Bridge support static displacement over-limit early warning correction method based on prediction model

Abstract

The invention relates to the technical field of bridge structure health monitoring and early warning, and discloses a bridge support static displacement overrun early warning correction method based on a prediction model, and the method comprises the following steps: S1, simulating the dynamic response of a support and an overall structure under various working conditions through building a coupling finite element model of the bridge support and the overall bridge structure; s2, displacement, vibration and temperature data of the support are collected in real time through a multi-source sensor, and data fusion and noise reduction processing are carried out; s3, establishing a multi-level prediction model based on a physical model and a data driving model, and predicting a future displacement trend of the support; and S4, monitoring a prediction error in real time through an error feedback mechanism. Prediction errors are corrected in real time through recursive Kalman filtering and an error feedback mechanism, an early warning threshold value is dynamically adjusted through Bayesian updating, the precision and reliability of early warning are improved in combination with a double triggering mechanism, the method adapts to changes under complex working conditions, and false alarms and missing alarms are reduced.

Description

Technical Field

[0001] The invention relates to the technical field of bridge structure health monitoring and early warning, and in particular to a correction method for bridge bearing static displacement over-limit early warning based on a prediction model. Background Art

[0002] As an important part of transportation infrastructure, the safety of bridges is directly related to the normal operation of the transportation system and public safety. Bridge bearings are key components in bridge structures, connecting the superstructure and substructure of the bridge, transferring loads and absorbing deformation. However, with the increase in the service life of the bridge and under the action of various environments and loads, the bearings may have abnormal displacement, which in turn affects the stability and safety of the overall bridge structure. Therefore, it is particularly important to monitor and warn the static displacement of bridge bearings.

[0003] In the prior art, the monitoring of bridge bearing displacement mainly relies on physical model prediction or simple data-driven models. These methods can provide prediction and early warning information of bearing displacement to a certain extent, but there are significant deficiencies in practical applications. Physical models such as finite element analysis rely on the theoretical design and material parameters of the bridge structure. Although they can better describe the mechanical properties of the bridge, they are prone to inaccuracy under complex working conditions. Although a single data-driven model can predict trends through historical data, it lacks an accurate description of structural mechanics, is difficult to cope with sudden changes in working conditions, and is easily disturbed by sensor noise and outliers.

[0004] In addition, most existing early warning systems use static threshold settings, that is, a fixed support displacement threshold is pre-set, and an early warning is triggered when the threshold is exceeded. However, this static threshold is difficult to adapt to the operating status of the bridge under different working conditions, especially under extreme environmental conditions. The static threshold often leads to false alarms or missed alarms, reducing the reliability of the early warning system. At the same time, the existing early warning system usually only considers the changes in the displacement of a single support, and does not combine the health status of the overall structure of the bridge, resulting in insufficient credibility of the early warning signal, which is easy to cause unnecessary maintenance or fail to detect potential safety hazards in a timely manner.

[0005] In order to solve the above problems, the present invention proposes a correction method for bridge bearing static displacement over-limit warning based on a prediction model. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a correction method for the static displacement over-limit warning of bridge bearings based on a prediction model, which solves the problems of low prediction accuracy of the static displacement of bridge bearings under complex working conditions and false alarms and omissions of the early warning system.

[0007] To achieve the above purpose, the present invention is implemented by the following technical scheme: A correction method for bridge bearing static displacement over-limit warning based on a prediction model comprises the following steps: S1. By establishing a coupled finite element model of the bridge bearing and the overall bridge structure, the dynamic response of the bearing and the overall structure under various working conditions is simulated; S2. Use multi-source sensors to collect the displacement, vibration and temperature data of the support in real time, and perform data fusion and noise reduction processing; S3. Establish a multi-level prediction model based on physical model and data-driven model to predict the future displacement trend of the bearing; S4, real-time monitoring of prediction errors through error feedback mechanism, and dynamic adjustment of prediction model parameters through adaptive correction algorithm; S5. Set dynamic thresholds and trigger warning signals when the static displacement of the bearing exceeds the limit. Combined with the overall structural health status of the bridge, generate maintenance recommendations.

[0008] Preferably, the step S1 specifically includes the following steps: S1.1. Construct a three-dimensional finite element model of the entire bridge, model the main beam, piers, abutments and supports through finite element analysis, and define the linear and nonlinear characteristics of the materials; S1.2. Establish a local refined model of the support, use nonlinear spring elements or contact elements, and define the elastic and plastic deformation behaviors of the support; S1.3, coupling the nonlinear mechanical behavior of the support with the overall bridge structure model, and realizing the displacement and force transmission between the support and the main beam through node setting; S1.4. Set the operating conditions of traffic load, wind load, earthquake action and temperature gradient change, and define the load size and distribution; S1.5. Use the Newmark-β integration method to perform dynamic response simulation and calculate the displacement, stress and strain of the bridge as a whole and the bearings under various working conditions.

[0009] Preferably, the step S3 specifically includes the following steps: S3.1. Construction and prediction of physical model: Using the finite element model of the bridge, the bearing displacement is preliminarily predicted through mechanical equations and structural response simulation. The working load, temperature, and material property parameters are input into the physical model to perform physical prediction of the bearing displacement. Statics analysis and dynamics analysis combined with the finite element method are used to predict the bearing displacement, and the displacement response values ​​under normal traffic loads, strong winds, and earthquakes are obtained. S3.2. Build an LSTM data-driven model: extract features from the historical monitoring data collected by sensors, use normalization methods to process the raw data, and use LSTM to train the data-driven model; S3.3, Model training and validation: Train the LSTM model using monitoring data, evaluate the model’s prediction effect through cross-validation, use mean square error as the loss function, adjust the model’s hyperparameters, optimize the model’s prediction accuracy, and use a split validation set to validate the model’s generalization ability; S3.4, Fusion of physical model and data-driven model: The Bayesian weighted fusion algorithm is used to fuse the prediction results of the physical model and the LSTM data-driven model. The weighting coefficient is dynamically adjusted according to the model performance under different working conditions, so that the fusion model can maintain high-precision prediction under various working conditions; S3.5. Prediction of future trend of bearing displacement: Use the fusion model to predict the displacement trend of the bearing in the future time period, output the changes in static and dynamic displacement of the bearing in a certain period of time in the future, and make a revised prediction based on the current monitoring data.

