Curtain wall glass cold bending stress analysis method and device based on multi-fidelity agent model

Abstract

The invention provides a curtain wall glass cold bending stress analysis method and device based on a multi-fidelity agent model, and belongs to the field of building curtain wall glass cold bending stress analys.The method comprises the steps that distribution of cold bending stress values obtained through a finite element model serves as low-fidelity data; taking the cold bending stress value of the test point obtained by the cold bending experiment as high-fidelity data, and fitting a multi-fidelity agent model; and carrying out cold bending stress prediction on curtain wall glass by adopting the multi-fidelity agent model. According to the method, the precision and the stability of predicting the cold bending stress of the glass are improved, and a theoretical basis is provided for reducing the probability of defective products in cold bending processing of the glass and reducing the potential safety hazard of the building curtain wall glass.

Description

Technical Field

[0001] The invention belongs to the field of cold bending stress analysis of building curtain wall glass, and in particular relates to a curtain wall glass cold bending stress analysis method device based on a multi-fidelity proxy model. Background Art

[0002] Glass cold bending is the main processing method to achieve the curved surface effect of curtain wall. Cold bending uses the elasticity of glass itself and mechanical pressure to make flat glass twist and deform within the permitted range to achieve the designed shape. However, the cold bending process will produce permanent residual stress, which is large and has complex and changeable distribution patterns. Once it fails and falls from a high altitude, it will cause serious harm. Therefore, in order to control the cold bending stress not to exceed the allowable value, it is necessary to accurately predict the magnitude and distribution of cold bending stress. Domestic and foreign scholars have conducted a lot of research on cold bending processing, and proposed that glass cold bending is a nonlinear large deformation problem, which is not suitable for analysis using the thin plate small deflection theory; it is proposed that under large deformation conditions, the influence of ambient temperature on glass cold bending stress can be ignored. Existing studies have analyzed the stress magnitude and distribution during cold bending from different angles, analyzed the deformation of glass during cold bending, and laid the foundation for the study of curtain wall glass cold bending. However, the glass mechanical properties parameters in these studies generally adopt the recommended values ​​of national standards, and do not consider the influence of different processes of different manufacturers on glass mechanical parameters. In addition, these studies do not consider the differences between engineering practice and finite element simulation, so the prediction accuracy is limited. Therefore, the above research has certain limitations. Summary of the invention

[0003] In response to the shortcomings of the existing technology, this application proposes a curtain wall glass cold bending stress analysis method and device based on a multi-fidelity proxy model, which improves the accuracy and stability of glass cold bending stress prediction, and provides a theoretical basis for reducing the probability of defective products in glass cold bending processing and reducing the safety hazards of building curtain wall glass.

[0004] In the first aspect, the present application proposes a curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model, including:

[0005] Step S1: while keeping the production process unchanged, prepare a standard glass sample, test the mechanical property parameters of the standard glass sample, and record the input pulse signal and the response result of the standard glass sample to the pulse signal under different placement states of the standard glass sample;

[0006] Step S2: performing fast Fourier transform on the input pulse signal and the response result of the standard glass sample to the pulse signal to obtain the natural frequency of the standard glass sample;

[0007] Step S3: Calculate the elastic modulus and Poisson's ratio of the standard glass sample according to the natural frequency, size and mass of the standard glass sample;

[0008] Step S4: Repeat steps S1 to S3 multiple times to obtain the average value of the elastic modulus and the average value of the Poisson's ratio;

[0009] Step S5: establishing a finite element model in a top-down manner, using the average value of the elastic modulus and the average value of the Poisson's ratio as inputs to the finite element model, and obtaining the distribution of the cold bending stress value;

[0010] Step S6: performing a cold bending test on the standard glass sample to obtain the cold bending stress value of the test point;

[0011] Step S7: using the distribution of cold bending stress values ​​obtained by the finite element model as low-fidelity data, and using the cold bending stress values ​​of the test points obtained by the cold bending experiment as high-fidelity data, to fit a multi-fidelity proxy model;

[0012] Step S8: using the multi-fidelity proxy model to predict the cold bending stress of the curtain wall glass.

[0013] The natural frequencies of the standard glass sample include a bending fundamental frequency and a torsional fundamental frequency.

[0014] The elastic modulus and Poisson's ratio of the standard glass sample are calculated as follows:

[0015] ;

[0016] ;

[0017] ;

[0018] Among them, m is the specimen mass of the standard glass sample, L is the length of the standard glass sample, b is the width of the standard glass sample, t is the thickness of the standard glass sample, and f t is the fundamental frequency of torsional vibration of the standard glass sample, f f is the bending fundamental frequency of the standard glass sample, E is the elastic modulus of the standard glass sample, μ is the Poisson's ratio of the standard glass sample, and G is the shear modulus of the standard glass sample.

[0019] The finite element model is established in a top-down manner, including: realizing displacement constraint by limiting the coordinates of the rubber gasket in the finite element model, and applying displacement load by changing the coordinates of the rubber gasket.

