Road and bridge real-time monitoring and maintenance system based on intelligent perception

Abstract

The invention relates to the technical field of bridge structure safety monitoring and maintenance, and discloses a road bridge real-time monitoring and maintenance system based on intelligent perception, which comprises a data acquisition module used for acquiring multi-dimensional data of a bridge operation state through a deployed multi-source sensor, the multi-source sensor comprises a vibration sensor, a strain sensor, a temperature and humidity sensor and image acquisition equipment; the data preprocessing module is used for carrying out cleaning, synchronization and feature extraction on the collected multi-dimensional data to obtain dynamic features, environment features and visual features; and a data fusion module. According to the method, the comprehensiveness and accuracy of bridge monitoring are improved through multi-source data fusion, the reliability of an evaluation result is ensured in combination with physical constraint optimization, health trend prediction and early warning functions are realized based on a time sequence model, scientific decision support is provided for bridge maintenance, and the monitoring and management efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge structure safety monitoring and maintenance, and in particular to a road bridge real-time monitoring and maintenance system based on intelligent perception. Background Art

[0002] As road bridges become increasingly important in transportation networks, real-time monitoring and maintenance of bridge operating conditions have become key technical means to ensure traffic safety. Traditional bridge monitoring mainly relies on manual inspections and a single type of sensor acquisition technology, such as monitoring bridge dynamic response through vibration sensors, measuring structural deformation through strain sensors, or observing crack expansion through visual equipment. These methods provide data support for bridge maintenance to a certain extent, but with the increase in the service life of bridges and the complexity of the operating environment, it is difficult to meet the comprehensive, real-time and efficient assessment needs of bridge health status by relying solely on a single monitoring method.

[0003] Existing technologies perform health assessments based on data collected by single-source sensors, but the limitations of a single data source make it difficult for monitoring results to fully reflect the operating status of the bridge. In addition, existing methods mostly use linear analysis models, which make it difficult to capture the nonlinear coupling relationship between multi-source data, and thus cannot accurately characterize the health status of the bridge in a complex operating environment. At the same time, existing technologies lack effective constraints on the physical laws of bridges (such as vibration modal characteristics and thermal expansion effects), and it is easy for the assessment results to be inconsistent with the actual bridge behavior. In terms of health trend prediction, existing technologies lack specificity in capturing the trend of health status changes over time, especially when the bridge is fatigued or the environment changes suddenly, and cannot provide reliable early warnings. Summary of the invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a real-time monitoring and maintenance system for roads and bridges based on intelligent perception, which solves the problems in the existing technology of incomplete and inaccurate monitoring results and insufficient health status prediction ability caused by a single data source, linear analysis model and lack of physical constraints.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A road and bridge real-time monitoring and maintenance system based on intelligent perception, comprising:

[0006] A data acquisition module, used to collect multi-dimensional data of the bridge operation status through the deployed multi-source sensors, wherein the multi-source sensors include vibration sensors, strain sensors, temperature and humidity sensors, and image acquisition equipment;

[0007] The data preprocessing module is used to clean, synchronize and extract features of the collected multi-dimensional data to obtain dynamic features, environmental features and visual features;

[0008] The data fusion module is used to extract the coupling relationship between multi-dimensional data based on the constructed multi-source data relationship graph and graph neural network model, and generate fused global features;

[0009] The physical constraint optimization module is used to impose physical constraints based on the vibration modal characteristics and thermal expansion characteristics of the bridge on the fused global features to optimize the global features;

[0010] A health status assessment module, used to generate a health status score based on the optimized global features;

[0011] The health trend prediction module is used to predict the future health status of the bridge based on the time series model and generate maintenance recommendations.

[0012] Preferably, the data acquisition module further comprises:

[0013] The multi-source sensor deployment submodule is used to deploy vibration sensors, strain sensors, and temperature and humidity sensors according to the key stress positions of the bridge structure, and obtain images of bridge cracks through image acquisition equipment;

[0014] The data transmission submodule is used to transmit the collected data to the data preprocessing module via a wireless network or a wired network.

[0015] Preferably, the data preprocessing module further comprises:

[0016] The data cleaning submodule is used to remove outliers and fill in missing data during sensor acquisition;

[0017] The data synchronization submodule is used to synchronize the time of data with different sampling frequencies;

[0018] The feature extraction submodule is used to extract power spectral density features from dynamic features, extract normalized statistical features from environmental features, and extract the width and length of cracks from visual features.

[0019] Preferably, the data fusion module includes:

[0020] A data relationship graph construction submodule is used to construct a relationship graph including multi-dimensional data nodes and data association edges based on the correlation of multi-source data, and the weight of the edge is calculated according to the information sharing amount between the node data;

[0021] A graph neural network submodule, used for performing multi-level message transmission and aggregation processing on node features in the relationship graph to extract nonlinear coupling features between multi-source data;

[0022] The global feature generation submodule is used to generate global features describing the health status of the bridge through feature aggregation method.

[0023] Preferably, the physical constraint optimization module includes:

[0024] The vibration modal constraint submodule is used to calibrate the global features based on the vibration modal characteristics of the bridge structure dynamics;

[0025] Thermal expansion constraint submodule, used to correct the relationship between temperature and humidity characteristics and strain characteristics based on the thermal expansion formula of bridge materials;

[0026] The comprehensive optimization submodule is used to adjust the weight parameters of the global features according to the optimization objective function, where the optimization objective function includes the data-driven loss term, the modal constraint loss term and the thermal expansion constraint loss term.

