A structural deformation prediction method and system for a construction engineering construction stage

Abstract

The present application relates to the technical field of building construction, and particularly relates to a structure deformation prediction method and system for a building construction stage. The structure deformation prediction system for the building construction stage comprises an active excitation response feedback module, an equivalent stiffness coefficient analysis module, a modal consistency correction module and a structure deformation prediction module. Through the four mechanisms of "active excitation + data inversion + physical consistency correction + modal constraint", the present application realizes high sensitivity acquisition, high precision identification and high reliability correction of the structure stiffness in the construction stage, and effectively solves the problems of insufficient response, non-identifiable parameters, non-unique results and cumulative prediction deviation of the traditional passive monitoring method.

Description

Technical Field

[0001] This invention relates to the field of building construction technology, and specifically to a method and system for predicting structural deformation during the construction phase of building engineering. Background Technology

[0002] During construction, structures undergo continuous deformation and dynamic response under the influence of factors such as self-weight, construction loads, support transformation, and temperature changes. To understand the stress and deformation state of the structure during construction, engineering practices typically employ monitoring equipment such as total stations, displacement gauges, strain gauges, or accelerometers to record the displacement, strain, or vibration data of the structure. Based on this monitoring data, the parameters of the finite element model (FEM) during construction are corrected to improve the model's ability to predict deformation in subsequent construction phases. Existing methods often employ passive monitoring, which involves fitting the monitored displacement or strain to correct the stiffness coefficients of components based on the structure's response under natural conditions. However, because the loads on the structure during construction are usually random, have limited amplitude, and act in a single direction, it is often difficult to elicit sufficient responses in key components, resulting in low sensitivity of monitoring data to changes in structural stiffness. Furthermore, multiple different parameter combinations often produce similar responses, easily leading to non-unique parameter identification results and difficulty in accurately reflecting the true stiffness distribution of the structure. Parameters obtained by passive fitting methods typically only characterize the equivalent stiffness under the current conditions and are insufficient to reflect the stiffness evolution of the structure in subsequent construction steps. Therefore, when performing multi-process predictions based on the finite element model modified with these parameters, the model error may gradually accumulate, affecting the judgment of the actual deformation of the structure. Summary of the Invention

[0003] This invention applies active excitation obtained through particle swarm optimization to the structure during the construction phase to acquire response data with high sensitivity to the stiffness coefficient of the components. A feedforward neural network is used to perform a preliminary inversion of the equivalent stiffness coefficient, followed by physical consistency correction based on displacement error using a differentiable finite element model. Furthermore, the random subspace method is employed to extract the actual low-order frequencies and mode shapes of the construction structure from the acceleration signal of the active excitation response, and these are used as global modal constraints to further correct the initially corrected stiffness coefficient. This results in obtaining a true, global, and consistent stiffness coefficient, significantly improving the uniqueness and accuracy of the stiffness coefficient inversion. This ensures that the finite element model with corrected stiffness coefficients can accurately predict the deformation of the subsequent construction structure.

[0004] This invention provides a method for predicting structural deformation during the construction phase of building engineering, including: When the stiffness coefficient adjustment trigger condition is met, an active excitation operation is performed on the construction structure through an active excitation scheme, and structural displacement data and structural acceleration data in response to the active excitation operation are collected. The active excitation scheme is optimized under the constraint of the sensitivity of the monitoring response. The active excitation scheme is concatenated with the corresponding structural displacement data to form excitation response data. The active excitation scheme includes several excitation parameter vectors. The excitation response data is then fed into a stiffness coefficient inversion network for processing, outputting an equivalent stiffness coefficient dataset. This dataset includes the equivalent stiffness coefficients of each component in the finite element model corresponding to the construction structure. The equivalent stiffness coefficients characterize the proportional relationship between the actual stiffness coefficients and the original design stiffness coefficients. The equivalent stiffness coefficient dataset is then loaded into the finite element model. By analyzing the difference between the finite element model's response to the active excitation scheme and the structural displacement data, physical consistency correction is performed on the equivalent stiffness coefficient dataset to obtain an initial corrected equivalent stiffness coefficient dataset. Based on the structural acceleration data of the active excitation operation response, the actual low-order modal data corresponding to the construction structure are determined. Based on the constraints of the actual low-order modal data, global modal consistency correction is performed on the initial correction equivalent stiffness coefficient dataset to obtain the final correction equivalent stiffness coefficient dataset. The final corrected equivalent stiffness coefficient dataset is loaded into the finite element model corresponding to the construction structure. Subsequent construction load conditions are input, and finite element analysis is performed to obtain the displacement data of each element in the finite element model, thereby realizing the prediction of structural deformation.