[0020] The multi-fidelity proxy model is calculated as follows:

[0021] ;

[0022] in, is the predicted value of cold bending stress of the multi-fidelity proxy model, is the maximum likelihood estimation result of the multi-fidelity surrogate model, c is the correlation vector between the observed data and the predicted data, C is the covariance matrix of the multi-fidelity surrogate model, and y is the predicted value of the low-fidelity surrogate model. The covariance matrix of the multi-fidelity surrogate model is calculated as follows:

[0023] ;

[0024] in, is the correlation coefficient between low-fidelity data and low-fidelity data, is the correlation coefficient between high-fidelity data and low-fidelity data, is the correlation coefficient between high-fidelity data and high-fidelity data, is the correlation coefficient between low-fidelity data and high-fidelity data.

[0025] The predicted value of the low-fidelity proxy model is obtained by the response of the low-fidelity proxy model to the corresponding measurement point. The low-fidelity proxy model is calculated as follows:

[0026] ;

[0027] ;

[0028] Among them, 1 represents a vector with n rows and 1 column, all elements of which are 1. An estimate of the mean of the low-fidelity surrogate model, An estimate of the variance of the low-fidelity surrogate model, Number of low-fidelity samples, x l For low-fidelity sample input, Low-fidelity proxy model correlation coefficient vector, is the sample response value of the low-fidelity data.

[0029] In the second aspect, the present application proposes a curtain wall glass cold bending stress analysis device based on a multi-fidelity proxy model, comprising:

[0030] A data acquisition module is used to make a standard glass sample while keeping the production process unchanged, perform mechanical property parameter tests on the standard glass sample, and record the input pulse signal of the standard glass sample under different placement states and the response result of the standard glass sample to the pulse signal;

[0031] A frequency calculation module, used for performing fast Fourier transform on the input pulse signal and the response result of the standard glass sample to the pulse signal to obtain the natural frequency of the standard glass sample;

[0032] A modulus calculation module, used to calculate the elastic modulus and Poisson's ratio of the standard glass sample according to the natural frequency, size and mass of the standard glass sample;

[0033] An average value obtaining module is used to repeat the data acquisition module, the frequency calculation module and the modulus calculation module multiple times to obtain the average value of the elastic modulus and the average value of the Poisson's ratio;

[0034] The finite element construction module is used to establish the finite element model in a top-down manner, taking the average value of the elastic modulus and the average value of the Poisson's ratio as the input of the finite element model to obtain the distribution of the cold bending stress value;

[0035] Cold bending test module, used to perform cold bending test on standard glass samples to obtain the cold bending stress value of the test point;

[0036] A model fitting module is used to fit a multi-fidelity proxy model by taking the distribution of cold bending stress values ​​obtained by the finite element model as low-fidelity data and the cold bending stress values ​​of the test points obtained by the cold bending experiment as high-fidelity data;

[0037] The stress prediction module is used to predict the cold bending stress of the curtain wall glass using the multi-fidelity proxy model.

[0038] In a third aspect, the present application proposes an electronic device comprising: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model.

[0039] In a fourth aspect, the present application proposes a computer-readable storage medium storing executable instructions, which, when executed, enable a processor to execute the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model.

[0040] In a fifth aspect, the present application proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model.

[0041] Beneficial effects:

[0042] The present application proposes a curtain wall glass cold bending stress analysis method and device based on a multi-fidelity proxy model, which improves the prediction accuracy of the curtain wall glass cold bending stress and improves the prediction stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Flow chart of the cold bending stress analysis method of curtain wall glass based on multi-fidelity proxy model in the embodiment of the present application;

[0044] Figure 2 Schematic diagram of the bending fundamental frequency experiment principle of the embodiment of the present application;

[0045] Figure 3 Schematic diagram of the torsional fundamental frequency experiment principle of the embodiment of the present application;

[0046] Figure 4 Bending and torsion test experiments of the embodiments of the present application;

[0047] Figure 5 Schematic diagram of cold bending processing in an embodiment of the present application;

[0048] Figure 6 Stress cloud diagram of the finite element calculation results of the embodiment of the present application;

[0049] Figure 7 Displacement cloud diagram of the finite element calculation results of the embodiment of the present application;

[0050] Figure 8 Schematic diagram of grid-independence verification with a grid density of 20-27 in an embodiment of the present application;

[0051] Fig. 9 Schematic diagram of grid-independence verification with a grid density of 32-42 in an embodiment of the present application;

[0052] Fig.10 Schematic diagram of the strain measurement experiment scheme of the embodiment of the present application;

[0053] Fig.11 Schematic diagram of glass cold bending experiment in the embodiment of the present application;

[0054] Fig.12 Scanning results of displacement changes of the main diagonal line of the glass in the embodiment of the present application;

[0055] Fig.13 The comparison between the experimental results and the finite element simulation results of the embodiment of the present application is shown in the figure;

[0056] Fig.14 Schematic diagram of low-fidelity sample variable reconstruction in an embodiment of the present application;

[0057] Fig.15 3D diagram of the finite element model of the embodiment of the present application;

[0058] Fig.16 3D diagram of the experimental model of the embodiment of the present application;

[0059] Fig.17 A 3D image of a multi-fidelity model of an embodiment of the present application;

