Mechanical test and simulation method for thin-walled steel structure based on digital twinning

Abstract

The present application provides a kind of based on digital twinning thin-walled steel structure mechanics test and simulation method, comprising the following steps: S1: build thin-walled steel structure entity geometric information reconstruction module;S2: build the modal information identification module of the thin-walled steel structure entity;S3: development modal stress field analysis algorithm, and based on the modal stress field analysis algorithm builds the stress field correction module of the thin-walled steel structure;S4: threshold is set for the start-stop switch of the stress field correction module;S5: overall the geometric information reconstruction module, the modal information identification module and the stress field correction module and start-stop node function constructed, realize the high fidelity finite element simulation of the thin-walled steel structure.The present application provides a kind of based on digital twinning thin-walled steel structure mechanics test and simulation method, realizes the high fidelity finite element modeling analysis of thin-walled steel structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical test numerical simulation and health monitoring of building steel structure, and particularly relates to a mechanical test and simulation method for thin-walled steel structure based on digital twinning. BACKGROUND

[0002] In engineering, local or overall buckling of thin-walled steel is usually allowed, and the post-buckling strength of the thin-walled steel is utilized to improve material utilization. The buckling behavior of the thin-walled steel is extremely susceptible to factors such as material properties of the component, residual stress, geometric defects, boundary conditions and component damage. In the simulation analysis of such structures, the assumptions and fitting of the above factors are not accurate enough, which brings errors to the calculation results, and the bearing capacity of the structure cannot be accurately predicted. In order to avoid the interference of the above factors on the simulation results, in addition to the traditional means such as material property test, processing residual stress detection, initial geometric defect determination and damage identification to improve the modeling accuracy, algorithm-driven simulation error identification and compensation is also an effective means. However, due to the large number of influencing factors, there are still problems such as complicated detection process, one-sided compensation mechanism and unsatisfactory optimization effect. In addition, in the process of virtual modeling, it is a very challenging task to create a mechanical model that has exactly the same geometric, physical characteristics and constraint conditions as the structure entity. Therefore, by means of the concept of digital twinning, a simulation method is proposed, which can realize automatic three-dimensional modeling by mapping the structure physical entity and realize high-fidelity finite element simulation of steel structure by comprehensively utilizing multiple error compensation mechanisms. It has very important significance for the safe operation, management and maintenance of thin-walled steel structure. SUMMARY

[0003] In view of the deficiencies in the prior art, the present application provides a mechanical test and simulation method for thin-walled steel structure based on digital twinning, which utilizes point cloud three-dimensional reconstruction, machine learning algorithm and operational modal analysis to realize high-fidelity finite element modeling and analysis of thin-walled steel structure.

[0004] In order to achieve the above purpose, the present application provides a mechanical test and simulation method for thin-walled steel structure based on digital twinning, comprising the steps of:

[0005] S1: building a thin-walled steel structure entity geometric information reconstruction module, using a point cloud data processing algorithm to realize automatic reconstruction of a thin-walled steel structure finite element three-dimensional geometric model;

[0006] S2: building a thin-walled steel structure entity modal information identification module, using OMA method to analyze the excitation response of the thin-walled steel structure and then extract the real modal parameters of the thin-walled steel structure;

[0007] S3: develop a modal stress field analysis algorithm, and build a stress field correction module of the thin-walled steel structure based on the modal stress field analysis algorithm, and fit the stress field of the structure by inputting the measured modal parameters by using a machine learning algorithm;

[0008] S4: set a threshold value for the start-stop switch of the stress field correction module, set a start-stop limit for the structure stress field correction according to the parameter form of the joint index and the simulation accuracy requirement;

[0009] S5: coordinate the constructed geometric information reconstruction module, modal information identification module and stress field correction module with the start-stop node function, and realize high-fidelity finite element simulation of the thin-walled steel structure.

[0010] Preferably, in the S1 step, the thin-walled steel structure is three-dimensionally scanned, point cloud modeling and segmentation, thin-walled component neutral axis automatic extraction, and stamp mapping and splicing are performed; wherein the thin-walled component neutral axis extraction includes point cloud slicing, slice centroid extraction, centroid coordinate mapping and neutral axis fitting related procedures.