[0005] As a preferred approach, an active excitation scheme is optimized under the constraint of monitoring response sensitivity, specifically including the following schemes: Construct several simulated incentive schemes, combine all simulated incentive schemes into a population set, and set the maximum number of iterations; The amount of information corresponding to each simulated incentive scheme is used as the fitness of the simulated incentive scheme. Based on the fitness of the simulated incentive scheme, the population set is iteratively updated using a population optimization algorithm until the maximum number of iterations is reached. The simulated incentive scheme with the highest fitness is then output as the active incentive scheme. The amount of information corresponding to the simulated stimulus scheme is calculated as follows: The simulated excitation scheme is loaded into the finite element model, and the baseline response displacement data of each monitoring point is obtained by analysis. Then, the equivalent stiffness coefficients are traversed, and data perturbation is performed on the equivalent stiffness coefficients. The perturbation response displacement data of each monitoring point is then obtained by analysis, and the difference between the baseline response displacement data and the corresponding perturbation response displacement data of the monitoring point is calculated and recorded as the sensitivity. This process continues until the equivalent stiffness coefficients are traversed. Based on the sensitivity, a sensitivity matrix is ​​constructed. The sensitivity matrix stores the sensitivity obtained by the i-th monitoring point after performing data perturbation on the j-th equivalent stiffness coefficient in the i-th row and j-th column. The product of the transpose of the sensitivity matrix and the sensitivity matrix is ​​denoted as the information matrix, and the determinant of the information matrix is ​​denoted as the information content corresponding to the simulation excitation scheme.

[0006] As a preferred aspect, physical consistency correction is performed on the equivalent stiffness coefficient dataset by examining the difference between the finite element model's response to the active excitation scheme and the structural displacement data, resulting in an initial corrected equivalent stiffness coefficient dataset. This process specifically includes the following steps: The equivalent stiffness coefficient dataset is loaded into the finite element model, and the response of the finite element model to the active excitation scheme is analyzed to obtain the theoretical displacement data value. The difference between the theoretical displacement data value and the structural displacement data is calculated and denoted as the error vector. Then, the product of the transpose of the error vector and the error vector is calculated to obtain the displacement error loss value. The partial derivative of the displacement error loss value with respect to each equivalent stiffness coefficient in the equivalent stiffness coefficient dataset is calculated to obtain the displacement error gradient vector corresponding to all equivalent stiffness coefficients. Based on the displacement error gradient vector, the equivalent stiffness coefficients in the equivalent stiffness coefficient dataset are adjusted by the gradient descent method until the equivalent stiffness coefficient dataset converges, realizing physical consistency correction and obtaining the initial corrected equivalent stiffness coefficient dataset.

[0007] As a preferred aspect, the actual low-order modal data corresponding to the construction structure are determined based on the structural acceleration data of the active excitation operation response, specifically including the following steps: After aligning the structural acceleration data of all monitoring points according to time, perform an average summation operation to obtain motion response data. Then, process the motion response data using the random subspace method to obtain fixed frequencies and corresponding mode shapes. Arrange all fixed frequencies in descending order of frequency, and select the top E fixed frequencies and their corresponding mode shapes to form the actual low-order modal data corresponding to the construction structure; the E value is determined by the operator.

[0008] As a preferred aspect, global modal consistency correction is performed on the initial corrected equivalent stiffness coefficient dataset based on constraints from actual low-order modal data to obtain the final corrected equivalent stiffness coefficient dataset. This process specifically includes the following steps: The initial calibration equivalent stiffness coefficient dataset is loaded into the finite element model, and theoretical low-order modal data are obtained through finite element analysis. The difference between the theoretical low-order modal data and the actual low-order modal data is calculated and denoted as the modal difference vector. The product of the transpose of the modal difference vector and the modal difference vector is denoted as the modal error loss value. The partial derivative of the modal error loss value with respect to each equivalent stiffness coefficient in the initial calibration equivalent stiffness coefficient dataset is calculated to obtain the modal error gradient vector corresponding to all equivalent stiffness coefficients. Based on the modal error gradient vector, the equivalent stiffness coefficients in the initial calibration equivalent stiffness coefficient dataset are adjusted using the gradient descent method until the initial calibration equivalent stiffness coefficient dataset converges, thereby achieving global modal consistency correction and obtaining the final calibration equivalent stiffness coefficient dataset.

[0009] As a preferred approach, training the stiffness coefficient inversion network includes the following steps: Several stiffness coefficient inversion training samples are obtained, including structural displacement data of excitation schemes and corresponding responses. These samples are labeled using an equivalent stiffness coefficient dataset. All labeled stiffness coefficient inversion training samples are combined into a stiffness coefficient inversion training set, which is then used to train the stiffness coefficient inversion network.

[0010] As a preferred aspect, the swarm optimization algorithm used in the process of optimizing the active incentive scheme is the particle swarm optimization algorithm.

[0011] This invention also provides a structural deformation prediction system for the construction phase of building engineering, comprising: The active excitation response feedback module is used to perform active excitation operation on the construction structure through an active excitation scheme when the stiffness coefficient adjustment trigger condition is met, and to collect structural displacement data and structural acceleration data in response to the active excitation operation. The active excitation scheme is optimized under the constraint of the sensitivity of the monitoring response. The equivalent stiffness coefficient analysis module is used to concatenate the active excitation scheme with the corresponding structural displacement data to form excitation response data. The active excitation scheme includes several excitation parameter vectors. The excitation response data is fed into the stiffness coefficient inversion network for processing, and an equivalent stiffness coefficient dataset is output. The equivalent stiffness coefficient dataset includes the equivalent stiffness coefficients of each component in the finite element model corresponding to the construction structure. The equivalent stiffness coefficients represent the proportional relationship between the actual stiffness coefficients and the original design stiffness coefficients. The equivalent stiffness coefficient dataset is loaded into the finite element model. The difference between the finite element model's response to the active excitation scheme and the structural displacement data is used to perform physical consistency correction on the equivalent stiffness coefficient dataset to obtain the initial corrected equivalent stiffness coefficient dataset. The modal consistency correction module is used to determine the actual low-order modal data corresponding to the construction structure based on the structural acceleration data of the active excitation operation response. Based on the constraints of the actual low-order modal data, global modal consistency correction is performed on the initial correction equivalent stiffness coefficient dataset to obtain the final correction equivalent stiffness coefficient dataset. The structural deformation prediction module is used to load the final corrected equivalent stiffness coefficient dataset into the finite element model corresponding to the construction structure, input subsequent construction load conditions, perform finite element analysis operations, obtain the displacement data of each element in the finite element model, and realize structural deformation prediction.

