Bridge Safety Early Warning Method and System Based on the Fusion of Data Model and Physical Model

Abstract

The present invention belongs to the technical field of bridge engineering, and specifically discloses a bridge safety warning method and system based on the fusion of data model and physical model. It includes: in the first stage, a neural network monitoring point excitation-response mapping relationship model established based on the full-bridge finite element model in the initial state is used to calculate the response information corresponding to the on-site excitation, and the response monitoring signal is compared to determine whether the structure is damaged from the perspective of the full bridge; in the second stage, data expansion from the monitoring points to the non-monitoring points is carried out, and the damage location and damage level are determined by comparing and setting thresholds; in the third stage, the adjustment parameters of the model are calculated according to the excitation-response signals of the sensors at the monitoring points, the finite element structure model is updated, and by combining the state information carried by the structure model, the remaining bearing capacity of the structure is calculated, and the service life of the component is determined. The present invention provides a safety assessment method that integrates monitoring data and physical models for structural health monitoring, which has clear physical meaning, high real-time performance, and reliable results.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge engineering, and more specifically, relates to a bridge safety early warning method and system based on the fusion of a data model and a physical model. Background Art

[0002] With the opening and operation of cross-sea cluster projects such as the Hong Kong-Zhuhai-Macao Bridge in recent years, "factory-based" and "standardized" prefabricated and assembled structures have been increasingly widely used in the construction of bridge engineering. Subsequently, the issue is how to achieve safety assessment and early warning for bridge structures with continuous spans of dozens of kilometers through a complete set of "standardized" and "intelligent" health monitoring systems. In the structural safety assessment of bridge model correction, there are mainly the following two problems:

[0003] Firstly, during the service process of prestressed concrete bridges, the environmental effects in the offshore area are prominent, and the non-linear degree of load types is high. In the offshore environment, reinforced concrete is prone to chloride ion corrosion, which changes the acidity and alkalinity of the concrete body. Affected by this, compared with inland areas, the prestress relaxation loss of offshore bridge structures is aggravated. For load types, stay cables are subjected to wind-induced vibration, the thermal expansion and contraction effect caused by temperature difference, and the dynamic evolution of the structure caused by vehicle driving. Under the action of such complex environmental factors and load conditions, the real-time performance of structural calculation will decrease.

[0004] Secondly, the physical model information calculated by structural components in different parts is diverse, and the heterogeneity of the information affects the calculation speed of structural safety assessment and the real-time generation of assessment results. For the diversity of physical model information, finite element models of different scales should be used for targeted analysis, and corresponding safety assessment methods should be proposed for evaluation.

[0005] Therefore, aiming at the performance degradation causes and superimposed loads of offshore prestressed concrete bridges, it is not only a development trend but also a necessity to standardize and design a complete set of on-site non-destructive monitoring systems, and be able to infer the damage information of unmonitored points of the structure based on the information collected at on-site monitoring points by the non-destructive monitoring system, and excavate the potential loads of the structure. And due to the significant "factory-based" and "standardized" degree of offshore prestressed concrete bridge structures, the "standardization" of the monitoring process is also very crucial. This helps to improve the speed of mapping data to the physical model, establish a bridge operation and maintenance model integrating data and physics, evaluate the service status of the structure, and put forward credible maintenance opinions and suggestions.

[0006] In recent years, with the development of science and technology, numerous monitoring means and evaluation systems have been evolving towards enhancing the "intelligence" in the management and maintenance process of bridge engineering. In the field of digital-physical integration, the technological development in aspects related to model updating is as follows: Currently, model updating tends to use computer-aided means to establish the mapping relationship between sensor signals (excitation, response information) and model information (stiffness matrix, structural vibration mode), thereby updating the finite element model to analyze the structural state or directly quantifying and calculating the structural safety for state evaluation. The main analysis means at the present stage is to establish the mapping relationship between signals and model information through means such as PSO and BP. Its analysis speed is relatively fast, but the physical and mechanical information between input and output during the analysis process is not clear, and it is difficult for the network to respond promptly to incorrect information. In the field of safety evaluation, the analytic hierarchy process and expert evaluation method are mainly adopted to evaluate the safety state of bridge structures, supplemented by big data analysis methods such as neural networks. This method can effectively achieve rapid evaluation through the analysis methods of cloud computing and servers, but the evaluation method lacks the ability of dynamic evolution. Most importantly, during the safety evaluation process, usually only a single type of load is evaluated, without considering the situation of dynamic coupling of multiple load types through physical relationships, lacking accuracy in practical engineering applications.

[0007] However, it is difficult to achieve real-time processing of the structural information of the on-site physical model during the current evaluation process, analyze the actual stress state of the structure in the current state, evaluate the deterioration trend of materials, and calculate the remaining bearing capacity of the structure and stress redistribution. A fully functional structural health monitoring system for offshore prestressed concrete bridges urgently needs to be established to achieve real-time structural updating and safety state evaluation in multi-source data fusion. Summary of the Invention

[0008] Aiming at the above defects or improvement requirements of the prior art, the present invention provides a bridge safety early warning method and system based on the fusion of data model and physical model. It realizes real-time analysis of sensor monitoring information through a surrogate model, processes and mines the structural model information, evaluates the structural safety status, and feeds back and maps the calculation results of the surrogate model to the established finite element model to achieve real-time updating of the finite element model, display the current state of the structure, and improve the accuracy of structural performance analysis. Through finite element analysis, not only can the location and damage level of potential damage be discovered, but also multiple different types of loads can be applied to the finite element model in advance, expanding the neural network analysis data set, providing a reliable and real-time updated training set for neural network pre-training, forming a real-time excitation-load mapping relationship with a complete physical relationship, and improving the accuracy of the output data of the neural network.

[0009] To achieve the above object, according to one aspect of the present invention, a bridge safety early warning method based on the fusion of data model and physical model is proposed, including the following steps:

[0010] Step S1, construct a multi-scale finite element model of the bridge structure, calculate the structural response by inputting simulated excitation data into the finite element model, and generate a data set including excitation signals, response signals, and structural model information;

[0011] Step S2, construct a neural network model, and train the neural network model using the corresponding data sets according to different calculation purposes of the neural network. The neural network model includes: neural network F-R, neural network R-R, and neural network FR-FE;

[0012] Step S3, the first safety assessment stage: use neural network F-R to calculate the excitation-monitoring point response at the on-site monitoring points, obtain the calculated response based on the finite element model in the initial state, and determine whether the bridge structure is abnormal or damaged;

[0013] Step S4, the second safety assessment stage: use neural network R-R to analyze the response information of the input monitoring points in Step S3, expand to obtain the calculated response of non-monitoring points, and perform classification of the monitoring threshold levels of the calculated response to determine the location and damage level of the components defined as damaged;

[0014] Step S5, the third safety assessment stage: analyze the excitation signal-response signal collected on-site through neural network FR-FE, obtain the correction parameters for updating the finite element model, and calculate the overall bearing capacity level and remaining service life of the bridge based on the updated multi-scale finite element model.

[0015] As a further preference, Step S1 includes the following steps:

[0016] Step S11, divide the construction sections that need to establish refined models according to the most unfavorable positions of the bridge and the positions of on-site embedded sensors;

[0017] Step S12, divide the finite element structure into multiple components according to the sensor data type and layout points, establish finite element models at the component scale and the full-bridge scale according to the characteristic information provided by the engineering drawings, and use different mesh scales in the same finite element model to simulate the monitored components and non-monitored components to establish a multi-scale finite element model;

[0018] Step S13, use the multi-scale finite element model to simulate and analyze the main loads and performance changes of the bridge, and establish a data set by exporting the calculation information and excitation information at different positions of the initial structural finite element model.

