An ai-driven bridge parameter hierarchical decoupling adaptive correction method

Abstract

The application discloses a bridge parameter layered decoupling adaptive correction method based on AI driving and belongs to the technical field of bridge construction control; the method is divided into four layers of identification processes according to the physical sources of effects; the first layer utilizes the instantaneous loading monitoring data of a hanging basket to combine an agent model and MCMC sampling to identify the instantaneous stiffness parameters of components; the second layer decouples to obtain the temperature expansion coefficient based on the day-night temperature-displacement time sequence monitoring; the third layer carries out isothermal time screening and temperature effect deduction on a long time sequence monitoring data set to construct a creep likelihood function to solve the ultimate creep coefficient; the fourth layer identifies the stiffness of connecting components based on the response of the eccentric working condition of the hanging basket; the method is synchronously updated to a finite element model to realize bridge linear prediction and construction closed-loop adaptive regulation and control; the application decouples and identifies different physical effect parameters respectively, reduces the mutual interference between parameters, improves the model parameter identification accuracy, effectively reduces the construction linear prediction deviation and is suitable for the cantilever pouring construction control of a large-span bridge.

Description

Technical Field

[0001] This invention belongs to the field of bridge construction control technology, and in particular relates to an AI-driven hierarchical decoupling adaptive correction method for bridge parameters. Background Technology

[0002] The finite element model (FEM) is a core tool for simulation analysis and alignment control during bridge construction. An accurate FEM model depends not only on correct structural geometry modeling and load input, but also on the accurate values ​​of key physical parameters within the model. However, many key parameters in the bridge structure FEM model—including the elastic modulus of concrete, the self-weight coefficient of components, the creep coefficient, the coefficient of thermal expansion, and boundary condition parameters such as support constraint stiffness and temporary consolidation stiffness—exhibit significant uncertainties during the construction phase and continue to evolve with construction progress, environmental changes, and material age. Traditional methods typically determine these parameters based on design specification recommendations or finite samples from material tests, treating them as fixed constants throughout the construction process. This leads to a gradual, systematic deviation between simulation results and actual field measurements, severely limiting the accuracy of construction alignment control.

[0003] To overcome the aforementioned problems, academia and engineering have developed finite element model correction techniques. For example, some methods employ finite element model correction based on inverse substructures, establishing the objective function by decomposing the overall structural test modes into independent substructure test modes; other methods identify the actual stiffness of continuous beam bridges based on static load tests; and still others use genetic algorithms to optimize finite element model parameters. In the field of digital twins, dynamic update methods for bridge digital twins based on multi-fidelity learning and sequential Bayesian inference have also emerged. However, most existing AI methods unify the identification of various physical parameters within the same framework, failing to address the problem of non-unique identification caused by parameter response coupling—AI models can fit data, but cannot distinguish which changes are caused by the elastic modulus and which by creep. Existing technologies suffer from the following three fundamental defects: (i) Parameter coupling leads to non-unique identification results: Existing methods usually package all parameters to be identified (elastic modulus, creep coefficient, temperature expansion coefficient, boundary constraint stiffness, etc.) into the same objective function for unified optimization. Since different parameters may produce similar sensitivity patterns to the response of the same structure—for example, the effects of reducing the elastic modulus and reducing the cross-sectional stiffness on the displacement response are almost the same—mathematically, there are a large number of local optima. The identification results heavily depend on the initial assumed values ​​and lack uniqueness guarantees.

[0004] (ii) The time-domain characteristics of responses from different parameters are confused: the elastic modulus affects the instantaneous elastic response—deformation that can be completed within minutes to hours after the formwork is moved forward or concrete is poured; while creep affects the long-term time-varying response—slow deformation that lasts for weeks or even months. The temperature effect is superimposed on both, continuously interfering with the monitoring data in a diurnal fluctuation manner. Mixing the instantaneous response, temperature response, and long-term time-varying response in the same dataset for identification is equivalent to using the same ruler to measure three physical phenomena at three different time scales, and the identification results will inevitably be mutually contaminated.

[0005] (III) The handling of temperature effects faces the dilemma of "correcting uncertainty with uncertainty": When dealing with the effects of temperature, existing technologies usually rely on the theoretical or empirical values ​​of material thermal parameters (thermal conductivity, specific heat capacity, etc.) to establish temperature correction models. However, these thermal parameters themselves also have significant uncertainties. Using uncertain parameters to correct equally uncertain monitoring data not only fails to improve the identification accuracy, but may also introduce additional systematic errors.