[0011] Preferably, in the S2 step, the excitation response of the thin-walled steel structure is analyzed, and for the non-time-varying signal under stationary excitation, the frequency, mode shape and damping ratio of the thin-walled steel structure are identified and calculated from the frequency response function by using the frequency domain decomposition method; for the case where the response signal is a time-varying signal, a time-frequency method is adopted to analyze the related parameters.

[0012] Preferably, in the S3 step, the development of the modal stress field analysis algorithm further includes the steps of:

[0013] S31: using the geometric information reconstruction module and the modal information identification module, relying on the assumed initial stress field, formulating a loading scheme to generate original samples, and selecting appropriate samples into the training set with reference to the joint index;

[0014] S32: using a machine learning algorithm to build an algorithm framework that meets the functional requirements, and introducing the training set with sufficient samples into the algorithm framework to train the algorithm framework;

[0015] S33: selecting a structure or component that has not been analyzed for simulation and modal measurement, and introducing the results in the form of a test set into the trained algorithm framework for testing.

[0016] Preferably, in the S3 step, the training set creation method of the modal stress field analysis algorithm is: firstly, a large number of initial stress field samples are obtained by using stress field scanning, technical regulations and residual stress research data, and the initial stress field samples are introduced into a finite element three-dimensional geometric model to generate a final analysis model; a large number of stress field and modal simulation samples are obtained by using the analysis model, and the simulation samples meeting the requirements are selected by using the joint index and threshold value and are included in the training set; the above process is repeated, and the training set is continuously expanded by replacing the structure or component and the loading scheme.

[0017] Preferably, in the S5 step, the high-fidelity finite element simulation refers to creating a three-dimensional geometric model by using the geometric information reconstruction module, giving the initial stress field by using the modal information identification module and the modal stress field analysis algorithm, adjusting the simulation stress field that needs to be corrected by using the stress field correction module, and verifying the improvement of simulation accuracy by using the corrected simulation modal.

[0018] The present application has the following beneficial effects due to the adoption of the above technical solutions:

[0019] 1. The thin-walled steel structure finite element can be automatically three-dimensionally modeled, and the initial geometric defects of the thin-walled component can be considered in the model;

[0020] 2. The initial stress field of the structure that is difficult to measure can be analyzed through the measurable structure modal, so as to ensure that the key simulation parameters of the thin-walled component are of sufficient accuracy;

[0021] 3. The simulation analysis takes the physical entity as a data source, and fully considers the influence of the real physical state of the thin-walled steel structure on the mechanical properties of the structure;

[0022] 4. The stress field of the simulation analysis is corrected by using the real modal at each loading step, the technical concept of virtual and real interaction of digital twin is highlighted, and the influence of the factors that cannot be quantified on the stress state of the structure is avoided;

[0023] 5. The multiple compensation mechanisms of the thin-walled steel structure simulation are planned by using the digital twin concept, and the fidelity of the structure simulation is greatly improved by the comprehensive application of multiple technologies.

[0024] 6. Although the present application aims to improve the simulation accuracy of the thin-walled steel structure, focuses on how to consider the initial defects of the structure in the simulation model, part of the functions (such as the geometric information reconstruction module) of the present application can provide reference for modeling and defect identification of other types of structures, and has expandability. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The flowchart of the thin-walled steel structure mechanical test and simulation method based on digital twin of the embodiments of the present application;

[0026] Figure 2 This is a flowchart illustrating the development of a modal stress field analysis algorithm and the construction of a stress field correction module according to an embodiment of the present invention.

[0027] Figure 3 This is a schematic diagram of the test apparatus for conducting axial compression tests and vibration response tests on thin-walled steel components according to an embodiment of the present invention;

[0028] Figure 4 This is a flowchart illustrating the modular architecture and practical process of the digital twin-driven integrated simulation technology for thin-walled steel components, as described in this invention.

[0029] Figure 5 This is a flowchart illustrating the process of creating and collecting training set samples for the modal stress field analysis algorithm according to an embodiment of the present invention. Detailed Implementation

[0030] The following is based on the attached diagram. Figures 1-5 The present invention provides preferred embodiments and describes them in detail to enable a better understanding of the functions and features of the present invention.