[0019] As a further preference, in Step S13, the structural response is calculated by inputting simulated excitation data into the excitation monitoring points of the finite element model, that is, a suitable degradation model is selected for different structural models, a geometric parameter degradation model of the bridge is established, and the degradation range of the finite element model calculation parameters compared with the initial model is determined.

[0020] As a further preference, in step S2, by inputting simulated loads into the excitation monitoring points of the full-bridge scale finite element model in the initial state, the calculated data of the response monitoring points is extracted as the training data set of the neural network F-R;

[0021] If the bridge structure is judged to be intact, the response signals of the non-monitoring points in the neural network R-R data set are obtained by inputting the simulated loads into the full-bridge scale finite element model in the initial state; if the bridge structure is judged to be damaged, its non-monitoring point response signals are obtained by inputting the simulated loads into the modified full-bridge scale finite element model, and according to the evaluation status of the first safety assessment stage, the corresponding node response data is extracted to obtain the monitoring point response - non-monitoring point response training set;

[0022] Based on the multi-scale finite element model, the excitation sensor data and response sensor data in each finite element model node are integrated to obtain the FR-FE training set.

[0023] As a further preference, step S3 includes the following steps:

[0024] Step S31, divide the signals collected on-site by the sensors of the on-site monitoring points for excitation - response into two parts. One part of the monitoring point excitation is used in the training process of the neural network F-R, and the other part of the monitoring point response is used as the validation set for the neural network F-R;

[0025] Step S32, input the excitation monitoring point signals in step S31 into the full-bridge finite element model to obtain the neural network F-R, and obtain the calculated response signals of the response monitoring points based on the initial finite element model;

[0026] Step S33, compare the calculated response signals of the response monitoring points in step S32 with the on-site monitoring response signals in step S31, calculate the mean square error of the measured data and calculated data of all response monitoring points, and determine whether the signals obtained by calculating through the full-bridge scale finite element model in the initial state are consistent with the on-site monitoring point signals;

[0027] Step S34, if the signal comparison error is within the given error range, it indicates that the offshore bridge structure has not suffered damage macroscopically and the macro safety state of the structure is good; if the signal comparison error exceeds the given range, it indicates that the structure can be judged to be damaged by calculating the on-site sensor signals through the neural network F-R at the macro scale;

[0028] Step S35, use the neural network F-R to calculate the excitation of the on-site monitoring points - the response of the monitoring points, and obtain the calculated response based on the finite element model in the initial state to determine whether the bridge structure is abnormal or damaged.

[0029] As a further preference, step S4 includes the following steps:

[0030] Step S41: Calculate the current structural model state evaluation parameters based on the calculated responses of all structural nodes at the full-bridge scale obtained from the neural network R-R and the set warning threshold and divide the values of the state evaluation parameters into damage levels and define the thresholds for each state evaluation parameter ;

[0031] Step S42: Arrange the structural damage levels in step S41 in descending order and analyze whether the state evaluation parameters at the most dangerous position of the structure exceed the given warning threshold range;

[0032] Step S43: If the state evaluation parameters exceed the warning threshold range, it is necessary to locate the damage position and its corresponding damage level; conversely, if the state evaluation parameters of all nodes do not exceed the warning threshold range, all nodes of the entire structure are in a safe state.

[0033] As a further preference, in step S41, the state evaluation parameters are divided into: <2.0%, 2.0%< <5.0%, 5.0%< <20.0%, >20.0% four levels.

[0034] Step S5 includes the following steps:

[0035] Step S51: Input the excitation-monitoring point response signal of the on-site sensor monitoring points into the neural network FR-FE and extract the weight matrix of the hidden layer of the neural network;

[0036] Step S52: Based on the inversion function in step S2, calculate the calculation parameters of the multi-scale finite element model, real-time correct the finite element model, use the corrected multi-scale finite element model to calculate the remaining load-bearing capacity of the structure, and evaluate the overall safety information of the structure;

[0037] Step S53: Based on the parameters of the corrected vectorized finite element model and the parameters of the original vectorized finite element model, calculate the material degradation index at each node, the structural cross-sectional area degradation index and the structural cross-sectional moment of inertia degradation index , deduce the degradation model of the material properties in the degradation index according to the on-site environmental factor influence data monitored by the on-site sensor, and predict the remaining service life of the material according to the material degradation index and fit , Derive the correlation between the reduction of cross-section properties and time from the time-varying curve to obtain the degradation function of the cross-section.

[0038] Step S54: For the evaluation of the refined model part of the finite element model, according to the material degradation index , the structural cross-sectional area degradation index and the structural cross-sectional moment of inertia degradation index correct the finite element component scale model, and combine the calculation results of the finite element model to evaluate the component safety information and calculate the remaining bearing capacity of the component.

[0039] As a further optimization, the following steps are also included:

[0040] Directly correct the finite element model using the structural damage information directly monitored by the on-site sensors in S3. That is, the final correction of the finite element model needs to be based on the consistency between the structural damage and the on-site monitored damage. Update the finite element model according to the degradation coefficients of the cross-section and materials analyzed from the monitoring signals in S5 and the adjustment parameters of the corrected structural damage, and use the updated finite element structural model to generate a new dataset for training the neural network model and correcting the finite element model.

[0041] According to another aspect of the present invention, a bridge safety early warning system based on the fusion of data model and physical model is also provided, including:

[0042] The first main control module is used to construct a multi-scale finite element model of the bridge structure, calculate the structural response by inputting simulated excitation data into the finite element model, and generate a dataset.

[0043] The second main control module is used to construct a neural network model and train the neural network model using the dataset. The neural network model includes: neural network F-R, neural network R-R, and neural network FR-FE.

[0044] The third main control module is used to calculate the excitation-monitoring point response of the on-site monitoring points using the neural network F-R, obtain the calculated response based on the initial state finite element model, and compare and analyze to determine whether the bridge structure is abnormal or damaged.

[0045] The fourth main control module is used to analyze the monitoring point response information input in the third main control module using the neural network R-R, expand to obtain the calculated response of non-monitoring points, and perform grading to determine the damage level and the location of the damage parameters exceeding the predetermined threshold.

[0046] The fifth main control module is used to analyze the excitation signal-response signal collected on-site through neural network FR-FE, obtain the correction parameters for updating the finite element model, and calculate the overall bearing capacity level and remaining service life of the bridge structure based on the updated multi-scale finite element model.

[0047] Generally speaking, compared with the prior art, the above technical solutions conceived by the present invention mainly have the following technical advantages:

[0048] 1. The three-stage bridge safety warning method based on the fusion of data model and physical model of the present invention can quickly and comprehensively analyze on-site sensor signals, perform threshold warning on the structural health status through excitation signals and response signals, and correct the finite element model to comprehensively and carefully evaluate the safety performance and service life of the overall structure / components, improving the risk assessment efficiency and ensuring the accuracy of the assessment results.

[0049] 2. The present invention can realize the real-time dynamic update of the structural model on the finite element calculation platform or the structural information management platform, facilitating the supervision party to obtain information for judging the structural state during the interval of regular structural safety inspections. Similarly, after regular structural safety inspections, the data reliability of both the platform model and the inspection report content can be verified by comparing the inspection report data.