[0012] The present invention has the following advantages: This invention applies active excitation obtained through particle swarm optimization to the structure during the construction phase to acquire response data with high sensitivity to the stiffness coefficient of the components. A feedforward neural network is used to perform a preliminary inversion of the equivalent stiffness coefficient, followed by physical consistency correction based on displacement error using a differentiable finite element model. Furthermore, the random subspace method is employed to extract the actual low-order frequencies and mode shapes of the construction structure from the acceleration signal of the active excitation response, and these are used as global modal constraints to further correct the initially corrected stiffness coefficient. This results in obtaining a true, global, and consistent stiffness coefficient, significantly improving the uniqueness and accuracy of the stiffness coefficient inversion. This ensures that the finite element model with corrected stiffness coefficients can accurately predict the deformation of the subsequent construction structure. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the structural deformation prediction system for the construction phase of building engineering, as used in an embodiment of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0015] Example 1: A method for predicting structural deformation during the construction phase of building engineering, comprising: During the construction phase, different construction projects have corresponding construction structures, such as the tower columns of a bridge project or the scaffolding structure of a residential building project. Displacement and acceleration sensors installed at various monitoring points on the construction structure continuously acquire structural displacement and acceleration data. It's important to note that these different monitoring points are determined by the operators based on the finite element model of the construction structure. They are generally evenly distributed within the installable area of ​​the construction structure, and the number of monitoring points is determined by the dimensions of the construction structure. The structural displacement data is the displacement distance data at the corresponding monitoring point location, and the structural acceleration data is a continuous acceleration signal data that varies over time. During construction... Due to operational and environmental influences, self-vibration can occur. Therefore, structural displacement data is generally collected after steady-state displacement. Structural displacement data can describe the structural deformation of the construction structure under construction operations. During the construction phase, in order to ensure the normal progress of construction, subsequent construction load conditions are usually input into the finite element model corresponding to the construction structure, and the subsequent structural deformation is predicted based on finite element analysis. Here, structural deformation refers to the displacement data of each element in the finite element model. The finite element analysis operation uses existing tools, such as ETABS, SAP2000, PKPM, etc. The finite element model needs to input the stiffness coefficient data and boundary conditions of each element. When the stiffness coefficient adjustment trigger condition is met, an active excitation operation is performed on the construction structure through an active excitation scheme, and structural displacement data and structural acceleration data in response to the active excitation operation are collected. The active excitation scheme is optimized under the constraint of the sensitivity of the monitoring response. The stiffness coefficient adjustment trigger condition can be the completion of a key process, or the difference between the actual monitored structural displacement data and the structural deformation predicted by the finite element model exceeds expectations. When the stiffness coefficient adjustment trigger condition is met, it means that the stiffness coefficient of the construction structure has changed under the influence of construction process or environment (such as temperature change). The structural deformation prediction by the finite element model may be inaccurate, affecting the judgment of the construction situation. Therefore, it is necessary to perform stiffness coefficient adjustment. A controllable construction load is applied to the construction structure through a manually set active excitation scheme, which can generate sufficient and directional independent response in key components, significantly improving the sensitivity of the monitoring point to the stiffness coefficient. The obtained structural displacement data and structural acceleration data also have a high information content on the stiffness coefficient, which can greatly reduce the non-uniqueness of stiffness coefficient inversion. The excitation scheme can be implemented by applying controllable loads to different positions through a servo loader. Active excitation schemes are concatenated with corresponding structural displacement data to form excitation response data. The active excitation scheme includes several excitation parameter vectors, such as force application location, direction, magnitude, and frequency. The excitation response data is then processed in a stiffness coefficient inversion network to output an equivalent stiffness coefficient dataset. This dataset includes the equivalent stiffness coefficients of each component in the finite element model corresponding to the construction structure. The equivalent stiffness coefficients represent the proportional relationship between the actual stiffness coefficients and the original design stiffness coefficients. The product of the equivalent stiffness coefficients and the original design stiffness coefficients is the stiffness coefficient set in the finite element model corresponding to the construction structure. By dividing the construction structure into several components and parameterizing the stiffness coefficients of these components in the finite element model using equivalent stiffness coefficients, the overall stiffness change is expressed in a low-dimensional way. This reduces the number of parameters, ensures the solvability of the identification problem, and reflects the main trend of actual component stiffness changes. The original design stiffness coefficients are determined by the operator based on... Based on design drawings and expert calculations, the stiffness coefficient inversion network is set up using a feedforward neural network. It generally includes an input layer, several hidden layers, and an output layer. To ensure the uniformity of the excitation response data, the active excitation scheme will fix the number of excitation parameter vectors, usually 10. If the number of excitation parameter vectors in the active excitation scheme optimized under the constraint of monitoring response sensitivity is less than 10, the remaining positions are set to 0. The equivalent stiffness coefficient dataset is loaded into the finite element model. The difference between the finite element model's response to the active excitation scheme and the structural displacement data is used to perform physical consistency correction on the equivalent stiffness coefficient dataset to obtain the initial corrected equivalent stiffness coefficient dataset. Since the equivalent stiffness coefficient dataset is obtained through the stiffness coefficient inversion network, it is a kind of data fitting. Through the physical consistency correction operation, the local, fast, and possibly incomplete physical initial stiffness estimate given by deep learning in real time can be optimized in small steps to satisfy the finite element dynamic equation and physical constraints. Based on the structural acceleration data of the active excitation operation response, the actual low-order modal data corresponding to the construction structure are determined. This actual low-order modal data includes low-order frequency data and corresponding mode shape data. Based on the constraints of this actual low-order modal data, global modal consistency correction is performed on the initial corrected equivalent stiffness coefficient dataset to obtain the final corrected equivalent stiffness coefficient dataset. It should be noted that the construction structure possesses corresponding fixed frequencies and mode shapes. This is because structural dynamics are determined by solving a series of eigenvalues ​​from the equations of motion. A set of fixed frequencies and mode shapes represents a natural oscillation mode, which is a comprehensive representation of the overall structural stiffness and overall mass distribution. In a fixed frequency range, the frequencies with the smallest values ​​are called lower-order natural frequencies (such as first, second, and third orders). The first-order frequency reflects the average level of overall bending stiffness, the second-order frequency often reflects symmetrical / anti-symmetrical bending modes, and the third-order frequency may reflect overall torsional stiffness. Lower-order modes are most sensitive to stiffness and are therefore most suitable as "stiffness constraints" during the construction phase. A mode shape is a "displacement vector field" that represents the characteristic deformation mode of a structure at a certain natural frequency, indicating "what shape this mode mainly has" (such as overall oscillation in the X direction, oscillation in the Y direction, torsion, etc.). The mode shape specifically represents the response of a fixed frequency at different monitoring points. The final corrected equivalent stiffness coefficient dataset is loaded into the finite element model corresponding to the construction structure. Subsequent construction load conditions are input, and finite element analysis is performed to obtain the displacement data of each element in the finite element model, thereby realizing the prediction of structural deformation. This invention applies active excitation obtained through particle swarm optimization to the structure during the construction phase to acquire response data with high sensitivity to the stiffness coefficient of the components. A feedforward neural network is used to perform a preliminary inversion of the equivalent stiffness coefficient, followed by physical consistency correction based on displacement error using a differentiable finite element model. Furthermore, the random subspace method is employed to extract the actual low-order frequencies and mode shapes of the construction structure from the acceleration signal of the active excitation response, and these are used as global modal constraints to further correct the initially corrected stiffness coefficient. This results in obtaining a true, global, and consistent stiffness coefficient, significantly improving the uniqueness and accuracy of the stiffness coefficient inversion. This ensures that the finite element model with corrected stiffness coefficients can accurately predict the deformation of the subsequent construction structure.

