Bridge overload early warning method and system

Abstract

The invention discloses a bridge overload early warning method and system. A front bridge is adopted to identify and predict spatial and temporal distribution of vehicle loads on a target bridge; after the spatial and temporal distribution of the vehicle on the target bridge is determined, target bridge monitoring data is utilized to calibrate the finite element model, the response of the structure is determined, and the abnormal strain, vertical displacement and anti-overturning stability coefficient of the structure are analyzed, so that multi-level early warning is performed on the state of the structure. According to the technical scheme, the problems that in bridge load monitoring, a traditional method is high in data isolation, poor in real-time performance and insufficient in anti-interference capacity are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge safety monitoring, and in particular relates to a bridge overload early warning method and system. Background Art

[0002] In recent years, with the continuous development of transportation and the extended service life of bridges, bridge damage and even collapse caused by overloading have become a common occurrence. Traditional monitoring systems often rely on a single sensing method, making it difficult to provide early warnings when a vehicle is overloaded. While existing technologies offer early warnings based on temperature, displacement, or single strain information, these methods do not fully capture vehicle load distribution, and their warning models lack robustness, making them unable to effectively address the risk of overloading in complex traffic environments. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a bridge overload early warning method and system.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A bridge overload early warning method, comprising:

[0006] Step S1, obtaining the vehicle load distribution on the leading bridge and the target bridge;

[0007] Step S2: Using the vehicle load distribution of the front bridge and the target bridge, a spatial mapping prediction model of vehicles on the two bridges is constructed; the vehicle information on the front bridge identified by the FBG sensor is input into the trained prediction model to predict the vehicle load distribution on the target bridge;

[0008] Step S3: After determining the temporal and spatial distribution of vehicles on the target bridge, the finite element model is calibrated using the target bridge monitoring data, and the abnormal strain, vertical displacement and anti-overturning stability coefficient of the structure are analyzed to provide a multi-level early warning of the structure status.

[0009] As a preferred method, the vehicle load distribution of the preceding bridge and the target bridge is used to construct a spatial mapping prediction model of vehicles on the two bridges using the GWO-GRU neural network.

[0010] As a preference, the front bridge uses fiber Bragg gratings (FBG) and cameras for long-term vehicle information monitoring, and the target bridge uses visual sensors combined with multiple FBG sensors for short-term detection to assist in building a vehicle space mapping model and calibrating the finite element model (FEM).

[0011] Preferably, in the leading bridge, FBG strain data is used to identify vehicle loads, and the camera captures license plate LP information to assign a unique identifier to each identified vehicle load; by matching the LP information identified on the two bridges, the vehicle load on the target bridge is determined based on the vehicle data on the leading bridge.

[0012] The present invention also provides a bridge overload warning system, comprising:

[0013] a first processing device for obtaining vehicle load distribution on the leading bridge and the target bridge;

[0014] The second processing device is used to construct a spatial mapping prediction model of vehicles on the two bridges using the vehicle load distribution of the preceding bridge and the target bridge; the vehicle information on the preceding bridge identified by the FBG sensor is input into the trained prediction model to predict the vehicle load distribution on the target bridge;

[0015] The third processing device is used to calibrate the finite element model using the monitoring data of the target bridge after determining the temporal and spatial distribution of vehicles on the target bridge, analyze the abnormal strain, vertical displacement and anti-overturning stability coefficient of the structure, and provide multi-level early warning of the structure's status.

[0016] Preferably, the second processing device utilizes the vehicle load distribution of the leading bridge and the target bridge and adopts a GWO-GRU neural network to construct a spatial mapping prediction model of vehicles on the two bridges.

[0017] As a preference, the front bridge uses fiber Bragg gratings (FBG) and cameras for long-term vehicle information monitoring, and the target bridge uses visual sensors combined with multiple FBG sensors for short-term detection to assist in building a vehicle space mapping model and calibrating the finite element model (FEM).

[0018] Preferably, FBG strain data is used to identify vehicle loads, and the camera captures license plate LP information to assign a unique identifier to each identified vehicle load; by matching the LP information identified on the two bridges, the vehicle load on the target bridge is determined based on the vehicle data on the preceding bridge.