[0006] Therefore, there is an urgent need for an adaptive correction method for bridge parameters based on hierarchical decoupling of monitoring data. This method can overcome the shortcomings of existing technologies, such as multi-parameter coupling leading to non-unique identification, mutual confusion of time-domain characteristics of different parameter responses, and reliance on uncertain thermal parameters for temperature effect processing. This method can achieve high-precision online identification of key physical parameters of bridges and dynamic updating of finite element models. Summary of the Invention

[0007] The purpose of this invention is to provide an AI-driven hierarchical decoupling adaptive correction method for bridge parameters, addressing the problems of non-unique identification due to multi-parameter coupling, confusion between parameters with different time-domain characteristics, and reliance on uncertain thermal parameters for temperature effect processing in existing bridge finite element model correction methods. This invention provides an AI-driven hierarchical decoupling adaptive correction method for bridge parameters to achieve high-precision, high-reliability online identification of bridge structural parameters and continuous dynamic updating of the finite element model. Instead of identifying various parameters together, it categorizes them into different levels according to their physical nature and response time-domain characteristics. Each level uses clean monitoring data matching the time-domain characteristics of its parameters for independent identification. The identification results of the upper level provide prior constraints for the lower level, thus fundamentally solving the problems of parameter coupling and time-domain confusion. Finally, the bridge structural parameters identified after hierarchical decoupling are substituted into the finite element model for forward modeling, ensuring continuous convergence of the error between the theoretical calculation values ​​and the measured monitoring data. This achieves synchronous evolution of the mechanical behavior of the virtual digital bridge and the physical bridge, significantly improving the calculation accuracy of the finite element model and the reliability of parameter identification.

[0008] To achieve the above objectives, the following technical solution is adopted: This invention provides an AI-driven hierarchical decoupling adaptive correction method for bridge parameters, comprising the following steps: S1. The parameters to be identified in the bridge finite element model are divided into four independent levels according to the response time domain characteristics and spatial influence range. The identification results of each level serve as the prior known conditions for the identification of the next level. The first level is the stiffness parameter including the elastic modulus of concrete and the self-weight coefficient of the component. The second level is the temperature expansion coefficient. The third level is the ultimate value of the creep coefficient. The fourth level is the boundary constraint stiffness including the horizontal shear stiffness of the support and the lateral stiffness of the pier. S2. Identify the stiffness parameters of the first layer, collect instantaneous load response data under construction conditions, construct a local finite element sub-model, embed the Gaussian process surrogate model or Bayesian neural network as the AI ​​replacement model of the local finite element sub-model into the Bayesian inference framework, input the measured construction load, and identify the posterior probability distribution of the elastic modulus and self-weight coefficient. S3. Identify the temperature expansion coefficient of the second layer, establish a linear regression model, and solve for the optimal estimate and confidence interval of the temperature expansion coefficient. S4. Identify the creep coefficient of the third layer, construct a time-temperature-displacement dataset, and construct an isothermal time series based on the intelligent filtering rules of isothermal conditions; use the elastic modulus obtained from the first layer identification as a known parameter to perform Bayesian inference and output the posterior probability distribution of the ultimate value of the creep coefficient. S5. Identify the stiffness of the fourth-layer boundary constraints, construct a weighted objective function of displacement and shear deformation, and use an optimization algorithm to solve for the optimal estimate of the constraint stiffness. S6. Inject the identification results from steps S2 to S5 into the finite element model of the entire bridge to update the virtual digital bridge, and predict the alignment of subsequent construction segments and calculate control commands.

[0009] Preferably, in step S2, the process of acquiring the instantaneous load response data includes: The structural displacement and strain data were collected before and after three working conditions: concrete pouring, formwork relocation, and prestressing tensioning. The displacement and strain increments of the final value relative to the initial value were calculated to obtain the instantaneous load response data. The initial and final values ​​of each measuring point were read within a period when the temperature change of the beam did not exceed ±0.5℃.

[0010] Preferably, in step S2, the construction process of the AI ​​alternative model includes: Using a small number of finite element forward modeling results from the local finite element sub-model as training samples, a Gaussian process surrogate model or a Bayesian neural network is trained to construct a nonlinear mapping between the parameters to be identified and the structural response; the AI ​​substitution model replaces the forward modeling calculation of the local finite element sub-model within the Markov chain Monte Carlo sampling loop.

[0011] Preferably, in step S2, the Bayesian inference specifically includes: A likelihood function is constructed based on the error between instantaneous load response data and AI alternative model predictions. The probability distribution based on design specification values ​​and material test results is used as the prior distribution. The Markov chain Monte Carlo method is used to solve for the posterior distribution, and the posterior mean and 95% confidence interval of the elastic modulus and the posterior mean and 95% confidence interval of the component self-weight coefficient are output. After the first layer of identification of each segment is completed, the new posterior distribution is fused with the historical identification results through Bayesian recursion to continuously narrow the uncertainty interval of the parameters.

[0012] Preferably, in step S3, within 24 hours after the completion of each newly poured beam segment, the representative temperature and corresponding displacement of the beam body are continuously recorded at a frequency of once per hour; the representative temperature of the beam body is the volume-weighted average of the temperature values ​​measured by temperature sensors embedded in the core area, near the surface and bottom plate of the cross section. The established linear regression model The expression is as follows: ; in, The initial temperature at the moment the pouring was completed; This is the initial displacement; This is due to random measurement error; It is the coefficient of thermal expansion; The temperature represents the beam.

