A method, system, and apparatus for Bayesian model correction driven by a large model

Abstract

This invention provides a large-scale model-driven Bayesian model correction method, system, and apparatus, relating to the field of Bayesian model correction technology. The method includes: establishing a finite element model of the target building; performing sensitivity analysis on each layer group; inputting the target building's attribute information, evolution information, and environmental information into a large-scale damage prediction model, outputting the damage degree ratio of each layer group; then, combining the corresponding relationships, converting them into the benchmark mean of corrected material parameters, and constructing a prior distribution for Bayesian model correction; constructing a Bayesian posterior distribution; executing the MCMC sampling method; based on the statistical characteristics of the posterior distribution, extracting the optimal estimates of the corrected material parameters for each layer group; and performing dual validity verification on the correction model using the optimal estimates: if the verification passes, the optimal estimates are used as the final corrected material parameters to update the finite element model. This solution allows prior knowledge to better reflect the actual structural state, improving the physical rationality and accuracy of parameter correction.

Description

Technical Field

[0001] This invention relates to the field of Bayesian model correction technology, and in particular to a method, system and apparatus for Bayesian model correction driven by a large model. Background Technology

[0002] Current Bayesian model correction methods are mainly based on Bayes' theorem, treating the material parameters of the building structure (to be corrected) as random variables and incorporating engineering information through prior distributions. In traditional methods, the prior distributions of material parameters often rely on design specifications or empirical values, which are difficult to fully reflect the performance degradation characteristics of the structure due to environmental effects and load evolution during long-term service, resulting in large deviations in the correction results. Existing methods mostly only utilize structural dynamic response data (such as natural frequencies), and lack effective integration of multi-source data such as building attribute information, service evolution information, and environmental factors, failing to fully explore their indicative role in structural performance degradation. Existing methods usually do not consider the differentiated impact of damage levels in different layer groups on the structural dynamic characteristics, leading to unreasonable weighting of parameters when constructing the likelihood function, thereby reducing the correction accuracy.

[0003] Furthermore, patent application CN118862565A, entitled "A Method and System for Structural Damage Identification Based on Convolutional Neural Network Surrogate Model and Bayesian Model," discloses the following: creating prior, likelihood, and posterior distributions of damage parameters through Bayesian modeling; generating samples using an inverse sampling strategy, performing finite element analysis, and constructing a dataset; designing and training a CNN surrogate model, combining it with the Metropolis-Hasting algorithm to form the AdaMH algorithm model, using this model to extract posterior samples, and correcting biases through error augmentation. However, this approach does not correlate with the actual damage state of the structure, leading to a disconnect between prior information and engineering reality. Moreover, assigning the same penalty to the frequency deviation of layer groups with different damage levels fails to highlight the parameter sensitivity of high-damage regions.

[0004] Furthermore, patent application CN106897717A, entitled "A Method for Correcting Bayesian Models Based on Multiple Tests Using Environmental Excitation Data," discloses: analyzing structural acceleration data collected under environmental excitation multiple times to obtain the natural frequencies and mode shapes of the structure measured in each test, and calculating the uncertainties of these modal parameters, represented by a covariance matrix; based on the structural modal parameters and their covariance matrices obtained from multiple tests, constructing an objective function based on Bayesian theory, and obtaining the optimal values ​​of the model parameters of the finite element model that needs correction through optimization of the objective function. However, the prior distribution of this scheme is artificially assumed, and different prior distribution assumptions may lead to drastically different correction results, which can easily affect the reliability of the posterior results.

[0005] Furthermore, patent application CN114282398A, entitled "A Bridge Health Monitoring System and Method Based on Big Data," discloses the following: using hyperbolic and power function models to obtain characteristic parameters of the constitutive relationship of concrete; and using the big data analysis software WEKA to determine the massive amounts of bridge monitoring data. However, this scheme not only has high computational complexity but also relies entirely on data quality, resulting in poor adaptability.

[0006] Therefore, there is an urgent need for new Bayesian model correction methods to improve model accuracy and practicality. Summary of the Invention

[0007] The purpose of this invention is to provide a method, system, and apparatus for large model-driven Bayesian model correction, so as to solve at least one of the above-mentioned technical problems existing in the prior art.

[0008] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a large model-driven Bayesian model correction method, comprising the following steps:

[0009] Step 1: Collect the target building's attribute information, evolution information, and environmental information. Based on the attribute information, divide the target building into multiple independent floor groups according to a preset floor grouping method to ensure that the stress state and damage characteristics of each floor group are consistent, and avoid the impact of differences in characteristics within the floor group on the accuracy of subsequent parameter correction. The attribute information includes the construction year, number of floors, structural type, location of functional floors, and concrete strength grade of each floor. The evolution information includes the year of use, cumulative load change (in the past 3 years), and load change rate. The environmental information includes regional annual precipitation, average humidity, and seismic intensity.

[0010] In one feasible implementation, the floor grouping method includes:

[0011] Step 11: Divide the floors with the same concrete strength grade into a group to obtain the initial material layer group, thereby ensuring that the core material parameters of each layer group are consistent.

[0012] Step 12: The total number of layers can be optimally grouped. Based on the principle of divisibility, from the array Choose a value as Then perform integer division operations to obtain evenly divided groups;

[0013] Step 13: Adjust the uniform layer group based on the building height grouping requirements.

[0014] In one feasible implementation, the building height grouping requirements include:

[0015] Layers 1-3, each layer forming a separate group;

[0016] Floors 4-9 should be divided into at least 3 groups;

[0017] Floors 10-27 should be divided into at least 4 groups;

[0018] If there are more than 27 floors, they should be divided into at least 6 groups.

[0019] Step 2: Establish the initial finite element model of the target building. Based on the concrete strength grade, set the material parameters of the finite element model (or use a surrogate model with the same material parameters). Assign the material parameters to each layer group so that the main material parameters of each layer group can be adjusted independently and specifically. The material parameters include the elastic modulus. and density .