[0050] 3. The present invention generates a training set by real-time updating the finite element model to establish a neural network model for rapid processing of on-site signals, realizes rapid processing of on-site signals for different purposes, and evaluates the structure in three stages. For different safety assessment situations of the structure in each assessment stage, the detected damage and potential damage can be quantitatively and timely fed back by setting warning thresholds.

[0051] 4. The safety assessment process proposed by the present invention calculates the remaining service life of structural components based on the statistical prediction value of load excitation monitored by sensors, and evaluates the remaining bearing capacity of the structure based on the corrected finite element model. Among them, the finite element model serves as a reliable physical and mechanical basis, and multi-scale simulation shows the damage information of the structure through the finite element model at the component scale. The safety state of the bridge as a whole and its components is evaluated through the structural state variables (stiffness matrix, structural vibration mode) revealed by the finite element model, and the physical meaning of the assessment process is clear and the assessment results are reliable. Description of the Drawings

[0052] Figure 1 is a schematic flow chart of a bridge safety warning method based on the fusion of data model and physical model according to an embodiment of the present invention;

[0053] Figure 2 is a schematic diagram of a multi-scale finite element model established with a prestressed simply supported beam as an example according to an embodiment of the present invention;

[0054] Figure 3 This is the flowchart of the first-stage safety assessment related to the embodiments of the present invention;

[0055] Figure 4 This is the flowchart of the second-stage safety assessment related to the embodiments of the present invention;

[0056] Figure 5 This is the flowchart of the third-stage safety assessment related to the embodiments of the present invention. Detailed implementation manners

[0057] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0058] Embodiment 1

[0059] As Figures 1 to 5 shown, a bridge safety early warning method based on the fusion of a data model and a physical model provided in this embodiment adopts a safety assessment method for structural assessment in three stages. In the first stage, a neural network monitoring point excitation-response mapping relationship model established based on the initial state full-bridge finite element model is used to quickly calculate the response information obtained by exciting the on-site sensors, and the monitoring signals are compared to determine whether the structure is damaged from the perspective of the whole bridge. After determining damage in the first stage, it enters the second stage. Through data expansion from the monitoring points to the non-monitoring points, the damage location and damage level are determined by comparing and setting thresholds. Whether or not it enters the second stage, it will enter the third stage. First, the adjustment parameters of the model are calculated based on the excitation-response signals of the monitoring point sensors, the finite element structure model is updated, the remaining bearing capacity of the structure is calculated by a numerical calculation method for calculating the membership degree of the structure, and the service life of the component is determined by an uncertainty method of probability statistics.

[0060] In this embodiment, a neural network is selected as the technical means to replace the healthy initial finite element model in the first-stage assessment and establish a mapping relationship between the structural excitation at the set monitoring points and the structural response at the monitoring points; in the second stage, based on the updated finite element model, the neural network is used to calculate and deduce the data at the non-monitoring points through the monitoring point data to expand the data positions of the non-monitoring points; the third stage is used to analyze the excitation-response signals of the on-site sensors and calculate the calculation parameters of the finite element model through an inverse function. And based on the damage directly monitored by the on-site sensors, the full-bridge and component-scale finite element models are comprehensively updated.

[0061] In this embodiment, a neural network is intended to be used to generate monitoring excitation data into model response information, which can improve the analysis speed of on-site sensor monitoring excitation information. A neural network is selected to establish a mapping relationship. In the first stage, it replaces the initial state multi-scale finite element model to calculate external excitation. In the second stage, it replaces the updated finite element model to establish a mapping relationship between the monitored point-unmonitored point information. In the third stage, the neural network establishes a data-physical model that reveals the relationship between the "excitation signal-response signal" and the "finite element model parameters" for correcting the finite element model. Using a neural network can improve the analysis speed of monitoring signals.

[0062] The safety assessment method for the multi-scale finite element model of the bridge structure provided in this embodiment can realize real-time analysis of sensor monitoring information through a surrogate model, process and mine the structural model information, evaluate the structural safety status, and map the calculation results of the surrogate model back to the established finite element model to realize real-time correction of the finite element model, display the current state of the structure, and improve the accuracy of structural performance analysis. Through finite element analysis, not only can the location and damage level of potential damage be discovered, but also various different types of loads can be applied to the finite element model in advance to expand the neural network analysis data set, provide a reliable and real-time updated training set for neural network pre-training, form a real-time excitation-load mapping relationship with a perfect physical relationship, and improve the accuracy of the neural network output data.

[0063] In addition, this embodiment relates to a structural safety assessment system, including: a multi-scale finite element model established at any time, a surrogate model based on deep learning established at any time, a neural network for the excitation-response mapping relationship of on-site monitoring points, a neural network for the response-response mapping relationship between on-site monitoring points and unmonitored points, and a loss function of the surrogate model calculated based on the surrogate model analysis information and on-site monitoring information.

[0064] In any of the above embodiments, the structural monitoring excitation information mainly includes vehicle load, temperature load, wind load, and seismic action. The structural monitoring response information mainly includes the structural stress, strain, and overall deflection of the bridge monitoring points (monitoring information related to prestress, etc.) obtained through other non-destructive testing methods or embedded means. When establishing the offshore bridge structure model, the multi-scale finite element module establishes structural models of different scales according to the detection parts of different sensors and the force characteristics of the structure, and establishes the mapping relationship between the finite element structural model and the sensor monitoring signal through numerical methods, including: the full-bridge scale finite element model describes the overall behavior and force-bearing mode of the bridge structure; based on the established full-bridge scale model, calculate the stress distribution and deformation of the overall structure. The main purpose is to analyze the distribution of mechanical parameters (stress, strain, deformation state, etc.) and structural behavior (downward deflection, collapse, etc.) of the structure under external loads, which serves as the calculation basis for the neural network numerical model; the refined finite element structural model represents the microscopic description and simulation of damage on structural components, and establishes the connection between the calculation parameters of the full-bridge scale finite element model to assist in correcting the full-bridge scale finite element model.

[0065] Based on the signals monitored on-site by sensors, analyze the data of unmonitored points through the expansion of monitoring points, and set threshold warnings through static mechanical indexes (such as displacement, stress, strain, etc.).

[0066] The finite element model is corrected in real time through a neural network, and the safety state of the physical structure is evaluated based on the finite element model, including: the mechanical evaluation indexes of the structure, mainly the load-bearing capacity of the multi-scale finite element model, and mainly the flexural, shear, and tensile load-bearing capacities in terms of direction. The remaining load-bearing capacity of the structure is mainly to deduce the remaining service life of the material according to the material degradation model and deduce the degradation function of the cross-section according to the fitted cross-section degradation parameters.

[0067] Based on the combination of any of the above solutions, this embodiment is implemented through the following technical solutions:

[0068] S1: Establish a bridge model at the initial moment to correct the multi-scale finite element model and ensure that the model information is consistent with the actual engineering structure. In terms of the modeling scale, divide the elements into parts that need to be divided into macro-scale structures and refined structures according to the layout positions of the monitoring points. In the same finite element model, use different mesh scales to simulate the monitored components and non-monitored components to establish a multi-scale finite element model. Calculate the structural response by inputting simulated excitation data to the monitoring points of the finite element model, and pre-generate a dataset for subsequent neural network training to provide model support for subsequent neural network training.

[0069] S2: Establish a neural network and use the multi-scale finite element model to generate a dataset for pre-training. The data of [excitation monitoring point signal - all node response signals - finite element model parameters] generated by using the established multi-scale finite element model through input simulation excitation data is used for the pre-training of the neural network required for each evaluation stage. The neural network F-R quickly calculates the relationship between the excitation monitoring point signal and the monitoring point response signal. The neural network R-R expands the monitoring point response signal - non-monitoring point response signal through real-time signals. The neural network FR-FE analyzes the information of [excitation signal - response signal] and [finite element model parameters]. The calculation process of the model needs to meet the requirements of in-situ bridge experiments or scaled-down tests to verify the applicability of the neural network.

