A method for fusing measured data and mechanical simulation data

Abstract

The application relates to a kind of fusion methods of measured data and mechanical simulation data, comprising: obtaining actual site information and measured data, extracting measured response, calling large language model to generate large parameter matrix to be optimized, constructing finite element simulation model and solving, generating simulation response;The response difference of simulation response and measured response is calculated, and the parameter value of large parameter matrix is optimized;Determine whether iteration is terminated, if yes, then determine whether the fusion result output by the current finite element simulation model conforms to the actual situation, otherwise, update the finite element simulation model and perform the next iteration;If the fusion result conforms to the actual situation, then apply equivalent load to the finite element simulation model obtained by iteration, obtain fusion data, and use it for pavement structure performance evolution analysis, otherwise, modify the parameter selection to be optimized, regenerate the finite element simulation model and update the large parameter matrix, and reiterate.Compared with the prior art, the application has the advantages of being able to improve data quality and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of airport engineering pavement structure performance evolution analysis, and particularly relates to a fusion method of measured data and mechanical simulation data. BACKGROUND

[0002] Research, analysis and prediction of the evolution law of the performance of the airport runway structure are important foundations for the design of the runway structure and material science, long-term durability operation of the pavement structure and operation safety of the flight area. At present, the analysis of the performance of the pavement structure is divided into the following methods:

[0003] 1. A pavement structure model is constructed by using mechanical knowledge, and the evolution law of the performance of the structure is analyzed by obtaining an explicit expression of different parameters. This method is based on mechanical laws, has high theoreticality and interpretability, and is not easily affected by randomness. Since the pavement structure is relatively complex, a constitutive model cannot be constructed, and therefore this method often relies on a simulation numerical model, such as a finite element method, to construct the model.

[0004] 2. A large amount of measured data is used to analyze the change law of the performance of the structure by using a statistical method, and the value of a future related index is predicted to thereby inversely deduce the structure state of the pavement. This method relies on the statistical law obtained from a large amount of data, reflects the actual situation of a complex system, and can reflect the influence caused by the non-homogeneity of materials, construction quality and defects, etc. for which a simulation model cannot be constructed. Related technologies include linear regression analysis, time series analysis, random process analysis and other methods. With the continuous development of artificial intelligence, various supervised learning and unsupervised learning are also applied to the analysis of the performance of the pavement structure.

[0005] However, the above methods also have disadvantages, such as: in the construction process of the simulation model, some assumptions made by humans may be wrong, for the sake of solving efficiency, some mechanical relationships may be simplified, and many conditions of the real scene cannot be considered, which will affect the results; in the data analysis process, the data analysis results are easily affected by data noise, measurement sensor abnormalities and other factors.

[0006] Mechanical simulation and data analysis each have advantages and disadvantages. The combination of data analysis and mechanical simulation, the use of measured data and a structure simulation model to obtain more high-quality fusion data can provide a more scientific and reasonable basis for subsequent analysis of the evolution of the performance of the structure, and form more reliable and highly interpretable analysis results.

[0007] Chinese patent CN114117840A discloses a structure performance prediction method based on simulation and test data hybrid driving. In static test, limited measurement data and simulation data are combined to establish a fusion model. A large amount of simulation data and a small amount of test data are used to improve the accuracy of the proxy model constructed by hybrid data. The method improves the model fidelity, reconstructs the structure state field, and realizes health prediction in spatial dimension by using a multi-precision deep neural network model to adaptively learn the linear and nonlinear relationship between test data and simulation data. In health monitoring test, the fusion model is corrected by using an optimization algorithm. The fusion model constructed by the method fuses health monitoring data and can reflect the true physical state of the structure to realize health prediction in time dimension. However, the method needs to model and solve large structures when generating simulation data, which has high threshold and takes a long time, and cannot be used in actual structure monitoring. Moreover, the search method used is not conducive to large-scale iteration, making it difficult to obtain high-quality solutions. In addition, the method is difficult to comprehensively evaluate the response difference of the structure, and the optimization results cannot reflect the true physical state of the structure.

[0008] Therefore, there is a need for a simulation and test data fusion method with high computational efficiency, which can effectively obtain high-quality solutions and the obtained optimization results can truly reflect the physical state, to improve the possibility of practical application. SUMMARY

[0009] The purpose of the present application is to provide a fusion method of measured data and mechanical simulation data, which fuses measured data and simulation data, establishes a simulation large model with mechanical knowledge supervision and measured data correction, and fully improves data quality to provide a basis for pavement structure performance evolution research under the fusion of mechanical simulation and data analysis.

