing, and organize the intra-cell directivity parameter θ' i,j and the end face directivity parameter θ' e,i obtained after standardization processing into a unified multi-scale structure tensor, denoted as structure cell end face parameter set Θ e ;

[0011] 4) take the structure cell end face parameter set Θ e formed in step 3) as input, construct a structure pose mapping model, and output a predicted value of the intra-cell directivity parameter using the structure pose mapping model and form a grid sub-cell pose angle set

[0012] 5) obtain pose deflection angle θ global ;

[0013] 6) according to the pose deflection angle θ global obtained in step 5), run a nonlinear dynamic finite element simulation to generate modal coordinate vector λ = [λ1, λ2,..., λ k ] and perturbation sample ε partial ;

[0014] 7) according to the grid sub-cell pose angle set Θ i obtained in step 4), and according to the modal coordinate vector λ = [λ1, λ2,..., λ k ] and the perturbation sample ε partial obtained in step 6), construct a modal response prediction model f modal , and obtain the modal coordinate vector prediction value modal output by the modal response prediction model f

[0015] 8) use the modal coordinate vector prediction value ​The modal superposition formula is applied to obtain the structural strain field

[0016] Preferably, the sub-section Z e,i in step 2) has a first end interface Specifically, the first end interface is:

[0017] Each sub-section Z e,i has two end faces along the main bearing direction as two end interfaces, and the first end interface and the last end interface are sequentially arranged along the force transmission direction.

[0018] Preferably, the global reference coordinate system O G = X G , Y G , Z G is a right-handed Cartesian rectangular coordinate system, the origin O G is located on the end face of the fixed end of the structure, the X G axis is along the main bearing direction of the whole structure, the Y G axis and the Z G axis are orthogonal to the X G axis, and form an orthogonal rectangular coordinate system.

[0019] Preferably, the method for obtaining the attitude deflection angle θ global in step 5) is specifically:

[0020] 51) Extract the normal direction of each grid sub-element G L from the local coordinate system O i,j , and denote it as n local ;

[0021] 52) Construct a rotation transformation matrix T L between the local coordinate system O G and the global reference coordinate system O LG ;

[0022] 53) Convert n local through the matrix T LG to obtain the global normal vector n global in the global reference coordinate system O G ;

[0023] 54) Determine the attitude deflection angle θ global using the global normal vector n global in the global reference coordinate system O G .

[0024] Preferably, the method for determining the attitude deflection angle θ global in step 54) is specifically:

[0025] ​θ global = cos -1 (dot(n global ,n ref ) / (‖n global ‖×‖n ref ‖))

[0026] where n ref is the reference vector in the global reference frame O G under ideal static state.

[0027] Preferably, the loss function of the structure pose mapping model in step 4) is:

[0028]

[0029] where, is the predicted value of the intracell directivity parameter output by the structure pose mapping model.

[0030] Preferably, the method of generating the modal coordinate vector λ = [λ1, λ2,..., λ k ] and the perturbation sample ε partial in step 6) is:

[0031] 61) Construct a set of bias perturbation parameters Δ to represent the potential source of structural performance changes, specifically:

[0032] Δ = {E, Δm, F, ζ, k c}

[0033] where E is the material elastic modulus; Δm is the mass offset value of the i-th sub-section; F is the excitation amplitude of the i-th sub-section; ζ is the damping ratio of the i-th sub-section; k c is the connection stiffness of the i-th sub-section.

[0034] 62) According to the pose deflection angle θ global in step 5) and the set of bias perturbation parameters Δ obtained in step 61), run a nonlinear dynamic finite element simulation to generate the modal coordinate vector λ = [λ1, λ2,..., λ k ] and the perturbation sample ε partial ; where k is the number of modes.

[0035] Preferably, the set of perturbation parameters Δ in step 61) has N groups, which are generated in the parameter space by Latin hypercube sampling, and the number of network training samples N is in the range of 500-10,000.