[0020] Step 3: (Using finite element analysis tools) Perform sensitivity analysis on the material parameters of each layer group to obtain the relative sensitivity coefficient of the natural frequency to the material parameters. Select the material parameters with a relative sensitivity coefficient greater than a preset threshold as the correction material parameters. .

[0021] In one feasible implementation, the specific formula for the sensitivity analysis includes:

[0022] ;

[0023] in, Indicates the first Natural frequencies calculated by finite element method For the Modified material parameters of the layer group The relative sensitivity coefficient is such that the larger the absolute value of the coefficient, the more significant the influence of the modified material parameter on the natural frequency of the finite element calculation.

[0024] Step 4: Input the target building's attribute information, evolution information, and environmental information into the damage prediction model, and output the damage degree ratio of each layer group; then combine it with the preset correspondence to convert it into the benchmark mean of the corrected material parameters, and set (reasonable) prior variance to construct the prior distribution of the Bayesian model correction.

[0025] In one feasible implementation, the specific method for constructing the large-scale damage prediction model includes:

[0026] Step a1: In the sample library, search for the attribute information, evolution information, and environmental information of multiple target structures that have the same structure as the target building, and construct a standardized feature vector;

[0027] Step a2: After standardizing the feature vectors and unifying the data format, use it as a dataset; calculate the structural aging index based on the design service life, years of use, and structural type; calculate the load variation coefficient based on the design load and the load variation of each floor group; and statistically analyze the damage degree ratio of each floor group (from historical inspection reports). (Values ​​range from 0 to 1);

[0028] For example, if the safety level of the first layer group is bu, then the damage ratio is between 0.26 and 0.5.

[0029] The specific formula for calculating the structural aging index includes:

[0030] Structural aging index = (Years of use / Design service life) × Structural form correction coefficient; whereby the specific value of the structural form correction coefficient can be:

[0031] Steel structure: 0.8 (strong corrosion resistance and aging resistance);

[0032] Brick-concrete structure: 1.2 (masonry materials age quickly);

[0033] Framework structure: 1.0 (stable performance).

[0034] The specific calculation formula for the load variation coefficient includes:

[0035] Load variation coefficient = cumulative load variation / design load;

[0036] The degree of damage can be calculated according to the relevant provisions of the "Standard for Reliability Appraisal of Civil Buildings" GB 50292-2015, or it can be viewed from existing test and appraisal reports;

[0037] Step a3 (Constructing an initial prediction network based on an open-source large model): The standardized feature vector obtained in step a1 is fused with the structural aging index and load change coefficient calculated in step a2 to construct a multi-source fusion feature vector as the model input. The damage degree ratio obtained in step a2 is used as the supervision label to perform supervised fine-tuning on the open-source large model, thereby obtaining a damage prediction large model adapted to the target building.

[0038] In one feasible implementation, the specific method for constructing the correspondence includes:

[0039] Step b1: Construct a formula for calculating the baseline mean of the corrected material parameters. The specific formula includes:

[0040] ;

[0041] ;

[0042] in, This represents the baseline mean of the elastic modulus; This represents the initial value of the elastic modulus; This represents the change in the elastic modulus; This represents the elastic modulus correction factor, in order to take into account the stress differences of the layers and optimize the spatial distribution of parameters; This represents the baseline mean density. Indicates the initial density value; Indicates the amount of density change; Indicates the density correction factor; This indicates an index of elastic modulus degradation. Indicates density degradation index;

[0043] Step b2: Take half of the damage level as... and The results showed the correlation between the proportion of damage and the degradation index;

[0044] Of course, in other implementation methods, the measured engineering values ​​can also be used as... and The result;

[0045] Step b3: Based on the position of the layer group, set and ;

[0046] For example, since the load on the bottom layer is large and the load on the top layer is small, then for the bottom layer, and The value range can be 0.95-0.98; for the middle layer, and It can take the value 1.0; for the top layer, and The value can range from 0.98 to 1.0 to optimize the parameter space distribution.

[0047] In one feasible implementation, the prior distribution modified by the Bayesian model is specifically expressed as follows:

[0048] ,Right now ;

[0049] ,Right now ;

[0050] ;

[0051] in, Represents a normal distribution; Indicates the first The predicted value of the current elastic modulus of the layer group; Indicates the first The predicted value of the current density of the layer group; Indicates the first The variance of the elastic modulus of the layer group; Indicates the first The variance of the density of the layer group; This represents the fundamental variance coefficient. For large-scale damage prediction models with specialized data, the larger the data volume and the smaller the coefficient value, the more reliable the prior information. For example:

[0052] If the data volume is ≥500 buildings, then =0.05;

[0053] If the data volume is between 300 and 499 buildings, then =0.08;

[0054] If the data volume is between 100 and 299 buildings, then =0.12;

[0055] If the data volume is less than 100 buildings, then =0.15;

[0056] In this way, the uncertainty of material parameters can be described and corrected using a normal distribution, and adaptive adjustments can also be made.

[0057] Step 5: Using the front of the target building Using the measured natural frequency as the correction target, a weighted Gaussian likelihood function is constructed. A weighting strategy based on the proportion of layer damage is introduced, and the prior distribution is fused with the weighted Gaussian likelihood function to construct a (complete) Bayesian posterior distribution.

[0058] In one feasible implementation, the specific formula for the Bayesian posterior distribution includes:

[0059] ;

[0060] in, This represents the Bayesian posterior distribution, which is the probability distribution of the corrected material parameters after the measured natural frequency correction. This represents the Gaussian likelihood function, i.e., given a set of modified material parameters. The degree of matching between the natural frequencies calculated by the finite element method and the measured natural frequencies is calculated. This represents the prior probability, used to initially understand the corrected material parameters before the measured natural frequency correction; Indicates the measured natural frequency The total probability of occurrence is a normalization constant.