[0060] Fig.18 Schematic diagram of a multi-fidelity proxy model prediction method according to an embodiment of the present application;

[0061] Fig.19 Schematic diagram of the relative error of 1:21 in the embodiment of the present application;

[0062] Fig. 20 Schematic diagram of the relative error of 22:41 in the embodiment of the present application;

[0063] Fig.21 Schematic diagram of average evaluation index of all high-fidelity sample points in the embodiment of the present application;

[0064] Fig. 22 Schematic diagram of the limit evaluation index of all high-fidelity sample points in the embodiment of the present application;

[0065] Fig.23 The embodiment of the present application is a schematic diagram of relative root mean square error;

[0066] Fig.24 Schematic diagram of relative maximum error of an embodiment of the present application;

[0067] Fig.25 Schematic diagram of the coefficient of determination of an embodiment of the present application. DETAILED DESCRIPTION

[0068] The specific implementation of the present application is further described in detail below in conjunction with the drawings and examples.

[0069] In recent years, domestic and foreign scholars have enriched the research related to cold bending of glass in many aspects. Some scholars have studied the mechanical properties and calculation models of monolithic glass and laminated glass. Some scholars have studied the cold bending process of monolithic glass based on thin plate theory and conjugate beam method. Some scholars have used finite element method to study the residual stress generated in the glass manufacturing process and proposed the corresponding cold bending failure theory. Some scholars have used fiber optic sensors to analyze the cold bending stress of glass, effectively improving the experimental measurement accuracy. Zhao Xi'an of China Academy of Building Research studied the deformation capacity of glass cold bending and gave the design formula, which provided an important reference for the formulation of national standards for curtain wall glass. Gu Yanbin and others used the small deflection theory to study the cold bending deformation of laminated glass and proposed a solution for solving the large deflection deformation of cold-bent glass. Tong Liping's team at Zhengzhou University used the weighted residual value method to conduct nonlinear mechanical analysis of tempered glass based on the large deflection theory of elastic thin plates. Zhao Jianan of Tongji University analyzed the stress of insulating glass and gave suggestions on the domestic curtain wall glass design standards.

[0070] However, in these studies, scholars have ignored the impact of the glass manufacturing process on the mechanical performance parameters. Since the finite element simulation and experimental calculation use the same mechanical performance parameters, the impact caused by the difference between the actual value and the recommended value cannot be reflected, but this undoubtedly reduces the prediction accuracy. In addition, the finite element method can quickly analyze the cold bending stress of glass, but the finite element simulation is an approximate treatment of the actual process, which is different from the actual engineering and also affects the prediction accuracy; the experiment can truly simulate the actual engineering, but it is expensive and time-consuming. In order to inherit the advantages of these two schemes and make up for the shortcomings, the finite element data and experimental data are comprehensively considered, and it is believed that the experimental data is closer to the actual engineering than the finite element data.

[0071] Therefore, in order to improve the prediction accuracy of glass cold bending stress while saving costs, a multi-fidelity proxy model was used to integrate finite element data and experimental data to analyze the stress value and distribution law of curtain wall glass under cold bending state, and a curtain wall glass cold bending stress analysis method based on multi-fidelity proxy model was proposed.

[0072] Embodiment 1:

[0073] This embodiment proposes a curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model. Figure 1 As shown, including:

[0074] Step S1: while keeping the production process unchanged, prepare a standard glass sample, test the mechanical property parameters of the standard glass sample, and record the input pulse signal and the response result of the standard glass sample to the pulse signal under different placement states of the standard glass sample;

[0075] In this embodiment, the dynamic elastic properties of the elastic material sample are determined by the mechanical properties parameters, geometric shape and mass of the sample. Therefore, the mechanical properties parameters of the material can be determined by the geometric size, mass and mechanical resonance frequency of the sample. Figure 2 , Figure 3 As shown, Figure 2 This is a schematic diagram of the bending fundamental frequency experiment principle. Figure 3 Schematic diagram of the torsional fundamental frequency experiment principle.

[0076] exist Figure 4 The impulse force signal is collected by the LC-01A impact hammer and transmitted to the DH5956 data acquisition system for storage and processing; the acceleration signal is collected by the PCB352C04 acceleration sensor and transmitted to the DH5956 data acquisition system for storage and processing via the YE5852 charge amplifier; finally, the computer processes the signal through fast Fourier transform to obtain the bending fundamental frequency f of the material f and the torsional fundamental frequency f t .

[0077] Step S2: performing fast Fourier transform on the input pulse signal and the response result of the standard glass sample to the pulse signal to obtain the natural frequency of the standard glass sample;

[0078] Specifically, first, the elastic modulus of the material is obtained by a bending fundamental frequency experiment, as shown in the figure; the shear modulus of the material is obtained by a torsional fundamental frequency experiment. The experimental object is a glass specimen provided by the manufacturer. The average value is taken after multiple measurements to obtain the parameters of the glass specimen.