[0013] Preferably, in step S4, the intelligent screening rules for isothermal conditions include: Rule 1, Strict Isothermal Matching: For any two moments and ,like If the beams at the two time points represent equal temperatures, then... This is the temperature tolerance threshold, initially set to ±0.5℃; for The beam's shape at any given moment represents its temperature; for The beam's shape at any given moment represents its temperature; Rule 2, Timing Pairing Constraint: Only pair times within the same segment that are different dates but whose temperatures meet the matching conditions; Rule 3, Isothermal Sequence Construction: Sort all isothermal moments in chronological order, and the displacement increment between two adjacent moments is the pure creep deformation; Rule 4, Tolerance Adaptive Relaxation: If the number of available data points that strictly meet the tolerance threshold is less than the preset number, the tolerance threshold will be automatically relaxed, and the temperature deviation introduced by the relaxation will be corrected by first-order linear correction using the temperature expansion coefficient identified in the second layer.

[0014] Preferably, in step S4, within 14 days after the completion of each newly poured beam segment, the displacement and representative temperature of the beam segment are continuously recorded at a frequency of once per hour to obtain a time-temperature-displacement dataset. Using the recommended creep coefficient range as the prior distribution for Bayesian inference, the creep coefficient expression is as follows: ; in, This represents the ultimate value of the creep coefficient. The time function is calculated based on the loading age, ambient humidity, and theoretical thickness of the component. For the current calculation time, For the concrete loading age, The duration of the load application. The calculations were performed according to concrete specifications based on ambient humidity, theoretical thickness of the component, and loading age.

[0015] Preferably, in step S5, the instantaneous loading test is performed using the case of unilateral eccentric forward movement of the hanging basket; a weighted objective function is constructed. The expression is as follows: ; in, The parameters to be identified are the boundary constraint stiffness; The measured values ​​of beam displacement under eccentric loading of the formwork. Input stiffness into the finite element model The calculated values ​​of the beam displacement model are obtained. This represents the measured value of the shear deformation of the support. Input stiffness into the finite element model The calculated values ​​of the support shear deformation model are obtained. The weighting coefficient is determined based on the accuracy of displacement and shear deformation measurements, and its value ranges from 0.5 to 1.0. The optimization algorithm is either a multi-starting-point gradient optimization algorithm or a particle swarm optimization algorithm.

[0016] Preferably, the data acquisition time window in step S5 is from 10 minutes before the start of the forward movement to 30 minutes after the forward movement is completed, with a total window duration of no more than 120 minutes and a temperature change of less than 0.5℃ within the window; Multiple independent identifications were performed using data from the single-sided forward movement of the hanging basket in multiple construction segments. The statistical average value was taken as the final estimate of the boundary constraint stiffness, and the standard deviation was calculated as an uncertainty measure.

[0017] Preferably, steps S2 to S6 are repeated after each new segment is completed, so as to achieve continuous online updating of parameters and adaptive evolution of the finite element model of the whole bridge.

[0018] Therefore, compared with the prior art, the present invention adopts the above-mentioned AI-driven hierarchical decoupling adaptive correction method for bridge parameters, which has the following advantages: (1) This invention abandons the traditional approach of mixing all parameters together for unified optimization, and innovatively performs hierarchical decoupling according to the physical nature of the parameters and the time-domain characteristics of the response—each layer identifies only 2 to 3 parameters, and uses the pure monitoring data that is most sensitive to the parameters of that layer. This strategy fundamentally eliminates the response coupling between parameters, ensuring the uniqueness and physical interpretability of the identification results.

[0019] (2) The third layer of this invention uses isothermal conditions for intelligent screening, directly extracting the displacement increment from the monitoring data at the same temperature, thus naturally offsetting the thermal expansion and contraction components. When data is insufficient, the temperature expansion coefficient identified in the second layer is used for linear compensation for small temperature differences.

[0020] (3) The monitoring data used in the first and fourth layers of this invention—concrete pouring, formwork movement, prestressing tensioning, and unilateral eccentric forward movement of the formwork—are all derived from the necessary procedures in the bridge construction process. No additional load tests, waiting time, or special equipment are required for parameter identification.

[0021] (4) Each layer of this invention adopts a Bayesian inference framework, outputting the posterior probability distribution of the parameters (rather than a single point estimate), and using the complete posterior distribution of the parameters in the upper layer as the prior known condition for the identification in the lower layer. The parameter uncertainty is passed down level by level among the four layers and narrows layer by layer, finally giving the mean and confidence interval of each parameter, providing complete probabilistic information for construction control decisions.