[0061] In one feasible implementation, the specific expression of the weighted Gaussian likelihood function includes:

[0062] ;

[0063] in, Indicates the first The natural frequency calculated by the first-order finite element method; Indicates the first Measured natural frequency; Indicates the first The variance of the natural frequency calculated by the first-order finite element method; Indicates the first The damage weight based on the natural frequency of the first order is specifically expressed as follows:

[0064] ;

[0065] in, P j Indicates the first The damage degree ratio of the layer group (obtained from step 4) is as follows: Since the corrected material parameters of the high-damage layer group have a more significant impact on the dynamic characteristics, the heavier the penalty for frequency deviation, the larger the likelihood function value. This indicates that the natural frequency calculated by the finite element method corresponding to the current corrected material parameters is closer to the measured natural frequency, thus providing an optimization target for parameter correction.

[0066] In this way, the error term of each natural frequency is weighted and averaged with the weights of its sensitive layer group to obtain the equivalent weight. This equivalent weight is then substituted into the bias penalty term of the Gaussian likelihood function to achieve the matching between the single-order frequency error and the equivalent weight, thereby achieving the technical effect that the more significant the impact, the heavier the penalty.

[0067] Step 6: Based on the Bayesian posterior distribution, select and execute the corresponding MCMC sampling method according to the dimension of the corrected material parameters; synchronously iterate to verify the convergence of the sampling results; until the sampling results converge, obtain the posterior distribution of the corrected material parameters for each layer group.

[0068] In one feasible implementation, the specific selection method for the MCMC sampling method in step 6 includes:

[0069] When the dimension of the corrected material parameter is less than the first dimension threshold, the (conventional) Metropolis-Hastings algorithm (MH) is selected for efficient calculation.

[0070] When the dimension of the modified material parameter is greater than or equal to the first dimension threshold and less than the second dimension threshold, the (conventional) adaptive Metropolis algorithm (AM) is selected to dynamically adjust the search strategy;

[0071] When the dimension of the corrected material parameters is greater than or equal to the second dimension threshold, the (conventional) Hamiltonian Monte Carlo (HMC) or NUTS algorithm is selected to utilize gradient information to accelerate convergence.

[0072] In one feasible implementation, the specific selection method of the MCMC sampling algorithm further includes calculating the effective sample size (i.e., ESS, the lower the value, the worse the sampling efficiency) after the sampling reaches a preset minimum sample size; after selecting MH, if the effective sample size is lower than a preset proportion threshold (preferably 0.3) of the total sample size, then automatically switching to AM; after selecting HMC, if the effective sample size is lower than a preset proportion threshold (preferably 0.2) of the total sample size, then automatically switching to NUTS.

[0073] In one feasible implementation, the specific verification method for the convergence of the sampling results includes verification using the Gelman-Rubin statistic. When the statistics corresponding to all parameters to be corrected are less than a preset threshold (preferably 1.1), the sampling results are determined to have reached a convergence state.

[0074] Step 7: Extract the optimal estimated value of the modified material parameters for each layer group from the posterior distribution of the modified material parameters; perform double validity verification on the finite element model corrected using the optimal estimated value; if the verification passes, update the finite element model with the optimal estimated value as the final modified material parameters.

[0075] In one feasible implementation, the specific verification conditions for the dual validity verification include:

[0076] For the first three natural frequencies, the relative error between the natural frequencies calculated by the finite element method and the measured natural frequencies. ≤5%, to ensure that the dynamic characteristics of the corrected finite element model are consistent with reality;

[0077] The post-test material degradation rate falls within the degradation range corresponding to the damage level, so as to avoid correcting material parameters in violation of physical laws.

[0078] The specific formulas for calculating relative error and material degradation rate include:

[0079] ;

[0080] ;

[0081] ;

[0082] in, The optimal estimate of the corrected material parameters can be specifically expressed as: ;in, Indicates the first The optimal estimate of the elastic modulus; Indicates the first The optimal estimate of the density;

[0083] Indicates the first The material change in the elastic modulus of the layer group; Indicates the first The optimal estimate of the elastic modulus of the layer group; Indicates the first The material degradation rate of the elastic modulus of the layer group;

[0084] Indicates the first The material variation in the density of the layer group; Indicates the first The optimal estimate of the density of the layer group; Indicates the first Material degradation rate of layer density;

[0085] In one feasible implementation, the calibrated finite element model is applied to the structural dynamic response analysis or seismic performance assessment of the target building, and a building maintenance plan is formulated based on the damage level and deterioration rate of each floor group of the target building.

[0086] Secondly, based on the same inventive concept, this application also provides a large model-driven Bayesian model correction system, including a data acquisition module, a data processing module and a result generation module;

[0087] The data acquisition module is used to collect attribute information, evolution information, and environmental information of the target building. The attribute information includes the year of construction, number of floors, structural type, location of functional floors, and concrete strength grade of each floor. The evolution information includes the year of use, cumulative load change, and load change rate. The environmental information includes regional annual precipitation, average humidity, and seismic intensity.

[0088] The data processing module includes a grouping unit, a finite element unit, a sensitivity analysis unit, a priori unit, a correction unit, a posterior unit, and a calibration unit.

[0089] The grouping unit divides the target building into multiple independent layer groups based on attribute information and according to a preset floor grouping method;

[0090] The finite element units are used to establish the initial finite element model of the target building. Based on the concrete strength grade, the material parameters of the finite element model are set; these material parameters are assigned to each layer group; the material parameters include the elastic modulus. and density ;

[0091] The sensitivity analysis unit is used to perform sensitivity analysis on the material parameters of each layer group, obtain the relative sensitivity coefficient of the natural frequency to the material parameter, and select the material parameter with a relative sensitivity coefficient greater than a preset threshold as the correction material parameter. ;

[0092] The prior unit is used to input the attribute information, evolution information and environmental information of the target building into the damage prediction model, output the damage degree ratio of each layer group, and then combine it with the preset correspondence to convert it into the benchmark mean of the corrected material parameters, and set a reasonable prior variance to construct the prior distribution corrected by the Bayesian model.