[0079] Step S3: Calculate the elastic modulus and Poisson's ratio of the standard glass sample according to the natural frequency, size and mass of the standard glass sample;

[0080] The elastic modulus and Poisson's ratio of glass materials are calculated as follows:

[0081] (1)

[0082] (2)

[0083] (3)

[0084] Among them, m is the specimen mass of the standard glass sample, L is the length of the standard glass sample, b is the width of the standard glass sample, t is the thickness of the standard glass sample, and f t is the fundamental frequency of torsional vibration of the standard glass sample, f f is the bending fundamental frequency of the standard glass sample, E is the elastic modulus of the standard glass sample, and μ is the Poisson's ratio of the standard glass sample.

[0085] After fast Fourier transform, the bending fundamental frequency f is obtained f =914.77HZ, torsion fundamental frequency f t =2585.74HZ; the calculated elastic modulus is 68.81MPa and Poisson's ratio is 0.257.

[0086] Step S4: Repeat steps S1 to S3 multiple times to obtain the average value of the elastic modulus and the average value of the Poisson's ratio;

[0087] In this embodiment, steps S1 to S3 are repeated multiple times, and the average value is taken from multiple experiments. The elastic modulus and Poisson's ratio are applied to finite element calculation and experimental calculation. In the experimental calculation, the elastic modulus and Poisson's ratio are used to convert the tested strain results into stress results, and the strain values ​​measured by the strain gauge sensor are calculated as stress values ​​and filled in Table 1. The calculation formula is as follows:

[0088] (4)

[0089] in, is the stress value, E is the elastic modulus of glass, is the Poisson's ratio of glass, ε0, ε 45 , ε 90 are the measured values ​​of the strain rosette in three directions.

[0090] Step S5: establishing a finite element model in a top-down manner, using the average value of the elastic modulus and the average value of the Poisson's ratio as inputs to the finite element model, and obtaining the distribution of the cold bending stress value;

[0091] In this embodiment, the cold bending process is simplified, and the schematic diagram is as follows: Figure 5 During the cold bending process, the BC side is constrained, the AB and CD1 sides are free, and the AD1 side is subjected to displacement load, so that the position of the glass corner point D1 reaches the design requirement D2, and the maximum cold bending displacement is dZ; this displacement is maintained for a period of time to allow the glass to be cold bent into shape.

[0092] This paper studies the cold bending process of semi-tempered glass by directly establishing a finite element model from the bottom up, and accurately controls the forces on nodes and units. During the cold bending process, there is no relative movement between the glass and the rubber gasket; considering the influence of different sizes on the model, the two units are coupled. In order to simplify the nonlinear problem of the supporting structure and improve the calculation accuracy, the displacement constraint is realized by limiting the coordinates of the rubber gasket in the finite element model, and the displacement load is applied by changing the coordinates of the rubber gasket.

[0093] The finite element calculation results are as follows Figure 6 , Figure 7 As shown, Figure 6 is the stress cloud diagram, Figure 7 is the displacement cloud map; the grid-independent verification is as follows Figure 8 , Fig. 9 As shown, Figure 8 The grid density is 20-27, Fig. 9 The mesh density is 32-42, and the case with a mesh density of 27×20 has a relative deviation of 5.813×10-4, and the case with a mesh density of 42×32 has a relative deviation of 0.0014. The calculation results of the SHELL181 unit have a relative deviation of 0.0071 from the calculation results of the SHELL63 unit; while the unit input parameter settings, Newton-Laplace algorithm settings, and solver parameter settings will not affect the results. These relative deviations are all less than 0.01, indicating the stability of the finite element simulation.

[0094] Step S6: performing a cold bending test on the standard glass sample to obtain the cold bending stress value of the test point;

[0095] Strain measurement experimental scheme is as follows Fig.10As shown, point D is the corner point of the cold-bent glass, and edge BC is the cold-bent fixed edge. Considering the high economic cost of the custom test bench, it is necessary to record as much stress data in the experiment as possible. The black dots in the figure represent the 41 strain gauge flower sticking positions. From the finite element simulation and previous experimental experience, it can be seen that the stress at the edge is greater than the stress in the middle, so the edge measurement points are arranged more densely than in the middle. This point arrangement method ensures that there are other measurement points around each measurement point, and the greater the stress value calculated by the finite element, the denser the measurement points are, and at the same time, there is no need to use a large number of strain gauges. The rectangle with a cross-hatching line in the upper left corner represents a basic grid unit. The shaded band on the diagonal BD represents the three-dimensional scanning area.

[0096] Test bench Fig.11 As shown, the AD beam is hinged at A, and the CD beam is hinged at C. By adjusting the support bolt at D, the AD beam rotates around A, and the gusset plate L1 applies a displacement load to the glass; at the same time, the BC beam does not move, and the gusset plate L2 applies a displacement constraint to the glass; the glass is cold-bent loaded.

[0097] Before the experiment, check whether the glass sample is damaged, measure the geometric dimensions, mark the measurement point position according to the experimental plan, and paste the strain gauge. The strain gauge and the wire are connected through the terminal block to ensure the stability of the connection. All strain gauges are connected to the red wire in the ε0 direction, the black wire in the ε45 direction, and the blue wire in the ε90 direction; connect to 3 strain gauges, such as Fig.11 As shown in the figure, the strain rosette attached to the glass sample plays a role in temperature compensation. Finally, an opaque tape is attached to the diagonal line where the cold bending point is located as the 3D scanning area.