[0022] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein: Figure 1 This is a flowchart of an AI-driven bridge parameter hierarchical decoupling adaptive correction method according to the present invention. Figure 2 This is a schematic diagram of the first-layer instantaneous stiffness parameter identification process of the present invention; Figure 3 This is a schematic diagram of the second-layer temperature-displacement monitoring data acquisition and linear fitting of the present invention; Figure 4 This is a schematic diagram of the intelligent screening process under the third layer of isothermal conditions in this invention; Figure 5 This is a schematic diagram illustrating the isothermal time matching principle of the third layer of the present invention; Figure 6 This is a flowchart of the closed-loop adaptive construction control process of the present invention, which includes monitoring, identification, updating, prediction, and control. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0026] Figure 1 This is a flowchart of an AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to the present invention. Figure 1 As shown, an AI-driven hierarchical decoupling adaptive correction method for bridge parameters includes the following steps: S1. The parameters to be identified in the bridge finite element model are divided into four independent levels according to the response time domain characteristics and spatial influence range. Each level is back-identified using pure monitoring data that matches the physical properties of the parameters at that level. The identification results of the upper level are used as the prior known conditions for the identification of the lower level. The first level is the stiffness parameter including the elastic modulus of concrete and the self-weight coefficient of the component. The second level is the temperature expansion coefficient. The third level is the ultimate value of the creep coefficient. The fourth level is the boundary constraint stiffness including the horizontal shear stiffness of the support and the lateral stiffness of the pier. Taking the cantilever casting construction of a prestressed concrete continuous beam bridge (span arrangement of 75m+125m+75m) as an example, sensor deployment.

[0027] Sensors are installed at the following locations in each cantilever section: Two displacement observation points are set up on the top and bottom surfaces of the beam segment (using machine vision intelligent measuring instruments or precision level instruments). One temperature sensor (accuracy ±0.2°C) is installed in the core area of ​​the beam section, near the surface (5cm from the surface), and on the bottom plate. A longitudinal displacement sensor is installed on the top surface of block 0; Displacement gauges were installed at the supports to measure shear deformation. All sensors are connected to the same data acquisition unit to achieve unified time synchronization and synchronous data acquisition, with a timestamp accuracy of ≤1 second.

[0028] S2. Identify the stiffness parameters of the first layer, collect instantaneous load response data under construction conditions, construct a local finite element sub-model, embed a Gaussian process surrogate model or a Bayesian neural network as an AI substitute model for the local finite element sub-model into the Bayesian inference framework, input the measured construction load, and identify the posterior probability distribution of the elastic modulus and self-weight coefficient; Figure 2 The diagram shown is a schematic of the first-layer instantaneous stiffness parameter identification process of the present invention.

[0029] S2.1 Data Acquisition: For each cantilevered segment during bridge construction, structural displacement and strain data were collected before and after the following three key working conditions: Working condition 1: Concrete pouring - The initial value is the stable reading before the start of pouring, and the final value is the reading after the completion of pouring; Operating Condition 2: Forward movement of the hanging basket – the initial value is the stable reading before the forward movement begins, and the final value is the reading after the forward movement is completed; Working condition 3: Prestressed tensioning - The initial value is the stable reading before tensioning begins, and the final value is the reading after tensioning is completed.

[0030] During the data acquisition period, the temperature variation of the beam was kept within ±0.5°C to ensure that the acquired data represented a pure elastic response and to eliminate interference from daily temperature variations and creep effects. Using the initial values ​​for each load condition as a baseline, the displacement and strain increments relative to the initial values ​​were calculated as the instantaneous load response data.

[0031] S2.2 Forward Model Establishment: Establish a local finite element sub-model for the current construction segment, using measured construction loads (wet weight of concrete, self-weight of hanging basket, prestressing tension, etc.) as input, and concrete elastic modulus E and component self-weight coefficient γ as parameters to be identified, to calculate the theoretical instantaneous displacement δ. model (E, γ) and theoretical instantaneous strain ε model (E, γ).

[0032] S2.3 Bayesian Inference and Recognition: Constructing the Likelihood Function Based on the error between the measured instantaneous response and the model prediction: ; in, This is the instantaneous load response dataset collected in the first layer. For the number of measurement points, To measure the standard deviation of noise; This represents the total number of measurement points involved in the identification under instantaneous operating conditions. It is the circumference constant; The standard deviation of displacement observation noise; This represents the measured displacement increment at the k-th measuring point; For input parameters AI replaces the theoretical displacement increment of the k-th measurement point output by the model / local finite element sub-model; It is an exponential function; To perform a series multiplication operation on all n measurement points, each measurement point is assumed to be subject to independent Gaussian noise.

[0033] The probability distribution based on design specification values ​​and material test results is used as the prior distribution: ; ; in, The nominal elastic modulus of concrete given in the design documents is used as the prior mean. It follows a normal distribution; The posterior distribution is solved using the Markov Chain Monte Carlo (MCMC) method. Output the posterior mean Ê of the elastic modulus E and its 95% confidence interval [E]. L E U ], where E L E is the lower limit of the elastic modulus E. U The upper limit of the elastic modulus E; and the self-weight coefficient. posterior mean and its 95% confidence interval.