[0016] The active stimulus scheme is optimized under the constraint of monitoring response sensitivity, and includes the following schemes: Several simulated stimulus schemes are constructed, each containing several simulated stimulus parameter vectors. These vectors include terms such as force application location, force application direction, force magnitude, and force application frequency. Each term in the simulated stimulus parameter vector is randomly assigned a value under constraints; for example, the force application location must be on the surface of the construction structure, the application direction must ensure the force acts on the construction structure, and the force magnitude must be within a controllable range and not affect the construction structure. The number of simulated stimulus parameter vectors in each scheme is fixed, but during optimization, there may be a case where all stimulus parameter vectors are 0, used to represent active stimulus schemes with different numbers of stimulus parameter vectors. All simulated stimulus schemes are grouped into a population set, and a maximum number of iterations is set. The amount of information corresponding to each simulated incentive scheme is used as the fitness of the simulated incentive scheme. Based on the fitness of the simulated incentive scheme, the population set is iteratively updated using the particle swarm optimization algorithm until the maximum number of iterations is reached, and the simulated incentive scheme with the highest fitness is output as the active incentive scheme. The amount of information corresponding to the simulated stimulus scheme is calculated as follows: The simulated excitation scheme is loaded into the finite element model, and the baseline response displacement data of each monitoring point is obtained by analysis. Then, the equivalent stiffness coefficients are traversed, and data perturbation is performed on the equivalent stiffness coefficients. Data perturbation generally refers to increasing the equivalent stiffness coefficient by 1%. The perturbation response displacement data of each monitoring point is then obtained by analysis, and the difference between the baseline response displacement data of the monitoring point and the corresponding perturbation response displacement data is calculated, which is denoted as sensitivity. Sensitivity characterizes the correlation between the change of the equivalent stiffness coefficient and the change of structural deformation. This process continues until the equivalent stiffness coefficients are traversed. Based on the sensitivity, a sensitivity matrix is ​​constructed. The sensitivity matrix stores the sensitivity obtained by the i-th monitoring point after performing data perturbation on the j-th equivalent stiffness coefficient in the i-th row and j-th column. The product of the transpose of the sensitivity matrix and the sensitivity matrix is ​​denoted as the information matrix. The information matrix stores the linear correlation of the response directions of different equivalent rigid parameters in the sensitivity matrix. The smaller the linear correlation of the response directions of different equivalent rigid parameters, the easier it is to invert the equivalent rigid parameters. The determinant of the information matrix is ​​denoted as the information content corresponding to the simulation excitation scheme. The determinant of the information matrix is ​​the product of all the eigenvalues ​​of the information matrix. Each column of the sensitivity matrix represents the response direction of an equivalent rigid parameter. Each column vector is regarded as the mapping direction of the equivalent rigid parameter space in the monitoring response space. When these directions are linearly correlated, the parameter identifiability decreases. The determinant of the information matrix is ​​equal to the square of the volume of the parallel polyhedron formed by these sensitivity directions. When the determinant is 0, the volume of the parallel polyhedron is zero, indicating the existence of unidentifiable parameter directions. When the determinant is large, it indicates that the sensitivity directions are linearly independent and well distributed, which is beneficial to improving the accuracy of equivalent rigid parameter inversion.