[0010] The purpose of the present application can be achieved by the following technical solutions:

[0011] A fusion method of measured data and mechanical simulation data, the fused data being used for pavement structure performance evolution analysis, the fusion method comprising the following steps:

[0012] S1, obtaining actual site information and measured data;

[0013] S2, extracting measured response by using cross-power spectrum processing based on the measured data;

[0014] S3, calling a large language model to generate a large parameter matrix to be optimized based on the actual site information, and constructing a finite element simulation model;

[0015] S4, solving the finite element simulation model to generate simulation response;

[0016] S5, calculating a response difference value of the simulation response and the measured response;

[0017] S6, optimizing parameter values of the large parameter matrix based on the response difference value;

[0018] S7, judging whether a preset iteration number is reached or a response difference value reaches a preset threshold, if yes, judging whether a fusion result output by a current finite element simulation model conforms to an actual situation, otherwise, returning to step S3 to update the finite element simulation model based on the optimized parameter values of the large parameter matrix and performing next iteration;

[0019] S8, if the fusion result conforms to the actual situation in step S7, taking the finite element simulation model obtained through iteration as a fusion model, applying an equivalent load to the fusion model to obtain fusion data, otherwise, modifying parameter selection to be optimized, returning to step S3 to regenerate the finite element simulation model and update the large parameter matrix, and re-performing iteration.

[0020] The finite element simulation model is specifically constructed as follows:

[0021] According to computing power, basic site conditions and target design parameter set;

[0022] Based on actual site information, natural language description is performed to obtain description text;

[0023] The parameter set and the description text are input into a large language model, and the large language model performs the following steps to generate a simulation model:

[0024] Model physical quantities in the description text are automatically located and parameters are extracted;

[0025] Mandatory parameters not mentioned in the description text are identified, and the mandatory parameters are inferred to complete or call related knowledge bases to assign default values to the mandatory parameters;

[0026] The extracted parameters and the completed mandatory parameters are uniformly converted, and rationality is verified;

[0027] Based on the verified parameters, a large parameter matrix to be optimized is set, a finite element simulation model is established in combination with a finite element modeling method of a pavement slab, and a working condition definition statement is generated.

[0028] The large parameter matrix to be optimized includes material parameters, geometric parameters and boundary condition parameters, wherein the material parameters include elastic modulus, Poisson's ratio and density, the geometric parameters include cross-sectional width, thickness and length, and the boundary condition parameters include foundation support stiffness and joint spring stiffness.

[0029] In constructing the finite element simulation model, the structure was divided into multiple sub-regions based on mechanical principles and the installation of sensing optical fibers. Parameters were set for each region. The structure was divided into two layers: horizontal and vertical. The horizontal layer represents the horizontal direction of the structure, and the vertical layer represents the depth direction. In the vertical layer, the thickness of the layers was adjusted to ensure that the sensing optical fibers passed evenly through the parameter adjustment area. In the horizontal layer, the partitions were densified within a preset range near the load-bearing area. Simultaneously, measurement lines were set in the finite element model according to the location of the on-site optical fiber sensing devices for subsequent data extraction.

[0030] In step S5, before calculating the response difference, the simulation response is aligned using the linear interpolation method of simulation data. Specifically:

[0031] For the simulation response time series { , }, to the time point set { Data alignment, where, For any time point in the simulation time series, For time points The corresponding simulated response value, For any point in the set of time points of the measured data;

[0032] Find the corresponding interval index such that: ;

[0033] By using linear interpolation, the adjustment parameter λ is calculated, and then the interpolated result is calculated using the adjustment parameter λ to achieve data alignment.

[0034] ,

[0035] ,

[0036] in, This is the interpolated response value of the simulation data at the corresponding time point of the measured data after interpolation.

[0037] The method for calculating the response difference is as follows: using the mode shape as a physical quantity characterizing the response, and quantifying the response difference based on the modal confidence criterion.

[0038] ,

[0039] ,

[0040] in, Indicates the response difference. For the first The weight of the MAC value corresponding to each test line. Indicates the first modal shape of the i-th test line according to the simulation response a and the modal shape of the measured response b the calculated MAC value, respectively represent the modal shape of the i-th test line modal shape of the i-th test line a and b k the conjugate transpose, n the order, L the number of test lines.

[0041] The response difference value is calculated by taking the modal shape and the frequency difference value as the physical quantity representing the response, and quantifying the modal shape based on the modal confidence criterion to obtain the response difference value after weighting:

[0042] ,

[0043] ,

[0044] ,

[0045] ,

[0046] wherein, the response difference value is composed of the MAC value and the frequency difference value , , respectively represent the weight of the MAC value and the frequency difference value of the i-th test line, the number of test lines; L modal shape of the i-th test line according to the simulation response and the modal shape of the measured response the calculated MAC value, a respectively represent the modal shape of the i-th test line b modal shape of the i-th test line and a b k the conjugate transpose, n the order; the frequency difference value of the i-th test line, the order of the i-th test line, the order of the i-th test line, the order of the i-th test line, k the order of the i-th test line, the order of the i-th test line, the order of the i-th test line. k the order of the i-th test line, the order of the i-th test line. the order of the i-th test line. k the order of the i-th test line. ​​​​

[0047] The parameter value optimization of the large parameter matrix based on the response difference value is specifically: adopting a search algorithm, taking the reciprocal of the response difference value as an optimization objective function, setting the parameters to be optimized as a population, converting the optimization process of the parameter value range into a process of different degrees of reduction of the initial value of the parameter based on a reduction factor, and optimizing the reduction factor.