[0036] Preferably, the loss function of the modal response prediction model in step 7) is defined as:

[0037]

[0038] wherein n is the maximum value of the modal number k, and n ranges from 5 to 20; is a loss value, used to measure the difference between the predicted value of the modal response prediction model and the true value.

[0039] Preferably, step 8) obtains the overall strain field of the structure by inversion , specifically:

[0040]

[0041] wherein φ k (x) is the strain mode function of the kth mode at the position x in the global reference coordinate system O G ; is the strain at position x under the kth mode.

[0042] Compared with the prior art, the advantages of the present application mainly lie in the following aspects:

[0043] 1. High-precision real-time inversion: Through structure pose reconstruction and modal coordinate inversion technology, the present application can realize high-precision real-time inversion of full-field deformation of rigid-flexible coupled structures in the absence of sensors, overcoming the problems of insufficient real-time performance and accuracy of traditional methods.

[0044] 2. No need for comprehensive sensor deployment: The present application can fully invert the full-field deformation through digital twin technology, combined with near-end sensor data and simulation analysis, in the absence of sensors at the far end, providing an innovative solution for traditional methods.

[0045] 3. Wide applicability: The present application is not only applicable to dynamic deformation inversion of large rigid-flexible coupled structures, but also can be applied to deformation monitoring of other complex dynamic systems, having wide application prospects. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is a structure pose reconstruction flowchart of the present application;

[0047] Figure 2 is a structure deformation field inversion flowchart of the present application. DETAILED DESCRIPTION

[0048] The present application relates to a high-stability structure deformation performance inversion method, suitable for performance analysis and evaluation of rigid-flexible coupled structures during dynamic deformation, especially in complex and extreme environments. The characteristics and advantages of the present application will become clearer and more explicit with the following more detailed description of the present application.

[0049] Step one, divide the whole structure into several finite structure units, the structure unit refers to the local area obtained by dividing the whole structure, which has clear geometric boundaries and connection relationship, and the constituent material can be rigid (such as metal beam) or flexible (such as composite material unfolding arm). Set the whole structure contains E structure units (that is, each part in the whole structure is mapped to a structure unit), each structure unit is marked as S e , where e = 1, 2, …, E. The structure unit division result provides basic structure unit division unit input for subsequent multi-scale modeling and structure posture feature extraction.

[0050] Step two, further divide each structure unit into N e subsections Z e,i , in order to improve the resolution and modeling accuracy of local deformation field. Wherein, i = 1, 2, …, N e . The subsection is an intermediate structure level, which is the scope of subsequent "end face directivity parameter" and "intra-unit directivity parameter" extraction, and its number will be used to construct the grouping index of input tensor.

[0051] Step three, mesh each subsection to obtain mesh sub-unit G i,j , each subsection Z e,i contains M i mesh sub-units, where j = 1, 2, …, M i . The mesh division of this step can use tetrahedral element, hexahedral element or other suitable element type for the topological characteristics of the structure. The mesh sub-unit is the basic unit for extracting local posture feature θ i,j , and its number will be used as one of the spatial resolution dimensions of the neural network output tensor.

[0052] Step four, define each subsection Z e,i has two end faces along the main bearing direction as two end interfaces, which are the first end interface and the last end interface These two interfaces refer to the two boundary interfaces of the starting end and the ending end of the subsection in its main bearing direction, which are usually the boundary node surfaces in its geometric model, used to connect adjacent subsections or connect with loading / constraint boundaries. In order to simplify the model input and improve real-time performance, the present application only selects the first end interface as the posture recognition reference surface. The first end interface definition will be used as the geometric reference surface of the end face directivity parameter θ e,i , which is used to extract the input reference coordinates of the posture change direction subsequently.