[0017] By analyzing the discrepancy between the finite element model's response to the active excitation scheme and the structural displacement data, physical consistency correction is performed on the equivalent stiffness coefficient dataset to obtain an initial corrected equivalent stiffness coefficient dataset. This process includes the following steps: The equivalent stiffness coefficient dataset is loaded into the finite element model. The response of the finite element model to the active excitation scheme is analyzed to obtain theoretical displacement data values. The difference between the theoretical displacement data values ​​and the structural displacement data is calculated and denoted as the error vector. Then, the product of the transpose of the error vector and the error vector itself is calculated, which is the sum of squares of the errors corresponding to all monitoring points, to obtain the displacement error loss value. The partial derivative of the displacement error loss value with respect to each equivalent stiffness coefficient in the equivalent stiffness coefficient dataset is calculated to obtain the displacement error gradient vector corresponding to all equivalent stiffness coefficients. Based on the displacement error gradient vector, the equivalent stiffness coefficients in the equivalent stiffness coefficient dataset are adjusted using the gradient descent method until the equivalent stiffness coefficient dataset converges. Physical consistency correction is performed to obtain an initial corrected equivalent stiffness coefficient dataset. It should be noted that since the analysis and calculation of the finite element model can be regarded as a mapping from stiffness coefficients to displacement response, the simplest form is KU=F, where K is the stiffness coefficient matrix, U is the displacement data, and F is the load. All of these are differentiable in the finite element analysis process, which is easily achieved in differentiable programming. Therefore, the partial derivatives of the displacement error loss value with respect to each equivalent stiffness coefficient in the equivalent stiffness coefficient dataset can be calculated using existing functions to obtain the displacement error gradient vector corresponding to all equivalent stiffness coefficients. This allows for the correction of the equivalent stiffness coefficient dataset, and convergence generally occurs when the changes in the equivalent stiffness coefficient dataset in the most recent iteration meet expectations.

[0018] The actual low-order modal data corresponding to the construction structure are determined based on the structural acceleration data of the active excitation operation response, specifically including the following steps: After aligning the structural acceleration data from all monitoring points according to time, an average summation operation is performed to obtain motion response data. This data is then processed using the random subspace method to obtain fixed frequencies and corresponding mode shapes. The random subspace method processes the motion response data by constructing a time-dependent Hankel matrix and extracting the state-space structure from the output sequence through orthogonal decomposition or projection operations. Subsequently, singular value decomposition is performed on the constructed system projection matrix to extract the stable state subspace, from which the system's state matrix and output matrix are obtained. By performing eigenvalue decomposition on the state matrix, the system's inherent characteristics, including natural frequencies and damping ratios, can be directly obtained. Simultaneously, multiplying the output matrix by the eigenvector yields the corresponding mode shapes. Since the entire process does not rely on external excitation information and only relies on response data to extract stable modes, it can effectively identify the low-order natural frequencies and mode shapes of the structure under actual operating conditions. This description is only a brief overview; for specific operational methods, please refer to existing technologies, which will not be elaborated upon here. Arrange all fixed frequencies in descending order of frequency, and select the top E fixed frequencies and their corresponding mode shapes to form the actual low-order modal data corresponding to the construction structure; the E value is determined by the operator.