[0019] This invention addresses the challenges of traditional bridge load monitoring methods, such as high data isolation, poor real-time performance, and insufficient anti-interference capabilities, thereby improving the accuracy and reliability of bridge overload warnings. By introducing a multi-bridge-linked load spatiotemporal mapping mechanism and intelligent optimization algorithms, it achieves greater flexibility and robustness in vehicle load distribution prediction, structural response analysis, and multi-level warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0021] Figure 1 This is a flow chart of a bridge overload warning method according to an embodiment of the present invention;

[0022] Figure 2 Schematic diagram of the vehicle space mapping model based on GWO-GRU NN;

[0023] Figure 3 Schematic diagram of different vehicle loading modes; (a) is loading mode I, (b) is loading mode III

[0024] Figure 4 Place schematics for the bridge and sensors;

[0025] Figure 5 Schematic diagram of simulated traffic flow on the two bridges;

[0026] Figure 6 Schematic diagram of the normal distribution of vehicle simulation parameters; (a) vehicle load, (b) vehicle speed

[0027] Figure 7 Schematic diagram of vehicle information identified on two bridges; (a) vehicle speed, (b) vehicle load

[0028] Figure 8 Schematic diagram of error distribution for identifying vehicle speed and load;

[0029] Figure 9 Schematic diagram of model validation results;

[0030] Figure 10 Schematic diagram of predicted vehicle load during the training period and prediction period;

[0031] Figure 11 Estimated load effects under two loading modes; (a) loading mode I, (b) loading mode II. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] Example 1:

[0035] like Figure 1 As shown, an embodiment of the present invention provides a bridge overload warning method. To facilitate early warning of potential overload on a target bridge, a pre-bridge is used to identify and predict the spatiotemporal distribution of vehicle loads on the target bridge. By installing monitoring and inspection systems on both bridges, effective early warnings can be provided for the target bridge. Two different sensing systems are installed on the pre-bridge and target bridges, respectively. The pre-bridge uses fiber Bragg gratings (FBGs) and cameras for long-term vehicle information monitoring, while the target bridge uses a visual sensor combined with multiple FBG sensors for short-term monitoring, assisting in the construction of a vehicle spatial mapping model and calibrating the finite element model (FEM). On the pre-bridge, FBG strain data is used to identify vehicle loads, while the camera captures license plate (LP) information, enabling the assignment of a unique identifier to each identified vehicle load. By matching the LP information identified on the two bridges, the vehicle load on the target bridge can be initially determined based on the vehicle data on the pre-bridge. Given the well-established engineering application foundation of current LP visual recognition algorithms, this embodiment of the present invention avoids explicit simulation of LP identification during the model construction phase, focusing instead on subsequent analysis based on the assumption that the vehicle LP is known, ensuring that the research addresses the core issues at hand. Using vehicle load data from the two bridges, a spatial mapping prediction model for vehicles on the two bridges was established using a GWO-GRU neural network. Once the prediction model is complete, the vehicle information on the front bridge identified by the FBG sensor can be input into the trained prediction model to predict the vehicle load distribution on the target bridge. After determining the spatiotemporal distribution (loading pattern) of vehicles on the target bridge, the finite element model is calibrated using the target bridge monitoring data to determine the structural response and analyze the abnormal strain, vertical displacement, and anti-overturning stability coefficient of the structure, thereby providing a multi-level early warning of the structural status, including:

[0036] Step 1: Vehicle load identification based on mechanical model

[0037] 1. Strain data acquisition and strain signal decoupling using influence surface theory

[0038] In order to obtain multi-lane vehicle load information on the front bridge, a mechanical model for identifying vehicle speed, wheelbase, load, and position on the bridge was established using the bending strain and shear strain of the long-term monitoring system. In this embodiment of the present invention, monitoring units are only set on two cross sections near the bridge supports. The number of monitoring units for each road section is determined according to the number of lanes. FBG shear strain sensors are installed on each monitoring unit, and FBG bending strain sensors are only installed on the monitoring units closest to the bridge supports. The mechanical model includes three aspects: data decoupling, the correlation between the two types of strains and the shaft parameters, and the correlation between the bending strain and the axle weight. Within the framework of the influence surface theory, the strain response characteristics caused by the single lane load are separated from the collected data and expressed as:

[0039]

[0040] Among them, ε i represents the measured strain, represents the separation strain on the i-th beam. The total number of lanes is denoted as N. η iN represents the lateral distribution coefficient of the i-th lane, specifically quantifying its sensitivity to the vehicle load applied on the N-th lane.