[0010] Preferably, the step S4 specifically includes the following steps: S4.1. Comparison between predicted value and actual value: Compare the predicted value of the support displacement output by the prediction model with the actual support displacement value collected by the sensor in real time to calculate the prediction error; S4.2. Establish an error feedback mechanism: record each prediction error through the error feedback mechanism, and feed the error back to the prediction model to correct the model parameters; S4.3, Recursive Kalman filter error correction: Recursive Kalman filter is used for error correction; S4.4. Adaptive adjustment of model parameters: Dynamically adjust material parameters in the physical model and hyperparameters in the data-driven model based on the corrected errors.

[0011] Preferably, the calculation formula of the prediction error in step S4.1 is as follows: Where e(t) is the prediction error, is the support displacement value actually measured by the sensor, are the support displacement values ​​predicted by the model.

[0012] Preferably, the step S5 specifically includes the following steps: S5.1. Set initial warning threshold: Based on historical monitoring data and structural design specifications, set the initial warning threshold of the static displacement of the bearing; S5.2. Dynamically adjust the threshold: Use the Bayesian update rule to dynamically adjust the warning threshold based on the real-time monitored support displacement data; S5.3, triggering of over-limit warning signal: when the predicted value of static displacement of the bearing exceeds the dynamically set threshold, the system automatically triggers the warning signal to remind the maintenance personnel that the bearing may have over-limit displacement; S5.4. Generate maintenance recommendations: The system generates maintenance recommendations based on the bearing displacement warning and the health status of the overall bridge structure; S5.5. Subsequent monitoring and feedback: After the early warning is triggered, the system continues to monitor the bearing displacement and the overall health of the bridge in real time, and compares the latest monitoring data with the early warning results.

[0013] Preferably, the dynamic threshold setting formula in step S5.2 is: , T(t) is the warning threshold after dynamic adjustment, is the initial warning threshold, α is the dynamic adjustment factor, and X(t) is the support displacement data monitored in real time.

[0014] Preferably, in step S5.3, the warning signal triggering mechanism for the bearing displacement exceeding the limit adopts a dual triggering condition: When the predicted support displacement When the dynamic threshold T(t) is exceeded and the overall structural health score of the bridge is lower than the preset threshold, an early warning signal is triggered.

[0015] Preferably, in step S3.2, the input features of the LSTM data-driven model include support displacement history data, support surface temperature, vibration frequency of the bridge main beam and wind load parameters.

[0016] Preferably, in step S4.4, the model parameters to be adaptively adjusted include: Material elastic modulus and yield stress in physical models; Learning rate, number of hidden units, and history length in data-driven models.

[0017] The present invention provides a correction method for bridge bearing static displacement over-limit warning based on a prediction model. It has the following beneficial effects: 1. The present invention significantly improves the accuracy of bearing displacement prediction by introducing a fusion method of physical model and data-driven model. It also organically combines mechanical analysis with historical data learning by adopting the physical model of finite element analysis and the time series data model of LSTM, using the Bayesian weighted fusion algorithm, effectively improving the prediction ability under complex working conditions and ensuring that the bearing displacement prediction is more accurate and robust.

[0018] 2. The present invention adopts an error feedback mechanism and a recursive Kalman filter to monitor and correct the prediction error in real time. The material parameters of the physical model and the hyperparameters of the data-driven model are adaptively adjusted through the recursive Kalman filter to ensure that the model can dynamically respond to changes in actual working conditions and optimize the prediction accuracy in real time. This ensures that the model still has good adaptability in long-term monitoring and significantly reduces the risk of error accumulation.

[0019] 3. The present invention introduces the Bayesian update rule to dynamically adjust the warning threshold according to the real-time monitored support displacement data, so that the threshold can be flexibly changed according to the current working conditions, avoiding the errors caused by static threshold setting, especially in extreme working conditions or long-term bridge aging. It significantly reduces false alarms and missed alarms and improves the warning accuracy of the system.

[0020] 4. The present invention adopts a dual trigger mechanism, which not only takes into account the static displacement exceeding the limit of the bearing, but also combines the health score of the overall structure of the bridge, ensuring that the warning is triggered only when the static displacement of the bearing exceeds the limit and the overall health score of the bridge is lower than the preset threshold, avoiding unnecessary warning signals caused by a single displacement exceeding the limit, significantly improving the reliability of the warning, and reducing false alarms caused by a single displacement fluctuation or sensor error. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Please see attached Figure 1 The embodiment of the present invention provides a correction method for bridge bearing static displacement over-limit warning based on a prediction model, comprising the following steps: S1. By establishing a coupled finite element model of the bridge bearing and the overall bridge structure, the dynamic response of the bearing and the overall structure under various working conditions is simulated; Step S1 specifically includes the following steps: S1.1. Construct a three-dimensional finite element model of the entire bridge, model the main beam, piers, abutments and supports through finite element analysis, and define the linear and nonlinear characteristics of the materials; S1.2. Establish a local refined model of the support, use nonlinear spring elements or contact elements, and define the elastic and plastic deformation behaviors of the support; S1.3, coupling the nonlinear mechanical behavior of the support with the overall bridge structure model, and realizing the displacement and force transmission between the support and the main beam through node setting; S1.4. Set the operating conditions of traffic load, wind load, earthquake action and temperature gradient change, and define the load size and distribution; S1.5. Use the Newmark-β integration method to perform dynamic response simulation and calculate the displacement, stress and strain of the bridge as a whole and the bearings under various working conditions.