[0031] like Figure 3 As shown, the development and application of this comprehensive simulation technology are illustrated using a thin-walled steel component 3, which can be installed on the experimental device shown in the diagram. The experimental device is set on the ground 4 and includes a clamp 2 and a loading device 1. Since this comprehensive simulation technology prepares a training set and completes algorithm training for the stress field correction module in advance, the development process of each module of the comprehensive simulation technology is explained first, and then the application process of this technology is illustrated using a high-fidelity simulation analysis of a single thin-walled component under stepwise static loading as an example.

[0032] The following is a detailed description of the digital twin-driven integrated simulation technology for shielded steel structures, as illustrated in the embodiments of this application.

[0033] Please see Figures 1-5 An embodiment of the present invention provides a method for mechanical testing and simulation of thin-walled steel structures based on digital twins, comprising the following steps:

[0034] S1: Construct a module for reconstructing the geometric information of thin-walled steel structures;

[0035] For those already installed Figure 3 The thin-walled steel component 3 on the experimental device shown was subjected to a refined point cloud scan, and the scan results were processed into a component-level point cloud model. This point cloud model was segmented along the local tensile axis, and the resulting point cloud slices were unfolded into two-dimensional images through voxelization. After filtering the images, the centroid extraction algorithm was used to obtain the centroid position of the slice and remap it to three-dimensional space. Then, the neutral axis curve of the component was fitted with the centroid of each slice, and a finite element three-dimensional geometric model was generated from it.

[0036] S2: Construct a module for recognizing the modal information of thin-walled steel structures;

[0037] The OMA method is used to obtain the modal parameters of the installed components in real time. First, the components are excited using an exciter, and the response is measured using an accelerometer. The power spectral density function matrix is ​​obtained from the measured response. This matrix is ​​then subjected to singular value decomposition (SVD) at each discrete frequency point using the frequency domain decomposition method. The frequency parameters are obtained based on the frequency corresponding to the position of the maximum singular value, and the corresponding mode shape can be obtained from the measured frequency. Since the input—the mode—needs to be described as completely as possible when creating the machine learning algorithm, an enhanced frequency domain decomposition method can also be used to transform the power spectral density of the response to the time domain through inverse Fourier transform for analysis. Then, the autocorrelation and cross-correlation functions are calculated to obtain the damping ratio of this mode. Due to the limited data sample size, only the top 8 modal frequencies and mode shapes of each measured mode are retained.

[0038] S3: Develop a modal stress field analysis algorithm and build a stress field correction module for thin-walled steel structures based on the modal stress field analysis algorithm;

[0039] Specifically, the constructed geometric information reconstruction module and modal information recognition module are combined with the initial stress samples to generate a training set. Then, the training set is imported into a machine learning framework based on a convolutional neural network for training, and the trained algorithm is tested.

[0040] Step S3 further includes the following steps:

[0041] S31: Using the geometric information reconstruction module and the modal information recognition module, based on the assumed initial stress fields, a loading scheme is proposed to generate original samples, and suitable samples are selected for inclusion in the training set with the joint index as a reference.

[0042] like Figure 5 As shown, the geometric information reconstruction module is first used to obtain the three-dimensional geometric model of the installed component. Then, a large number of initial stress fields are derived from measured data, specifications, and literature to obtain different analysis models. Static analysis and modal analysis are performed on each analysis model. The modal analysis results are then combined with the measured modal results obtained based on the modal parameter identification module to obtain the joint index of {frequency difference, MAC}. The frequency difference is taken as the mean of the relative errors of the frequencies of each modality. The MAC calculation formula is (considering only the top 8 modes of vibration intensity):

[0043]

[0044] The test mode vector refers to the mode shape vector of test mode j, which in this context refers to the simulation mode in the finite element simulation. The mode shape vector of the compatible analytical mode j refers to the measured mode derived from the excitation response analysis of the structure.

[0045] The threshold is preset as: the relative error of frequency is lower than 0.1 and the MAC is not lower than 0.95; the threshold is adjusted according to the specific situation of the obtained sample number. By comparing the joint index result with the threshold, it is evaluated whether the corresponding simulation result can be included in the training set. By replacing the component and modifying the loading scheme and repeating the above steps, the capacity of the training set elements is continuously expanded.