[0019] Based on the constraints of actual low-order modal data, global modal consistency correction is performed on the initial corrected equivalent stiffness coefficient dataset to obtain the final corrected equivalent stiffness coefficient dataset. The specific steps include the following: The initial calibration equivalent stiffness coefficient dataset is loaded into the finite element model, and theoretical low-order modal data is obtained through finite element analysis. It should be noted that finite element analysis can actually be understood as solving the equation (K–ω²M)φ=0, where K is the stiffness coefficient matrix, ω is the fixed frequency, M is the mass matrix, and φ is the mode shape. The theoretical low-order modal data is composed of the first E fixed frequencies and their corresponding mode shapes after being arranged in descending order of frequency from the obtained fixed frequencies. The difference between theoretical low-order modal data and actual low-order modal data is calculated and denoted as the modal difference vector. The product of the transpose of the modal difference vector and the modal difference vector is denoted as the modal error loss value, which is the sum of squares of each term in the modal difference vector. The partial derivative of the modal error loss value with respect to each equivalent stiffness coefficient in the initial calibration equivalent stiffness coefficient dataset is calculated to obtain the modal error gradient vector corresponding to all equivalent stiffness coefficients. Based on the modal error gradient vector, the equivalent stiffness coefficients in the initial calibration equivalent stiffness coefficient dataset are adjusted using the gradient descent method until the initial calibration equivalent stiffness coefficient dataset converges, achieving global modal consistency correction and obtaining the final calibration equivalent stiffness coefficient dataset. Similarly, the solution for modal data in finite element analysis is also differentiable, so parameter correction can also be performed using partial derivatives.

[0020] Training the stiffness coefficient inversion network involves the following steps: Several stiffness coefficient inversion training samples were obtained. These samples included excitation schemes and corresponding structural displacement data. These samples were acquired by operators through actual excitation experiments on different construction structures and labeled using an equivalent stiffness coefficient dataset. This equivalent stiffness coefficient dataset consisted of data measured by operators based on the construction structures corresponding to the stiffness coefficient inversion training samples. All labeled stiffness coefficient inversion training samples were combined into a stiffness coefficient inversion training set. This training set was then used to analyze the stiffness coefficient system. The stiffness coefficient inversion network is trained. During training, the difference between the predicted output of the stiffness coefficient inversion network and the equivalent stiffness coefficient dataset labeled with the stiffness coefficient inversion training samples is used to construct the loss value through the method of MSE. The parameters of the stiffness coefficient inversion network are updated iteratively in the direction of the minimum loss value using the gradient descent method. It is judged whether the accuracy of the stiffness coefficient inversion network meets the expectations. If the accuracy of the stiffness coefficient inversion network meets the expectations, the trained stiffness coefficient inversion network is output. Otherwise, the stiffness coefficient inversion network is trained again using the stiffness coefficient inversion training set.

[0021] Example 2: A structural deformation prediction system for the construction phase of building engineering, such as... Figure 1 As shown, it includes: The active excitation response feedback module is used to perform active excitation operations on the construction structure through an active excitation scheme when the stiffness coefficient adjustment trigger condition is met. It collects structural displacement and acceleration data in response to the active excitation operation. The active excitation scheme is optimized under the constraint of monitoring response sensitivity. The stiffness coefficient adjustment trigger condition can be the completion of a key construction process, or the difference between the actual monitored structural displacement data and the structural deformation predicted by the finite element model exceeding expectations. When the stiffness coefficient adjustment trigger condition is met, it indicates that under the influence of construction processes or environmental factors (such as temperature changes), the construction structure's response to... Since the stiffness coefficient has changed, predicting structural deformation using the finite element model may become inaccurate, affecting the judgment of the construction situation. Therefore, it is necessary to adjust the stiffness coefficient. By using an actively set excitation scheme to apply controllable construction loads to the construction structure, sufficient and directional independent responses can be generated in key components, which significantly improves the sensitivity of the monitoring points to the stiffness coefficient. The obtained structural displacement and acceleration data also have a high information content on the stiffness coefficient, which can greatly reduce the non-uniqueness of stiffness coefficient inversion. The excitation scheme here can be achieved by applying controllable loads to different positions through a servo loader. The equivalent stiffness coefficient analysis module is used to concatenate the active excitation scheme with the corresponding structural displacement data to form excitation response data. The active excitation scheme includes several excitation parameter vectors, which include force application location, direction, magnitude, and frequency. The excitation response data is then fed into the stiffness coefficient inversion network for processing, outputting an equivalent stiffness coefficient dataset. This dataset includes the equivalent stiffness coefficients for each component in the finite element model corresponding to the construction structure. The equivalent stiffness coefficients represent the proportional relationship between the actual stiffness coefficients and the original design stiffness coefficients. The product of the equivalent stiffness coefficients and the original design stiffness coefficients is the stiffness coefficient set in the finite element model corresponding to the construction structure. The original design stiffness coefficients are calculated by operators based on design drawings and expert experience. The stiffness coefficient inversion network is based on a feedforward neural network and generally includes an input layer. The system consists of a hidden layer and an output layer. To ensure the uniformity of the excitation response data, the active excitation scheme fixes the number of excitation parameter vectors, typically 10. If the number of excitation parameter vectors in the optimized active excitation scheme is less than 10 under the constraint of monitoring response sensitivity, the remaining positions are set to 0. The equivalent stiffness coefficient dataset is loaded into the finite element model. The physical consistency correction is performed on the equivalent stiffness coefficient dataset based on the difference between the finite element model's response to the active excitation scheme and the structural displacement data, resulting in an initially corrected equivalent stiffness coefficient dataset. This is because the equivalent stiffness coefficient dataset is obtained through stiffness coefficient inversion network processing, which is a data fitting process. Through the physical consistency correction operation, the local, fast, and potentially incomplete physical stiffness initial estimate given by deep learning in real time can be optimized in small steps to satisfy the finite element dynamic equations and physical constraints. The modal consistency correction module is used to determine the actual low-order modal data corresponding to the construction structure based on the structural acceleration data of the active excitation operation response. The actual low-order modal data includes low-order frequency data and corresponding mode shape data. Based on the constraints of the actual low-order modal data, global modal consistency correction is performed on the initial corrected equivalent stiffness coefficient dataset to obtain the final corrected equivalent stiffness coefficient dataset. It should be noted that the construction structure has corresponding fixed frequencies and mode shapes. This is because the structural dynamics are determined by solving a series of eigenvalues ​​from the equations of motion. A set of fixed frequencies and mode shapes represents a natural oscillation mode, which are related to the overall stiffness of the structure and its overall dynamic range. The overall mass distribution is reflected in the lowest frequencies at a fixed frequency, which are called the lower-order natural frequencies (such as first, second, and third orders). The first-order frequency reflects the average level of the overall bending stiffness, the second-order frequency often reflects symmetrical / anti-symmetrical bending modes, and the third-order frequency may reflect the overall torsional stiffness. The lower-order modes are most sensitive to stiffness and are therefore most suitable as "stiffness constraints" during the construction phase. The mode shape is a "displacement vector field" that represents the characteristic deformation mode of the structure at a certain natural frequency, indicating "what shape this mode mainly has" (such as overall oscillation in the X direction, oscillation in the Y direction, torsion, etc.), and the mode shape specifically represents the response of a fixed frequency at different monitoring points. The structural deformation prediction module is used to load the final corrected equivalent stiffness coefficient dataset into the finite element model corresponding to the construction structure, input subsequent construction load conditions, perform finite element analysis operations, obtain the displacement data of each element in the finite element model, and realize structural deformation prediction.