[0098] The measurement results of the strain gauge in the cold bending test are shown in Table 1; the scanning results of the displacement change of the main diagonal line of the glass are shown in Fig.12 As shown in the figure, the experimental results are compared with the finite element simulation results. Fig.13 As shown, this shows that there is a certain deviation between the finite element results and the actual situation, and it is necessary to comprehensively consider the influence of the finite element results and experimental results.

[0099] Table 1 Experimental stress results

[0100] Measuring point number Stress value (MPa) Measuring point number Stress value (MPa) Measuring point number Stress value (MPa) 1 7.212 15 6.396 29 6.818 2 8.376 16 6.360 30 6.789 3 8.672 17 5.643 31 6.490 4 8.749 18 5.368 32 6.266 5 8.226 19 6.370 33 5.496 6 8.188 20 5.737 34 7.405 7 7.855 21 5.987 35 8.347 8 6.811 22 5.926 36 8.383 9 4.492 23 5.941 37 8.585 10 6.751 24 6.018 38 8.761 11 6.947 25 6.179 39 8.541 12 6.926 26 6.615 40 8.155 13 7.023 27 6.469 41 7.001 14 6.788 28 6.534

[0101] Step S7: using the distribution of cold bending stress values ​​obtained by the finite element model as low-fidelity data, and using the cold bending stress values ​​of the test points obtained by the cold bending experiment as high-fidelity data, to fit a multi-fidelity proxy model;

[0102] Step S8: using the multi-fidelity proxy model to predict the cold bending stress of the curtain wall glass.

[0103] The multi-fidelity proxy model comprehensively considers data from different sources and represents them as low-fidelity data and high-fidelity data according to the authenticity of the data. The input variables of the data are represented as:

[0104] (5)

[0105] Among them, x l is the low-fidelity sample input, x h is the high-fidelity sample input. X is the input of the multi-fidelity proxy model. The response variable of the multi-fidelity proxy model data is expressed as:

[0106] (6)

[0107] in, is the response corresponding to the low-fidelity sample, is the response corresponding to the high-fidelity sample. Y is the response variable of the data of the multi-fidelity surrogate model.

[0108] The low-fidelity data is adjusted by the proportional coefficient ρ using the comprehensive correction function method, and the Gaussian process d(x) (with parameters ,and decided by , Optimize and estimate ) as compensation and fit the multi-fidelity proxy model.

[0109] (7)

[0110] The correlation between response variables is expressed as:

[0111] (8)

[0112] (9)

[0113] (10)

[0114] in, is the correlation coefficient between low-fidelity data and low-fidelity data, is the correlation coefficient between high-fidelity data and low-fidelity data, is the correlation coefficient between high-fidelity data and high-fidelity data, is the correlation coefficient between low-fidelity data and high-fidelity data; Low-fidelity proxy model variance, Gaussian process variance, Low-fidelity proxy model correlation coefficient vector, Gaussian process correlation coefficient vector.

[0115] The covariance matrix of the multi-fidelity surrogate model is expressed as:

[0116] (11)

[0117] The maximum likelihood estimation result of the kriging surrogate model composed of low-fidelity data (referred to as the low-fidelity surrogate model) is expressed as:

[0118] (12)

[0119] (13)

[0120] Among them, 1 represents a vector with n rows and 1 column, all elements of which are 1. An estimate of the mean of the low-fidelity surrogate model, An estimate of the variance of the low-fidelity surrogate model, The number of low-fidelity samples will Substituting into equations (13) and (14) we can obtain the response of the low-fidelity proxy model to the corresponding measurement point, which is the predicted value of the low-fidelity proxy model.

[0121] According to the Markov property, the high-fidelity data is accurate and all errors come from the low-fidelity data. The parameter formula is expressed as:

[0122] (14)

[0123] The maximum likelihood estimation result is expressed as:

[0124] (15)

[0125] (16)

[0126] in, An estimate of the mean of a Gaussian process, An estimate of the variance of the Gaussian process, Number of high-fidelity samples.

[0127] The correlation vector c between the observed data and the predicted data is expressed as:

[0128] (17)

[0129] The prediction function of the multi-fidelity surrogate model is expressed as:

[0130] (18)

[0131] in, is the predicted value of cold bending stress of the multi-fidelity proxy model, is the maximum likelihood estimation result of the multi-fidelity surrogate model, c is the correlation vector between the observed data and the predicted data, C is the covariance matrix of the multi-fidelity surrogate model, The table shows the predicted values ​​of the low-fidelity proxy model.

[0132] In the multi-fidelity proxy model, high-fidelity sample points are obtained by experiment, and low-fidelity sample points are obtained by finite element. Considering the influence of randomness on the cold bending process, the distribution of glass parameters simulated by finite element is shown in Table 2. Considering that the thickness of glass is generally 10mm or 12mm, and the frequency of use of the two in engineering is similar, it is represented by a two-point distribution with p=0.5.