[0070] S3: Achieve macroscopic early warning by calculating the excitation monitoring point signal - response monitoring point signal through the neural network F-R. Use the neural network F-R to calculate the excitation monitoring point - response monitoring point signal and obtain the calculated response based on the finite element model in the initial state. Compare whether there is a large difference between the structural calculated response and the on-site monitoring point response information to determine whether the bridge structure is abnormal or damaged.

[0071] S4: Achieve damage early warning at the most unfavorable position of components by expanding non-monitoring response signals from the response monitoring point signal through the neural network R-R. According to the calculation results of the initial refined finite element model and the design load, divide the numerical values of the non-monitoring point response signals into multiple levels. Evaluate the damage level of the calculated non-monitoring point response signals and locate the highest damage level (the most dangerous position).

[0072] S5: Use the neural network FR-FE data - physical model to analyze the relationship between [excitation signal - response signal] and [finite element model parameters], and real-time correct the multi-scale finite element model to evaluate the safety status of the structure / components. Calculation of the bearing capacity and safety assessment of the structure / components: At this evaluation stage, the remaining bearing capacity of the structure will be calculated from the overall structure and the refined components divided, the remaining service life will be analyzed, and the safety status of the structure will be evaluated. First, analyze the [excitation signal - response signal] collected on-site through the neural network FR-FE to obtain the [finite element model parameters] and update the finite element model. Use the multi-scale finite element model to calculate the overall bearing capacity level; isolate the refined component finite element model to evaluate the remaining service life.

[0073] The method further includes the following steps:

[0074] S6: Directly correct the finite element model using the structural damage information directly obtained by sensors. Monitor the structural damage information using on-site sensors to directly correct the multi-scale finite element model. The final correction result of the finite element model in this iteration step includes the direct damage read by the sensors and the correction result analyzed based on the excitation monitoring points and response monitoring points. Remove the influence of the directly corrected damaged part on the parameters using the finite element model parameters in S5, and update to obtain the finally corrected finite element model in this iteration step. Generate a data set using the updated finite element structural model for the pre-training of the subsequent neural network and the correction of the finite element model.

[0075] As a further improvement of the above solution, S1 specifically includes:

[0076] 1) Establish a finite element model based on engineering design drawings, and select appropriate element types, element lengths, and boundary conditions to establish finite element models at the full-bridge scale and component scale.

[0077] 2) Adjust the transition part of the two mesh division scales of the finite element model. Appropriately, the selected mesh scale and simulation method can be verified through a load test, and the finite element model can be adjusted to output the calculated response at the position of the response monitoring point.

[0078] 3) Use the multi-scale finite element model to simulate and analyze the main loads and performance changes of the bridge (vehicle load, temperature change, wind load, prestress loss, concrete performance degradation, etc.), and establish a neural network analysis data set by exporting the calculation information and excitation information at different positions of the initial structural finite element model. The structural monitoring excitation information mainly includes vehicle load, temperature load, wind load, and seismic action. The structural monitoring response information mainly includes the structural stress, strain, and overall deflection of the bridge monitoring points (monitoring information related to prestress, etc.) obtained through other non-destructive testing methods or embedded means.

[0079] As a further supplement to the above solution, as a significant environmental factor for offshore bridges compared to other bridges, the chloride ion monitoring density will directly act on the model information correction through material degradation models, etc., so it is not listed as a main load in the main evaluation process.

[0080] As a further supplement to the above solution, in order to obtain the training sets of different neural networks, data sets need to be established through different finite element models. For example, in the neural network F-R, the excitation of the monitoring points - the response of the monitoring points of the initial full-bridge scale finite element model needs to be obtained; in the neural network F-R, the response of the monitoring points - the response of the non-monitoring points of the refined part finite element model needs to be obtained.

[0081] Further, S2 specifically includes:

[0082] 1) Divide the data set in S1 into two parts. One part is used as the training set of the neural network, and the other part is used as the training set of the neural network. Flatten the initial data set to make it suitable as an appropriate signal input for the sensor in the time scale. If it is for the neural network F-R, only the excitation-response signal of the monitoring point needs to be monitored. If it is for the neural network R-R, the finite element model of the previous time step is required to calculate the response of non-monitoring points based on the response of the monitoring points.

[0083] 2) Select a suitable neural network method, analyze the parts divided into the training set and the validation set in 1), adjust the loss function according to the output data type, and obtain the initial pre-trained neural network model.

[0084] 3) A total of three neural networks need to be established in this stage. The neural network F-R establishes the mapping relationship between the excitation monitoring point signal and the response monitoring point signal based on the full-bridge scale model; the neural network R-R establishes the mapping between the response monitoring point and the non-monitoring point response based on the component part finite element model; the neural network FR-FE is used to analyze the excitation monitoring point signal - response monitoring point signal, calculate the calculation parameters of the finite element model, and is used for finite element model correction. The capital letters in the formed neural network FR-FE are abbreviations of force and response respectively, and the latter FE represents the finite element model.

[0085] 4) Analyze the physical meaning of the numerical values in the weight matrix of the neural network FR-FE, and propose a calculation function for back-inferring the structural model information from the weights. Improve the physical information inversion part of the neural network to verify the credibility of the neural network FR-FE.

[0086] 5) Use the part divided into the validation set in 1) to verify the neural network and check the accuracy of the inversion function calculated in 4). If the error of the inversion function in inferring the structural model information does not meet the preset range, return to 4) to re-deduce the physical information inversion function. Use the scaled model test or the full-scale bridge test to verify the applicability of the inversion function and adjust the parameters.

[0087] As a further supplement to the above solution, model interpretation methods such as SHAP, LIME, and feature attribution can be used for the neural network to improve the interpretability of the network and enhance the physical meaning of the code.

[0088] As a further supplement to the above solution, if it is determined that the full-bridge scale is damaged in several consecutive time steps, it can be judged that the structure is abnormally damaged or damaged at the full-bridge scale. In the next time step, the evaluation stage I can be directly skipped, and the evaluation stage II can be entered to locate the damage site and issue a threshold warning for the component scale model.

[0089] Furthermore, the specific content of S4 includes:

[0090] 1) Since the conclusion obtained in S3 is that the bridge structure has been damaged, the finite element component scale model can directly obtain the damage state of the structure according to the non-destructive monitoring devices arranged on site, and make a preliminary correction to the component scale model.

[0091] 2) Input the data of the monitoring points of the sensor response at the construction site into the neural network R-R, and calculate the calculated response data of the non-monitoring points. Expand the response signal from the monitoring points to the non-monitoring points.

[0092] 3) Extract the monitoring response point data in the calculated response data of the nodes, compare the calculated data with the sensor monitoring data, and calculate the current model response evaluation parameters . The physical meaning of the calculated parameter correction data is , is the response of the finite element model at the initial moment when calculating the stress, strain, deflection, etc. of the bridge when it is opened to traffic, represents the stress, strain, deflection and other response information of the current sensor signal. The state evaluation parameter is divided into <2.0%, 2.0%< <5.0%, 5.0%< <20.0%, >20.0% four levels.

[0093] As a further supplement to the above scheme, for the calculation of the component scale model state evaluation parameter , the denominator corresponds to the structure of the structure model corrected in the previous time step.