[0041] 2. Calculation of vehicle parameters

[0042] The bending and shear strains are used to develop a mechanical model that, combined with the shaft data, allows for accurate calculation of shaft parameters. The shaft velocity is determined from the time difference between the shear strain peaks recorded in the two monitoring sections, while the distance between the shafts is inferred from the time associated with the peak difference in bending strain. Both shaft velocity and distance can be estimated and expressed as:

[0043]

[0044] d=v*(t j+1 -t j ) (3)

[0045] Where S1 and S2 represent the corresponding monitoring units located in the two sections. The time instances in units S1 and S2 are denoted as t S1 and t S2 , represents the point where the maximum shear strain is observed. The variable L represents the distance between the two monitoring sections. The time corresponding to the peak bending strain of element S1 caused by the j-th axis and the (j+1)-th axis is t j and t j+1 express.

[0046] In summary, the separated bending strain is combined with the obtained axle parameters to estimate the vehicle load of a specific lane. Based on the bending strain, the axle load is calculated by taking the peak value, and its expression is as follows:

[0047]

[0048] Among them, λ i represents the lane-specific calibration factor for lane i. represents the recorded strain resulting from the loading event on the Mth axle of the monitoring unit located in lane i. M represents the axle count. L represents the span length of the bridge. The variable u M,2 Defines the longitudinal coordinate of the second axle when the Mth axle acts on the monitoring unit. i,M The dynamic load magnitude of the Mth axle when the i-th lane passes through the monitoring area is quantified. According to equation (4), the vehicle load of each lane can be effectively identified, thereby facilitating the determination of the vehicle load distribution of the bridge span.

[0049] Based on the vehicle load data obtained from the preceding bridge, the vehicle load distribution on the target bridge can be obtained by correlating the LP details of the vehicles on the two bridges.

[0050] Step 2: Vehicle information prediction based on data model

[0051] 1. GRU-based modeling

[0052] The GRU is a variant of the recurrent neural network (RNN) that has a more streamlined architecture than the long short-term memory (LSTM) model. The GRU's structure is relatively simple, with only two gates (update and reset), while other architectures typically have three gates (input, forget, and output). The output of the GRU unit is composed of these two gates, allowing it to retain or discard input information as needed.

[0053] The equations (5) to (8) show the equations for resetting the gate output, updating the gate output z t And the unit output h t Calculation:

[0054] r t =σ(W r ×[h t-1 ,x t ]) (5)

[0055] z t =σ(W z ×[h t-1 ,x t ]) (6)

[0056]

[0057] Among them, r t and z t are the outputs of the reset gate and update gate respectively. t is the current input. h t and ht-1 represents the hidden state at time t and t-1. Describes the candidate hidden state at time t. tan h is the hyperbolic tangent function. σ is the sigmoid activation function. W r ,W z and W h is the parameter matrix.

[0058] 2. Parameter optimization based on GWO

[0059] The GWO algorithm simulates the hierarchical structure and hunting behavior of a gray wolf population. The algorithm divides the gray wolf population into different hierarchies and assigns a fitness value to each wolf. The wolf with the highest fitness value is designated as α wolf, while the other wolves are divided into β, δ, and ω wolves according to their fitness values. The hunting process is initiated by α wolf, supported by β and δ wolves, and then ω wolf. The GWO algorithm is summarized as two main steps: (1) Target encirclement. When the algorithm is activated, the behavior of the wolf pack relative to the target can be described as:

[0060] D=|V×X P (t)-X(t)| (9)

[0061] X(t+1)=X P (t)-A×D (10)

[0062] Where D is the distance between the wolf and the prey, V and A are coefficient vectors, X(t) represents the position of the gray wolf in the tth iteration, and X P (t) is the position of the prey in the tth iteration; (2) Target tracking. Throughout the hunting process, the gray wolf adjusts its position according to the positions of the other three wolves.