[0048] The parameter value optimization of the large parameter matrix based on the response difference value is specifically: adopting a reinforcement learning algorithm, inputting the state , is the change amount of the MAC value in the response difference, is the frequency difference value in the response difference, and the corresponding action vector , is the number of parameter types to be estimated, is the spatial region dimension, the output action vector is split into a matrix corresponding to each parameter, and the output value is converted to obtain a specific physical quantity value, the reward function is composed of a difference quantization index and a step penalty, and is expressed as:

[0049] ,

[0050] ,

[0051] wherein, is the reward function for the difference quantization index, and a is a user-defined coefficient, is the search step, is the total reward function.

[0052] The judgment of whether the fusion result output by the current finite element simulation model conforms to the actual situation specifically includes parameter mutation checking and response mutation checking, wherein,

[0053] The parameter mutation checking is specifically: setting a threshold value δ, calculating the absolute value of the difference of each parameter and the parameters of its spatially adjacent upper, lower, left and right sub-regions, and finding the maximum value Δ, when Δ≤δ, it is determined that there is no mutation, otherwise, the parameter mutation checking fails;

[0054] The response mutation checking is specifically: performing continuity checking on the response data, removing the parabolic trend of the response data according to , wherein, is the residual for continuity checking, is the response data, is the parabolic data fitted according to the position point i ; the residual Adopt CUSUM detection to check whether there is a mutation point, if the upper CUSUM and the lower CUSUM exceed the threshold value, it is determined that mutation occurs, and the response mutation check does not pass;

[0055] When any one of the parameter mutation check and the response mutation check does not pass, it is determined that the fusion result of the current finite element simulation model output does not conform to the actual situation.

[0056] Compared with the prior art, the present application has the following beneficial effects:

[0057] 1. The present application introduces an integrated platform based on large language model interaction. The large language model trained for structural simulation calculation script can quickly respond to user natural language instructions, complete finite element calculation script generation, and integrate functions such as finite element calculation software calling and optimal parameter searching, reducing the complexity of the fusion process and facilitating rapid deployment and use in engineering sites.

[0058] 2. The present application introduces mutual power spectrum for signal processing of collected data, extracts mode shapes and natural frequencies to construct response difference functions, and introduces MAC (Modal Assurance Criterion) to quantify mode shape differences, achieving accurate capture of structural vibration response differences. Introducing this difference calculation method into the reward function of reinforcement learning can ensure that the result is close to the actual condition of the structure.

[0059] 3. When using a search algorithm for solving, in the search strategy, the present application uses a search scale factor instead of direct search, making the numerical scale more friendly, avoiding optimization of the optimizer for different dimensions / magnitudes, and naturally ensuring that the parameters meet the boundary conditions.

[0060] 4. The present application takes mechanical constraints as the core and measured data as the basis to ensure that the fusion result meets both the actual site structure state and the basic mechanical laws; through the dual control of "parameter mutation check" and "response mutation check", the abnormality of model setting and numerical output is simultaneously identified and located, ensuring the credibility and robustness of the simulation-data fusion result. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 The flowchart of the method of the present application;

[0062] Figure 2 The flowchart of constructing a finite element simulation model;

[0063] Figure 3 The parameter category and region division schematic diagram in the process of constructing a finite element simulation model in an embodiment;

[0064] Figure 4 The DDPG network operation flowchart;

[0065] Figure 5Flowchart for solving cycle using search algorithm

[0066] Figure 6 Flowchart for generating fused data

[0067] Figure 7 Diagram for test beam in Example 2

[0068] Figure 8 Diagram for measured response of layer 1 extracted using cross power spectrum in Example 2

[0069] Figure 9 Diagram for measured response of layer 2 extracted using cross power spectrum in Example 2

[0070] Figure 10 Diagram for measured response of layer 3 extracted using cross power spectrum in Example 2

[0071] Figure 11 Diagram for mode shape curve of layer 1 generated using initial values in Example 2

[0072] Figure 12 Diagram for mode shape curve of layer 2 generated using initial values in Example 2

[0073] Figure 13 Diagram for mode shape curve of layer 3 generated using initial values in Example 2

[0074] Figure 14 Flowchart for response difference calculation in Example 2

[0075] Figure 15 Diagram for optimal parameters determined in Example 2

[0076] Figure 16 Diagram for simulation model with optimal parameter values in Example 2

[0077] Figure 17 Comparison diagram of fused data and measured data in Example 2. DETAILED DESCRIPTION

[0078] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present application, and gives a detailed implementation and specific operation process, but the protection scope of the present application is not limited to the following embodiments.