[0094] Optionally, can be calculated according to the corrected structure model adjusted during the construction process, and the initial response of the structure can be recalculated.

[0095] Further, the S5 specifically includes:

[0096] 1) Input the excitation-monitoring point response signal of the on-site sensor monitoring points into the neural network FR-FE, and extract the weight matrix of the hidden layer of the neural network.

[0097] 2) Calculate the material degradation index at each node of the corrected vector finite element model parameters and the initial vector finite element model parameters based on the inversion function in S2 , the structural cross-sectional area degradation index and the structural section moment of inertia degradation index , and use the corrected multi-scale finite element model to calculate the remaining bearing capacity of the structure and evaluate the overall safety information of the structure.

[0098] 3) Based on the on-site environmental factors data obtained by on-site sensor monitoring, derive the material degradation model and analyze The significance of this model is to predict its remaining useful life. , With the time-varying curve, the correlation between the reduction of cross-sectional properties and time is derived, and its remaining service life is predicted.

[0099] 4) For the modification of the refined model, the model is first modified directly through the damage information obtained by on-site sensor monitoring, and then the degradation index is used to modify the model. , , The finite element component scale model is modified, and the component safety information is evaluated and the residual bearing capacity of the component is calculated based on the finite element model calculation results.

[0100] 5) Since the refined finite element model in S4 is directly corrected by the nondestructive testing results of the on-site sensors, the refined finite element model can be used to obtain the quantitative information of defects (concrete crack width, length, depth and location, prestressed steel bar section reduction and corrosion area) and obtain a quantitative damage assessment report.

[0101] 6) The modified component scale model is evaluated by the state parameters , the degree of degradation of its resistance is quantitatively evaluated based on the degradation model and the fitted degradation curve.

[0102] Furthermore, the S6 specifically includes:

[0103] 1) According to the modified finite element full-bridge scale model and component scale model described in S5, a new data set is generated according to the load type and performance change (vehicle load, temperature change, prestress loss, concrete performance degradation, etc.) in S1 to participate in the generation and iteration of neural network FR, neural network RR and neural network FR-FE;

[0104] 2) The updated neural network FR-FE not only needs to meet the pre-set loss function requirements, but also needs to add some finite element model information (such as boundary conditions or finite element unit types) as the loss function. If the ratio of the finite element model calculation parameters calculated by S2 to the calculation parameters of the corresponding position of the initial model is greater than 1, it means that the material performance at this position is better than the initial state. Then this parameter does not conform to physical common sense and needs to be corrected.

[0105] Furthermore, the model building and correction process is calculated through the cloud platform, and the collected sensor data will also be synchronized to the cloud server.

[0106] Furthermore, the final result of the evaluation process is a cloud document report on the structural state. The information directly corrected in the finite element model includes concrete crack information and prestressed tendon defects. The excitation part is mainly vehicle load, and the response part includes, but is not limited to, the displacement, inclination angle, and strain collected by on-site sensors, as well as the natural vibration frequency and modal information of the structure obtained from the synchronously corrected finite element model. The evaluation includes cable force and vehicle load.

[0107] Furthermore, the monitoring points of the bridge refer to: according to the actual structural type, by applying the measured vehicle load, etc. through the finite element model, select the key points with larger deflection and stress across the bridge, or the points that can reflect the dynamic characteristics.

[0108] Embodiment 2

[0109] The following combines specific drawings to describe the specific implementation cases of the present invention in detail, and elaborates on a safety assessment method for a bridge model correction structure of the present invention.

[0110] It should be noted that the finite element model used can be established through commercial finite element software such as Midas, Abaqus, ANSYS, etc., or through MATLAB. The use of finite element software is not restricted here, and the finite element only provides the mechanical calculation results of the load-excitation of the offshore bridge structure in the present invention, which is used to establish the neural network data set and the safety assessment in the third stage. In addition, the neural network is used in the present invention to directly obtain the calculation error between the calculated response and the measured response through the sensor excitation signal and the model calculation parameters instead of the finite element model. The neural network can have multiple branches such as CNN, RNN, GAN, etc. Each architecture method of the neural network has different characteristics during use, and this example does not make specific restrictions on this. The on-site monitoring load information obtained by this method does not involve the personal information of the vehicle owner, and uploading the information to the cloud will not violate personal privacy. The collected and uploaded monitoring information only includes the vehicle axle weight, the driving position of the vehicle, and the driving speed.

[0111] It should be noted that the method provided by the present invention is not limited to newly built offshore prestressed concrete bridges, and is also applicable to existing bridges that can provide complete bridge structure damage information in the regular monitoring report. For existing bridges, since the model information before completion and opening to traffic cannot be obtained, the structural model for structural correction and comparison reference is the structural model on the design drawing. Before performing the model update for the next time step, more detailed multiple iterations are required to update and evolve the initial finite element model to conform to the current working state.

[0112] Such as Figure 1As shown, the present invention provides a three-stage bridge safety warning method based on the fusion of data models and physical models. The following case is the evaluation process of a certain prestressed concrete continuous rigid frame bridge by this evaluation method, which specifically includes the following steps:

[0113] S1: The main load-bearing components of prestressed concrete are the reinforced concrete beam body, prestressed steel bars, and the bridge piers connected to the superstructure. Among them, the beam body and the bridge piers are divided into different construction segments according to the cross-sectional size. According to the most unfavorable positions calculated by the finite element model and the positions of the on-site embedded sensors, the construction segments that need to establish refined models are divided. The finite element structure is divided into multiple components according to the sensor data types and layout points, and a multi-scale finite element model is established according to the characteristic information provided by the engineering drawings. The full-bridge scale partial model in the multi-scale reveals the maximum bending moment, shear force, and deformation information of the cross-section within the full bridge range. The component scale part therein deeply calculates the stress and strain distribution of the cross-section, as well as the relevant mechanical models of the prestressed steel bars as the main load-bearing components and the surrounding concrete. In addition to the finite element model, a model reflecting from the component scale to the full-bridge scale needs to be established. The reaction relationship here is mainly the information conversion of the finite element models under different modeling scales.

[0114] S2: Pre-train a neural network to establish surrogate models representing different physical relationships. The neural network F-R is used to provide the physical relationship between the excitation monitoring points and the response monitoring points. The neural network R-R is used to provide the physical relationship between the response monitoring points and the responses of the unmonitored points. The neural network FR-FE is used to analyze the signals of the sensor excitation monitoring points and the response monitoring points, and directly generate the calculation parameters of the finite element model. Among them, the generalization process of the response of the monitored points to the response of the unmonitored points can be calculated by the finite element model established in S1.

[0115] S3: Safety assessment stage I: Use the neural network F-R to calculate the input of the monitoring points. At this time, the input data of the neural network F-R are all the excitation signals of the on-site sensors, and the calculated responses of the set response monitoring points are obtained based on the model information of the finite element model in the initial state. Calculate the overall mean square error MSE between the calculated output response and the measured response data of the response monitoring points, and analyze whether the physical information of the structure in the current response signal changes compared with the physical information in the finite element main model in the previous time step, so as to determine whether there are abnormalities or damages in the bridge structure.

[0116] S4: Safety assessment stage II: If it is determined in S3 that the bridge structure is not intact, the assessment stage enters the second stage. In this stage, the neural network R-R is used to calculate the signals at the input response monitoring points in S3 to obtain the calculated responses at non-monitoring points. Since the physical information has a correlation relationship under the constraint of the finite element model, the structural response evaluation parameters are divided into four levels according to the physical information of the initial refined finite element model. The calculated evaluation parameters are classified into a certain level and the damage is located, and the node safety assessment status parameters and the early warning positions are sorted out.