[0024] Specifically, in the correction process of the static displacement over-limit warning of the bridge bearing, it is crucial to accurately simulate the dynamic response of the bearing and the overall bridge structure under various working conditions. By establishing a coupled finite element model of the bridge bearing and the overall bridge structure, the mechanical interaction between the bearing and structural components such as the main beam, piers, and abutments can be effectively captured, and basic data for subsequent prediction models and error correction can be provided. This model not only needs to consider the coupling relationship between the bearing and the overall structure, but also needs to conduct a detailed analysis of the dynamic response under different working conditions (such as traffic loads, wind loads, earthquakes, etc.). To this end, the specific implementation steps include constructing a three-dimensional finite element model of the entire bridge, a local refined model of the bearing, processing the coupled mechanical behavior, and setting the working conditions. Finally, the dynamic response of the bearing and the overall bridge is simulated and calculated by numerical methods.

[0025] Detailed description: S1.1: Construct the overall three-dimensional finite element model of the bridge In this embodiment, the overall structure of the bridge is first modeled using finite element analysis software by collecting information such as the design drawings, material properties, and geometric parameters of the components of the bridge. The main components of the bridge include the main beam, piers, abutments, and supports, which are discretized using finite element beam units, shell units, and solid units. The geometric properties and material parameters of the main beam and piers need to be accurately defined in the model to ensure that the stress and deformation behavior of the bridge under different working conditions can be truly reflected.

[0026] Both the linear and nonlinear characteristics of the material need to be considered in the model, especially the plastic deformation that may occur in high stress areas (such as the connection between the bearing and the pier). On this basis, a three-dimensional finite element model of the entire bridge is constructed to lay the foundation for the subsequent coupling of the bearing and the overall bridge structure.

[0027] S1.2: Establish a local refined model of the support In this embodiment, nonlinear spring units or contact units are used to finely model the local mechanical behavior of the support. The support model needs to consider the elastic and plastic deformation characteristics of the support, especially the nonlinear behavior of the rubber support material, such as yield stress, elastic modulus, stiffness, and hardening effect under high load conditions. The deformation behavior of the support can be described by the following elastic-plastic mechanical relationship:

[0028] in, is the load response of the support, is the displacement of the support, is the stiffness matrix of the support, is the damping coefficient, is the time derivative of displacement. The support model not only needs to reflect its elastic deformation under conventional loads, but also needs to consider the stress-strain relationship entering the plastic region under overload conditions.

[0029] S1.3: Coupling of bearings to the overall bridge structure In this embodiment, the nonlinear mechanical behavior of the bearing is coupled with the overall bridge structure model through the substructure method. The specific method is to coordinate the displacement and force at the connection node between the bearing and the main beam, and realize mechanical coupling through node setting. This method ensures that the displacement response of the bearing can be transmitted through components such as the main beam and piers, simulating the interaction between the bearing and the overall bridge.

[0030] To achieve this coupling, the stiffness matrix of the local support model is The stiffness matrix of the overall bridge model Combined to form the coupling stiffness matrix of the whole bridge :

[0031] Through the displacement coordination conditions at the nodes, the mechanical behaviors of the supports and the entire bridge can be effectively coupled to ensure the accuracy and stability of the model.

[0032] S1.4: Working condition setting In this embodiment, a variety of working conditions are set, including traffic load, wind load, earthquake effect, temperature gradient change, etc. The size and distribution of these loads are defined according to the actual working environment of the bridge.

[0033] Traffic load: According to the design load level of the bridge, the dynamic and static vehicle loads on the bridge deck are set and applied to the main beam and the bridge deck.

[0034] Wind load: Horizontal wind load is applied according to the wind environment in the area where the bridge is located, mainly acting on the windward side of the bridge deck, main beam and piers.

[0035] Earthquake action: According to the earthquake intensity and code requirements, horizontal and vertical earthquake acceleration loads are set and applied to the foundation and piers of the bridge.

[0036] Temperature gradient: Set the temperature gradient change to consider the impact of day and night temperature difference or seasonal temperature change on the thermal stress of the bridge structure.

[0037] By setting these working conditions, the stress state of the overall bridge structure and supports can be simulated and analyzed according to the actual environment.

[0038] S1.5: Dynamic Response Simulation In this embodiment, the Newmark-β integration method or other numerical integration methods are used to perform time domain simulation calculations on the dynamic response of the bridge. The Newmark-β method is suitable for solving the dynamic equations of the bridge structure in the time domain:

[0039] in: M is the structural mass matrix, C is the damping matrix, K is the stiffness matrix, U is the node displacement vector, F(t) is the time-dependent external load.

[0040] This method iteratively calculates the dynamic response of the bridge structure under different working conditions to obtain the displacement, stress, strain and other parameters of the bridge and its bearings.

[0041] The specific steps include: Input traffic load, wind load, earthquake action and temperature gradient and other working conditions. The Newmark-β method is used to iteratively calculate the dynamic response of the bridge structure at different time steps. Output the displacement, stress and strain results of the bearing and the entire bridge under various working conditions.

[0042] Through this dynamic response simulation, the displacement and force characteristics of the bearing and the overall bridge structure under actual working conditions can be obtained.