[0053] Step five, define the end face directivity parameter to represent the change of the normal direction of the first end interface in the deformation process. The end face directivity parameter is marked as θ e,i , which represents the subsection Ze,i Homepage The normal deflection angle relative to the initial static state after loading. θ e,i The unit is angle (°) or radian (rad), used to quantify the overall attitude change trend of the structure at that location. This angle can be obtained through attitude sensors (such as inertial measurement units, fiber optic gyroscopes, etc.) deployed in the front end interface region or digital image processing methods (such as digital image correlation technology, DIC). If simulation calculation is used, it can be calculated by extracting the angle between the front end interface normal vector under the initial configuration and the current configuration. This parameter, as one of the core variables for modal recognition and neural network input, is used to establish the mapping relationship between macroscopic pose and microscopic response. End face directivity parameter θ e,i It will be used as an important component of the macroscopic attitude characteristics of the structural unit, and will be input into the structural attitude mapping model and modal response prediction model.

[0054] This invention primarily uses the parameters of the initial interface as the basis for feature extraction, but in further extensions to higher-order response modeling scenarios, the final interface can be introduced. normal parameter θ tail And by calculating the angle difference Δθ between the two ends. e =θ tail -θ head It is used to identify complex responses such as torsion, bending or non-uniform tension inside structural units, and has good scalability and engineering adaptability.

[0055] Step Six: Define the intra-element orientation parameters to reflect the local displacement gradient and nonlinear deformation trend within the structure. Extract the mesh sub-element G. i,j The spatial deflection angle of the principal normal direction during loading is called the element directivity parameter, denoted as θ. i,j , representing the mesh sub-cell G i,j The local pose relative to its sub-segment Z e,i Homepage The normal offset angle. This angle is mainly used to reflect the non-rigid response characteristics of the structure at the microscopic level, including microscale effects such as local warping, torsion, and shear offset.

[0056] Intra-cell directivity parameter θ i,j The methods of obtaining it include, but are not limited to:

[0057] High-density sensor extraction method: Deploy miniature attitude sensors (such as MEMS gyroscopes and fiber Bragg grating sensors) in key structural areas to collect minute angular offsets in real time;

[0058] Image processing method: Through high-resolution image acquisition system and digital image correlation technology (DIC), the displacement field change of the grid between different time frames is calculated;

[0059] Numerical simulation method: Based on the output of multi-body dynamics or finite element simulation platform, the normal change of nodes is discretely extracted to obtain the attitude information of each G i,j .

[0060] Intra-cell directivity parameter θ i,j characterizes the local nonlinear deformation behavior of the structure and will be used as an input feature in the construction of the multi-scale characterization tensor.

[0061] Step seven, convert the end face directivity parameter θ e,i obtained in step five and the intra-cell directivity parameter θ i,j obtained in step six into a unified global reference coordinate system O G , ensuring the physical consistency of the end face directivity parameter θ e,i and the intra-cell directivity parameter θ i,j .

[0062] Set the global reference coordinate system O G = X G , Y G , Z G , which is a right-handed Cartesian rectangular coordinate system defined as follows: the origin O G of the global reference coordinate system is located on the end face of the fixed end of the overall structure, the X G axis is along the main load direction of the structure, the Y G axis and the Z G axis are orthogonal to the X G axis, forming an orthogonal rectangular coordinate system; in engineering practice, the Z G axis can represent the direction of gravity and is often set to [0, 0, 1] as a reference direction for attitude.

[0063] This transformation is achieved through an attitude rotation matrix or a direction cosine matrix, ensuring that all local deflection quantities have clear physical meanings in a unified coordinate system, facilitating subsequent model input and feature fusion.

[0064] Step eight, normalize the end face directivity parameter and intra-cell directivity parameter in the global reference coordinate system O G obtained in step seven to improve the numerical stability and network learnability of the attitude parameters. This includes: linear normalization to the interval [-1, 1]; using sin(θ), cos(θ) instead of direct angle input to express its periodicity; if there is attitude signal jitter, sliding average or Kalman filter can be applied for noise reduction. Obtain the end face directivity parameter θ′ e,i and the intra-cell directivity parameter θ′i,j These standardized structural pose features serve as the core input tensors for the training and inference phases of the neural network model, enhancing numerical stability and convergence efficiency.