[0079] Embodiment 1

[0080] The present embodiment provides a fusion method of measured data and mechanical simulation data, and the fused data is used for pavement structure performance evolution analysis, such as Figure 1 As shown in the figure, the fusion method comprises the following steps:

[0081] S1, obtaining actual site information and measured data;

[0082] S2, based on the measured data, using mutual power spectrum processing to extract the measured response;

[0083] S3, based on the actual site information, calling a large language model to generate a to-be-optimized large parameter matrix, and constructing a finite element simulation model;

[0084] S4, solving the finite element simulation model to generate a simulation response;

[0085] S5, calculating the response difference between the simulation response and the measured response;

[0086] S6, optimizing the parameter values of the large parameter matrix based on the response difference;

[0087] S7, determining whether the preset number of iterations is reached or the response difference reaches the preset threshold, if yes, determining whether the fusion result output by the current finite element simulation model conforms to the actual situation, otherwise, returning to step S3 to update the finite element simulation model based on the parameter values of the optimized large parameter matrix and performing the next iteration;

[0088] S8, if the fusion result in step S7 conforms to the actual situation, the finite element simulation model obtained by iteration is taken as a fusion model, an equivalent load is applied to the fusion model, and fusion data is obtained, otherwise, the parameter selection to be optimized is modified, and step S3 is returned to regenerate the finite element simulation model and update the large parameter matrix, and reiterate.

[0089] In step S3, the large language model uses a specified prompt word template to generate a simulation file, the prompt word template is generated through a multi-stage prompt chain, an intermediate workpiece constraint and an automatic verifier, the site information described in natural language is extracted, inferred and integrated, and program statements understandable by ABAQUS are generated, so that the response of the simulation model is better adjusted, which provides support for subsequent rapid generation of finite element simulation models for simulation response data generation and modification of parameters under the action of the optimization method.

[0090] As shown in Figure 2 , constructing the finite element simulation model includes the following steps:

[0091] S31, according to the computing power, the basic situation of the site and the target design parameter set;

[0092] S32, based on the actual site information, a natural language description is obtained, and a description text is obtained;

[0093] S33, inputting the parameter set and the description text into the large language model, and the large language model performs the following steps to generate a simulation model:

[0094] S331 automatically locates physical quantities of the model in the description text, such as the thickness, modulus, Poisson's ratio and density information of the pavement, the ground response modulus under healthy conditions, the normal joint spring stiffness, the aircraft load speed, etc., and extracts the values ​​to obtain the extracted parameters.

[0095] S332, Identify necessary parameters not mentioned in the description text, such as grid type and grid size, and perform reasoning to complete the necessary parameters or call the pavement engineering industry specifications and standard knowledge base to assign them default values;

[0096] S333 performs a unified conversion on the extracted parameters and the required parameters for completion, and verifies their rationality (value range, relationship between variables);

[0097] S334, based on the verified parameters, sets the large parameter matrix to be optimized, combines the finite element modeling method of the track panel, establishes the finite element simulation model, and generates the working condition definition statement.

[0098] In this embodiment, the large parameter matrix to be optimized includes material parameters, geometric parameters, and boundary condition parameters, wherein the material parameters include the elastic modulus. Poisson's ratio and density Geometric parameters include cross-sectional width ,thickness and length Boundary condition parameters include foundation support stiffness. and joint spring stiffness There are a total of 8 parameters, as shown in equations (1)-(4).

[0099] (1)

[0100] (2)

[0101] (3)

[0102] (4)

[0103] in, For material parameters, For geometric parameters, These are the boundary condition parameters.

[0104] In constructing the finite element simulation model, the structure is divided into multiple sub-regions based on mechanical laws and the installation of sensing optical fibers, such as... Figure 3The structure is divided into two layers of horizontal and vertical directions, the horizontal direction is the horizontal direction of the structure, and the vertical direction is the depth direction of the structure. The parameters are set in each area, wherein, in the vertical structure layer, since the optical fiber is buried at different depths, the thickness of the layered structure is adjusted to make the sensing optical fiber uniformly pass through the parameter adjustment area (0.05m, 0.5m and 0.8m); in the horizontal structure layer, considering that the structure performance near the load loading area is easy to change, the partition is encrypted in the preset range near the load loading area, so as to better reflect the change of the structure performance.

[0105] According to the buried position of the on-site optical fiber sensing device, the measuring line is set in the finite element model for subsequent data extraction.

[0106] Combined with the obtained parameters and the setting of the to-be-optimized parameters, the simulation model generation module is used to establish a finite element simulation model by combining the finite element modeling method of the pavement slab, and generate a working condition definition statement that can be understood by a commercial finite element software. The module finally generates a.py file, which contains the description of the model and the ABAQUS subroutine call statement. Running the file can call ABAQUS to calculate the global response result, and extract the simulation response at the set measuring line position. The code of the pavement slab model generation component is as follows:

[0107] s=mdb.models["Model-1"].constrainedsketch(name="profile",sheetsize=200.0)

[0108] g,v,d,c=s.geometry,s.vertices,s.dimensions,s.constraints

[0109] s.setPrimaryObject(option=STANDALONE)

[0110] Then, in step S4, the commercial finite element software ABAQUS is called to perform response analysis. The modal analysis step is adopted, and the highest order mode is 10. The mode response is solved.