[0063] 3. Establishment of vehicle space mapping model based on GWO-GRU neural network

[0064] To establish a spatial mapping model of vehicle loads between the target bridge and the preceding bridge, vehicle information on both bridges must be obtained. For the preceding bridge, vehicle loads are identified using FBG strain data, which is integrated with the LP details identified by the camera, assigning a unique identifier to each vehicle load. Based on the information obtained from the preceding bridge, the vehicle load on the target bridge is inferred by matching the LPs of vehicles on the two bridges.

[0065] The spatiotemporal data of the vehicle loads on the identified front bridge and target bridge are used as input variables of the GRU model. The GWO algorithm is then used to optimize the key hyperparameters of the GRU network, including the number of hidden layer neurons, initial learning rate, and regularization coefficient. The flowchart for establishing the spatial mapping model is shown in the figure. Figure 2 shown.

[0066] Using the established vehicle space mapping model, real-time vehicle data is obtained from the front bridge and used to predict the vehicle load distribution on the target bridge. Assume that the vehicle information sequence passing through the front bridge is as follows:

[0067]

[0068] in, represents the load, speed, and lane position of the vehicle on the leading bridge at time t. This set of vehicle data is input into the mapping model to predict the arrival lane on the target bridge The goal of training the GWO-GRU neural network is to minimize the classification loss between the predicted arrival lane and the actual arrival lane. training samples, each sample consists of an input sequence and a corresponding target output, and the loss function Defined as cross entropy loss, it can be expressed as:

[0069]

[0070] Where c is the number of lanes, Describes the indicator function of the actual vehicle category, represents the predicted probability of different types of vehicles in lane c. The backpropagation algorithm and Adam optimizer are used to adjust the weight parameters of the GRU network to minimize the loss function. After the model training is completed, the input sequence from the training set is forward propagated through the GRU network. The resulting output is the probability distribution of various vehicle categories in the lane, which is expressed as:

[0071]

[0072] Among them, W fc and b fc are the weights and biases of the fully connected layer, respectively. Softmax is the function used to convert the output into a probability distribution. Ultimately, the prediction result is determined by selecting the lane with the highest probability as the arrival lane for the vehicle on the target bridge. Therefore, the model output can be expressed as:

[0073]

[0074] Here, argmax is a function that finds the maximum value in an array.

[0075] Step 3: Overload warning based on calibration FEM

[0076] To accurately detect overloaded vehicles on the target bridge at an early stage, the first step involves calibrating the FEM using strain sensors strategically placed on the target bridge. As the calibrated vehicles traverse each lane of the target bridge, a novel GRU neural network establishes a robust correlation between the strain response and the design parameters. Peak strain data obtained during the calibration test is fed into the neural network, leveraging the neural network's nonlinear mapping capabilities to generate refined design parameters for the FEM. By comparing the initial FEM model, the collected strain data, and the target strain, the network identifies the most effective adjustment parameters. Through rigorous training, the GRU neural network identifies new parameters that capture the complex nonlinear relationship between peak strain and the adjusted design parameters. This approach facilitates necessary modifications to the finite element model, ultimately improving its accuracy and reliability.

[0077] The predicted vehicle loads acting on the target bridge are divided into two different load modes, such as Figure 3 The first mode (Loading Mode I) is based on a maximum probability spatial distribution derived from the predicted vehicle speeds. The second mode (Loading Mode II) accounts for unfavorable load distribution, reducing vehicle spacing and concentrating vehicles in the middle of the span. This helps prevent potential damage to the bridge caused by low-probability distributions of vehicle loads.

[0078] These two loading modes were applied to the calibrated finite element model to evaluate the structural response. In addition to analyzing the maximum strain and maximum vertical displacement, the overturning stability factor was also considered to evaluate the bridge condition as described below:

[0079]

[0080] Among them, R Gi represents the support reaction force, μ represents the impact coefficient, vd i is the vertical distance from each support point to the rollover axis, and e is the vertical distance from the transverse lane position to the rollover axis. For Loading Mode I, three levels of warning thresholds are set based on different responses. In contrast, Loading Mode II only uses the third level warning threshold to detect abnormal responses.

[0081] Through real-time feedback, the degree of overload under the vehicle load pattern can be predicted, so as to determine whether the target bridge is overloaded. According to the early warning results, the bridge traffic flow can be intervened in time to avoid the occurrence of overload accidents.