[0093] The correction unit is used to correct the front of the target building. The measured intrinsic frequency is used as the correction target. A weighted strategy of the proportion of damage degree of fusion layer group is introduced to construct a weighted Gaussian likelihood function. The prior distribution is then fused with the weighted Gaussian likelihood function to construct a complete Bayesian posterior distribution.

[0094] The posterior unit is used to select and execute the corresponding MCMC sampling method based on the Bayesian posterior distribution and the dimension of the corrected material parameters; synchronously iterate to verify the convergence of the sampling results; and obtain the posterior distribution of the corrected material parameters for each layer group after the sampling results converge.

[0095] The calibration unit extracts the optimal estimated value of the corrected material parameters for each layer group from the posterior distribution of the corrected material parameters; performs dual validity verification on the finite element model corrected using the optimal estimated value; if the verification passes, the optimal estimated value is used as the final corrected material parameters to update the finite element model.

[0096] The result generation module is used to send the finite element model out.

[0097] In one feasible implementation, the large model-driven Bayesian model correction system further includes an application module for applying the finite element model to the structural dynamic response analysis or seismic performance assessment of the target building, and for developing a building maintenance plan based on the damage level and deterioration rate of each floor group of the target building.

[0098] Thirdly, based on the same inventive concept, this application also provides a large model-driven Bayesian model correction device, including a processor, a memory, and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to execute the Bayesian model correction method as described above. The bus connects the various functional components for transmitting information.

[0099] By adopting the above technical solution, the present invention has the following beneficial effects:

[0100] This invention provides a large-model-driven Bayesian model correction method, system, and device. It divides the building into independent layers according to building type and floor height, and configures independent elastic modulus and density parameters for each layer. This enables the structural model to be characterized in a layered and refined manner, allowing the parameters to accurately match the actual deterioration state of the corresponding layer. It avoids the parameter deviation of local damaged layers being masked by overall modeling, thereby significantly improving the matching degree between the corrected layer parameters and the actual damage distribution, laying the foundation for subsequent accurate correction of high-damage layers.

[0101] This solution uses a large-scale damage prediction model to transform the full life-cycle information of the target building structure into the proportion of damage levels in each layer group. Then, it combines this with a pre-defined correspondence to generate the initial degradation law of the corrected material parameters. This allows the prior distribution to be directly related to the physical characteristics of structural damage (e.g., the moderately damaged layer group corresponds to a clear elastic modulus degradation range), thereby avoiding the subjectivity and blindness of prior information and significantly improving the reliability of the correction starting point. This solution breaks through the limitations of static priors in traditional methods, making prior knowledge more consistent with the actual structural state and improving the physical rationality and accuracy of parameter correction.

[0102] This scheme employs a differentiated design of damage level weights (with lower weights for high-damage layers and heavier penalties for frequency deviations) to prioritize the constraint of high-damage layer parameters by the likelihood function. This ensures the accuracy of corrections in critical damage areas and prevents the parameter errors of high-damage layers from being masked by deviations in ordinary layers. The relative frequency error can be controlled within a preset threshold. After correction, the consistency between the finite element model and the actual structural dynamic characteristics is significantly improved, allowing Bayesian correction to better align with the physical laws of structural dynamic characteristics. This solves the problem of insufficient correction accuracy caused by the traditional likelihood function's equal penalty for frequency deviations in all layers. Attached Figure Description

[0103] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0104] Figure 1 A flowchart of a large model-driven Bayesian model correction method provided in an embodiment of the present invention;

[0105] Figure 2 A flowchart of the floor grouping method provided in an embodiment of the present invention;

[0106] Figure 3 A flowchart illustrating the specific construction method of the large-scale damage prediction model provided in this embodiment of the invention;

[0107] Figure 4 A diagram of a large model-driven Bayesian model correction system provided for an embodiment of the present invention. Detailed Implementation

[0108] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0109] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0110] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0111] The present invention will be further explained below with reference to specific embodiments.

[0112] It should also be noted that the specific embodiments or implementation methods described below are a series of optimized settings listed by the present invention to further explain the specific content of the invention, and these settings can be combined or used in conjunction with each other.

[0113] Example 1:

[0114] like Figure 1 As shown in the figure, this embodiment provides a large model-driven Bayesian model correction method, which includes the following steps:

[0115] Step 1: Collect the target building's attribute information, evolution information, and environmental information. Based on the attribute information, divide the target building into multiple independent floor groups according to a preset floor grouping method to ensure that the stress state and damage characteristics of each floor group are consistent, and avoid the impact of differences in characteristics within the floor group on the accuracy of subsequent parameter correction. The attribute information includes the construction year, number of floors, structural type (e.g., frame, shear wall, frame-shear wall, etc.), location of functional floors (e.g., transfer floor, equipment floor, etc.), and concrete strength grade of each floor (e.g., C30, C40, etc.). The evolution information includes the year of use, cumulative load change (in the past 3 years), and load change rate. The environmental information includes regional annual precipitation, average humidity, and seismic intensity.

[0116] Furthermore, such as Figure 2 As shown, the floor grouping method includes:

[0117] Step 11: Divide the floors with the same concrete strength grade into a group to obtain the initial material layer group, thereby ensuring that the core material parameters of each layer group are consistent.

[0118] Step 12: Based on the total number of floors Can be optimally grouped Based on the principle of divisibility, from Select Then perform integer division operations to obtain evenly divided groups;

[0119] Step 13: Adjust the uniform layer group based on the building height grouping requirements.