[0111] The simulation response and the measured response data points are often difficult to keep consistent in actual application. Therefore, in step S5, before calculating the response difference, the simulation response is processed by linear difference method to align the data, which is specifically:

[0112] Align the data of the simulation response time series{ , } to the time point set{ }, wherein, is any time point in the simulation time series, is the time point is the response value of the corresponding simulation response, is any point in the set of time points of the measured data;

[0113] find the corresponding interval index such that: ;

[0114] Using linear interpolation, calculate the adjustment parameter λ, and use the adjustment parameter λ to calculate the interpolated result, realize data alignment:

[0115] (5)

[0116] (6)

[0117] wherein, is the interpolated response value of the simulation data at the corresponding measured data time point after interpolation.

[0118] After data alignment, calculate the response difference value.

[0119] In one embodiment, modal shape is taken as the physical quantity representing the response, and the response difference is quantified based on modal assurance criterion (MAC):

[0120] (7)

[0121] (8)

[0122] wherein, represents the response difference value, is the weight of the MAC value corresponding to the i-th survey line, represents the MAC value calculated according to the modal shape of the simulation response and the modal shape of the measured response of the i-th survey line, a respectively represent the i-th modal shape b and of the i-th survey line, respectively represent the i-th modal shape a and b of the i-th survey line, k denotes conjugate transpose, n is the order, L is the number of survey lines.

[0123] In another embodiment, modal shape and frequency difference value are taken as the physical quantity representing the response, and after modal shape quantification based on modal assurance criterion, the response difference value is obtained by weighting:

[0124] (9)

[0125] (10)

[0126] (11)

[0127] (12)

[0128] in, The response difference is represented by the MAC value and the frequency difference. constitute, , The first The weights of the MAC value and frequency difference corresponding to each test line. L This refers to the number of survey lines; Indicates the first The test line is based on the modal shape of the simulated response. a and measured response mode shape b The calculated MAC value, They represent the first Modal vibration modes of the strip measurement line a and b of k A vector of order 1, where * denotes the conjugate transpose. n It is the order; For the first Frequency difference of the test lines For the first strip survey line k The difference in natural frequencies of the first order. For the first Simulation response of the test line k First natural frequency, For the first Measured response of the test line k The first natural frequency.

[0129] In one preferred embodiment, when calculating the modal confidence criterion (MAC) to quantify the response difference, the first-order mode shape that best approximates the measured response is selected for calculating the MAC value, i.e., k is set to 1. The MAC value ranges from [0,1]. When the MAC is close to 1, it indicates that the two modes are very similar, while when it is close to 0, it indicates that the difference is large.

[0130] The MAC value alone can only measure the relative similarity of the responses, while the difference calculated by Equation (9) comprehensively quantifies the difference between the measured response and the simulated response, providing a reference value for finding the balance point between the two.

[0131] After calculating the response difference, the parameter values ​​of the large parameter matrix are optimized based on the response difference.

[0132] In an embodiment, a search algorithm such as genetic algorithm can be used, taking the inverse of the difference as the optimization objective function, setting the parameters to be optimized as the population, converting the optimization process of the parameter value range into a process of different degrees of reduction of the initial value of the parameter based on the reduction factor, and optimizing the reduction factor. The reduction factor is set to (0, 1]. The algorithm uses the method of randomly generating a parameter matrix as the search starting point, and performs 50 iterations of search, with the population size of each search being 50. The entire optimization solution cycle is as shown in Figure 5 .

[0133] In another embodiment, a reinforcement learning algorithm is used, specifically the DDPG algorithm under the actor-critic algorithm framework. This framework consists of a reservoir, training and target actor networks, and training and target critic networks. The critic network gradually gives the Q value approximating the true value through learning of the Q value, and the actor network gradually updates the strategy under the guidance of the Q value to generate the optimal action according to the state.

[0134] The algorithm running framework is as shown in Figure 4 The state is input into the agent centered on the actor network, where is the change amount of the MAC value in the response difference, is the frequency difference in the response difference, and the corresponding action vector is generated, where is the number of parameter types to be estimated, is the spatial region dimension, the output action vector is split into a matrix corresponding to each parameter as shown in equation (13), and the output value is converted to obtain the specific physical quantity value, where the elastic modulus , the Poisson's ratio , and the density are calculated as shown in equations (14)-(16). For the generality of the method in the future, the parameter conversion method as shown in equations (17)-(18) is proposed. For parameters in the range (0, 1), the method shown in equation (17) is used for conversion, and for parameters with a larger value range, the method of equation (18) is used for conversion. The calculation of the physical quantities shown in equations (14)-(16) falls into this category.

[0135] , , (13)

[0136] (14)

[0137] (15)

[0138] (16)

[0139] (17)

[0140] (18)

[0141] wherein, , , are the output value vectors of the corresponding modulus, Poisson's ratio, density, are , the output values of each sub-region in a certain vector, , are the maximum and minimum values of the elastic modulus, , are the maximum and minimum values of the Poisson's ratio, , are the maximum and minimum values of the density, , , are the converted elastic modulus, Poisson's ratio and density, is the converted physical quantity, , are the maximum and minimum values of the corresponding physical quantity.

[0142] The reward function is composed of two parts of difference quantization index and step penalty, and is expressed as:

[0143] (19)

[0144] (20)

[0145] wherein, is the reward function for the difference quantization index, and a is a self-defined coefficient, is the search step, is the total reward function.