[0065] Step 9: Obtain the intra-cell directivity parameter θ′ from the standardization process in Step 8. i,j With end face pointing parameter θ′ e,i The organization is represented by a unified multi-scale structural tensor, which serves as the pose feature input for structural elements and is denoted as the set of structural element end-face parameters Θ. e This method is used to achieve full-field 3D deformation reconstruction in complex scenarios. Unlike traditional methods that only focus on nodal displacements, the above steps, through local directional modeling, can capture more detailed structural response features, effectively improving the model's robustness to uncertainties such as nonlinear deformation, boundary perturbations, and sensor loss. This multi-scale tensor structure Θ... e As a complete structural pose feature vector, it is input into the neural network model to reconstruct the internal local pose or predict modal coordinates.

[0066] Step 10, Sub-segment Z e,i and its grid sub-unit G i,j The attitude information is usually measured or defined in its own local coordinate system, from the local coordinate system O. L Extract each grid sub-cell G i,j The normal direction is denoted as n. local , representing the grid sub-cell G i,j In the current local coordinate system O L The orientation of the n. local This can originate from geometric normal vectors in the initial stage of structural modeling, attitude sensor measurement data (such as IMU), element normals in finite element simulation results, or surface vectors reconstructed by digital image processing algorithms. Mesh surface normal n local For subsequent normal vector transformation and attitude deflection angle θ global The basic data input for calculation.

[0067] Step 11: Construct a local coordinate system O L With global reference coordinate system O G The rotation transformation matrix T between LG This matrix describes the rigid body rotation relationship from local attitude to global attitude and can be constructed using Euler angles, direction cosine matrices, or quaternions. Transformation matrix T LG It is the core intermediate input for normalization and orientation angle calculation.

[0068] Step 12: Set the local coordinate system O L The normal direction n of the grid sub-cell local Through matrix T LGtransformed into global normal vector n in global reference coordinate system O G global , the specific formula is:

[0069] n global = T LG × n local

[0070] n global represents the sub-section Z e,i The spatial orientation in the global reference coordinate system is the basis for subsequent angle calculation. The global normal vector n global is the direct input quantity for calculating the pose deflection angle θ global .

[0071] Step thirteen, in the global reference coordinate system O G , define the reference vector n ref in the ideal static state, which is used to measure the degree of deviation. The definition of n ref is determined according to the main bearing direction of the sub-section Z e,i , which is often selected as: n ref =

[001] or other unit vectors of the main bearing direction. The reference vector n ref is the reference direction for calculating the pose deflection angle θ global , which participates in the neural network input or the construction of the physical consistency loss term.

[0072] Step fourteen, calculate the included angle between the global normal vector n global obtained in step twelve and the reference vector n ref obtained in step thirteen, to obtain the pose deflection angle θ global , the calculation formula is:

[0073] θ global = cos -1 (dot(n global ,n ref ) / (‖n global ‖×‖n ref ‖))

[0074] Where dot represents vector dot product, and ||·|| represents vector length. The angle is in degrees or radians, which is the rotation amount of the structural unit relative to the ideal direction, and is one of the input features of the deformation inversion model. The pose deflection angle θ global will be used as a measure of the pose change of the structure in the global coordinate system, and as one of the important input features in the neural network model or modal prediction process.

[0075] Step fifteen, with the structure unit end face parameter set Θ e ​For input, a neural network model is constructed to predict the local pose angles of its internal grid, which is a structure pose mapping model. The structure pose mapping model is a feedforward neural network (Multi-Layer Perceptron, MLP). The output unit points to the predicted value of the directional parameter and forms a set of grid cell pose angles

[0076] The mapping function f of the structure pose mapping model pose , is defined as follows:

[0077] Θ i =f pose (Θ e ; W; θ global )

[0078] Where W is a set of network parameters, the network structure is composed of multiple layers of perception, and the activation function uses ReLU or Tanh.

[0079] The loss function is:

[0080]

[0081] Where, is the predicted value of the neural network, i.e. the predicted value of the unit directional parameter output by the structure pose mapping model, and θ′ i,j is the unit directional parameter in the global reference coordinate system in step eight as a multi-scale representation of structural deformation input.