[0043] S2. Use multi-source sensors to collect displacement, vibration and temperature data of the support in real time, and perform data fusion and noise reduction processing; Specifically, in the correction method of the static displacement over-limit warning of the bridge bearing, the real-time acquisition and processing of multi-source sensor data is a key link. By acquiring the displacement, vibration and temperature data of the bearing in real time through the sensor, high-precision and synchronous data support can be provided for the subsequent model prediction and error correction. Since the sensor data is subject to noise interference and the amount of collected data is large, data fusion and noise reduction processing technology must be used to ensure the accuracy and consistency of the data. The purpose of data fusion is to integrate the data collected by different sensors to improve the reliability of the data; and noise reduction processing is used to eliminate the impact of environmental and equipment noise on the data, ensuring the quality of subsequent processing data. This embodiment describes in detail the specific steps of data acquisition, fusion and noise reduction processing of multi-source sensors.

[0044] Detailed description: In this embodiment, sensors are first arranged on the bearing and related bridge structures. The selected sensors include displacement sensors, acceleration sensors and temperature sensors to obtain static displacement, dynamic vibration and temperature data of the bearing.

[0045] Displacement sensors are installed at key positions of bridge supports to collect static displacement data of the supports. The sensors must have high precision and real-time performance to ensure that they can reflect the displacement changes of the supports under various working conditions.

[0046] Acceleration sensors are installed at the connection between the piers and the main beams of the bridge to monitor the vibration characteristics of the bearings and the bridge and collect acceleration data caused by vibration.

[0047] Temperature sensors are arranged around the support to monitor the temperature changes of the support and the surrounding environment and provide temperature compensation data.

[0048] The data collection frequency of all sensors is set according to the dynamic characteristics of the bridge to ensure that subtle changes in the bearings and bridge structure can be captured.

[0049] In this embodiment, a real-time data acquisition system based on multi-source sensors is constructed, which is responsible for synchronously acquiring displacement, vibration and temperature data from various sensors and storing the data in the form of time series.

[0050] The data collected by the sensor is defined as a time series dataset:

[0051] in, represents the sensor data collected by the i-th sensor at time t, and n is the total number of sensors.

[0052] Displacement data Represents the change of static displacement of the support with time; Vibration data Represents the dynamic acceleration of the support under different working conditions; Temperature data Provides real-time temperature information on the support surface.

[0053] The system is designed to ensure that data from all sensors can be collected synchronously and transmitted to the storage module of the monitoring center for subsequent data processing and analysis.

[0054] In this embodiment, wireless transmission technology is used to transmit the data collected by the sensor to the central monitoring system in real time. Data verification technology is used during the transmission process to ensure the integrity of the data during the transmission process. The real-time transmitted data will be stored in the database and marked with a timestamp to ensure the synchronization of the data during subsequent analysis.

[0055] The data collected by the sensor each time is stored in a fixed format, specifically: D(t)={X(t),t} Where D(t) is the complete data set collected at time t, including the data X(t) from multiple source sensors and the timestamp t.

[0056] In this embodiment, the data collected by the multi-source sensors are fused by data fusion technology. Since the displacement, vibration and temperature data come from different sensors, data fusion is required to ensure the uniformity and consistency of the data.

[0057] Adaptive Kalman filtering is used to fuse multi-source data. Kalman filtering can effectively filter out the noise in sensor data by dynamically adjusting the filter gain and perform spatiotemporal registration of each sensor data. The state space equation of Kalman filtering is as follows:

[0058] in: is the system state vector, representing the displacement and vibration of the bridge bearing; is the state transfer matrix, which represents the relationship between displacement and vibration; is the control input; is process noise and measurement noise, indicating the noise interference in the data; is the observation matrix, which represents the relationship between the sensor observation value and the state.

[0059] By adjusting the Kalman gain , Kalman filtering can adaptively correct sensor data according to noise characteristics and fuse data from different sensors to reduce errors. The calculation formula of Kalman gain is:

[0060] in, is the forecast error covariance, is the measurement noise covariance matrix.

[0061] In this embodiment, the data collected by the multi-source sensors is further subjected to noise reduction processing. The data collected by the sensors may contain environmental noise, measurement errors of the sensors themselves, and noise interference during the transmission process. In order to ensure the accuracy and reliability of the data, it is necessary to remove the noise through filtering technology.

[0062] In the noise reduction process, conventional noise reduction techniques such as low-pass filtering and Fourier transform are used in combination. The low-pass filter is used to filter out high-frequency noise signals in the sensor data, while the Fourier transform converts the data from the time domain to the frequency domain and removes the noise components of abnormal frequencies through frequency domain analysis.

[0063] The transfer function of the low-pass filter can be expressed as:

[0064] in, is the cutoff frequency of the filter, and s is a variable in the complex frequency domain. The filter can effectively filter out high-frequency noise exceeding the set frequency, thereby retaining the effective signals of the support displacement and vibration.

[0065] After noise reduction, the sensor data is cleaned and can be used for subsequent analysis and prediction model input.

[0066] S3. Establish a multi-level prediction model based on physical model and data-driven model to predict the future displacement trend of the bearing; The S3 step specifically includes the following steps: S3.1. Construction and prediction of physical model: Using the finite element model of the bridge, the bearing displacement is preliminarily predicted through mechanical equations and structural response simulation. The working load, temperature, and material property parameters are input into the physical model to perform physical prediction of the bearing displacement. Statics analysis and dynamics analysis combined with the finite element method are used to predict the bearing displacement, and the displacement response values ​​under normal traffic loads, strong winds, and earthquakes are obtained. S3.2. Build an LSTM data-driven model: extract features from the historical monitoring data collected by sensors, use normalization methods to process the raw data, and use LSTM to train the data-driven model; S3.3, Model training and validation: Train the LSTM model using monitoring data, evaluate the model’s prediction effect through cross-validation, use mean square error as the loss function, adjust the model’s hyperparameters, optimize the model’s prediction accuracy, and use a split validation set to validate the model’s generalization ability; S3.4, Fusion of physical model and data-driven model: The Bayesian weighted fusion algorithm is used to fuse the prediction results of the physical model and the LSTM data-driven model. The weighting coefficient is dynamically adjusted according to the model performance under different working conditions, so that the fusion model can maintain high-precision prediction under various working conditions; S3.5. Prediction of future trend of bearing displacement: Use the fusion model to predict the displacement trend of the bearing in the future time period, output the changes in static and dynamic displacement of the bearing in a certain period of time in the future, and make a revised prediction based on the current monitoring data.