[0146] In the optimization process, the parameter value solved each time is transmitted back to the finite element simulation model, the simulation response of the updated model is calculated, the new response difference value is calculated based on the simulation response obtained by the new calculation, and one cycle is completed. According to actual needs and experience, the iteration number or the response difference value threshold is set as the termination condition, and when the set termination condition is reached, the cycle is ended.

[0147] After the end of the cycle, the finite element simulation model parameters at this time and the initial setting value has changed greatly, the local changes of structure, geometry, boundary conditions are considered as the deviation from the actual site and ideal situation, at the same time, since the response calculation of finite element simulation model conforms to the mechanical constraints, therefore, the response data generated by the simulation model at this time should be able to meet the structure condition of actual site and mechanical law. But since the initial setting parameters, there is a certain assumption in the delineation of the region, that is, the material properties of adjacent regions may differ greatly, but in fact, its change should be continuous change rather than sudden change. Therefore, it is necessary to judge the mutation of parameters and generated data to determine whether the fusion result output by the current finite element simulation model conforms to the actual situation, which includes parameter mutation check and response mutation check:

[0148] 1. Parameter mutation check

[0149] The form of parameter is m x n size matrix, so the main difference between the adjacent elements in the matrix is that the difference between the adjacent elements in the matrix cannot be too large. Set threshold δ, calculate the absolute value of the difference between each parameter and its spatial adjacent upper and lower left and right sub-region parameters, and find the maximum value Δ, when Δ≤δ, it is determined that there is no mutation, otherwise, it is determined that there is a large local mutation, the parameter mutation check is not passed, the parameter assignment region is adjusted, the methods that can be used include local encryption or local fusion, and the above fusion link is performed again until the parameter no longer mutates.

[0150] 2. Response mutation check

[0151] The continuity of response data is checked, according to formula (21), the parabolic trend of response data is removed:

[0152] (21)

[0153] Wherein, is the residual for continuity check, is the response data, is the parabolic data fitted according to the position point i .

[0154] The residual is checked by CUSUM to see if there is a mutation point, the mean value and the standard deviation σ are calculated, the upper CUSUM and the lower CUSUM are calculated as shown in formula (22) and (23), if the upper CUSUM and the lower CUSUM exceed the threshold value (5σ), it is determined that there is a mutation, the response mutation check is not passed, the simulation model design is changed, and the above steps are repeated.

[0155] (22)

[0156] (23)

[0157] wherein, is the allowable offset, taking 0.5σ; is the data value at the current detection time point, is the value of the upper side CUSUM cumulative statistics at the nth moment, and when exceeding the threshold value (taking 5σ), it is determined that mutation occurs, is the value of the lower side CUSUM cumulative statistics at the nth moment.

[0158] When any one of the parameter mutation check and the response mutation check does not pass, it is determined that the fusion result of the current finite element simulation model output does not conform to the actual situation.

[0159] The specified load is applied to the final fusion model, and the response data is obtained, which can be considered as simulation and measured data fusion data, and the data conforms to the actual situation and the mechanical constraint, and can better evaluate and predict the pavement structure performance. The fusion data generation process is as shown in Figure 6 .

[0160] In summary, the core of the data fusion of the present application is the process of generating simulation data and fusing physical quantities of actual measurement data. The simulation data is generated, and the actual measurement data is connected in real time. The measured response and the simulation response respectively contain the uncertainty in the service process of the actual structure and the mechanical constraint realized by the finite element method. Through repeated iteration and optimization, the fusion data gradually approaches the balance point between the measured data and the simulation data. By using the optimization algorithm to continuously reduce the difference generated by the response difference calculation module, the difference between the measured and simulated response data is significantly reduced, ensuring that the finally fused data accurately reflects the comprehensive characteristics of both. At the same time, the mechanism supports the adaptive evolution of the model in the whole service cycle, ensuring that the fusion can always accurately reflect the actual operation state of the structure.

[0161] Example 2

[0162] This embodiment takes the fusion of the asphalt trabecular fiber measured data and the mechanical simulation law by the method described in Example 1 as an example for detailed description.

[0163] As shown in Figure 7 , the test specimen is a 0.8x0.1x0.3m asphalt simply supported beam. The local stiffness change of the material is changed by local heating to simulate the change of the structure performance in the actual use process of the structure. The fiber embedded in the specimen is used to obtain the vibration strain data under the hammering, and the fiber is embedded in the upper, middle and lower three layers of the beam to better cover each height of the structure. The initial material parameters, geometric parameters and boundary conditions of the specimen are known.

[0164] I. Initial finite element simulation model establishment

[0165] According to the actual burying condition of the optical fiber, the whole simply supported beam is divided into 3x12 regions. Specifically, the height direction is divided into 3 layers, and the length direction is divided into 12 regions along the beam. Among them, the regions in the beam are encrypted to accurately reflect the change of the structure performance. In terms of parameter selection, except for the material modulus, other parameters such as Poisson's ratio, density, plate bottom support, thickness and the like are set as constants to save computing resources and speed up the calculation speed. That is, in this embodiment, the shape and size of the large model parameters participating in the optimization are 3x12.