[0034] S2.4 Multi-segment fusion: As construction progresses, after the first layer of identification of each segment is completed, the new posterior distribution is fused with the historical identification results using Bayesian recursion to continuously narrow the uncertainty range of the parameters.

[0035] Introducing Gaussian process surrogate models or Bayesian neural networks as AI alternatives to finite element models—using a small number of finite element forward modeling results as training samples, constructing elastic modulus E and self-weight coefficient. By establishing a high-precision nonlinear mapping relationship between the model and the structural response, and then embedding the surrogate model into the MCMC sampling loop, the computation time for a single identification is reduced from several hours to minutes.

[0036] Before pouring concrete for segment N, record the initial displacement readings at each measuring point. After pouring, and provided the beam temperature change during the data collection period does not exceed ±0.5°C, record the displacement readings at each measuring point again and calculate the displacement increment before and after pouring. For the two conditions of formwork relocation and prestressing tensioning, collect the readings before and after pouring in the same manner and calculate the displacement increment.

[0037] Using the actual concrete volume (known), formwork self-weight (known), and prestressing tension (known) of the segment as load inputs, a local finite element sub-model of the segment is established. The elastic modulus E and self-weight coefficient γ are used as parameters to be identified, and the MCMC method (10,000 iterations, with the first 2,000 iterations being burn-in) is employed to solve for the posterior distribution. The output of the segment is... and .

[0038] S3. Identify the temperature expansion coefficient of the second layer, establish a linear regression model, and solve for the optimal estimate and confidence interval of the temperature expansion coefficient; such as Figure 3 The diagram shown is a schematic diagram of the second-layer temperature-displacement monitoring data acquisition and linear fitting of the present invention.

[0039] S3.1 Data Acquisition: Within 24 hours after the completion of each newly poured beam segment, data shall be continuously recorded at a frequency of once per hour. The beam represents temperature. (The volume-weighted average of measurements from at least three temperature sensors embedded in the core area, near the surface, and on the base plate of the cross-section is used.) Beam segment displacement at corresponding time .

[0040] Establish a temperature-displacement dataset for this segment over 24 hours. The expression is as follows: ; S3.2 Parameter Identification: Since the 24-hour time window is much smaller than the significant development period of creep (the deformation caused by creep within 24 hours is usually less than the sensor accuracy), the creep effect within this window can be safely ignored. The coefficient of thermal expansion is used as the metric. For the parameters to be identified, establish a linear regression model: ; in, The initial temperature at the moment the pouring was completed; This is the initial displacement; This is due to random measurement error; It is the coefficient of thermal expansion; The temperature represents the beam. Least squares estimation or Bayesian linear regression is used to solve the problem. The optimal estimate and its 95% confidence interval.

[0041] Within 24 hours of the completion of the Nth segment's casting, the representative temperature and corresponding displacement of the beam were recorded hourly, resulting in 24 data points. A linear regression model was established based on the temperature and displacement at the moment of casting completion, and solved using the least squares method.

[0042] S4. Identify the creep coefficient of the third layer, construct a time-temperature-displacement dataset, and filter and construct an isothermal time series based on isothermal condition intelligent screening rules; use the elastic modulus obtained from the first layer identification as a known parameter for Bayesian inference, and output the posterior probability distribution of the final value of the creep coefficient; such as Figure 4 The diagram shown is a schematic of the intelligent screening process under the third layer of isothermal conditions according to the present invention.

[0043] Note: To avoid contamination of creep identification by plastic shrinkage and autogenous shrinkage during concrete curing, a stress-free gauge is used to obtain the shrinkage strain increment during curing. The formula is as follows: ; in, Let t be the concrete drying shrinkage strain at time t, and the actual shrinkage strain after deducting the effect of temperature. The total strain reading (including shrinkage strain and temperature strain) is obtained by the stressless gauge at time t. This is the coefficient of linear thermal expansion of concrete; The temperature change at time t relative to the initial reference time; Then subtract this portion from the displacement increment, leaving the pure creep displacement. Alternatively, calculate the theoretical shrinkage strain according to the specification, subtract this shrinkage increment from the total displacement increment in S4.3, and obtain the pure creep deformation.

[0044] S4.1 Continuous monitoring data acquisition: Within 14 days after the completion of each newly poured beam segment, the displacement of the beam segment shall be continuously recorded at a frequency of once per hour. and represent temperature Establish a three-dimensional dataset of time-temperature-displacement. The expression is as follows: , where N≤14×24=336.