[0082] Implementation 1:

[0083] Taking two simply supported beam bridges with different spans as an example, continuous vehicle data is used to simulate random traffic flows on the two bridges. By identifying vehicle loads and predicting vehicle spatiotemporal information, an effective overload warning is issued for the target bridge.

[0084] Overview of Numerical Models

[0085] like Figure 4 As shown, two T-beam bridges with identical cross-sections but different spans represent the pre-designed bridge and the target bridge, respectively. The pre-designed bridge has a span of 30m, the target bridge has a span of 45m, and the distance between the two bridges is 2000m. Both bridges are 13.2m wide and 2.75m high. They are constructed of five T-beams. The bridges are made of C50 concrete, which has an elastic modulus of 3.45×10 4 MPa, specific gravity is 26kN / m 3 , with a Poisson's ratio of 0.2. Each unit span on the bridge is 2.5 meters. The previous bridge consisted of 12 units per main beam, for a total of 60 units in the entire structure. The target bridge also had 12 units per main beam, for a total of 90 units. To simulate realistic bridge damage, the stiffness of unit 4 in G1 (T-beam 1), unit 9 in G3, and unit 12 in G5 on the target bridge was reduced by 30%.

[0086] like Figure 4 As shown, a long-term monitoring system was installed on the lead bridge, while a short-term inspection system was installed on the target bridge. For the lead bridge, two monitoring segments, S1 and S2, were designed as monitoring units. These segments were located at the bridge entrances of the first and second units, respectively. Given that the bridge consists of three lanes, the corresponding units on beams G2, G3, and G4 were selected as monitoring units for each segment. In units S11 to S13, bending strain sensors were installed at the bottom of the T-beams, while shear strain sensors (two cross-mounted FBG strain sensors) were installed on one side. In units S21 to S23, only shear strain sensors were installed on the side of the units. A total of 15 sensors were installed on the lead bridge to monitor strain response. For the target bridge, five strain sensors were installed at the midspan of the bridge for finite element calibration. In addition, a camera was installed on each bridge for LP identification.

[0087] Random traffic flow simulation

[0088] The present invention utilizes the NaSch model, originally a two-dimensional CA model numbered 184, as a basic tool for implementing the Intelligent Driving Model (IDM) to manage vehicle motion. The simulation captures the lane-changing behavior of vehicles as they leave the leading bridge and approach the target bridge, reflecting the dynamic changes in traffic flow between the two structures. The lane-changing behavior is affected by several factors, including the lane, speed, and load of each vehicle when leaving the leading bridge, as well as the lane-changing rules. The time evolution of the simulated traffic flow on the leading bridge and the target bridge is shown in Figure 2. Figure 5 shown.

[0089] Figure 5 Shows simulated traffic flow on three lanes. Figure 5As can be seen in Figure 2, although the traffic volume on both bridges varies at any given moment, the overall trend remains consistent. The peak traffic volumes on each lane are approximately 680, 750, and 700 vehicles per hour (veh / h), respectively. On average, the traffic volume on all lanes is approximately 600 vehicles per hour (veh / h), showing slight fluctuations. A random 20-second sample of vehicle data was selected from the simulated traffic flow to analyze the normal distribution of vehicle loads and speeds on each lane. The results obtained from this data are summarized in Figure 6 middle. Figure 6 (a) and (b) describe the normal distribution of vehicle load and speed, respectively. Figure 6 The analysis of (a) shows that the peak probability densities of vehicle loads in lanes I, II, and III are approximately 201.8 kN, 254.3 kN, and 229.1 kN, respectively. This indicates that the vehicle load in lane II is greater than that in other lanes. Figure 6 In (b), the peak probability density of vehicle speed in lane I is approximately 12.31 m / s, while lanes II and III exhibit peak values of approximately 12.26 m / s and 12.36 m / s, respectively. This indicates that the speed distribution of the three lanes varies very little.

[0090] Vehicle load identification

[0091] The simulated traffic flow was integrated into the finite element model of the bridge under study, allowing the extraction of the unit response data measured by the sensors to determine the vehicle loads. Following the method outlined above for calculating vehicle axle loads from strain, the strain history data collected from the monitoring units were analyzed. In this process, a two-axle calibration vehicle (with axle loads of 5 kN and 9 kN, and an axle spacing of 6 m) was first driven over each lane of the bridge to determine the lateral distribution coefficient η. ij The measured strains are then decoupled using Equation (1). The two separated strains are then used to estimate the velocity and axle spacing. Once the vehicle position is accurately determined, the axle weight can be calculated using Equation (4). This calculation is crucial for understanding the load distribution and overall dynamics of the vehicle on the bridge.