[0022] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A structural deformation prediction method for a construction phase of a building project, characterized by, include: When the stiffness coefficient adjustment trigger condition is met, an active excitation operation is performed on the construction structure through an active excitation scheme, and structural displacement data and structural acceleration data in response to the active excitation operation are collected. The active excitation scheme is optimized under the constraint of the sensitivity of the monitoring response. Active excitation schemes are concatenated with corresponding structural displacement data to form excitation response data. The active excitation scheme includes an excitation parameter vector. The excitation response data is then fed into a stiffness coefficient inversion network for processing, outputting an equivalent stiffness coefficient dataset. This dataset includes the equivalent stiffness coefficients of each component in the finite element model corresponding to the construction structure. The equivalent stiffness coefficients characterize the proportional relationship between the actual stiffness coefficients and the original design stiffness coefficients. The equivalent stiffness coefficient dataset is then loaded into the finite element model. Physical consistency correction is performed on the equivalent stiffness coefficient dataset based on the difference between the finite element model's response to the active excitation scheme and the structural displacement data, resulting in an initial corrected equivalent stiffness coefficient dataset. Based on the structural acceleration data of the active excitation operation response, the actual low-order modal data corresponding to the construction structure are determined. Based on the constraints of the actual low-order modal data, global modal consistency correction is performed on the initial correction equivalent stiffness coefficient dataset to obtain the final correction equivalent stiffness coefficient dataset. The final corrected equivalent stiffness coefficient dataset is loaded into the finite element model corresponding to the construction structure. Subsequent construction load conditions are input, and finite element analysis is performed to obtain the displacement data of each element in the finite element model, thereby realizing the prediction of structural deformation. The active stimulus scheme is optimized under the constraint of monitoring response sensitivity, and includes the following schemes: Construct a simulated incentive scheme, combine all simulated incentive schemes into a population set, and set the maximum number of iterations; The amount of information corresponding to each simulated incentive scheme is used as the fitness of the simulated incentive scheme. Based on the fitness of the simulated incentive scheme, the population set is iteratively updated using a population optimization algorithm until the maximum number of iterations is reached. The simulated incentive scheme with the highest fitness is then output as the active incentive scheme. The amount of information corresponding to the simulated stimulus scheme is calculated as follows: The simulated excitation scheme is loaded into the finite element model, and the reference response displacement data of each monitoring point is obtained by analysis. Then, the equivalent stiffness coefficient is traversed, and data perturbation is performed on the equivalent stiffness coefficient. The perturbation response displacement data of each monitoring point is then obtained by analysis, and the difference between the reference response displacement data of the monitoring point and the corresponding perturbation response displacement data is calculated and denoted as sensitivity. Until the equivalent stiffness coefficients are traversed, a sensitivity matrix is ​​constructed based on the sensitivity. The i-th row and j-th column of the sensitivity matrix stores the sensitivity obtained by the i-th monitoring point after data perturbation is performed on the j-th equivalent stiffness coefficient. The product of the transpose of the sensitivity matrix and the sensitivity matrix is ​​denoted as the information matrix, and the determinant of the information matrix is ​​denoted as the information content corresponding to the simulation excitation scheme.