[0120] Furthermore, the building height grouping requirements include:

[0121] Layers 1-3, each layer forming a separate group;

[0122] Floors 4-9 should be divided into at least 3 groups;

[0123] Floors 10-27 should be divided into at least 4 groups;

[0124] If there are more than 27 floors, they should be divided into at least 6 groups.

[0125] Step 2: Establish the initial finite element model of the target building. Based on the concrete strength grade, set the material parameters of the finite element model (or use a surrogate model with the same material parameters). Assign the material parameters to each layer group so that the main material parameters of each layer group can be adjusted independently and specifically. The material parameters include the elastic modulus. and density .

[0126] Step 3: (Using finite element analysis tools) Perform sensitivity analysis on the material parameters of each layer group to obtain the relative sensitivity coefficient of the natural frequency to the material parameters. Select the material parameters with a relative sensitivity coefficient greater than a preset threshold as the correction material parameters. .

[0127] Furthermore, the specific formula for the sensitivity analysis includes:

[0128] ;

[0129] in, Indicates the first Natural frequencies calculated by finite element method For the Modified material parameters of the layer group The relative sensitivity coefficient is such that the larger the absolute value of the coefficient, the more significant the influence of the modified material parameter on the natural frequency of the finite element calculation.

[0130] Step 4: Input the target building's attribute information, evolution information, and environmental information (such as construction year, usage function, load conditions, environmental conditions, etc.) into the damage prediction model, and output the damage degree ratio of each layer group; then combine it with the preset correspondence to convert it into the benchmark mean of the corrected material parameters, and set a reasonable prior variance to construct the prior distribution of the Bayesian model correction.

[0131] Furthermore, such as Figure 3 As shown, the specific construction method of the large-scale damage prediction model includes:

[0132] Step a1: In the sample library, search for the attribute information, evolution information, and environmental information of multiple target structures that have the same structure as the target building, and construct a standardized feature vector;

[0133] Step a2: After standardizing the feature vectors and unifying the data format, use it as a dataset; calculate the structural aging index based on the design service life, years of use, and structural type; calculate the load variation coefficient based on the design load and the load variation of each floor group; and statistically analyze the damage degree ratio of each floor group (from historical inspection reports). (Values ​​range from 0 to 1);

[0134] For example, if the safety level of the first layer group is bu, then the damage ratio is between 0.26 and 0.5.

[0135] The specific formula for calculating the structural aging index includes:

[0136] Structural aging index = (Years of use / Design service life) × Structural form correction coefficient; whereby the specific value of the structural form correction coefficient can be:

[0137] Steel structure: 0.8 (strong corrosion resistance and aging resistance);

[0138] Brick-concrete structure: 1.2 (masonry materials age quickly);

[0139] Framework structure: 1.0 (stable performance).

[0140] The specific calculation formula for the load variation coefficient includes:

[0141] Load variation coefficient = cumulative load variation / design load;

[0142] The degree of damage can be calculated according to the relevant provisions of the "Standard for Reliability Appraisal of Civil Buildings" GB 50292-2015, or it can be viewed from existing test and appraisal reports;

[0143] Step a3: The standardized feature vector obtained in step a1 is fused with the structural aging index and load change coefficient calculated in step a2 to construct a multi-source fusion feature vector as the model input. The damage degree ratio obtained in step a2 is used as the supervision label to perform supervised fine-tuning on the open source large model to obtain a damage prediction large model adapted to the target building.

[0144] Furthermore, the specific method for constructing the correspondence includes:

[0145] Step b1: Construct a formula for calculating the baseline mean of the corrected material parameters. The specific formula includes:

[0146] ;

[0147] ;

[0148] in, This represents the baseline mean of the elastic modulus; This represents the initial value of the elastic modulus; This represents the change in the elastic modulus; This represents the elastic modulus correction factor (e.g., 0.95-0.98 for the bottom layer, 1.0 for the middle layer, and 0.98-1.0 for the top layer) to take into account the stress differences between the layers (i.e., the bottom layer has a larger load and the top layer has a smaller load) and optimize the spatial distribution of parameters. This represents the baseline mean density. Indicates the initial density value; Indicates the amount of density change; Indicates the density correction factor; This indicates an index of elastic modulus degradation. Indicates density degradation index;

[0149] Step b2: Take half of the damage level as... and The results showed the correlation between the damage degree ratio and the material degradation index;

[0150] Step b3: Based on the position of the layer group, set and ;

[0151] For example, since the load on the bottom layer is large and the load on the top layer is small, then for the bottom layer, and The value range can be 0.95-0.98; for the middle layer, and It can take the value 1.0; for the top layer, and The value can range from 0.98 to 1.0 to optimize the parameter space distribution.

[0152] Furthermore, the prior distribution modified by the Bayesian model is specifically expressed as follows:

[0153] ,Right now ;

[0154] ,Right now ;

[0155] ;

[0156] in, Represents a normal distribution; Indicates the first The predicted value of the current elastic modulus of the layer group; Indicates the first The predicted value of the current density of the layer group; Indicates the first The variance of the elastic modulus of the layer group; Indicates the first The variance of the density of the layer group; This represents the fundamental variance coefficient. For large-scale damage prediction models with specialized data, the larger the data volume and the smaller the coefficient value, the more reliable the prior information. For example:

[0157] If the data volume is ≥500 buildings, then =0.05;

[0158] If the data volume is between 300 and 499 buildings, then =0.08;

[0159] If the data volume is between 100 and 299 buildings, then =0.12;

[0160] If the data volume is less than 100 buildings, then =0.15;

[0161] In this way, the uncertainty of material parameters can be described and corrected using a normal distribution, and adaptive adjustments can also be made.

[0162] Step 5: Using the front of the target building Using the measured natural frequency as the correction target, a weighted Gaussian likelihood function is constructed. A weighting strategy based on the proportion of layer damage is introduced, and the prior distribution is fused with the weighted Gaussian likelihood function to construct a complete Bayesian posterior distribution.