[0082] Step sixteen, construct a set of bias disturbance parameters Δ, i.e. the potential source of structural performance changes, such as:

[0083] Δ={E,Δm,F,ζ,k c}

[0084] Where E is the material elastic modulus; Δm is the mass offset value of the i-th sub-section; F is the excitation amplitude of the i-th sub-section; ζ is the damping ratio of the i-th sub-section; k c is the connection stiffness of the i-th sub-section.

[0085] This set of disturbance parameters Δ is the training input variable for establishing the mapping relationship between strain observation and modal response. N sets of bias disturbance parameter sets Δ are generated in the parameter space by Latin hypercube sampling (LHS), and the typical network training sample number N: 500-10,000.

[0086] Step seventeen, according to the pose deflection angle θ global of step 14 and the bias disturbance parameter set Δ obtained in step 16, run a nonlinear dynamic finite element simulation to generate modal coordinate vectors λ=[λ1,λ2,...,λ k ] and disturbance samples εpartial where k is the number of modes.

[0087] The posture deflection angle θ obtained in step fourteen is introduced global As state auxiliary input, to enhance the structural state perception ability of the modal response prediction model. Run the nonlinear dynamic finite element simulation under each group of sample conditions, output: modal coordinate vector λ = [λ1, λ2,..., λ k ] and disturbance sample ε partial (i.e. the strain observation value corresponding to each sub-section); modal coordinate vector λ = [λ1, λ2,..., λ k ] represents the response eigenvalue of the structure at different modes, reflecting the main mode characteristics of the overall structure in the dynamic state.

[0088] Step eighteen, according to the grid sub-element posture angle set Θ i obtained in step 15, and according to the modal coordinate vector λ = [λ1, λ2,..., λ k ] and the disturbance sample ε partial obtained in step 17, construct the modal response prediction model f modal ; Construct the modal response prediction model based on radial basis function neural network (Radial Basis Function Neural Network, RBF-NN), which is used to replace the traditional inversion process based on modal participation factor solution.

[0089] Construct the modal response prediction model f modal :

[0090]

[0091] where Θ rbf is the network weight of the modal response prediction model;

[0092] The loss function is defined as:

[0093]

[0094] where n is the maximum value of the number of modes k, and the typical value range of n is 5-20. is the modal coordinate vector prediction value output by the modal response prediction model; is the loss value, which is used to measure the difference between the predicted value and the true value of the modal response prediction model.

[0095] The modal strain superposition formula is used to reconstruct the full-field strain distribution:

[0096]

[0097] where φ k(x) is the strain mode shape function at position x under the kth mode G (x) is the strain mode shape function at position x under the kth mode (x) is the strain mode shape function at position x under the kth mode

[0098] The modal response prediction model uses the disturbance sample ε partial As input features, the predicted modal principal coordinate vector for subsequent deformation field recovery. The process of structural deformation field inversion to obtain the prediction model is shown in Figure 2 .

[0099] The above variables are integrated to form a real-time input feature set in the operation stage, which drives the network model to complete the reconstruction of the full-field strain response of the structure. The structure pose reconstruction process is shown in Figure 1 .

[0100] The method effectively breaks through the limitation of traditional modal identification relying on high-density sensors and high-precision modeling. In the presence of disturbance parameters, uncertain connection conditions and missing remote sensors, the structural deformation response can still be efficiently and stably estimated.

[0101] Embodiment

[0102] Spacecraft rigid-flexible coupling structure deformation field inversion system construction and test based on the present application

[0103] 1. Model background and simulation object

[0104] A certain type of space deployment structure (such as a three-section deployment arm) is selected as the verification object. The structure is composed of carbon fiber flexible arm body and aluminum alloy rigid node, with a total length of 2.4m, containing 12 main structural units, each unit is divided into 10 sub-sections, and each section contains 8 grid sub-units.