[0045] in, The observation time corresponding to the i-th sample; The measured temperature corresponding to the i-th sample; This represents the long-term displacement monitoring value corresponding to the i-th sample. S4.2, Intelligent screening under isothermal conditions: In In this process, the following rules are used to select "isothermal pairs" with equal temperatures: Rule 1 (Strict Isothermal Matching): For any two moments and If satisfied If the two temperatures are equal, then they are considered equal. This is the temperature tolerance threshold, initially set at ±0.5°C; For example... Figure 5 The diagram shown is a schematic diagram of the isothermal time matching principle of the third layer of the present invention. Rule 2 (Time-series pairing constraint): Only pair times within the same segment that are different dates but have the same temperature; Rule 3 (Construction of Isothermal Sequence): Sort all isothermal moments in chronological order to construct an isothermal time series t. (1) <t (2) <... <t (m) In this sequence, the displacement increment between two adjacent time points This is pure creep deformation—because at the same temperature, thermal expansion and contraction contribute nothing; in, This represents the displacement monitoring value of the measuring point at time k+1. This represents the displacement monitoring value of the measuring point at time k.

[0046] Rule 4 (Adaptive Tolerance Relaxation): If fewer than 5 usable data points strictly meet ±0.5°C within 14 days, the tolerance threshold is automatically relaxed to ±1.0°C, and the temperature expansion coefficient identified in the second layer is utilized. A first-order linear correction is applied to the temperature deviation introduced by the relaxation: ; in, This represents the pure creep displacement increment after the temperature effect has been removed; This represents the total displacement increment obtained from instrument measurements.

[0047] S4.3 Construction of Pure Creep Deformation Curve: Through the above isothermal matching and correction, M pure creep deformation data points are obtained. A creep development curve was constructed with time as the horizontal axis and pure creep deformation as the vertical axis. Among them, The time interval for the k-th time step; This represents the pure creep displacement increment after the temperature effect has been deducted in the k-th step.

[0048] Creep constitutive model: The creep coefficient calculation model adopted, creep coefficient Expressed as: ; in, This is the ultimate value of the creep coefficient (the core parameter to be identified). The creep development time function is calculated and determined by the standard formula based on known factors such as loading age, ambient humidity, and theoretical thickness of the component. For the current calculation time, For the concrete loading age, The duration of the load application. The calculations were performed according to concrete specifications based on ambient humidity, theoretical thickness of the component, and loading age.

[0049] S4.5 Bayesian Inference and Identification: The Ê identified in the first layer is input as a known stiffness parameter into the finite element creep analysis module, in order to... For the parameters to be identified, use pure creep deformation data points. Construct the likelihood function for the observed values. The expression is as follows: ; in, The standard deviation of creep displacement observation noise; For the k-th sample, the measured pure creep displacement increment after deducting the temperature effect; Input creep ultimate coefficient The theoretical creep displacement increment calculated by the creep constitutive model; This represents the total number of valid creep samples.

[0050] The recommended creep coefficient range is used as the prior distribution: ,and The MCMC method is used to solve for the posterior distribution. Output posterior mean and its 95% confidence interval.

[0051] S4.6 Multi-segment recursive fusion: As construction progresses, for each newly added segment that completes 14 days of monitoring, the new posterior distribution is recursively fused with historical identification results using Bayesian methods to continuously improve... The recognition accuracy.

[0052] For 14 days after the completion of the Nth segment pouring, displacement and temperature were continuously recorded hourly. Data from 4:00 AM to 6:00 AM daily (the period with the least daily temperature fluctuation) was used as the priority matching source. [Settings...] , and filter isothermal moments from the data points.

[0053] Taking actual monitoring data as an example: at 5:00 AM on the first day, the temperature was 12.3°C and the displacement was 2.1 mm; at 5:00 AM on the eighth day, the temperature was 12.5°C and the displacement was 3.8 mm. The temperature difference between the two is 0.2°C ≤ 0.5°C, forming an isothermal pair. The displacement increment of 1.7 mm is pure creep deformation. If there are fewer than 5 strictly matched isothermal pairs within 14 days, the tolerance is relaxed to 1.0°C, and the following is utilized... Make corrections.

[0054] Substitute the obtained pure creep deformation data points into the standard creep model, and For the parameters to be identified, the posterior distribution is solved using the MCMC method, and the output is... and its 95% confidence interval.

[0055] S5. Identify the stiffness of the fourth-layer boundary constraints, construct a weighted objective function of displacement and shear deformation, and use an optimization algorithm to solve for the optimal estimate of the constraint stiffness. S5.1 Data Acquisition: The unilateral eccentric forward movement of the formwork during bridge construction was selected as the natural instantaneous loading test. Data was continuously collected within a time window of 10 minutes before the start of the unilateral forward movement and 30 minutes after the movement was completed (approximately 60-90 minutes in total). Longitudinal horizontal displacement at the root of the main beam (top surface of block 0) ; Shear deformation at the support ; Pier top rotation angle θ (used for data verification).