[0117] S5: Safety assessment stage III: Whether the assessment stage enters assessment stage II or not, it enters assessment stage III. In this assessment stage, the remaining bearing capacity of the structure will be quantitatively and qualitatively analyzed from the whole bridge and components. First, the finite element model is corrected by the neural network FR-FE. For the refined component part of the finite element model, the remaining bearing capacity quantitatively evaluated by the finite element model is used to calculate the remaining service life of the structure based on the material degradation model formulated in S1; for the whole-bridge finite element model, the overall bearing capacity level of the structure is directly calculated, the safety state is quantitatively evaluated, the model parameter correction coefficient is sorted out, the geometric property degradation model is fitted, and the load information collected on site is statistically analyzed to predict the exceeding probability of the load.

[0118] S6: The three-dimensional characteristics of the cracks in the concrete structure in the on-site monitoring information are used to correct the concrete beam model, the monitoring information of the prestressed tendons is used to correct the elastic modulus and the tensile force of the prestressed tendons, and the finite element refined structure model is directly corrected. And the calculation parameters of the whole-bridge model are adjusted through the corrected mesoscopic scale model. The degradation coefficients of the analysis section and the material obtained by back-calculation in S5 are divided by the equivalent structure adjustment parameters directly corrected, and the finite element model at this time step is updated. The updated finite element structure model is used for the pre-training of the subsequent neural network and the correction of the finite element model.

[0119] Step S1 includes 3 steps: S11~S13.

[0120] S11: Establish a multi-scale finite element model based on the structural construction drawings. Select the beam element as the main element of the whole-bridge scale model, use the solid element to establish the refined model of the concrete, use the bar element to simulate the prestressed steel bars in the concrete, and add the bond-slip model to simulate the interaction between the two.

[0121] S12: Analyze the bridge model and select appropriate degradation models for different structural models. For the concrete part, the steel bar part and the prestressed tendon part, different degradation models are selected based on the material properties, and an appropriate method is selected to establish the geometric parameter degradation model of the bridge. Further determine the degradation range of the finite element model calculation parameters compared with the initial model under the condition of updating each iteration step in the surrogate model in S4.

[0122] S13: Define the warning threshold for the response signal of each node evaluated by the threshold in S4, and define the prediction model for the remaining service life of components by statistically distributing the load information in S5.

[0123] For step S13, since the structure monitored in the design case is a continuous rigid frame, the key points are set at the intersection positions of 1 / 4, 1 / 2, and 3 / 4 of the beam length and the beam pier. Further, response monitoring points need to be densified at the 1 / 4L position of the side span and at the mid-span of the main span to improve the correction accuracy; if the bridge structure form monitored in the design is simply supported, the key points set in the influence matrix are at 1 / 4, 1 / 2, 3 / 4 of the beam length and the positions of the supports, and the key monitoring positions of strain, inclination, and deflection will be mainly set at the key points; if the bridge structure form is continuous, in addition to arranging key points on the main span, key points also need to be added at the side span positions.

[0124] As Figure 2 shown, step S2 includes: S21~S25.

[0125] S21: Generate an initial data set through the multi-scale finite element model established in S11. The data set is used for the pre-training process of three neural networks. By inputting simulated loads into the excitation monitoring points of the full-bridge scale finite element model in the initial state, the calculated data of the response monitoring points are extracted as the training data set F-R of the neural network F-R. If the bridge structure is judged to be intact, the response signals of the non-monitoring points in the neural network R-R data set are obtained by inputting the simulated loads into the full-bridge scale finite element model in the initial state; if the bridge structure is judged to be damaged, its non-monitoring point response signals are obtained by inputting the simulated loads into the corrected full-bridge scale finite element model. According to different bridge stage I evaluation states, the corresponding node response data are extracted to obtain the monitoring point response - non-monitoring point response training set; based on the model of the multi-scale finite element model, the excitation sensor data and response sensor data in each finite element model node are integrated to obtain the FR-FE training set.

[0126] S22: Train the neural network F-R through the generated data set F-R. This neural network is used to quickly calculate the excitation of the on-site sensors at the monitoring points - the response of the monitoring points. Based on the full-bridge scale finite element model in the initial state, input the model parameters and excitation monitoring point data into the neural network, and output the calculated response of the structure response monitoring points.

[0127] S23: Train the neural network R-R through the generated data set R-R. This neural network is used to quickly calculate the response of the on-site sensors at the monitoring points - the response of the non-monitoring points of the monitoring points. Based on the refined finite element model of the previous time step, input the model parameters and the response of the monitoring points into the neural network, and output the calculated response of all nodes of the structure. The purpose is to expand the non-monitoring point response through the monitoring point response.

[0128] S24: Train the neural network FR-FE with the generated dataset FR-FE. This neural network is used to analyze the correlation between data through neural network based on the excitation-response data pairs of on-site sensors at the monitoring points, and output the calculation parameters of the finite element model. Based on the finite element model of the previous time step, input the excitation monitoring point signal and the response monitoring point signal into the neural network, and output the calculation parameters of the finite element model.

[0129] S25: Adjust the finite element model and the neural network through full-scale bridge or scaled model tests.

[0130] For the first-stage evaluation phase of the structure in step S3, it includes: S31~S35.

[0131] S31: Divide the signals collected on-site from the excitation-response signals of the on-site monitoring points into two parts. One part of the excitation of the monitoring points is used as the input part of the neural network F-R, and the other part of the response of the monitoring points is used for comparison with the output of the neural network F-R.

[0132] S32: Input the excitation monitoring point signal in S31 into the full-bridge finite element model to obtain the neural network F-R, and obtain the calculated response signal of the response monitoring point based on the initial finite element model.

[0133] S33: Compare the calculated response signal of the response monitoring point in S32 with the on-site monitoring response signal in S31, calculate the mean square error of the measured data and the calculated data of all response monitoring points, and determine whether the signal obtained by the calculation of the initial state full-bridge scale finite element model is consistent with the on-site monitoring point signal.

[0134] S34: If the signal comparison error is within the given error range, it indicates that the offshore bridge structure has not suffered damage macroscopically and the macroscopic safety state of the structure is good; if the signal comparison error exceeds the given range, it indicates that the structure can be judged as damaged from the on-site sensor signals at the macroscopic scale.

[0135] For step S31, the neural network F-R is established by the full-bridge scale finite element model in the initial state of opening to traffic, and is used to quickly calculate the excitation-response of the monitoring points. Because when evaluating the structural state in stage I, the structure is defaulted to have no damage, so the comparison object is the initial undamaged bridge model. If in a certain time step, it is determined in stage I evaluation that damage occurs in the full-bridge scale, then there is no need for stage I evaluation in subsequent time steps, and directly enter stage II for damage location and parameter threshold warning of the bridge structure.

[0136] For the second-stage evaluation phase of the structure in step S4, it includes: S41~S45.

[0137] S41: As in S23, use the neural network R-R. Input the response monitoring point signals into the neural network R-R to obtain the calculated responses of all structural nodes at the full-bridge scale.

[0138] S42: Calculate the current structural model state evaluation parameters according to the warning threshold set in S13 , in accordance with <2.0%, 2.0%< <5.0%, 5.0%< <20.0%, >20.0% to divide the damage levels, and locate the positions of each non-zero parameter The non-zero parameters correspond to different damage levels.

[0139] S43: Arrange the structural warning levels in S42 in descending order, and analyze whether the responses at the most dangerous positions of the structure exceed the given warning range.