[0105] 2. Simulation modeling and disturbance parameter setting

[0106] A three-dimensional simulation model is established using nonlinear multi-body dynamics modeling software (such as ADAMS or Abaqus), and a set of disturbance parameters is introduced:

[0107] Elastic modulus E ∈ [125, GPa, 145, GPa];

[0108] Node mass disturbance \Deltam ∈ [-5%, +5%];

[0109] Excitation amplitude F ∈ [30, N, 50, N];

[0110] Damping ratio ζ ∈ [0.01, 0.08];

[0111] Contact stiffness k c ∈ [2×10 4 ,1×105 ]N / m.

[0112] 3. Pose mapping network training process

[0113] For each set of perturbation conditions, the end face normal deflection angle θ e,i is collected as the neural network input; the corresponding grid sub-element pose angle θ i,j is taken as the output label. A 4-layer feedforward neural network is used for training:

[0114] Learning rate: 1e-3;

[0115] Activation function: ReLU;

[0116] Batch size: 32;

[0117] Optimizer: Adam.

[0118] The average prediction error is <0.4°, and the training error converges stably.

[0119] 4. Modal agent network training

[0120] 500 sets of samples are generated by Latin hypercube sampling, and the structural part node strain value ε partial and the corresponding modal coordinate vector λ are extracted. A radial basis neural network is used for modeling, and the structure is:

[0121] Number of center nodes: 200;

[0122] Base function width parameter σ = 0.5;

[0123] The loss function MSE converges to <1×10 -3 .

[0124] 5. Strain field inversion effect verification

[0125] In the independent test set, the true strain field ε true (x) is compared with the predicted strain The average error is controlled at 4.8%; under a perturbation amplitude of ±15%, the system stably reconstructs the structural response of the distal region.

[0126] Although the present application has been disclosed with reference to the preferred embodiments above, it is not intended to limit the present application, and any person skilled in the art can make possible changes and modifications to the technical solutions of the present application using the methods and technical contents disclosed above without departing from the spirit and scope of the present application. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical solutions of the present application, are within the protection scope of the present application. In the case of no conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0127] That which is not described in detail in the specification of the present application is well known to those skilled in the art.

Claims

1. A method for inverting the deformation performance of highly stable structures, characterized in that, include: 1) Divide the overall structure into multiple structural units, and further divide each structural unit into N. e Sub-segment Z e,i Each sub-segment is divided into grids to obtain M. i Each grid sub-cell G i,j Where i = 1, 2, ..., N e j = 1, 2, ..., M i ; 2) Define the end-face pointing parameter θ e,i Subsegment Z e,i Homepage The normal deflection angle relative to the initial static state after loading; defining the directivity parameter θ within the element. i,j Represents the mesh sub-cell G i,j The local pose relative to its sub-segment Z e,i Homepage The normal offset angle; 3) Adjust the end-face pointing parameter θ e,i and the intra-cell directivity parameter θ i,j Switch to a unified global reference coordinate system O G The following is represented and standardized. The intra-cell directivity parameter θ′ obtained after standardization is... i,j With end face pointing parameter θ′ e,i The organization is a unified multi-scale structural tensor, denoted as the set of end-face parameters Θ of the structural elements. e ; 4) The set of end face parameters Θ of the structural unit formed in step 3) e Using the input, a structure attitude mapping model is constructed, and the predicted values ​​of the intra-cell directivity parameters are output using the structure attitude mapping model. And form a set of attitude angles for the mesh sub-units. 5) Obtain the attitude deflection angle θ global ; 6) Based on the attitude deflection angle θ obtained in step 5), global Run a nonlinear dynamics finite element simulation to generate the modal coordinate vector λ = [λ1, λ2, ..., λ]. k ] and perturbation sample ε partial ; 7) Based on the set of mesh sub-cell attitude angles Θ obtained in step 4), i And based on the modal coordinate vector λ=[λ1,λ2,...,λ] obtained in step 6), k ] and perturbation sample ε partial Construct a modal response prediction model f modal Obtain the modal response prediction model f modal Predicted output modal coordinate vector 8) Predict the modal coordinate vector values ​​obtained in step 7). The overall strain field of the structure is obtained by applying the modal superposition formula.