[0166] The simulation large model platform based on the large language model is combined with reinforcement learning, the large language model is embedded in the reinforcement learning environment, and the large language model, agent, environment and the like are integrated into the platform to realize integration, and the fusion process and result are visualized.

[0167] The platform analyzes the input, and the obtained finite element model parameter values are as follows: the density is set to 2200 kg / m 3 , the initial modulus E is 4.5e 8 Pa, the Poisson's ratio is 0.35, the boundary condition is simply supported, and other material parameters adopt inference or default values. The to-be-optimized parameter is set to I E x E, I E is a to-be-optimized parameter matrix, and the value range is limited to (0, 1], that is, the reduction of the initial modulus.

[0168] II. Introduction of measured response and difference calculation

[0169] The measured data of the optical fiber are processed, the signal processing method of solving the cross power spectrum CPSD is used to extract the mode shape, and three mode shape vectors are extracted. The size of a single mode shape is . As shown in Figure 8 , Figure 9 and Figure 10 .

[0170] Run the script file in the previous step to call the ABAQUS software to implement a finite element calculation, solve and extract the mode shape curve at the measuring line position. The measuring line setting is consistent with the measured data, that is, the burying depth is 0.0725, 0.03, 0, and the number of data points is 240. The specific mode shape curve is shown in Figure 11 , Figure 12 and Figure 13 .

[0171] Only the reward calculation of ΔMAC is performed, and the response difference is calculated according to the process as shown in Figure 14 .

[0172] III. Data fusion cycle

[0173] The parameter value is optimized by using the reinforcement learning DDPG algorithm. The episode is set to 500, the step is 100 in each episode, and the actor network is repeatedly trained. The final action value is output, and the modulus distribution in the beam shown by the final action is as shown in Figure 15 It can be seen that the modulus in the beam region decreases, which can correspond to the modulus drop caused by heating of the asphalt beam, and qualitatively verify the effectiveness of the fusion framework operation.

[0174] Four, fusion data generation

[0175] The parameter optimization result is assigned to the simulation model, and the construction of the simulation large model is completed. The optimal parameters determined by the algorithm framework do not have mutation, and by testing, the parameters are brought into the initial finite element simulation model, and the simulation model used for verification in the ABAQUS software is as shown in Figure 16 The input working condition is set to use the loading head static load, the fusion data is calculated, and as shown in Figure 17 It can be seen that the data is relatively continuous and does not have mutation, and by testing, the data is high-quality fusion data.

[0176] The reconstructed fusion data is compared with the strain data measured by the optical fiber under static load of the asphalt beam, it is found that the fusion data is relatively consistent with the measured data, and the noise level is low, which further verifies the effectiveness of the fusion framework.

[0177] The above describes the preferred embodiments of the present application in detail. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment on the basis of the prior art according to the concept of the present application shall be within the protection scope determined by the claims.

Claims

1. A method for fusing measured data and mechanical simulation data, characterized in that, The fused data is used for pavement structure performance evolution analysis, and the fusion method comprises the following steps: S1, obtaining actual site information and measured data; S2, based on the measured data, using mutual power spectrum processing to extract the measured response; S3, based on the actual site information, calling a large language model to generate a to-be-optimized large parameter matrix, and constructing a finite element simulation model: According to the computing power, the basic situation of the site and the target design parameter set; Based on the actual site information, a natural language description is obtained; The parameter set and the description text are input into the large language model, and the large language model performs the following steps to generate a simulation model: Automatically locate the model physical quantity in the description text and extract the parameters; Identify the necessary parameters not mentioned in the description text, infer and complete the necessary parameters or assign default values to them by calling related knowledge bases; Uniformly convert the extracted parameters and the completed necessary parameters, and check the rationality; Based on the parameters after checking, set the to-be-optimized large parameter matrix, establish the finite element simulation model combined with the finite element modeling method of the pavement slab, and generate a working condition definition statement; S4, solving the finite element simulation model to generate a simulation response; S5, calculating the response difference value between the simulation response and the measured response; S6, optimizing the parameter values of the large parameter matrix based on the response difference value; S7, determining whether the preset number of iterations is reached or the response difference value reaches the preset threshold, if yes, determining whether the fusion result output by the current finite element simulation model conforms to the actual situation, otherwise, returning to step S3 to update the finite element simulation model based on the parameter values of the optimized large parameter matrix for the next iteration; S8, if the fusion result in step S7 conforms to the actual situation, the finite element simulation model obtained by iteration is taken as the fusion model, and an equivalent load is applied to the fusion model to obtain fusion data, otherwise, the parameter selection to be optimized is modified, and step S3 is returned to regenerate the finite element simulation model and update the large parameter matrix for reiteration.

2. The method of claim 1, wherein, The to-be-optimized large parameter matrix includes material parameters, geometric parameters and boundary condition parameters, wherein the material parameters include elastic modulus, Poisson's ratio and density, the geometric parameters include cross-sectional width, thickness and length, and the boundary condition parameters include foundation support stiffness and joint spring stiffness.