[0133] Table 2 Distribution of finite element simulation input parameters

[0134] Random Parameters Numeric distributed Coefficient of variation Elastic modulus / GPa 68.81 Lognormal distribution 0.173 Poisson's ratio 0.257 Lognormal distribution 0.141 Length / m 3.193 normal distribution 0.030 Width / m 1.110 normal distribution 0.020 Displacement / m 0.085 normal distribution 0.025

[0135] The above six random variables are sampled as input variables according to the distribution law, and the corresponding results are calculated by calling ANSYSbatch through MATLAB. The coordinate values ​​and the first principal stress values ​​are taken as outputs, and the stress distribution in the glass plane during the cold bending process is considered. In order to make different variables meet the fitting requirements of the multi-fidelity proxy model, the input and output of the finite element are separated, augmented and merged as low-fidelity samples of the multi-fidelity proxy model, such as Fig.14 As shown, taking a finite element calculation as an example (31 calculations in this example), there are 6 input variables in one calculation, represented as a 1×6 vector, and 3 output variables, represented as a 374×3 matrix. In order to conform to the learning format of the multi-fidelity agent model and provide the predicted value of the full glass plane; the vector of the input variable is expanded to a matrix of the same dimension, and after separation and merging, it is as follows Fig.14 shown.

[0136] Only finite element data is used to construct a single-fidelity finite element proxy model, and only experimental data is used to construct a single-fidelity experimental proxy model. A multi-fidelity proxy model is constructed by comprehensively considering low-fidelity samples and high-fidelity samples obtained from experiments. The model comparison of the three proxy models under the conditions of this cold bending experiment is shown in Figure 2. Fig.15 , Fig.16 , Fig.17 As shown, Fig.15 This is a 3D diagram of the finite element model. Fig.16 This is a 3D diagram of the experimental model. Fig.17 This is a 3D diagram of the multi-fidelity model. The black dots in the figure represent the experimental results. The experimental results of the finite element proxy model are outside the surface, indicating that the finite element proxy model is a biased estimate of the high-fidelity sample; the experimental results of the experimental proxy model and the multi-fidelity proxy model are on the surface, indicating that the experimental proxy model and the multi-fidelity proxy model are unbiased estimates of the high-fidelity sample.

[0137] The default prediction of the proxy model refers to building a proxy model using a sample set with some sample points omitted, and comparing the prediction results of the default sample points with the existing results. The default prediction method of the multi-fidelity proxy model taking the m high-fidelity sample point as an example is expressed as follows: remove the m high-fidelity sample and other random n-1 non-repeated high-fidelity samples from the high-fidelity samples, use Nn high-fidelity samples and all low-fidelity samples to fit the proxy model, and predict and evaluate the m high-fidelity sample point, such as Fig.18 The default possible combinations are expressed as the number of rows. , a matrix with n columns.

[0138] In order to comprehensively evaluate these three proxy models, the evaluation indicators considered include: root mean square error (MSE), determination coefficient ( ), sample standard deviation (STD), relative root mean square error (RRMSE), prediction residual sum of squares (PRESS), and relative maximum error (RMAE) are as follows. The larger the determination coefficient or the smaller the other evaluation indicators, the higher the prediction accuracy of the proxy model.

[0139] (19)

[0140] (20)

[0141] (twenty one)

[0142] (twenty two)

[0143] (twenty three)

[0144] (twenty four)

[0145] in, is the high-fidelity sample value, is the high-fidelity sample mean, is the predicted value of the high-fidelity proxy model, and N is the number of high-fidelity samples. The relative deviations of the three proxy models at the high-fidelity sample points when n=1 and the relative deviations of the direct comparison between the finite element and the experiment are shown in Fig.19 , Fig. 20 As shown, Fig.19 This is a diagram of the relative error of 1:21. Fig. 20 This shows that the finite element proxy model cannot improve the prediction effect, the experimental proxy model is unstable, and the multi-fidelity proxy model can significantly improve the prediction effect.

[0146] The statistical evaluation indicators of the three proxy models are as follows: Fig.21 , Fig. 22 As shown, Fig.21 is the average evaluation index of all high-fidelity sample points, Fig. 22 The figure shows the limit evaluation index of all high-fidelity sample points. A comprehensive comparison of relative deviation and various evaluation indicators shows that the prediction accuracy and stability of the multi-fidelity proxy model are better than those of the finite element proxy model and the experimental proxy model.

[0147] Consider three default cases of multi-fidelity proxy model n=10, 20, 30 respectively; considering the limitation of calculation conditions, 20 groups of non-repeated default combinations are randomly selected for each high-fidelity sample point, and the limit evaluation index of these 20 groups is used as the evaluation index of the high-fidelity sample point. Fig.23 , Fig.24 , Fig.25 As shown, Fig.23 is a schematic diagram of relative root mean square error, Fig.24 is a schematic diagram of relative maximum error, Fig.25 Figure 1 is a schematic diagram of the coefficient of determination. As the number of high-fidelity samples increases, the outliers of the evaluation index decrease, the median of the relative root mean square error and the relative maximum error decrease, the median of the coefficient of determination increases, and the distribution of sample points becomes more concentrated. Taking the coefficient of determination as an example, as the number of high-fidelity samples increases, the worst coefficient of determination increases from 0.855 to 0.988. This shows that as the number of high-fidelity samples increases, the prediction accuracy of the multi-fidelity proxy model increases, and the prediction stability improves.