[0092] To simulate LP, each vehicle is assigned a unique identification number to distinguish it from all other vehicles. By comparing these IDs of vehicles crossing the two bridges, the vehicle load distribution on the target bridge can be obtained. The identified vehicle loads on the two bridges are shown in Figure 7 shown.

[0093] Figure 7Figures (a) and (b) depict the estimated vehicle speed and load on the leading bridge and target bridge, respectively. Based on the 20-second vehicle load time series data obtained by monitoring the leading bridge, it is observed that the vehicle queue gradually enters the target bridge within the time window of 146s to 216s, affected by the changes in driving speed and lane distribution. In order to clearly illustrate the load distribution characteristics of the two bridges, Figure 7 A time dimension truncation was used to display only the first 180 seconds of synchronous monitoring data. This approach emphasizes the dynamic load transfer process of the two bridges in the spatial domain. In the front bridge, the speed variation between the three lanes is relatively small, typically in the range of 10 to 15 m / s. The vehicle load distribution on lane III was observed to be 127.4 to 416.0 kN, on lane II 139.5 to 451.8 kN, and on lane I 122.4 kN to 319.9 kN. These identified load distributions are consistent with the Figure 6 The simulated vehicle information shown in (a) and (b) is closely related. The error estimation of vehicle speed and axle load distribution on all lanes is performed, as shown in Figure 8 shown.

[0094] like Figure 8 As shown, the differences in recognized vehicle speeds across lanes are significantly smaller than those in recognized vehicle loads. The maximum errors in recognized vehicle speeds for the three lanes were 4.96%, 4.70%, and 4.82%, respectively, all remaining below 5%. The peak recognition errors for the three lanes were 7.62%, 7.25%, and 7.20%, respectively, but the vast majority of these errors remained below 5%, with an average error of 4.24% across the three lanes. This finding demonstrates that this method effectively ensures the accuracy of vehicle information collected on the two bridges.

[0095] Vehicle spatiotemporal information prediction

[0096] The sample set of the GWO-GRU neural network is built using the spatiotemporal information of vehicles collected from the preceding bridge and the target bridge. The database includes 151 sets of vehicle data collected from these bridges, 70% of which are used for training the model and the remaining 30% for validation. The input data of the GWO-GRU neural network includes the arrival lane, speed and load information of the vehicle on the preceding bridge, while the output data corresponds to the arrival lane of the vehicle on the target bridge. The test results of the model are shown in Figure 2. Figure 9 shown.

[0097] Figure 9 The validation results of the trained model are given. By inputting the validation set into the established mapping model, the predicted arrival lane of each vehicle on the target bridge is generated. Figure 9It can be clearly seen that the predicted arrival lane corresponds closely to the actual lane, thus confirming the effectiveness of the established model. Subsequently, the vehicle information identified on the preceding bridge is input into the model to predict the spatiotemporal distribution of vehicles on the target bridge, as shown in Figure 2. Figure 10 shown.

[0098] Figure 10 The predicted vehicle loads for lanes I, II, and III over a 40-second training cycle and a 20-second prediction cycle are shown. During these two cycles, the load on lane I fluctuated between 120 and 330 kN. For lanes II and III, the predicted load fluctuations within the prediction cycle closely correlated with the observed load fluctuations during the training cycle, consistently ranging between 120 and 430 kN. The error calculation of the predicted results compared to the true values validated the accuracy of the proposed method. The maximum absolute errors for the three lanes were 6.69%, 7.38%, and 6.43%, respectively, and the average errors were 2.67%, 3.49%, and 3.24%, respectively.

[0099] The error at most time points is kept within 5%, and the overall average error is less than 4%. This shows that the proposed method can effectively predict the real-time arrival load of the target bridge.