[0163] Furthermore, the specific formula for the Bayesian posterior distribution includes:

[0164] ;

[0165] in, This represents the Bayesian posterior distribution, which is the probability distribution of the corrected material parameters after the measured natural frequency correction. This represents the Gaussian likelihood function, i.e., given a set of modified material parameters. The degree of matching between the natural frequencies calculated by the finite element method and the measured natural frequencies is calculated. This represents the prior probability, used to initially understand the corrected material parameters before the measured natural frequency correction; Indicates the measured natural frequency The total probability of occurrence is a normalization constant.

[0166] Furthermore, the specific expression of the weighted Gaussian likelihood function includes:

[0167] ;

[0168] in, Indicates the first The natural frequency calculated by the first-order finite element method; Indicates the first Measured natural frequency; Indicates the first The variance of the natural frequency calculated by the first-order finite element method; Indicates the first The damage weight based on the natural frequency of the first order is specifically expressed as follows:

[0169] ;

[0170] in, P j Indicates the first The degree of damage to the layers (obtained from step 4);

[0171] For example: in cases of severe injury The value is 0.8; for moderate damage, The value is 1; for minor damage, The value is 1.2; since the corrected material parameters of the high-damage layer group have a more significant impact on the dynamic characteristics, the heavier the penalty for frequency deviation, the larger the likelihood function value, which indicates that the natural frequency calculated by the finite element method corresponding to the current corrected material parameters is closer to the measured natural frequency, thus providing an optimization target for parameter correction.

[0172] Will Layers with a value ≥0.1 are designated as sensitive layer groups (i.e., a 10% change in the corrected material parameters of this layer group will result in a ≥1% change in the natural frequency), and damage weights are assigned only to sensitive layer groups to participate in the error calculation of the natural frequency of that order.

[0173] In this way, the error term of each natural frequency is weighted and averaged with the weights of its sensitive layer group to obtain the equivalent weight. This equivalent weight is then substituted into the bias penalty term of the likelihood function to achieve the matching between the single-order frequency error and the equivalent weight, thereby achieving the technical effect that the more significant the impact, the heavier the penalty.

[0174] Step 6: Based on the Bayesian posterior distribution, select and execute the corresponding MCMC sampling method according to the dimension of the corrected material parameters (which can be determined by the number of layers, with each layer containing 2 corrected material parameters); synchronously iterate to verify the convergence of the sampling results; until the sampling results converge, obtain the posterior distribution of the corrected material parameters for each layer.

[0175] Furthermore, the specific selection method for the MCMC sampling method in step 6 includes:

[0176] When the dimension of the corrected material parameter is less than the first dimension threshold, the Metropolis-Hastings algorithm (MH) is selected for efficient calculation.

[0177] When the dimension of the modified material parameter is greater than or equal to the first dimension threshold and less than the second dimension threshold, the adaptive Metropolis algorithm (AM) is selected to dynamically adjust the search strategy.

[0178] When the dimension of the modified material parameters is greater than or equal to the second dimension threshold, the Hamiltonian Monte Carlo (HMC) or NUTS algorithm is selected to utilize gradient information to accelerate convergence.

[0179] In one feasible implementation, the specific selection method of the MCMC sampling algorithm further includes calculating the effective sample size (i.e., ESS, the lower the value, the worse the sampling efficiency) after the sampling reaches a preset minimum sample size (e.g., 1000); after selecting MH, if the effective sample size is lower than a preset proportion threshold (preferably 0.3) of the total sample size, then automatically switching to AM; after selecting HMC, if the effective sample size is lower than a preset proportion threshold (preferably 0.2) of the total sample size, then automatically switching to NUTS.

[0180] In one feasible implementation, the specific verification method for the convergence of the sampling results includes verification using the Gelman-Rubin statistic. When the statistics corresponding to all parameters to be corrected are less than a preset threshold (preferably 1.1), the sampling results are determined to have reached a convergence state.

[0181] Step 7: Extract the optimal estimated value of the modified material parameters for each layer group from the posterior distribution of the modified material parameters; perform double validity verification on the finite element model corrected using the optimal estimated value; if the verification passes, update the finite element model (or surrogate model) with the optimal estimated value as the final modified material parameters, thereby completing the finite element model calibration.

[0182] Furthermore, the specific calculation formula for the dual validity verification includes:

[0183] For the first three natural frequencies, the relative error between the natural frequencies calculated by the finite element method and the measured natural frequencies. ≤5%, to ensure that the dynamic characteristics of the corrected finite element model are consistent with reality;

[0184] The post-test material degradation rate falls within the degradation range corresponding to the damage level, so as to avoid correcting material parameters in violation of physical laws.

[0185] The specific formulas for calculating the relative error and material degradation rate include:

[0186] ;

[0187] ;

[0188] ;

[0189] in, The optimal estimate of the corrected material parameters can be specifically expressed as: ;in, Indicates the first The optimal estimate of the elastic modulus; Indicates the first The optimal estimate of the density;

[0190] Indicates the first The material change in the elastic modulus of the layer group; Indicates the first The optimal estimate of the elastic modulus of the layer group; Indicates the first The material degradation rate of the elastic modulus of the layer group;

[0191] Indicates the first The material variation in the density of the layer group; Indicates the first The optimal estimate of the density of the layer group; Indicates the first Material degradation rate of layer density;

[0192] Furthermore, the calibrated finite element model is applied to the structural dynamic response analysis or seismic performance assessment of the target building, and a building maintenance plan is formulated based on the damage level and deterioration rate of each floor group of the target building.

[0193] Example 2:

[0194] like Figure 4 As shown, this embodiment provides a large model-driven Bayesian model correction system, including a data acquisition module, a data processing module, and a result generation module;

[0195] The data acquisition module is used to collect attribute information, evolution information, and environmental information of the target building. The attribute information includes the year of construction, number of floors, structural type, location of functional floors, and concrete strength grade of each floor. The evolution information includes the year of use, cumulative load change, and load change rate. The environmental information includes regional annual precipitation, average humidity, and seismic intensity.