[0148] This embodiment proposes a curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model. First, the dynamic elastic performance experiment of the glass is completed to obtain the elastic modulus and Poisson's ratio of the glass. Then, a finite element model is established to analyze the cold bending stress of the glass, and an experimental comparison is performed based on the finite element results. The finite element model data and the experimental model data are comprehensively considered to construct a multi-fidelity proxy model. Finally, the stability and accuracy of the multi-fidelity proxy model prediction are verified.

[0149] Embodiment 2:

[0150] This embodiment proposes a curtain wall glass cold bending stress analysis device based on a multi-fidelity proxy model, comprising: a data acquisition module, a frequency calculation module, a modulus calculation module, an average value obtaining module, a finite element construction module, a cold bending experiment module, a model fitting module and a stress prediction module, wherein the data acquisition module is connected to the frequency calculation module, the frequency calculation module is connected to the modulus calculation module, the modulus calculation module is connected to the average value obtaining module, the average value obtaining module is connected to the finite element construction module, the finite element construction module is connected to the cold bending experiment module, the cold bending experiment module and the finite element construction module are respectively connected to the model fitting module, and the model fitting module is connected to the stress prediction module;

[0151] A data acquisition module is used to make a standard glass sample while keeping the production process unchanged, perform mechanical property parameter tests on the standard glass sample, and record the input pulse signal of the standard glass sample under different placement states and the response result of the standard glass sample to the pulse signal;

[0152] A frequency calculation module, used for performing fast Fourier transform on the input pulse signal and the response result of the standard glass sample to the pulse signal to obtain the natural frequency of the standard glass sample;

[0153] A modulus calculation module, used to calculate the elastic modulus and Poisson's ratio of the standard glass sample according to the size and mass of the standard glass sample of the natural frequency of the standard glass sample;

[0154] An average value obtaining module is used to repeat the data acquisition module, the frequency calculation module and the modulus calculation module multiple times to obtain the average value of the elastic modulus and the average value of the Poisson's ratio;

[0155] The finite element construction module is used to establish the finite element model in a top-down manner, taking the average value of the elastic modulus and the average value of the Poisson's ratio as the input of the finite element model to obtain the distribution of the cold bending stress value;

[0156] Cold bending test module, used to perform cold bending test on standard glass samples to obtain the cold bending stress value of the test point;

[0157] A model fitting module is used to fit a multi-fidelity proxy model by taking the distribution of cold bending stress values ​​obtained by the finite element model as low-fidelity data and the cold bending stress values ​​of the test points obtained by the cold bending experiment as high-fidelity data;

[0158] The stress prediction module is used to predict the cold bending stress of the curtain wall glass using the multi-fidelity proxy model.

[0159] Embodiment 3:

[0160] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the curtain wall glass cold bending stress analysis method based on the multi-fidelity proxy model.

[0161] The electronic device may be a mobile phone, a computer or a tablet computer, etc., including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model as described in the embodiment is implemented. It is understood that the electronic device may also include an input / output (I / O) interface and a communication component.

[0162] The processor is used to execute all or part of the steps in the curtain wall glass cold bending stress analysis method based on the multi-fidelity proxy model as described in the above embodiment. The memory is used to store various types of data, which may include instructions of any application or method in the electronic device, as well as data related to the application.

[0163] The processor can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components, and is used to execute the curtain wall glass cold bending stress analysis method based on the multi-fidelity proxy model described in the above embodiment.

[0164] Embodiment 4:

[0165] This embodiment provides a computer-readable storage medium storing executable instructions. When the instructions are executed and implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0166] The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model described in each embodiment of the present application.

[0167] The aforementioned storage media include: flash memory, hard disk, multimedia card, card-type memory (for example, SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR abbreviation, memory data register) memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, CD, server, APP (Application, abbreviation of application software) application store and other media that can store program verification codes, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the above-mentioned curtain wall glass cold bending stress analysis method based on the multi-fidelity proxy model can be implemented.

[0168] Embodiment 5:

[0169] This embodiment provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model.

[0170] Based on such understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a computer program product.

[0171] The various embodiments in the present application are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0172] The protection scope of the present application is not limited to the above-mentioned embodiments. Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the scope and spirit of the present disclosure. If these changes and modifications fall within the scope of the claims of the present disclosure and their equivalents, the intention of the present disclosure also includes these changes and modifications.

Claims

1. A curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model, characterized in that: include: Step S1: while keeping the production process unchanged, prepare a standard glass sample, test the mechanical property parameters of the standard glass sample, and record the input pulse signal and the response result of the standard glass sample to the pulse signal under different placement states of the standard glass sample; Step S2: performing fast Fourier transform on the input pulse signal and the response result of the standard glass sample to the pulse signal to obtain the natural frequency of the standard glass sample; Step S3: Calculate the elastic modulus and Poisson's ratio of the standard glass sample according to the natural frequency, size and mass of the standard glass sample; Step S4: Repeat steps S1 to S3 multiple times to obtain the average value of the elastic modulus and the average value of the Poisson's ratio; Step S5: establishing a finite element model in a top-down manner, using the average value of the elastic modulus and the average value of the Poisson's ratio as inputs to the finite element model, and obtaining the distribution of the cold bending stress value; Step S6: performing a cold bending test on the standard glass sample to obtain the cold bending stress value of the test point; Step S7: using the distribution of cold bending stress values ​​obtained by the finite element model as low-fidelity data, and using the cold bending stress values ​​of the test points obtained by the cold bending experiment as high-fidelity data, to fit a multi-fidelity proxy model; Step S8: using the multi-fidelity proxy model to predict the cold bending stress of the curtain wall glass.