[0140] S44: If the state parameters exceed the warning threshold, it is necessary to locate the damage position and its corresponding damage level; if the state parameters do not exceed the warning range, all nodes of the entire structure are in a safe state.

[0141] For the third stage of the structural assessment phase, step S5 includes: S51~S55.

[0142] S51: Analyze the on-site sensor excitation-response signal dataset FR-FE through the neural network FR-FE to directly obtain the calculation parameters of the finite element model, and verify the rationality of the modified parameters of the finite element model through the back-calculation function in S24.

[0143] S52: Directly use the remaining load-bearing capacity of the finite element calculation model through the modified full-bridge finite element model, and check the structure according to the manufacturing specifications.

[0144] S53: Extract the refined modeling part with a small mesh size, and directly calculate the remaining load-bearing capacity of the component. Calculate the maximum bending moment, maximum shear force, and maximum tensile force that this part can withstand according to the stress condition of the component , , , and complete the evaluation of the remaining load-bearing capacity level of the component by comparing with the relevant regulations in the specifications.

[0145] S54: Extract the ratio of each calculation parameter to the calculation parameter of the initial finite element component model through the modified component finite element model, that is, the adjustment parameter . Use the degradation model of the structural materials determined in S12, and by comparing the extracted Determine the development state of the component in terms of material properties or geometric properties at this time, so as to determine the remaining service state of the material and calculate the overall service life of the component.

[0146] S55: Statistically analyze the external excitations (including vehicle loads, temperature changes, etc.) that the structure has endured from the previous time period to this time period. Summarize the finite element model information and integrate the bridge safety status assessment report.

[0147] Furthermore, for step S53, the purpose of this case is to determine the remaining load-bearing capacity of the finite element model corrected at this time step, and the evaluation objects are the maximum bending moment, the maximum shear force, and the maximum tensile force. , , . To achieve the same purpose, the principal stress of the element can be determined by applying virtual loads to the structure, comparing the principal stress with the material strength, and evaluating the warning level of the structure by setting different warning thresholds. Similarly, different evaluation methods can be used in this step, starting from the objects of safety assessment and adjusting the calculation process.

[0148] For step S54, for the vehicle loads in the initial few time steps, since the vehicle load information of this bridge is not stored on the cloud server, the vehicle load used to predict the remaining service life of the bridge can be replaced by using the vehicle load information of other bridges at this time.

[0149] For step S6, in S5, the quantitative evaluation results of each node are obtained by analyzing the excitation and response monitoring point data of the finite element model at the full-bridge scale, but the damage information directly obtained by sensors is still lacking. The finite element model is updated by simulating damage in the refined part. The improved neural network update here is based on the finite element model with damage.

[0150] Furthermore, the damage information included in the safety status assessment report integrated in S5 only comes from sensors, and the damage in the report is not simulated in the finite element model at this time. The damage simulation is carried out in S6.

[0151] Furthermore, the finite element model in S6 not only needs to simulate sensor damage, but also needs to remove the reduction of calculation parameters caused by damage on the basis of the finite element model parameters calculated in S5. Therefore, the calculation parameters of the finite element model also need to be adjusted. The specific method is to calculate the influence of damage on each node, and subtract the influence of damage from the model calculation parameters obtained in S5.

[0152] Specifically, the bridge to be evaluated is a physical bridge to be evaluated for building safety. In this embodiment, there are no restrictions on the structure and material properties of the bridge to be evaluated. The bridge to be evaluated includes, but is not limited to, bridges of various architectural forms and uses such as ancient architectural bridges and cross-sea bridges. The original bridge design drawing is a drawing that reflects various building data of the bridge, which is drawn based on the built bridge or obtained according to the historical building information of the bridge. The original bridge design drawing is marked with building data and building material information. The bridge connection point is the splicing site of multiple bridge components for the purpose of constructing the bridge body and achieving balanced load bearing and stress.

[0153] Embodiment 3

[0154] This embodiment provides a bridge safety early warning system based on the fusion of a data model and a physical model. This system is used to implement any combination of the above embodiments or solution methods, and includes:

[0155] The first main control module is used to construct a multi-scale finite element model of the bridge structure, calculate the structural response by inputting simulated excitation data into the finite element model, and generate a data set;

[0156] The second main control module is used to construct a neural network model and train the neural network model using the data set. The neural network model includes: neural network F-R, neural network R-R, and neural network FR-FE;

[0157] The third main control module is used to calculate the excitation-monitoring point response at the on-site monitoring points using neural network F-R, obtain the calculated response based on the finite element model in the initial state, and conduct a comparative analysis to determine whether the bridge structure is abnormal or damaged;

[0158] The fourth main control module is used to analyze the monitoring point response information input in the third main control module using neural network R-R, expand to obtain the calculated response at non-monitoring points, and conduct a grade division to determine the damage grade and the location of the damage parameters exceeding the predetermined threshold;

[0159] The fifth main control module is used to analyze the excitation signal-response signal collected on-site through neural network FR-FE, obtain the correction parameters for updating the finite element model, and calculate the overall bearing capacity level and the remaining service life of the bridge structure based on the updated multi-scale finite element model.

[0160] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. A bridge safety early warning method based on the fusion of data model and physical model, characterized in that: The following steps are involved: Step S1, constructing a multi-scale finite element model of the bridge structure, calculating the structural response by inputting simulated excitation data into the finite element model, and generating a data set including an excitation signal, a response signal and structural model information; Step S2, constructing a neural network model, and using corresponding data sets to train the neural network model according to different neural network calculation purposes, the neural network model includes: neural network FR, neural network RR and neural network FR-FE; Step S3, first safety assessment stage: using the neural network FR to calculate the on-site monitoring point excitation-monitoring point response, and obtain the calculated response based on the initial state finite element model to determine whether the bridge structure is abnormal or damaged; Step S4, second safety assessment stage: using neural network RR to analyze the monitoring point response information input in step S3, expanding the non-monitoring point calculation response, and performing calculation response monitoring threshold level classification to determine the location and damage level of the component defined as damaged; Step S5, the third safety assessment stage: analyzing the excitation signal-response signal collected on site through the neural network FR-FE, obtaining the correction parameters for updating the finite element model, and calculating the overall bearing capacity level of the bridge and the remaining service life of the structure based on the updated multi-scale finite element model; The method further comprises the following steps: The finite element model is directly corrected using the structural damage information directly monitored by the field sensor in S3, that is, the final correction of the finite element model needs to be based on the consistency between the structural damage and the field monitoring damage, and the finite element model is updated according to the degradation coefficient of the section and material analyzed by the monitoring signal in step S5 and the adjustment parameters of the corrected structural damage, and a new data set for training the neural network model and correcting the finite element model is generated using the updated finite element structural model; Step S5 includes the following steps: Step S51, input the monitoring point excitation-monitoring point response signal of the field sensor into the neural network FR-FE, and extract the weight matrix of the hidden layer of the neural network; Step S52, based on the inversion function in step S2, calculating the calculation parameters of the multi-scale finite element model, correcting the finite element model in real time, using the corrected multi-scale finite element model to calculate the residual bearing capacity of the structure, and evaluating the overall safety information of the structure; Step S53, calculating the material degradation index at each node based on the modified vector finite element model parameters and the initial vector finite element model parameters , structural cross-sectional area degradation index and the degradation index of the moment of inertia of the structural section According to the on-site environmental factors data obtained by on-site sensor monitoring, the degradation model of material properties in the degradation index is derived, and the material degradation index is used to determine the degradation model of the material properties in the degradation index. Predict the remaining useful life of materials, fit , With the time-varying curve, the correlation between the reduction of the cross-sectional properties and time is derived to obtain the degradation function of the cross-sectional properties; Step S54, evaluating the refined model part of the finite element model according to the material degradation index , structural cross-sectional area degradation index and the degradation index of the moment of inertia of the structural section The finite element component scale model is modified, and the component safety information is evaluated and the residual bearing capacity of the component is calculated based on the finite element model calculation results.