[0196] The data processing module includes a grouping unit, a finite element unit, a sensitivity analysis unit, a priori unit, a correction unit, a posterior unit, and a calibration unit.

[0197] The grouping unit divides the target building into multiple independent layer groups based on attribute information and according to a preset floor grouping method;

[0198] The finite element units are used to establish the initial finite element model of the target building. Based on the concrete strength grade, the material parameters of the finite element model are set; these material parameters are assigned to each layer group; the material parameters include the elastic modulus. and density ;

[0199] The sensitivity analysis unit is used to perform sensitivity analysis on the material parameters of each layer group, obtain the relative sensitivity coefficient of the natural frequency to the material parameter, and select the material parameter with a relative sensitivity coefficient greater than a preset threshold as the correction material parameter. ;

[0200] The prior unit is used to input the attribute information, evolution information and environmental information of the target building into the damage prediction model, output the damage degree ratio of each layer group, and then combine it with the preset correspondence to convert it into the benchmark mean of the corrected material parameters, and set a reasonable prior variance to construct the prior distribution corrected by the Bayesian model.

[0201] The correction unit is used to correct the front of the target building. Using the measured intrinsic frequency as the correction target, a weighted strategy of the proportion of damage degree of fusion layer group is introduced to construct a weighted Gaussian likelihood function, and the prior distribution is fused with the weighted Gaussian likelihood function to construct a complete Bayesian posterior distribution;

[0202] The posterior unit is used to select and execute the corresponding MCMC sampling method based on the Bayesian posterior distribution and the dimension of the corrected material parameters; synchronously iterate to verify the convergence of the sampling results; and obtain the posterior distribution of the corrected material parameters for each layer group after the sampling results converge.

[0203] The calibration unit extracts the optimal estimated value of the corrected material parameters for each layer group from the posterior distribution of the corrected material parameters; performs dual validity verification on the finite element model corrected using the optimal estimated value; if the verification passes, the optimal estimated value is used as the final corrected material parameters to update the finite element model.

[0204] The result generation module is used to send the finite element model out.

[0205] Furthermore, the large model-driven Bayesian model correction system also includes an application module for applying the finite element model to the structural dynamic response analysis or seismic performance assessment of the target building, and for developing a building maintenance plan based on the damage level and deterioration rate of each floor group of the target building.

[0206] Example 3:

[0207] This embodiment provides a large model-driven Bayesian model correction device, including a processor, a memory, and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to execute the Bayesian model correction method as described above. The bus connects the various functional components for information transmission.

[0208] In another embodiment, this solution can also be implemented using an integrated device, which may include corresponding modules that perform one or more steps in the various embodiments described above. A module may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented through some combination thereof.

[0209] The processor executes the various methods and processes described above. For example, the method implementations in this scheme can be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some implementations, part or all of the software program can be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above can be performed. Alternatively, in other implementations, the processor can be configured to execute one of the methods described above by any other suitable means (e.g., by means of firmware).

[0210] This device can be implemented using a bus architecture. A bus architecture can include any number of interconnect buses and bridges, depending on the specific application of the hardware and overall design constraints. The bus connects various circuits, including one or more processors, memory, and / or hardware modules. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuitry, external antennas, etc.

[0211] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Component (EISA) buses, etc. Buses can be divided into address buses, data buses, control buses, etc.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A large model driven Bayesian model revision method, characterized in that, include: Step 1: Collect attribute information, evolution information, and environmental information of the target building; Based on attribute information, the target building is divided into multiple independent floor groups according to a preset floor grouping method; Step 2, establish the initial finite element model of the target building, set the material parameters of the finite element model according to the concrete strength grade; assign the material parameters to each layer group; the material parameters include elastic modulus and density ; Step 3: Perform sensitivity analysis on the material parameters of each layer group to obtain the relative sensitivity coefficient of the natural frequency to the material parameters. Select the material parameters with a relative sensitivity coefficient greater than a preset threshold as the correction material parameters. ; Step 4: Input the target building's attribute information, evolution information, and environmental information into the damage prediction model, and output the damage degree ratio of each layer group; then combine it with the preset correspondence to convert it into the benchmark mean of the corrected material parameters, and set the prior variance to construct the prior distribution of the Bayesian model correction. The specific methods for constructing the correspondence include: Step b1: Construct a formula for calculating the baseline mean of the corrected material parameters. The specific formula includes: ； ； in, This represents the baseline mean of the elastic modulus; This represents the initial value of the elastic modulus; This represents the change in the elastic modulus; This represents the elastic modulus correction factor; This represents the baseline mean density. Indicates the initial density value; Indicates the amount of density change; Indicates the density correction factor; This indicates an index of elastic modulus degradation. Indicates density degradation index; Step b2: Take half of the damage level as... and The results showed the correlation between the damage degree ratio and the material degradation index; Step b3: Based on the position of the layer group, set and ; Step 5: Using the front of the target building Using the measured natural frequency as the correction target, a weighted Gaussian likelihood function is constructed. A weighting strategy based on the proportion of layer damage is introduced, and the prior distribution is fused with the weighted Gaussian likelihood function to construct a Bayesian posterior distribution. Step 6: Based on the Bayesian posterior distribution, select and execute the corresponding MCMC sampling method according to the dimension of the corrected material parameters; synchronously iterate to verify the convergence of the sampling results; until the sampling results converge, obtain the posterior distribution of the corrected material parameters for each layer group; Step 7: Extract the optimal estimated value of the modified material parameters for each layer group from the posterior distribution of the modified material parameters; perform double validity verification on the finite element model corrected using the optimal estimated value; if the verification passes, update the finite element model with the optimal estimated value as the final modified material parameters.