[0046] S32: using a machine learning algorithm to construct an algorithm framework meeting the functional requirements, and introducing a sample sufficient training set into the algorithm framework to train the algorithm framework;

[0047] The input of the model is a multi-dimensional vector constructed with {frequency, mode shape vector, damping ratio}, where frequency, mode shape vector and damping ratio are characteristics that can represent modal information, and the dimensions of the characteristics are 1, n and 1 respectively. The data matrix constructed is a sparse matrix. The purpose of model training is to find the mapping relationship of the modal corresponding to the stress field cloud map. A multi-layer neural network is needed to process the input data and obtain the encoding vector corresponding to the stress field cloud map. The training process includes two neural networks, namely the modal analysis network and the stress field cloud map feature extraction convolutional neural network. The latter is used to extract the features of the stress field cloud map, and the former uses the extracted features as input labels for training. The modal analysis network training forms the mapping of the input to the stress field cloud map feature vector. Both the modal analysis network and the convolutional neural network are implemented through python compilation.

[0048] S33: selecting a structure or component that has not been analyzed to perform simulation and modal measurement respectively, and introducing the results in the form of a test set into the trained algorithm framework for testing.

[0049] Selecting a thin-walled steel component that has not been analyzed, selecting different loading schemes for simulation and test, repeating the steps of creating a training set to obtain the corresponding test set. Re-input the test set into the modal analysis multi-layer neural network obtained by training, and the network output obtains the feature vector of the stress field cloud map, and the feature vector is input to the inverse coding to obtain the corresponding stress field cloud map. The stress field cloud map obtained by the neural network is compared with the stress field cloud map obtained by the traditional algorithm to calculate the similarity, and finally the accuracy of the neural network algorithm is obtained. If the test result does not meet the requirements, restart step S3 and repeat steps S31 to S33 until the algorithm that meets the requirements is obtained.

[0050] S4: setting a threshold for the start-stop switch of the stress field correction module, setting a start-stop limit for the structure stress field correction start according to the parameter form of the joint index and the simulation accuracy requirement;

[0051] The threshold setting should first be based on the specific form of the joint index. In this embodiment, the joint index of {frequency difference, MAC} is selected. Secondly, a reasonable threshold should be set while taking into account both calculation accuracy and correction efficiency. That is, it is necessary to improve the fidelity of finite element simulation and minimize the number of stress correction starts to ensure optimization efficiency. Finally, the threshold selection when creating the training set should be referenced to prevent the start and stop thresholds from exceeding the limit of the training set threshold.

[0052] Furthermore, the application process of this technology is illustrated by taking the high-fidelity simulation analysis of a single thin-walled component under progressive static loading as an example.

[0053] like Figure 4 As shown, the thin-walled steel component 3 is first installed on... Figure 3 The test bench is shown, and the finite element three-dimensional geometric model of the thin-walled steel component 3 is realized using the pre-built geometric information reconstruction module (module 1). The pre-loaded component is processed by the modal parameter identification module (module 2) to obtain the corresponding parameters, and then the initial stress field of the component is obtained by the modal-stress field analysis algorithm and applied to the finite element model. Then, material properties, boundaries and mesh generation are introduced to obtain an analysis model with the same initial stress field as the thin-walled steel structure. The load and load step are set according to the loading scheme of the analysis model and the finite element analysis is completed. Then, the component on the test bench is loaded in stages according to the loading scheme. After each loading is completed, joint index calculation and stress field correction judgment are performed. When there is a need for correction, the stress field correction module (module 3) first corrects the simulated stress field corresponding to the loading step. After the stress field is corrected, modal simulation is immediately performed, and the simulated mode is checked with the measured mode to verify the optimization of the simulation fidelity. After the test of this loading step is completed, the next loading step is executed. The modal determination, index calculation, start and stop determination and stress field correction steps are repeated until all loading schemes are completed, and the simulation of the entire static loading process of this thin-walled steel member 3 ends.