[0067] In step S3.2, the input features of the LSTM data-driven model include historical data of bearing displacement, bearing surface temperature, vibration frequency of the bridge girder, and wind load parameters.

[0068] Specifically, in the correction method of the static displacement over-limit warning of the bridge bearing, predicting the future displacement trend of the bearing is one of the key steps. In order to obtain more accurate prediction results, a multi-level prediction method combining physical models and data-driven models is adopted. The physical model provides a basic prediction of the bearing displacement based on mechanical equations and structural response simulation through finite element analysis of the bridge; while the data-driven model uses historical monitoring data to perform time series prediction based on the long short-term memory network (LSTM). Through the combination of the two, the mechanical properties of the bridge structure and the dynamic characteristics of the data drive can be considered at the same time to ensure the prediction accuracy under various working conditions. This embodiment describes in detail the construction and fusion process of the multi-level prediction model.

[0069] Detailed Description In this embodiment, the construction of the physical model is based on the finite element model of the bridge, and its goal is to simulate the displacement changes of the bearings under different working conditions through mechanical equations and structural response analysis.

[0070] First, the physical model needs to input the following parameters: Working loads: including traffic loads, wind loads and earthquake effects; Temperature: Consider the impact of temperature gradient changes on bearing materials and bridge structures; Material properties: such as elastic modulus E and yield stress of the support material wait.

[0071] For normal traffic loads, strong winds and earthquakes, static and dynamic analysis combined with finite element method are used for prediction. The basic form of the mechanical equation is: KU=F Among them, K is the structural stiffness matrix, U is the displacement vector, and F is the external load vector. Through static analysis and dynamic response calculation, the displacement response value of the support under different working conditions can be obtained. The prediction results of the physical model provide a preliminary displacement prediction basis for the subsequent data-driven model.

[0072] In this embodiment, the LSTM data-driven model is based on the historical monitoring data of the sensor and captures its dynamic change law by learning the time series characteristics of the support displacement.

[0073] First, feature extraction is performed on historical monitoring data. The input features include: Support displacement history data: past support static and dynamic displacement records; Support surface temperature: historical changes in support temperature, used to compensate for the effects of temperature on material properties; Vibration frequency of the bridge girder: reflects the dynamic impact of the bridge on the support when it vibrates; Wind load parameters: As an input of environmental factors, it is used to reflect the impact of wind on support displacement.

[0074] These data are normalized to ensure that the data set has good statistical properties. The structure of the LSTM model includes an input layer, multiple hidden layers, and an output layer. During the training process, the future support displacement is predicted through time series input. The state update equation of the LSTM unit is as follows:

[0075] in, They are input gate, forget gate, and output gate respectively. is the state of the memory cell, is the output of the hidden layer.

[0076] The LSTM model learns the temporal characteristics of support displacement through the above formula, captures its long-term dependencies, and outputs displacement prediction values ​​at future moments.

[0077] In this embodiment, the historical monitoring data collected by the sensor is used to train the LSTM model. During the training process, the mean square error is used as the loss function to calculate the error between the model prediction value and the actual measurement value. The specific formula is as follows:

[0078] in, is the predicted value of the LSTM model, is the actual measurement value of the sensor, and N is the number of data samples.

[0079] During the model training process, the hyperparameters of the LSTM model, including the learning rate, number of hidden layer units, and historical sequence length, are adjusted to ensure the convergence and prediction accuracy of the model. The prediction effect of the model is evaluated through cross-validation, and the generalization ability of the model is tested using a split validation set to ensure that it still has high prediction accuracy under complex working conditions.

[0080] In this embodiment, a Bayesian weighted fusion algorithm is used to fuse the prediction results of the physical model and the LSTM data-driven model. The physical model provides a displacement prediction basis based on mechanical calculations, while the LSTM data-driven model uses historical monitoring data to capture the temporal characteristics of displacement. Through fusion, the advantages of both can be combined to further improve the accuracy of the prediction.

[0081] The basic formula of Bayesian weighted fusion is:

[0082] in, is the predicted value after fusion, is the predicted value of the physical model, is the predicted value of the LSTM data-driven model, and is the weighting coefficient.

[0083] Weighting coefficient and The posterior probability of historical data is used to update the prediction results, ensuring that the weights can be adaptively adjusted under different working conditions. This fusion algorithm ensures that the prediction results of the physical model and the data-driven model are reasonably integrated under complex working conditions, thereby improving the robustness of the overall prediction.

[0084] In this embodiment, the fusion model is used to predict the future displacement trend of the support. Specifically, the fusion model can output the changes in the static displacement and dynamic displacement of the support in the future time period. This prediction process combines the current monitoring data and historical data, and can correct the errors in the model prediction in real time.

[0085] Through the following prediction formula, the fusion model outputs the displacement trend of the support at the future time t+Δt:

[0086] Among them, X(t) is the monitoring data at the current moment, θ is the parameter vector of the model, is the predicted value at the current moment. By combining the current monitoring data, the predicted value can be dynamically corrected to obtain a more accurate bearing displacement prediction result.