2. The structural deformation prediction method for construction phase of building engineering according to claim 1, characterized in that, By analyzing the discrepancy between the finite element model's response to the active excitation scheme and the structural displacement data, physical consistency correction is performed on the equivalent stiffness coefficient dataset to obtain an initial corrected equivalent stiffness coefficient dataset. This process includes the following steps: The equivalent stiffness coefficient dataset is loaded into the finite element model, and the response of the finite element model to the active excitation scheme is analyzed to obtain the theoretical displacement data value. The difference between the theoretical displacement data value and the structural displacement data is calculated and denoted as the error vector. Then, the product of the transpose of the error vector and the error vector is calculated to obtain the displacement error loss value. The partial derivative of the displacement error loss value with respect to each equivalent stiffness coefficient in the equivalent stiffness coefficient dataset is calculated to obtain the displacement error gradient vector corresponding to all equivalent stiffness coefficients. Based on the displacement error gradient vector, the equivalent stiffness coefficients in the equivalent stiffness coefficient dataset are adjusted by the gradient descent method until the equivalent stiffness coefficient dataset converges, realizing physical consistency correction and obtaining the initial corrected equivalent stiffness coefficient dataset.

3. The structural deformation prediction method for construction phase of building engineering according to claim 2, characterized in that, The actual low-order modal data corresponding to the construction structure are determined based on the structural acceleration data of the active excitation operation response, specifically including the following steps: After aligning the structural acceleration data of all monitoring points according to time, perform an average summation operation to obtain motion response data. Then, process the motion response data using the random subspace method to obtain fixed frequencies and corresponding mode shapes. Arrange all fixed frequencies in descending order of frequency, and select the top E fixed frequencies and their corresponding mode shapes to form the actual low-order modal data corresponding to the construction structure.

4. The structural deformation prediction method for construction phase of building engineering according to claim 3, characterized in that, Based on the constraints of actual low-order modal data, global modal consistency correction is performed on the initial corrected equivalent stiffness coefficient dataset to obtain the final corrected equivalent stiffness coefficient dataset. The specific steps include the following: The initial calibration equivalent stiffness coefficient dataset is loaded into the finite element model, and theoretical low-order modal data are obtained through finite element analysis. The difference between the theoretical low-order modal data and the actual low-order modal data is calculated and denoted as the modal difference vector. The product of the transpose of the modal difference vector and the modal difference vector is denoted as the modal error loss value. The partial derivative of the modal error loss value with respect to each equivalent stiffness coefficient in the initial calibration equivalent stiffness coefficient dataset is calculated to obtain the modal error gradient vector corresponding to all equivalent stiffness coefficients. Based on the modal error gradient vector, the equivalent stiffness coefficients in the initial calibration equivalent stiffness coefficient dataset are adjusted using the gradient descent method until the initial calibration equivalent stiffness coefficient dataset converges, thereby achieving global modal consistency correction and obtaining the final calibration equivalent stiffness coefficient dataset.

5. The structural deformation prediction method for construction phase of building engineering according to claim 4, characterized in that, Training the stiffness coefficient inversion network involves the following steps: Acquire stiffness coefficient inversion training samples, which include structural displacement data of excitation schemes and corresponding responses, and label them using an equivalent stiffness coefficient dataset. Combine all labeled stiffness coefficient inversion training samples into a stiffness coefficient inversion training set, and train the stiffness coefficient inversion network using the stiffness coefficient inversion training set.

6. The structural deformation prediction method for construction phase of building engineering according to claim 5, wherein, The swarm optimization algorithm used in the process of optimizing the active incentive scheme is the particle swarm optimization algorithm.

7. A structural deformation prediction system for the construction phase of building engineering, characterized in that, The system applies a structural deformation prediction method for the construction phase of building engineering as described in any one of claims 1-6, including: The active excitation response feedback module is used to perform active excitation operations on the construction structure through an active excitation scheme when the stiffness coefficient adjustment trigger condition is met, and to collect structural displacement data and structural acceleration data in response to the active excitation operation. The active excitation scheme is optimized under the constraint of the sensitivity of the monitoring response. The equivalent stiffness coefficient analysis module is used to concatenate the active excitation scheme with the corresponding structural displacement data to form excitation response data. The active excitation scheme includes an excitation parameter vector. The excitation response data is fed into the stiffness coefficient inversion network for processing, and an equivalent stiffness coefficient dataset is output. The equivalent stiffness coefficient dataset includes the equivalent stiffness coefficients of each component in the finite element model corresponding to the construction structure. The equivalent stiffness coefficients represent the proportional relationship between the actual stiffness coefficients and the original design stiffness coefficients. The equivalent stiffness coefficient dataset is loaded into the finite element model. The difference between the finite element model's response to the active excitation scheme and the structural displacement data is used to perform physical consistency correction on the equivalent stiffness coefficient dataset to obtain the initial corrected equivalent stiffness coefficient dataset. The modal consistency correction module is used to determine the actual low-order modal data corresponding to the construction structure based on the structural acceleration data of the active excitation operation response. Based on the constraints of the actual low-order modal data, global modal consistency correction is performed on the initial correction equivalent stiffness coefficient dataset to obtain the final correction equivalent stiffness coefficient dataset. The structural deformation prediction module is used to load the final corrected equivalent stiffness coefficient dataset into the finite element model corresponding to the construction structure, input subsequent construction load conditions, perform finite element analysis operations, obtain the displacement data of each element in the finite element model, and realize structural deformation prediction.

Images