2. The method for inverting the deformation performance of a highly stable structure according to claim 1, characterized in that, The sub-segment Z mentioned in step 2) e,i Homepage Specifically: Define each sub-segment Z e,i The two end faces along the main load-bearing direction serve as the two end interfaces, and the two end faces along the force transmission direction serve as the first end interfaces. With end interface 3. The method for inverting the deformation performance of a highly stable structure according to claim 2, characterized in that, The global reference coordinate system O G =X G ,Y G Z G Let O be a right-handed Cartesian coordinate system with the origin O. G Located on the end face of the fixed support end of the structure, X G The axis is along the overall main load-bearing direction of the structure, Y G Axis, Z G Axis and X G The axes are pairwise orthogonal, forming an orthogonal rectangular coordinate system.

4. The method for inverting the deformation performance of a highly stable structure according to claim 3, characterized in that, Step 5) Obtain the attitude deflection angle θ global The method is as follows: 51) From the local coordinate system O L Extract each grid sub-cell G i,j The normal direction is denoted as n. local ; 52) Construct a local coordinate system O L With global reference coordinate system O G The rotation transformation matrix T between LG ; 53) n local Through matrix T LG Transform to obtain the global reference coordinate system O. G The global normal vector n global ; 54) Using the global reference coordinate system O G The global normal vector n global Determine the attitude deflection angle θ global .

5. The method for inverting the deformation performance of a highly stable structure according to claim 4, characterized in that, In step 54), the attitude deflection angle θ is determined. global The method is as follows: i global =cos -1 (dot(n global ,n ref ) / (‖n global ‖×‖n ref ‖)) Where, n ref For the global reference coordinate system O G The reference vector in the ideal static state.

6. A method for inverting the deformation performance of a highly stable structure according to any one of claims 3-5, characterized in that, The loss function of the structure-attitude mapping model in step 4) is: in, This refers to the predicted values ​​of the intra-cell directivity parameters output by the structural attitude mapping model.

7. The method for inverting the deformation performance of a highly stable structure according to claim 6, characterized in that, Step 6) Generate the modal coordinate vector λ = [λ1, λ2, ..., λ k ] and perturbation sample ε partial The method is as follows: 61) Construct a set of deviation disturbance parameters Δ to represent the potential sources of structural performance changes, specifically: Δ={E,Δm,F,ζ,k c } Where E is the material's elastic modulus; Δm is the mass offset value of the i-th sub-segment; F is the excitation amplitude of the i-th sub-segment; ζ is the damping ratio of the i-th sub-segment; k c Let be the connection stiffness of the i-th sub-segment. 62) Based on the attitude deflection angle θ from step 5). global Using the deviation disturbance parameter set Δ obtained in step 61), run a nonlinear dynamic finite element simulation to generate the modal coordinate vector λ=[λ1,λ2,...,λ k ] and perturbation sample ε partial Where k is the number of modes.

8. The method for inverting the deformation performance of a highly stable structure according to claim 7, characterized in that, The perturbation parameter set Δ in step 61) has N sets, which are generated in the parameter space by Latin hypercube sampling. The number of network training samples N ranges from 500 to 10,000.

9. The method for inverting the deformation performance of a highly stable structure according to claim 7, characterized in that, In step 7), the loss function of the modal response prediction model is defined as: Where n is the maximum value of the number of modes k, and the value of n ranges from 5 to 20; The loss value is used to measure the degree of difference between the predicted value and the actual value of the modal response prediction model.

10. The method for inverting the deformation performance of a highly stable structure according to claim 9, characterized in that, Step 8) Inversion to obtain the overall strain field of the structure The method is as follows: Where, φ k (x) represents the global reference coordinate system O. G The strain mode shape function of the k-th mode at the lower coordinate x; Let x be the strain at position x in the k-th mode.

Images