[0056] The above time window does not exceed 120 minutes, the daily temperature variation is negligible (<0.5°C), and the deformation of the creep effect at this time scale is much smaller than the sensor accuracy. Therefore, the collected data is pure instantaneous elastic response data.

[0057] S5.2 Forward Model Establishment: Establish a full-bridge finite element sub-model that includes the elastic deformation of the piers and the shear deformation of the supports, using the first layer of identified... and the third layer of recognition Given the parameters, the boundary constraint stiffness (including the horizontal shear stiffness of the support) and pier body thrust stiffness The parameter to be identified is the measured hanging basket load, which is used as input to calculate the theoretical longitudinal displacement of the pier top. and theoretical support shear deformation .

[0058] S5.3 Objective Function and Optimization Solution: Constructing the Objective Function The expression is as follows: ; in, The parameters to be identified are the boundary constraint stiffness; The measured values ​​of beam displacement under eccentric loading of the formwork. Input stiffness into the finite element model The calculated values ​​of the beam displacement model are obtained. This represents the measured value of the shear deformation of the support. Input stiffness into the finite element model The calculated values ​​of the support shear deformation model are obtained; λ is the weighting coefficient of displacement and shear deformation, determined based on the measurement accuracy of both (range 0.5~1.0). Because... The dimensions are 2 to 4, and the solution is obtained using multi-starting-point gradient optimization algorithm or particle swarm optimization algorithm. The optimal estimate .

[0059] S5.4 Multi-condition statistical averaging: Multiple independent identifications are performed using the single-sided forward movement data of the hanging basket in multiple construction segments. The statistical average value is taken as the final estimate of the boundary constraint stiffness, and its standard deviation is calculated as an uncertainty measure.

[0060] Under the condition of unilateral forward movement of the formwork (each cantilever segment of this bridge will undergo unilateral forward movement once), within a time window of 10 minutes before the formwork begins to move forward and 30 minutes after it reaches its final position, the longitudinal displacement of the top surface of block 0 and the shear deformation of the support are continuously collected. The measured formwork load is used as input. and For the parameters to be identified, the objective function is solved using the particle swarm optimization algorithm (population size 50, number of iterations 200). The minimum value is determined by multiple independent identifications using data from the forward movement of the basket in all completed segments, followed by statistical averaging.

[0061] S6. Inject the identification results from steps S2 to S5 into the finite element model of the entire bridge to update the virtual digital bridge, and predict the alignment of subsequent construction segments and calculate control commands.

[0062] All parameters obtained from the first to fourth layers of identification — , , , , —And its uncertainty range, are all injected into the full-bridge finite element model, completing a full update of the "virtual digital bridge". The updated finite element model is used for subsequent construction segment alignment prediction and control command calculation, realizing closed-loop adaptive construction control of "monitoring → identification → updating → prediction → control". Figure 6The diagram shown is a flowchart of the closed-loop adaptive construction control process of the present invention, which includes monitoring, identification, updating, prediction, and control.

[0063] In summary, the embodiments of the present invention provide an AI-driven hierarchical decoupling adaptive correction method for bridge parameters, which enables high-precision and high-reliability online identification of bridge structural parameters and continuous dynamic updating of the finite element model.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An AI-driven hierarchical decoupling adaptive correction method for bridge parameters, characterized in that, Includes the following steps: S1. The parameters to be identified in the bridge finite element model are divided into four independent levels according to the response time domain characteristics and spatial influence range. The identification results of each level serve as the prior known conditions for the identification of the next level. The first level is the stiffness parameter including the elastic modulus of concrete and the self-weight coefficient of the component. The second level is the temperature expansion coefficient. The third level is the ultimate value of the creep coefficient. The fourth level is the boundary constraint stiffness including the horizontal shear stiffness of the support and the lateral stiffness of the pier. S2. Identify the stiffness parameters of the first layer, collect instantaneous load response data under construction conditions, construct a local finite element sub-model, embed the Gaussian process surrogate model or Bayesian neural network as the AI ​​replacement model of the local finite element sub-model into the Bayesian inference framework, input the measured construction load, and identify the posterior probability distribution of the elastic modulus and self-weight coefficient. S3. Identify the temperature expansion coefficient of the second layer, establish a linear regression model, and solve for the optimal estimate and confidence interval of the temperature expansion coefficient. S4. Identify the creep coefficient of the third layer, construct a time-temperature-displacement dataset, and construct an isothermal time series based on the intelligent filtering rules of isothermal conditions; use the elastic modulus obtained from the first layer identification as a known parameter to perform Bayesian inference and output the posterior probability distribution of the ultimate value of the creep coefficient. S5. Identify the stiffness of the fourth-layer boundary constraints, construct a weighted objective function of displacement and shear deformation, and use an optimization algorithm to solve for the optimal estimate of the constraint stiffness. S6. Inject the identification results from steps S2 to S5 into the finite element model of the entire bridge to update the virtual digital bridge, and predict the alignment of subsequent construction segments and calculate control commands.

2. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S2, the process of acquiring the instantaneous load response data includes: The structural displacement and strain data were collected before and after three working conditions: concrete pouring, formwork relocation, and prestressing tensioning. The displacement and strain increments of the final value relative to the initial value were calculated to obtain the instantaneous load response data. The initial and final values ​​of each measuring point were read within a period when the temperature change of the beam did not exceed ±0.5℃.

3. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S2, the construction process of the AI ​​alternative model includes: Using a small number of finite element forward modeling results from the local finite element sub-model as training samples, a Gaussian process surrogate model or a Bayesian neural network is trained to construct a nonlinear mapping between the parameters to be identified and the structural response; the AI ​​substitution model replaces the forward modeling calculation of the local finite element sub-model within the Markov chain Monte Carlo sampling loop.

4. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S2, the Bayesian inference specifically includes: A likelihood function is constructed based on the error between instantaneous load response data and AI alternative model predictions. The probability distribution based on design specification values ​​and material test results is used as the prior distribution. The Markov chain Monte Carlo method is used to solve for the posterior distribution, and the posterior mean and 95% confidence interval of the elastic modulus and the posterior mean and 95% confidence interval of the component self-weight coefficient are output. After the first layer of identification of each segment is completed, the new posterior distribution is fused with the historical identification results through Bayesian recursion to continuously narrow the uncertainty interval of the parameters.

5. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S3, within 24 hours after the completion of each newly poured beam segment, the representative temperature and corresponding displacement of the beam are continuously recorded at a frequency of once per hour. The representative temperature of the beam is a volume-weighted average of the temperature measurements taken by temperature sensors embedded in the core area, near the surface, and bottom plate of the cross section. The established linear regression model The expression is as follows: ； in, The initial temperature at the moment the pouring was completed; This is the initial displacement; This is due to random measurement error; It is the coefficient of thermal expansion; The temperature represents the beam.

6. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S4, the intelligent screening rules for isothermal conditions include: Rule 1, Strict Isothermal Matching: For any two moments and ,like If the beams at the two time points represent equal temperatures, then... This is the temperature tolerance threshold, initially set to ±0.5℃; for The beam's shape at any given moment represents its temperature; for The beam's shape at any given moment represents its temperature; Rule 2, Timing Pairing Constraint: Only pair times within the same segment that are different dates but whose temperatures meet the matching conditions; Rule 3, Isothermal Sequence Construction: Sort all isothermal moments in chronological order, and the displacement increment between two adjacent moments is the pure creep deformation; Rule 4, Tolerance Adaptive Relaxation: If the number of available data points that strictly meet the tolerance threshold is less than the preset number, the tolerance threshold will be automatically relaxed, and the temperature deviation introduced by the relaxation will be corrected by first-order linear correction using the temperature expansion coefficient identified in the second layer.

7. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S4, within 14 days after the completion of each newly poured beam segment, the displacement and representative temperature of the beam segment are continuously recorded at a frequency of once per hour to obtain a time-temperature-displacement dataset. Using the recommended creep coefficient range as the prior distribution for Bayesian inference, the creep coefficient expression is as follows: ； in, This represents the ultimate value of the creep coefficient. The time function is calculated based on the loading age, ambient humidity, and theoretical thickness of the component. For the current calculation time, For the concrete loading age, The duration of the load application. The calculations were performed according to concrete specifications based on ambient humidity, theoretical thickness of the component, and loading age.

8. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, In step S5, the instantaneous loading test is performed using the unilateral eccentric forward movement of the hanging basket; a weighted objective function is constructed. The expression is as follows: ； in, The parameters to be identified are the boundary constraint stiffness; The measured values ​​of beam displacement under eccentric loading of the formwork. Input stiffness into the finite element model The calculated values ​​of the beam displacement model are obtained. This represents the measured value of the shear deformation of the support. Input stiffness into the finite element model The calculated values ​​of the support shear deformation model are obtained. The weighting coefficient is determined based on the accuracy of displacement and shear deformation measurements, and its value ranges from 0.5 to 1.

0. The optimization algorithm is either a multi-starting-point gradient optimization algorithm or a particle swarm optimization algorithm.

9. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, The data acquisition time window for step S5 is from 10 minutes before the forward movement begins to 30 minutes after the forward movement is completed, with a total window duration of no more than 120 minutes and a temperature change of less than 0.5℃ within the window. Multiple independent identifications were performed using data from the single-sided forward movement of the hanging basket in multiple construction segments. The statistical average value was taken as the final estimate of the boundary constraint stiffness, and the standard deviation was calculated as an uncertainty measure.

10. The AI-driven hierarchical decoupling adaptive correction method for bridge parameters according to claim 1, characterized in that, Steps S2 to S6 are repeated after each new segment is completed, enabling continuous online updates of parameters and adaptive evolution of the finite element model of the entire bridge.

Images