2. The curtain wall glass cold bending stress analysis method based on multi-fidelity proxy model according to claim 1 is characterized in that: The natural frequencies of the standard glass sample include a bending fundamental frequency and a torsional fundamental frequency.

3. The curtain wall glass cold bending stress analysis method based on multi-fidelity proxy model according to claim 1 is characterized in that: The elastic modulus and Poisson's ratio of the standard glass sample are calculated as follows: ； ； ； Among them, m is the specimen mass of the standard glass sample, L is the length of the standard glass sample, b is the width of the standard glass sample, t is the thickness of the standard glass sample, and f t is the fundamental frequency of torsional vibration of the standard glass sample, f f is the bending fundamental frequency of the standard glass sample, E is the elastic modulus of the standard glass sample, μ is the Poisson's ratio of the standard glass sample, and G is the shear modulus of the standard glass sample.

4. The curtain wall glass cold bending stress analysis method based on multi-fidelity proxy model according to claim 1 is characterized in that: The finite element model is established in a top-down manner, including: realizing displacement constraint by limiting the coordinates of the rubber gasket in the finite element model, and applying displacement load by changing the coordinates of the rubber gasket.

5. The curtain wall glass cold bending stress analysis method based on multi-fidelity proxy model according to claim 1 is characterized in that: The multi-fidelity proxy model is calculated as follows: ； in, is the predicted value of cold bending stress of the multi-fidelity proxy model, is the maximum likelihood estimation result of the multi-fidelity surrogate model, c is the correlation vector between the observed data and the predicted data, C is the covariance matrix of the multi-fidelity surrogate model, and y is the predicted value of the low-fidelity surrogate model. The covariance matrix of the multi-fidelity surrogate model is calculated as follows: ； in, is the correlation coefficient between low-fidelity data and low-fidelity data, is the correlation coefficient between high-fidelity data and low-fidelity data, is the correlation coefficient between high-fidelity data and high-fidelity data, is the correlation coefficient between low-fidelity data and high-fidelity data.

6. The curtain wall glass cold bending stress analysis method based on multi-fidelity proxy model according to claim 1 is characterized in that: The predicted value of the low-fidelity proxy model is obtained by the response of the low-fidelity proxy model to the corresponding measurement point. The low-fidelity proxy model is calculated as follows: ； ； Among them, 1 represents a vector with n rows and 1 column, all elements of which are 1. An estimate of the mean of the low-fidelity surrogate model, An estimate of the variance of the low-fidelity surrogate model, Number of low-fidelity samples, Enter values ​​for the low-fidelity samples, Low-fidelity proxy model correlation coefficient vector, is the sample response value of the low-fidelity data.

7. Curtain wall glass cold bending stress analysis device based on multi-fidelity proxy model, characterized in that: include: A data acquisition module is used to make a standard glass sample while keeping the production process unchanged, perform mechanical property parameter tests on the standard glass sample, and record the input pulse signal of the standard glass sample under different placement states and the response result of the standard glass sample to the pulse signal; A frequency calculation module, used for performing fast Fourier transform on the input pulse signal and the response result of the standard glass sample to the pulse signal to obtain the natural frequency of the standard glass sample; A modulus calculation module, used to calculate the elastic modulus and Poisson's ratio of the standard glass sample according to the natural frequency, size and mass of the standard glass sample; An average value obtaining module is used to repeat the data acquisition module, the frequency calculation module and the modulus calculation module multiple times to obtain the average value of the elastic modulus and the average value of the Poisson's ratio; The finite element construction module is used to establish the finite element model in a top-down manner, taking the average value of the elastic modulus and the average value of the Poisson's ratio as the input of the finite element model to obtain the distribution of the cold bending stress value; Cold bending test module, used to perform cold bending test on standard glass samples to obtain the cold bending stress value of the test point; A model fitting module is used to fit a multi-fidelity proxy model by taking the distribution of cold bending stress values ​​obtained by the finite element model as low-fidelity data and the cold bending stress values ​​of the test points obtained by the cold bending experiment as high-fidelity data; The stress prediction module is used to predict the cold bending stress of the curtain wall glass using the multi-fidelity proxy model.

8. An electronic device, characterized in that: include: One or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: It stores executable instructions, which, when executed, enable the processor to execute the curtain wall glass cold bending stress analysis method based on a multi-fidelity proxy model as described in any one of claims 1 to 6.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method for cold bending stress analysis of curtain wall glass based on a multi-fidelity proxy model as described in any one of claims 1 to 6 is implemented.

Images