3. The method of claim 1, wherein the measured data and the mechanical simulation data are fused by using a weighted average method. In the process of constructing the finite element simulation model, the structure is divided into multiple sub-regions combined with the mechanical law and the sensing optical fiber embedding condition, and parameter setting is performed in each region, wherein the structure is divided into two layers of horizontal and vertical directions, the horizontal direction is the horizontal direction of the structure, and the vertical direction is the depth direction of the structure, in the vertical structure layer, the thickness of the layer is adjusted to make the sensing optical fiber pass through the parameter adjustment region uniformly, and in the horizontal structure layer, the partition is densified within the preset range near the load loading area; at the same time, according to the embedding position of the on-site optical fiber sensing device, the measuring line is set in the finite element model synchronously, which is used for subsequent data extraction.

4. The method of claim 1, wherein, In step S5, before calculating the response difference value, the simulation data linear difference method is used for data alignment processing of the simulation response, specifically: aligning the time series of simulation responses { , } to the set of time points { } where, is any time point in the time series of simulation responses, is the time point corresponding to the response value of the simulation response, is any point in the set of time points of the measured data; Finding the corresponding interval index such that: ; The adjustment parameter λ is calculated by using linear difference value, and the interpolated result is calculated by using the adjustment parameter λ to realize data alignment: ， ， wherein, is the interpolated response value of the simulation data at the time point of the measured data after interpolation.

5. The method of claim 1, wherein, The response difference calculation method is: taking modal vibration mode as the physical quantity representing the response, and quantifying the response difference based on modal confidence criterion: ， ， wherein, denotes the response difference, is the MAC value of the i-th measurement line, is the weight of the MAC value of the i-th measurement line, denotes the MAC value of the i-th measurement line, denotes the modal shape of the i-th measurement line according to the simulated response, a and the modal shape of the i-th measurement line according to the measured response, b is the MAC value calculated from the modal shape of the i-th measurement line, denote the modal shape of the i-th measurement line, and the modal shape of the i-th measurement line, a and b are the i-th eigenvector of the modal shape of the i-th measurement line, k denotes the conjugate transpose, n is the order, L is the number of measurement lines.

6. The method of claim 1, wherein, The response difference calculation method is: taking modal vibration mode and frequency difference as the physical quantity representing the response, and quantifying the modal vibration mode based on modal confidence criterion, and then obtaining the response difference by weighting: ， ， ， ， in, The response difference is represented by the MAC value and the frequency difference. constitute, , The first The weights of the MAC value and frequency difference corresponding to each test line. L This refers to the number of survey lines; Indicates the first The test line is based on the modal shape of the simulated response. a and measured response mode shape b The calculated MAC value, They represent the first Modal vibration modes of the strip measurement line a and b of k A vector of order 1, where * denotes the conjugate transpose. n It is the order; For the first Frequency difference of the test lines For the first strip survey line k The difference in natural frequencies of the first order. For the first Simulation response of the test line k First natural frequency, For the first Measured response of the test line k The first natural frequency.

7. The method of claim 5 or 6, wherein the method further comprises: The parameter value optimization of the large parameter matrix based on the response difference is specifically: adopting a search algorithm, taking the reciprocal of the response difference as the optimization objective function, setting the parameters to be optimized as the population, converting the optimization process of the parameter value range into the process of different degrees of reduction of the initial value of the parameter based on the reduction factor, and optimizing the reduction factor.

8. The method of claim 6, wherein the measured data and the mechanical simulation data are fused by using a weighted average method. The parameter value optimization of the large parameter matrix based on the response difference value is specifically: using a reinforcement learning algorithm, inputting a state to an agent with an actor network as the core wherein, is a change amount of the MAC value in the response difference, is a frequency difference value in the response difference, and an action vector is generated wherein, is a parameter type number to be estimated, is a divided spatial region dimension, the output action vector is split into a matrix corresponding to each parameter, and the output value is converted to obtain a specific physical quantity value, and a reward function is composed of a difference quantization index and a step penalty, and is represented as: ， ， wherein, is a reward function for the difference quantization index, and a is a self-defined coefficient, is a search step size, is a total reward function.

9. The method of claim 1, wherein, The judgment whether the fusion result output by the current finite element simulation model conforms to the actual situation specifically includes parameter mutation check and response mutation check, wherein, The parameter mutation check is specifically: setting a threshold δ, calculating the absolute value of the difference of each parameter and the parameters of the adjacent upper, lower, left and right sub-regions in space, and finding the maximum value Δ, when Δ≤δ, it is determined that there is no mutation, otherwise, the parameter mutation check fails; The response mutation check specifically comprises: performing continuity check on the response data, removing a parabolic trend of the response data according to , wherein, is a residual for the continuity check, is the response data, is a parabolic data fitted according to the position points i ; the residual is checked by using a CUSUM detection to see if there is a mutation point, if both the upper CUSUM and the lower CUSUM exceed the threshold, it is determined that a mutation occurs, and the response mutation check fails. When any of the parameter mutation check and the response mutation check fails, it is determined that the fusion result output by the current finite element simulation model does not conform to the actual situation.

Images