[0054] In this embodiment, firstly, based on the concept that digital twins should be driven by physical entity data sources, a basic module was built to acquire and analyze the geometric and modal information of the structural entity. This provides a channel for automatically incorporating measured physical space information into the simulation, improving the fidelity of the simulation model. Secondly, machine learning algorithms are used to establish the connection between modalities and stress fields, fitting the difficult-to-measure stress field distribution with easily obtainable structural modal parameters, thus improving the efficiency of simulation modeling. Furthermore, the application of integrated simulation technology emphasizes the real-time feedback of the physical structural state to the simulation process and actively regulates the simulation process based on the measured structural state, highlighting the interactivity of the virtual and physical twins.

[0055] The application is described in detail above with reference to the drawings. Those skilled in the art can make various changes to the application according to the above description. Therefore, some details in the embodiments should not be regarded as limiting the application, and the scope of protection of the application is defined by the appended claims.

Claims

1. A method for mechanical test and simulation of thin-walled steel structure based on digital twinning, comprising the steps of: S1: building a thin-walled steel structure entity geometry information reconstruction module, and using a point cloud data processing algorithm to realize automatic reconstruction of a thin-walled steel structure finite element three-dimensional geometric model; S2: building a thin-walled steel structure entity modal information identification module, and using an OMA method to analyze the excitation response of the thin-walled steel structure and then extract the real modal parameters of the thin-walled steel structure; S3: developing a modal stress field analysis algorithm, building a stress field correction module of the thin-walled steel structure based on the modal stress field analysis algorithm, and using a machine learning algorithm to fit the structure stress field from the input measured modal parameters; S4: setting a threshold value for the start-stop switch of the stress field correction module, setting a start-stop limit for the structure stress field correction based on the parameter form of the joint index and the simulation accuracy requirement; S5: coordinating the geometry information reconstruction module, the modal information identification module and the stress field correction module and the start-stop node function to realize high-fidelity finite element simulation of the thin-walled steel structure; wherein In the step S3, the development of the modal stress field analysis algorithm further comprises the steps of: S31: using the geometry information reconstruction module and the modal information identification module, relying on the assumed initial stress field, formulating a loading scheme to generate original samples, and selecting appropriate samples into the training set with reference to the joint index; S32: using a machine learning algorithm to build an algorithm framework that meets the functional requirements, and introducing a sufficient training set of samples into the algorithm framework to train the algorithm framework; S33: selecting a structure or component that has not been analyzed for simulation and modal measurement, and introducing the results in the form of a test set into the trained algorithm framework for testing.

2. The digital-twin-based thin-walled steel structure mechanical test and simulation method according to claim 1, characterized in that, In the step S1, the thin-walled steel structure is subjected to three-dimensional scanning, point cloud modeling and segmentation, automatic extraction of the neutral axis of the thin-walled component, and mapping and splicing; wherein the extraction of the neutral axis of the thin-walled component includes point cloud slicing, slice centroid extraction, centroid coordinate mapping and neutral axis fitting related procedures.

3. The digital-twin-based thin-walled steel structure mechanical test and simulation method according to claim 1, characterized in that, In the step S2, the excitation response of the thin-walled steel structure is analyzed, and for the non-time-varying signal under stationary excitation, the frequency, mode shape and damping ratio of the thin-walled steel structure are identified and calculated from the frequency response function using the frequency domain decomposition method; for the case where the response signal is time-varying, the time-frequency method is adopted to analyze the related parameters.

4. The digital-twin-based thin-walled steel structure mechanical test and simulation method according to claim 1, characterized in that, In the step S3, the training set creation method of the modal stress field analysis algorithm is: first, a large number of initial stress field samples are obtained by stress field scanning, technical regulations and residual stress research data, and the initial stress field samples are introduced into the finite element three-dimensional geometric model to generate a final analysis model; a large number of stress field and modal simulation samples are obtained by using the analysis model, and the simulation samples that meet the requirements are selected into the training set by using the joint index and threshold value; The above process is repeated, and the training set is continuously expanded by replacing the structure or component and the loading scheme.

5. The method according to claim 4, wherein, In the S5 step, the high-fidelity finite element simulation refers to creating a three-dimensional geometric model by using the geometric information reconstruction module, giving the initial stress field by using the modal information identification module and the modal stress field analysis algorithm, adjusting the simulation stress field that needs to be corrected by using the stress field correction module, and checking the improvement of simulation accuracy by using the corrected simulation modal.

Images