[0087] S4, real-time monitoring of prediction errors through error feedback mechanism, and dynamic adjustment of prediction model parameters through adaptive correction algorithm; The S4 step specifically includes the following steps: S4.1. Comparison between predicted value and actual value: Compare the predicted value of the support displacement output by the prediction model with the actual support displacement value collected by the sensor in real time to calculate the prediction error; S4.2. Establish an error feedback mechanism: record each prediction error through the error feedback mechanism, and feed the error back to the prediction model to correct the model parameters; S4.3, Recursive Kalman filter error correction: Recursive Kalman filter is used for error correction; S4.4. Adaptive adjustment of model parameters: Dynamically adjust material parameters in the physical model and hyperparameters in the data-driven model based on the corrected errors.

[0088] The calculation formula for the prediction error in step S4.1 is as follows: , where e(t) is the prediction error, is the support displacement value actually measured by the sensor, are the support displacement values ​​predicted by the model.

[0089] In step S4.4, the model parameters that are adaptively adjusted include: Material elastic modulus and yield stress in physical models; Learning rate, number of hidden units, and history length in data-driven models.

[0090] Specifically, in the correction method of the static displacement over-limit warning of bridge bearings, real-time monitoring of prediction errors and dynamic adjustment of model parameters are key links to improve prediction accuracy. By establishing an error feedback mechanism, the system can compare the model prediction value with the actual observation value of the sensor in real time, obtain the error and use the adaptive correction algorithm to iteratively optimize the model. Specifically, correcting the prediction error through a recursive Kalman filter can effectively reduce the impact of the prediction error on the model accuracy, while dynamically adjusting the material parameters in the physical model and the hyperparameters in the data-driven model to ensure that the model can continue to adapt to changing working conditions. The following is the detailed technical implementation of this step.

[0091] Detailed Description In this embodiment, the output of the prediction model is first compared with the actual observation data in real time. Specifically, the support displacement value predicted by the model is compared with the actual support displacement value collected by the sensor, so as to calculate the prediction error e(t).

[0092] The calculation formula for the prediction error is:

[0093] in: e(t) is the prediction error at time t; δsactual(t)\delta_s^{actual}(t)δsactual(t) is the support displacement value actually measured by the sensor; are the support displacement values ​​predicted by the model.

[0094] Through this real-time error calculation, it is possible to capture the deviation of the model when predicting future displacement trends and identify the differences between the model and actual working conditions.

[0095] In this embodiment, an error feedback mechanism is established to record and process the error information in each prediction. This mechanism feeds back the prediction error e(t) to the model each time, prompting the model to adjust according to the actual monitoring value and error information. The specific steps are as follows: Error recording: After each model prediction output, the calculated error value e(t) is recorded in the feedback system.

[0096] Feedback process: Through the feedback mechanism, the error value is fed back to the physical model and data-driven model, so that the model can perceive the size and direction of its own prediction error.

[0097] Through this feedback method, the system can gradually correct the model parameters so that subsequent prediction results can better conform to the actual situation.

[0098] S4.3: Error Correction of Recursive Kalman Filter In this embodiment, in order to further correct the prediction error, a recursive Kalman filter is used to correct the error. The recursive Kalman filter can dynamically adjust the prediction value in real time according to the historical error and sensor noise.

[0099] The state update formula of Kalman filter is:

[0100] in: is the predicted value of the support displacement after correction; is the Kalman gain, which indicates the influence of the current error on the next prediction; e(t) is the prediction error at the current moment.

[0101] Kalman Gain The calculation formula is:

[0102] in, is the error covariance, is the observation matrix, is the measurement noise covariance.

[0103] Through the recursive Kalman filter, the future prediction results can be adjusted according to the current error after each prediction step, ensuring that the model reduces error accumulation during the gradual learning and correction process.

[0104] In this embodiment, based on the error corrected by Kalman filtering, the system adaptively adjusts the parameters of the physical model and the data-driven model. The adjusted parameters include: Material parameters in the physical model: Elastic modulus EThe elastic modulus of the bearing material directly affects the displacement prediction accuracy of the bearing. When the prediction error indicates that the model's response to the bearing displacement is insufficient or excessive, this parameter is adaptively adjusted to correct the mechanical response of the model.

[0105] Yield stress The yield stress of the support in the plastic region determines its deformation behavior under high load conditions. Based on error feedback, this parameter is adjusted in time to match the actual support material properties.

[0106] Hyperparameters in data-driven models: Learning rate: The learning rate of the LSTM model controls the update step size of the model. By adaptively adjusting the learning rate, the learning speed of the model can be accelerated or slowed down to adapt to different working conditions.

[0107] Number of hidden layer units: The number of hidden layer units of the LSTM model determines the model's ability to remember historical data. If the error is large, the system may increase the number of hidden layer units to enhance the model's ability to learn complex data features.

[0108] History sequence length: The history sequence length determines how much historical data the LSTM model can use to make future predictions. When the working conditions are complex and the prediction error is large, appropriately increasing the history sequence length can improve the model's ability to predict future displacements.

[0109] The purpose of adaptive adjustment is to ensure that the physical model and data-driven model can adapt to different working conditions in a timely manner and make reasonable adjustments to each prediction error, thereby improving future prediction accuracy. S5. Set dynamic thresholds and trigger warning signals when the static displacement of the bearing exceeds the limit. Combined with the overall structural health status of the bridge, generate maintenance recommendations.

[0110] Step S5 specifically includes the following steps: S5.1. Set initial warning threshold: Based on historical monitoring data and structural design specifications, set the initial warning threshold of the static displacement of the bearing; S5.2. Dynamically adjust the threshold: Use the Bayesian update rule to dynamically adjust the warning threshold based on the real-time monitored support displacement data; S5.3, triggering of over-limit warning signal: when the predicted value of static displacement of the bearing exceeds the dynamically set threshold, the system automatically triggers the warning signal to remind the maintenance personnel that the bearing may have over-limit displacement; S5.4. Generate maintenance recommendations: The system generates maintenance recommendations based on the bearing displacement warning and the health status of the overall bridge structure; S5.5. Subsequent monitoring and feedback: After the early warning is triggered, the system continues to monitor the bearing displacement and the overall health of the bridge in real time, and compares the latest monitoring data with the early warning results.