[0100] Overload warning

[0101] To analyze and predict the load effects of vehicle loads on the actual structure, a GRU neural network was used to calibrate the finite element model of the target bridge. The GRU neural network established correlations between peak strain and design parameters. The elastic moduli of the main beam, cross beam, and virtual beam were selected as initial adjustment parameters. Validation data showed that the error between the calculated results of the calibrated finite element model and the measured values of the actual model was within 5%, accurately reflecting the true dynamic characteristics of the structure. Therefore, this calibrated finite element model can be considered the optimal model for identifying overload conditions on the target bridge.

[0102] Based on the predicted vehicle load condition I, vehicle load condition II can be determined by reducing the distance between vehicles to 3 to 4 meters and concentrating the vehicles in the middle span area of the bridge. These two load conditions are input into the calibrated finite element model (FEM) to calculate the load effect of the entire structure. The analysis includes the maximum strain, anti-overturning stability coefficient and the maximum vertical displacement of the structure. In terms of strain response, three levels of warning thresholds are set, namely 0.0021, 0.0027 and 0.003; for the anti-overturning stability coefficient, the warning thresholds are 3.25, 2.75 and 2.5 respectively; in terms of vertical displacement, the warning thresholds are set to 0.063m, 0.081m and 0.09m. The maximum strain, anti-overturning stability coefficient and maximum vertical displacement of the structure under the two load conditions are extracted from the calibrated finite element model, as shown in Figure 2. Figure 11 shown.

[0103] Figure 11 (a) shows the maximum strain, anti-overturning stability factor and maximum displacement of the structure under load case I; Figure 11 (b) shows the corresponding load effect under load case II. In addition, Figure 11 The structural response based on actual vehicle distribution is also shown. Figure 11 It can be observed that when the vehicle passes over the target bridge under Load Condition I, all structural responses remain within the safe range. However, under Load Condition II, the structural strain response exceeds the third-level warning threshold at 9.17 seconds and 11.17 seconds; the displacement response also exceeds the third-level warning threshold at 9.17 seconds, 11.17 seconds, and 17.17 seconds. The anti-overturning stability coefficient does not fall below the warning threshold throughout the entire prediction period. These results indicate that under Vehicle Load Condition II, adverse loads can cause the structural strain and displacement responses to exceed the warning limits at certain moments.

[0104] Figure 11 The actual response values for two actual load conditions are also presented. As can be seen from the figure, with the exception of the strain response at 7.16 seconds, all other responses are highly consistent with the actual values. At 7.16 seconds, the error in the predicted strain value reached 13.6%. Further analysis shows that this large error is primarily due to inaccurate vehicle position recognition. These results verify that the proposed method has good early warning capabilities under both Conditions I and II.

[0105] This invention is used to achieve early identification and early warning of vehicle overloads. To achieve this goal, two bridges equipped with different sensor systems were used as research objects. Three models were constructed to identify vehicle loads, predict vehicle loads, and estimate load effects. The results demonstrate the effectiveness, reliability, and feasibility of this invention. The main technical benefits are as follows:

[0106] (1) This invention integrates mechanical, data, and physical models to accurately predict vehicle load distribution and its effects, enabling early identification of overloaded vehicles. This proactive approach helps identify risks and prevent potential damage, ensuring the overall safety of bridge structures.

[0107] (2) A vehicle space mapping model based on the GMO-GRU neural network was constructed, which comprehensively considers factors such as vehicle speed, load, and driving lane to accurately predict the vehicle arrival lane on the target bridge. This model effectively improves the accuracy of the vehicle load distribution prediction on the target bridge.

[0108] (3) The present invention proposes predictive load conditions and adverse load conditions, and sets reasonable warning thresholds accordingly. Its goal is to minimize the potential impact of extreme events on the structure while ensuring structural safety under normal driving conditions.

[0109] (4) This invention uses predicted load effects as the basis for overload risk assessment, rather than relying solely on vehicle loads. This approach is not only applicable to newly constructed bridges, but can also serve as an effective monitoring and early warning tool for older bridges under high load levels.

[0110] Example 2:

[0111] An embodiment of the present invention further provides a bridge overload warning system, comprising:

[0112] a first processing device for obtaining vehicle load distribution on the leading bridge and the target bridge;

[0113] The second processing device is used to construct a spatial mapping prediction model of vehicles on the two bridges using the vehicle load distribution of the preceding bridge and the target bridge; the vehicle information on the preceding bridge identified by the FBG sensor is input into the trained prediction model to predict the vehicle load distribution on the target bridge;

[0114] The third processing device is used to calibrate the finite element model using the monitoring data of the target bridge after determining the temporal and spatial distribution of vehicles on the target bridge, analyze the abnormal strain, vertical displacement and anti-overturning stability coefficient of the structure, and provide multi-level early warning of the structure's status.