2. A bridge safety early warning method based on the fusion of data model and physical model according to claim 1, characterized in that: Step S1 includes the following steps: Step S11, dividing the construction section where a refined model needs to be established according to the most unfavorable position of the bridge and the position of the pre-buried sensors on site; Step S12, dividing the finite element structure into multiple components according to the sensor data type and layout points, establishing component scale and full bridge scale finite element models according to the characteristic information provided by the engineering drawings, using different grid scales in the same finite element model to simulate monitoring components and non-monitoring components, and establishing a multi-scale finite element model; Step S13, using a multi-scale finite element model to simulate and analyze the main loads and performance changes of the bridge, and establishing a data set by exporting the calculation information and excitation information of different positions of the initial structural finite element model.

3. A bridge safety early warning method based on the fusion of data model and physical model according to claim 2, characterized in that: In step S13, the structural response is calculated by inputting simulated excitation data into the finite element model excitation monitoring point, that is, a suitable degradation model is selected for different structural models, a bridge geometric parameter degradation model is established, and the degradation range of the finite element model calculation parameters compared with the initial model is determined.

4. The bridge safety early warning method based on the fusion of data model and physical model according to claim 1 is characterized in that: In step S2, by inputting simulated loads to the excitation monitoring points in the full-bridge scale finite element model in the initial state, the calculated data of the response monitoring points are extracted as a training data set for the neural network FR; If the bridge structure is judged to be intact, the non-monitoring point response signals of the neural network RR data set are obtained by inputting the simulated load into the full-bridge scale finite element model in the initial state; if the bridge structure is judged to be damaged, its non-monitoring point response signals are obtained by inputting the simulated load into the revised full-bridge scale finite element model, and according to the assessment status of the first safety assessment stage, the corresponding node response data are extracted to obtain the monitoring point response-unmonitoring point response training set; Based on the model of the multi-scale finite element model, the excitation sensor data and response sensor data in each finite element model node are integrated to obtain the FR-FE training set.

5. The bridge safety early warning method based on the fusion of data model and physical model according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31, dividing the signals collected on-site from the excitation-response signals of the on-site monitoring point sensors into two parts, one part of the monitoring point excitation is used as a training process of the neural network FR, and the other part of the monitoring point response is used as a verification set with the neural network FR; Step S32, inputting the excitation monitoring point signal in step S31 into the full-bridge finite element model to obtain a neural network FR, and obtaining a response monitoring point calculation response signal based on the initial finite element model; Step S33, comparing the calculated response signal of the response monitoring point in step S32 with the on-site monitoring response signal in step S31, calculating the mean square error between the measured data and the calculated data of all the response monitoring points, and determining whether the signal obtained by calculating the initial state full-bridge scale finite element model of the verification signal is consistent with the on-site monitoring point signal; Step S34: If the signal comparison error is within the given error range, it means that the offshore bridge structure has not been damaged at a macroscopic level, and the macroscopic safety status of the structure is good; if the signal comparison error exceeds the given range, it means that the structure can be judged as damaged at a macroscopic level through the calculation of the field sensor signal by the neural network FR; Step S35 uses the neural network FR to calculate the on-site monitoring point excitation-monitoring point response, and obtains the calculated response based on the initial state finite element model to determine whether the bridge structure is abnormal or damaged.

6. The bridge safety early warning method based on the fusion of data model and physical model according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41, calculate the current structural model state assessment parameter according to the calculation response of all structural nodes of the full bridge scale obtained by the neural network RR and the set warning threshold , and evaluate the state parameters The damage level is divided according to the value of The threshold value of Step S42: Arrange the structural damage levels in step S41 according to their magnitude, and analyze the state assessment parameters of the most dangerous position of the structure. Whether it exceeds the given warning threshold range; Step S43: If the status evaluation parameter If the warning threshold is exceeded, the damage position and its corresponding damage level need to be located; otherwise, if all node status evaluation parameters If the warning threshold is not exceeded, all nodes in the entire structure are in a safe state.

7. The bridge safety early warning method based on the fusion of data model and physical model according to claim 6 is characterized in that: In step S41, the state evaluation parameter Divided into: <2.0%, 2.0%< <5.0%, 5.0%< <20.0%%, >20.0% four levels.

8. A bridge safety early warning system based on the fusion of data model and physical model, characterized in that: include: The first main control module is used to construct a multi-scale finite element model of the bridge structure, calculate the structural response by inputting simulated excitation data into the finite element model, and generate a data set; A second main control module is used to construct a neural network model and use the data set to train the neural network model, wherein the neural network model includes: a neural network FR, a neural network RR, and a neural network FR-FE; The third main control module is used to calculate the on-site monitoring point excitation-monitoring point response using the neural network FR, obtain the calculated response based on the initial state finite element model, and conduct comparative analysis to determine whether the bridge structure is abnormal or damaged; The fourth main control module is used to use the neural network RR to analyze the response information of the monitoring points input in the third main control module, expand the calculated responses of the non-monitoring points, and classify them to determine the damage level and the location of the damage parameter exceeding the predetermined threshold; The fifth main control module is used to analyze the excitation signal-response signal collected on site through the neural network FR-FE to obtain the correction parameters for updating the finite element model, and calculate the overall bearing capacity level of the bridge and the remaining service life of the structure based on the updated multi-scale finite element model; The fifth main control module is also used to perform the following steps: Step S51, input the monitoring point excitation-monitoring point response signal of the field sensor into the neural network FR-FE, and extract the weight matrix of the hidden layer of the neural network; Step S52, based on the inversion function in step S2, calculating the calculation parameters of the multi-scale finite element model, correcting the finite element model in real time, using the corrected multi-scale finite element model to calculate the residual bearing capacity of the structure, and evaluating the overall safety information of the structure; Step S53, calculating the material degradation index at each node based on the modified vector finite element model parameters and the initial vector finite element model parameters , structural cross-sectional area degradation index and the degradation index of the moment of inertia of the structural section According to the on-site environmental factors data obtained by on-site sensor monitoring, the degradation model of material properties in the degradation index is derived, and the material degradation index is used to determine the degradation model of the material properties in the degradation index. Predict the remaining useful life of materials, fit , With the time-varying curve, the correlation between the reduction of the cross-sectional properties and time is derived to obtain the degradation function of the cross-sectional properties; Step S54, evaluating the refined model part of the finite element model according to the material degradation index , structural cross-sectional area degradation index and the degradation index of the moment of inertia of the structural section Modify the finite element component scale model, combine the finite element model calculation results, evaluate the component safety information, and calculate the component residual bearing capacity; The system is also configured to perform the following steps: The finite element model is directly corrected using the structural damage information directly monitored by the on-site sensors in the third main control module. That is, the final correction of the finite element model needs to be based on the consistency between the structural damage and the on-site monitored damage. The finite element model is updated according to the degradation coefficients of the cross-section and the material analyzed by the monitoring signal in the fifth main control module and the adjustment parameters of the corrected structural damage. The updated finite element structural model is used to generate a new data set for training the neural network model and correcting the finite element model.

Images