[0111] The dynamic threshold setting formula in step S5.2 is: , where T(t) is the warning threshold after dynamic adjustment, is the initial warning threshold, α is the dynamic adjustment factor, and X(t) is the support displacement data monitored in real time.

[0112] In step S5.3, the early warning signal triggering mechanism for bearing displacement exceeding the limit adopts dual triggering conditions: When the predicted support displacement When the dynamic threshold T(t) is exceeded and the overall structural health score of the bridge is lower than the preset threshold, an early warning signal is triggered.

[0113] Specifically, in the correction method of the static displacement over-limit warning of bridge bearings, the setting of dynamic thresholds and the triggering of warning signals are important means to ensure bridge safety. The system sets the initial warning threshold and dynamically adjusts the threshold according to the real-time monitoring data, so that when the static displacement of the bearing exceeds the limit, a warning signal is issued in time, and maintenance recommendations are generated in combination with the health status of the overall bridge structure. This mechanism triggers warnings through dual conditions, reducing false alarms while ensuring that the bridge is in a healthy and safe operating state.

[0114] Detailed Description In this embodiment, the initial warning threshold of the static displacement of the bearing is set based on the historical monitoring data and design specifications of the bridge. The setting of the initial warning threshold needs to take into account the following factors: Historical monitoring data: Based on the support displacement data monitored in the past, analyze the displacement variation range under normal operation, extract common displacement fluctuations, and set a reasonable initial threshold.

[0115] Structural design specifications: Based on the maximum allowable displacement specified in the bridge structure design, ensure that the set initial threshold does not exceed the design limit to ensure safe operation of the bridge.

[0116] Set the initial threshold It is a preliminary warning standard for the static displacement of the bearing and serves as a benchmark for subsequent dynamic adjustments.

[0117] In this embodiment, in order to make the early warning system adapt to the actual operation status of the bridge, the Bayesian update rule is used to dynamically adjust the early warning threshold. The system adjusts the initially set early warning threshold based on the real-time monitoring support displacement data and the current health status. The formula is as follows:

[0118] in: T(t) is the warning threshold after dynamic adjustment at time t; is the initial warning threshold; α is a dynamic adjustment factor that controls the speed and amplitude of threshold adjustment; X(t) is the support displacement data monitored in real time.

[0119] The Bayesian update rule ensures that the system can adjust the threshold in a timely manner according to the real-time monitoring of the support displacement data, avoiding the false alarm or missed alarm situation that may be caused by the static threshold setting.

[0120] In this embodiment, the early warning system uses dual trigger conditions to improve the accuracy of the early warning. The system will simultaneously monitor the static displacement prediction value of the bearing and the health status of the overall bridge structure. Only when both conditions are met at the same time will the early warning signal be triggered.

[0121] The specific trigger conditions are as follows: The static displacement of the support exceeds the dynamic threshold: when the predicted static displacement of the support When the dynamically adjusted warning threshold T(t) is exceeded, it serves as a preliminary warning condition.

[0122] The overall structural health score of the bridge is lower than the preset threshold: The system combines the overall structural health score of the bridge, which is based on the monitoring data of other parts of the bridge (such as stress and vibration of piers and main beams). When the overall health score of the bridge is lower than the set threshold, it serves as the second warning condition.

[0123] Only when the predicted bearing displacement exceeds the dynamic threshold and the bridge health score is lower than the preset value will the system issue a warning signal to alert maintenance personnel that the bearing may have excessive displacement problems.

[0124] In this embodiment, after the warning signal is triggered, the system automatically generates maintenance suggestions based on the bearing displacement warning situation and the overall structural health status of the bridge. The maintenance suggestions include: Inspection content: The system will recommend detailed inspection of key parts such as supports, piers, main beams, etc., especially areas with abnormal displacement and stress concentration.

[0125] Emergency treatment measures: If the bearing displacement exceeds the limit seriously, the system will give corresponding emergency treatment measures, such as load limit or bridge closure, to ensure the safe operation of the bridge.

[0126] Subsequent monitoring recommendations: Based on the current warning situation, the system will recommend increasing the monitoring frequency of key bridge parts and obtaining more real-time data for subsequent tracking.

[0127] The generated maintenance recommendations are sent in the form of reports to bridge management and maintenance personnel so that they can take appropriate actions.

[0128] In this embodiment, after the warning signal is triggered, the system will continue to monitor the displacement of the support and the health status of the overall bridge structure in real time, and compare the latest monitoring data with the warning results. The process includes the following steps: Real-time monitoring: The system continuously obtains real-time data from sensors, especially parameters such as displacement, vibration and temperature of the support.

[0129] Early warning assessment: Compare the latest data with the early warning conditions. If the bearing displacement returns to normal or the bridge health score rises back to the safe range, the system automatically releases the early warning state; otherwise, the early warning state is maintained and a reminder is issued.

[0130] Feedback mechanism: By comparing with historical warning data, the accuracy of warnings is evaluated, and warning parameters and thresholds are adjusted to further optimize system performance.

[0131] Through subsequent monitoring and feedback mechanisms, the system can dynamically adjust its threshold settings and early warning strategies to ensure long-term monitoring of the safe operation of the bridge.