2. The Bayesian model correction method according to claim 1, characterized in that, The floor grouping method includes: Step 11: Divide the floors with the same concrete strength grade into a group to obtain the initial material layer group; Step 12: Based on the total number of floors Can be optimally grouped Based on the principle of divisibility, select a value from 2, 3, and 4 as the divisibility factor. Then perform integer division operations to obtain evenly divided groups; Step 13: Adjust the uniform layer group based on the building height grouping requirements.

3. The Bayesian model correction method according to claim 2, characterized in that, The building height grouping requirements include: Layers 1-3, each layer forming a separate group; Floors 4-9 should be divided into at least 3 groups; Floors 10-27 should be divided into at least 4 groups; If there are more than 27 floors, they should be divided into at least 6 groups.

4. The Bayesian model correction method according to claim 1, characterized in that, The attribute information includes the year of construction, number of floors, structural type, location of functional floors, and concrete strength grade of each floor; the evolution information includes the year of use, cumulative load change, and load change rate; the environmental information includes regional annual precipitation, average humidity, and seismic intensity.

5. The Bayesian model correction method according to claim 1, characterized in that, The specific construction method of the large-scale damage prediction model includes: Step a1: In the sample library, search for the attribute information, evolution information, and environmental information of multiple target structures that have the same structure as the target building, and construct a standardized feature vector; Step a2: After unifying the data format of the standardized feature vectors, use them as a dataset; calculate the structural aging index based on the design service life, years of use, and structural type; calculate the load variation coefficient based on the design load and the load variation of each floor group; and statistically analyze the damage degree ratio of each floor group. ; Step a3: Combine the standardized feature vector, structural aging index and load change coefficient to construct a multi-source fusion feature vector as the model input, use the damage degree ratio as the supervision label, perform supervised fine-tuning on the open source large model, and obtain a damage prediction large model adapted to the target building.

6. The Bayesian model correction method according to claim 1, characterized in that, The specific expression for the weighted Gaussian likelihood function includes: ； in, Indicates the first The natural frequency calculated by the first-order finite element method; Indicates the first Measured natural frequency; Indicates the first The variance of the natural frequency calculated by the first-order finite element method; Indicates the first The damage weight based on the natural frequency of the first order is specifically expressed as follows: ； in, P j Indicates the first The degree of damage to the layers.

7. The Bayesian model correction method according to claim 6, characterized in that, The specific verification conditions for the dual validity verification include: For the first three natural frequencies, the relative error between the natural frequencies calculated by the finite element method and the measured natural frequencies. ≤5%; The post-test material degradation rate falls within the degradation range corresponding to the damage level.

8. A large-model-driven Bayesian model correction system, characterized in that, It includes a data acquisition module, a data processing module, and a result generation module; The data acquisition module is used to collect attribute information, evolution information, and environmental information of the target building. The attribute information includes the year of construction, number of floors, structural type, location of functional floors, and concrete strength grade of each floor. The evolution information includes the year of use, cumulative load change, and load change rate. The environmental information includes regional annual precipitation, average humidity, and seismic intensity. The data processing module includes a grouping unit, a finite element unit, a sensitivity analysis unit, a priori unit, a correction unit, a posterior unit, and a calibration unit. The grouping unit divides the target building into multiple independent layer groups based on attribute information and according to a preset floor grouping method; The finite element units are used to establish the initial finite element model of the target building. Based on the concrete strength grade, the material parameters of the finite element model are set; these material parameters are assigned to each layer group; the material parameters include the elastic modulus. and density ; The sensitivity analysis unit is used to perform sensitivity analysis on the material parameters of each layer group, obtain the relative sensitivity coefficient of the natural frequency to the material parameter, and select the material parameter with a relative sensitivity coefficient greater than a preset threshold as the correction material parameter. ; The prior unit is used to input the attribute information, evolution information and environmental information of the target building into the damage prediction model, output the damage degree ratio of each layer group, and then combine it with the preset correspondence to convert it into the benchmark mean of the corrected material parameters, and set the prior variance to construct the prior distribution of the Bayesian model correction. The specific methods for constructing the correspondence include: Step b1: Construct a formula for calculating the baseline mean of the corrected material parameters. The specific formula includes: ； ； in, This represents the baseline mean of the elastic modulus; This represents the initial value of the elastic modulus; This represents the change in the elastic modulus; This represents the elastic modulus correction factor; This represents the baseline mean density. Indicates the initial density value; Indicates the amount of density change; Indicates the density correction factor; This indicates an index of elastic modulus degradation. Indicates density degradation index; Step b2: Take half of the damage level as... and The results showed the correlation between the damage degree ratio and the material degradation index; Step b3: Based on the position of the layer group, set and ; The correction unit is used to correct the front of the target building. Using the measured natural frequency as the correction target, a weighted strategy of the proportion of damage degree of fusion layer group is introduced to construct a weighted Gaussian likelihood function, and the prior distribution is fused with the weighted Gaussian likelihood function to construct a Bayesian posterior distribution; The posterior unit is used to select and execute the corresponding MCMC sampling method based on the Bayesian posterior distribution and the dimension of the corrected material parameters; synchronously iterate to verify the convergence of the sampling results; and obtain the posterior distribution of the corrected material parameters for each layer group after the sampling results converge. The calibration unit extracts the optimal estimated value of the corrected material parameters for each layer group from the posterior distribution of the corrected material parameters; performs dual validity verification on the finite element model corrected using the optimal estimated value; if the verification passes, the optimal estimated value is used as the final corrected material parameters to update the finite element model. The result generation module is used to send the finite element model out.

9. A Bayesian model correction device driven by a large model, characterized in that, It includes a processor, a memory, and a bus. The memory stores instructions and data read by the processor. The processor is used to call the instructions and data in the memory to execute the Bayesian model correction method as described in any one of claims 1-7. The bus connects the functional components and is used to transmit information.

Images