[0115] As an implementation method of the embodiment of the present invention, the second processing device uses the vehicle load distribution of the leading bridge and the target bridge and adopts the GWO-GRU neural network to construct a spatial mapping prediction model of vehicles on the two bridges.

[0116] As an implementation method of an embodiment of the present invention, the front bridge uses fiber Bragg gratings (FBG) and cameras for long-term vehicle information monitoring, and the target bridge uses visual sensors combined with multiple FBG sensors for short-term detection to assist in constructing a vehicle space mapping model and calibrating the finite element model (FEM).

[0117] As an implementation method of the present invention, FBG strain data is used to identify vehicle loads, and the camera captures license plate LP information and assigns a unique identifier to each identified vehicle load; by matching the LP information identified on the two bridges, the vehicle load on the target bridge is determined based on the vehicle data on the preceding bridge.

[0118] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A bridge overload warning method, characterized in that: include: Step S1, obtaining the vehicle load distribution on the leading bridge and the target bridge; Step S2: Using the vehicle load distribution of the front bridge and the target bridge, a spatial mapping prediction model of vehicles on the two bridges is constructed; the vehicle information on the front bridge identified by the FBG sensor is input into the trained prediction model to predict the vehicle load distribution on the target bridge; Step S3: After determining the temporal and spatial distribution of vehicles on the target bridge, the finite element model is calibrated using the target bridge monitoring data, and the abnormal strain, vertical displacement and anti-overturning stability coefficient of the structure are analyzed to provide a multi-level early warning of the structure status.

2. The bridge overload warning method according to claim 1, characterized in that: The vehicle load distribution on the preceding bridge and the target bridge is used to construct a spatial mapping prediction model for vehicles on the two bridges using the GWO-GRU neural network.

3. The bridge overload warning method according to claim 2, characterized in that: The front bridge uses fiber Bragg gratings (FBG) and cameras for long-term vehicle information monitoring, while the target bridge uses a visual sensor combined with multiple FBG sensors for short-term detection to assist in building a vehicle space mapping model and calibrating the finite element model (FEM).

4. The bridge overload warning method according to claim 3, characterized in that: In the front axle, FBG strain data is used to identify vehicle loads, and the camera captures license plate LP information to assign a unique identifier to each identified vehicle load; By matching the LP information identified on the two bridges, the vehicle load on the target bridge is determined based on the vehicle data on the preceding bridge.

5. A bridge overload warning system, characterized in that: include: a first processing device for obtaining vehicle load distribution on the leading bridge and the target bridge; The second processing device is used to construct a spatial mapping prediction model of vehicles on the two bridges using the vehicle load distribution of the preceding bridge and the target bridge; the vehicle information on the preceding bridge identified by the FBG sensor is input into the trained prediction model to predict the vehicle load distribution on the target bridge; The third processing device is used to calibrate the finite element model using the monitoring data of the target bridge after determining the temporal and spatial distribution of vehicles on the target bridge, analyze the abnormal strain, vertical displacement and anti-overturning stability coefficient of the structure, and provide multi-level early warning of the structure's status.

6. The bridge overload warning system according to claim 5, characterized in that: The second processing device uses the vehicle load distribution of the leading bridge and the target bridge and adopts the GWO-GRU neural network to build a spatial mapping prediction model of vehicles on the two bridges.

7. The bridge overload warning system according to claim 6, characterized in that: The front bridge uses fiber Bragg gratings (FBG) and cameras for long-term vehicle information monitoring, while the target bridge uses a visual sensor combined with multiple FBG sensors for short-term detection to assist in building a vehicle space mapping model and calibrating the finite element model (FEM).

8. The bridge overload warning system according to claim 7, characterized in that: FBG strain data is used to identify vehicle loads, and the camera captures license plate LP information to assign a unique identifier to each identified vehicle load; By matching the LP information identified on the two bridges, the vehicle load on the target bridge is determined based on the vehicle data on the preceding bridge.

Images