[0132] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A correction method for bridge bearing static displacement over-limit warning based on a prediction model, characterized in that: The following steps are involved: S1. By establishing a coupled finite element model of the bridge bearing and the overall bridge structure, the dynamic response of the bearing and the overall structure under various working conditions is simulated; S2. Use multi-source sensors to collect the displacement, vibration and temperature data of the support in real time, and perform data fusion and noise reduction processing; S3. Establish a multi-level prediction model based on physical model and data-driven model to predict the future displacement trend of the bearing; S4, real-time monitoring of prediction errors through error feedback mechanism, and dynamic adjustment of prediction model parameters through adaptive correction algorithm; S5. Set dynamic thresholds and trigger warning signals when the static displacement of the bearing exceeds the limit. Combined with the overall structural health status of the bridge, generate maintenance recommendations.

2. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 1 is characterized in that: The S1 step specifically includes the following steps: S1.

1. Construct a three-dimensional finite element model of the entire bridge, model the main beam, piers, abutments and supports through finite element analysis, and define the linear and nonlinear characteristics of the materials; S1.

2. Establish a local refined model of the support, use nonlinear spring elements or contact elements, and define the elastic and plastic deformation behaviors of the support; S1.3, coupling the nonlinear mechanical behavior of the support with the overall bridge structure model, and realizing the displacement and force transmission between the support and the main beam through node setting; S1.

4. Set the operating conditions of traffic load, wind load, earthquake action and temperature gradient change, and define the load size and distribution; S1.

5. Use the Newmark-β integration method to perform dynamic response simulation and calculate the displacement, stress and strain of the bridge as a whole and the bearings under various working conditions.

3. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 1 is characterized in that: The S3 step specifically includes the following steps: S3.

1. Construction and prediction of physical model: Using the finite element model of the bridge, the bearing displacement is preliminarily predicted through mechanical equations and structural response simulation. The working load, temperature, and material property parameters are input into the physical model to perform physical prediction of the bearing displacement. Statics analysis and dynamics analysis combined with the finite element method are used to predict the bearing displacement, and the displacement response values ​​under normal traffic loads, strong winds, and earthquakes are obtained. S3.

2. Build an LSTM data-driven model: extract features from the historical monitoring data collected by sensors, use normalization methods to process the raw data, and use LSTM to train the data-driven model; S3.3, Model training and validation: Train the LSTM model using monitoring data, evaluate the model’s prediction effect through cross-validation, use mean square error as the loss function, adjust the model’s hyperparameters, optimize the model’s prediction accuracy, and use a split validation set to validate the model’s generalization ability; S3.4, Fusion of physical model and data-driven model: The Bayesian weighted fusion algorithm is used to fuse the prediction results of the physical model and the LSTM data-driven model. The weighting coefficient is dynamically adjusted according to the model performance under different working conditions, so that the fusion model can maintain high-precision prediction under various working conditions; S3.

5. Prediction of future trend of bearing displacement: Use the fusion model to predict the displacement trend of the bearing in the future time period, output the changes in static and dynamic displacement of the bearing in a certain period of time in the future, and make a revised prediction based on the current monitoring data.

4. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 1 is characterized in that: The S4 step specifically includes the following steps: S4.

1. Comparison between predicted value and actual value: Compare the predicted value of the support displacement output by the prediction model with the actual support displacement value collected by the sensor in real time to calculate the prediction error; S4.

2. Establish an error feedback mechanism: record each prediction error through the error feedback mechanism, and feed the error back to the prediction model to correct the model parameters; S4.3, Recursive Kalman filter error correction: Recursive Kalman filter is used for error correction; S4.

4. Adaptive adjustment of model parameters: Dynamically adjust material parameters in the physical model and hyperparameters in the data-driven model based on the corrected errors.

5. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 4 is characterized in that: The calculation formula of the prediction error in step S4.1 is as follows: , where e(t) is the prediction error, is the support displacement value actually measured by the sensor, are the support displacement values ​​predicted by the model.

6. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 1 is characterized in that: The step S5 specifically includes the following steps: S5.

1. Set initial warning threshold: Based on historical monitoring data and structural design specifications, set the initial warning threshold of the static displacement of the bearing; S5.

2. Dynamically adjust the threshold: Use the Bayesian update rule to dynamically adjust the warning threshold based on the real-time monitored support displacement data; S5.3, triggering of over-limit warning signal: when the predicted value of static displacement of the bearing exceeds the dynamically set threshold, the system automatically triggers the warning signal to remind the maintenance personnel that the bearing may have over-limit displacement; S5.

4. Generate maintenance recommendations: The system generates maintenance recommendations based on the bearing displacement warning and the health status of the overall bridge structure; S5.

5. Subsequent monitoring and feedback: After the early warning is triggered, the system continues to monitor the bearing displacement and the overall health of the bridge in real time, and compares the latest monitoring data with the early warning results.

7. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 6 is characterized in that: The dynamic threshold setting formula in step S5.2 is: , where T(t) is the warning threshold after dynamic adjustment, is the initial warning threshold, α is the dynamic adjustment factor, and X(t) is the support displacement data monitored in real time.

8. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 6 is characterized in that: In step S5.3, the early warning signal triggering mechanism for the bearing displacement exceeding the limit adopts dual triggering conditions: When the predicted support displacement When the dynamic threshold T(t) is exceeded and the overall structural health score of the bridge is lower than the preset threshold, an early warning signal is triggered.

9. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 3 is characterized in that: In the step S3.2, the input features of the LSTM data-driven model include support displacement history data, support surface temperature, vibration frequency of the bridge main beam and wind load parameters.

10. The correction method for bridge bearing static displacement over-limit warning based on prediction model according to claim 4 is characterized in that: In step S4.4, the model parameters to be adaptively adjusted include: Material elastic modulus and yield stress in physical models; Learning rate, number of hidden units, and history length in data-driven models.

Images