[0027] Preferably, the vibration modal characteristics are determined by the vibration frequency of the bridge structure, and the vibration frequency is positively correlated with the stiffness and mass of the bridge. During the optimization process, the vibration modal characteristics are ensured to be consistent with the actual dynamic characteristics of the bridge.

[0028] Preferably, the health status assessment module includes:

[0029] A health index generation submodule, used to generate a bridge health status score according to the global characteristics, wherein the health status score ranges from 0 to 1;

[0030] The indicator visualization submodule is used to graphically display the current health status of the bridge.

[0031] Preferably, the health trend prediction module includes:

[0032] The time series model submodule is used to predict the future trend of bridge health status scores through the long short-term memory network model;

[0033] The early warning generation submodule is used to generate health early warning information according to the downward trend of the health status score.

[0034] Preferably, the maintenance recommendations are generated based on historical data and future predicted values ​​of health status scores, with priority given to repairing critical bridge parts with health status scores below a preset threshold.

[0035] Preferably, the global feature is generated by a fusion representation including data collected by all sensors, and the fusion representation can simultaneously reflect the dynamic behavior, environmental impact and visual damage status of the bridge.

[0036] The present invention provides a real-time monitoring and maintenance system for roads and bridges based on intelligent perception. It has the following beneficial effects:

[0037] 1. The present invention collects vibration signals, strain signals, temperature and humidity parameters and image information through multi-source sensors, and comprehensively constructs a multi-dimensional data basis for the operation status of the bridge. The data fusion module combines the characteristics of the graph neural network to fully explore the nonlinear coupling relationship between multi-source data and generate global feature representations. This deep fusion of multi-source information avoids the one-sidedness of traditional single data source monitoring methods, greatly improves the comprehensiveness and accuracy of monitoring results, and has significant advantages, especially in complex operating environments.

[0038] 2. The present invention combines the vibration modal characteristics, thermal expansion characteristics and other physical laws of the bridge with the data-driven model through a physical constraint optimization module to optimize the global feature representation. This module ensures that the bridge health status assessment results are not only consistent in a mathematical sense, but also have physical rationality in engineering applications. This innovative method effectively compensates for the possible erroneous judgments that may occur in pure data-driven models in nonlinear and complex environments, making the assessment results more reliable, especially showing higher robustness under conditions of long-term fatigue damage to the bridge or extreme environmental changes.

[0039] 3. The present invention uses a long short-term memory network to accurately predict the future health status of the bridge. Combined with the dynamic abnormal trend recognition function, the module can promptly detect the significant downward trend of the bridge health score and issue early warning information to provide managers with scientific maintenance decision support. This function not only reduces the risk of sudden bridge failures, but also realizes the active and intelligent maintenance of bridges, greatly improving the efficiency and effectiveness of bridge safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0041] Figure 2 It is a structural schematic diagram of a data acquisition module of the present invention;

[0042] Figure 3 It is a structural schematic diagram of a data preprocessing module of the present invention;

[0043] Figure 4 It is a schematic diagram of the data fusion structure of the present invention;

[0044] Figure 5 It is a schematic diagram of the structure of the physical constraint optimization module of the present invention;

[0045] Figure 6 This is a structural diagram of the health status assessment module of the present invention;

[0046] Figure 7 It is a structural schematic diagram of the health trend prediction module of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Please refer to the attached Figure 1 The embodiment of the present invention provides a road and bridge real-time monitoring and maintenance system based on intelligent perception, including:

[0049] A data acquisition module, used to collect multi-dimensional data of the bridge operation status through the deployed multi-source sensors, wherein the multi-source sensors include vibration sensors, strain sensors, temperature and humidity sensors, and image acquisition equipment;

[0050] In this embodiment, the specific implementation of the data acquisition module is as follows:

[0051] Generally speaking, the data acquisition module continuously collects multi-dimensional data including vibration signals, strain signals, environmental parameters and crack images through sensors installed in key parts of the bridge structure.

[0052] Specifically, vibration sensors are used to measure the vibration response of a bridge under external loads (such as vehicle traffic, wind load). As an option, vibration sensors can be installed at the mid-span position or support area of ​​the bridge to obtain representative vibration data. This data is usually collected in the form of a time series, containing information such as amplitude and frequency.

[0053] In one possible implementation, strain sensors are placed in the main stress-bearing areas of the bridge, such as the lower edge of the beam with greater tension or the surface of the pier in the compression area. The data collected by the strain sensors is used to quantify the deformation of the bridge, and the sampling frequency is set according to the working conditions of the bridge. For example, the sampling frequency can be increased when a heavy-loaded vehicle passes by to better capture transient stress changes.

[0054] Meanwhile, the temperature and humidity sensors are used to measure environmental parameters, and their placement is usually selected in the area near the outer surface of the beam body on the upper part of the bridge. For example, in some embodiments, the temperature and humidity sensors can be arranged in the area near the outer edge of the bridge cross section to more accurately reflect the impact of the external environment on the performance of the bridge material.

[0055] The image acquisition device is installed in areas where cracks are likely to appear, such as the bridge deck pavement, the bottom of the beam, and near the expansion joint. Specifically, the image acquisition device can capture cracks, peeling or rust on the bridge surface in real time through a high-resolution camera. In some embodiments, the camera equipment carried by the drone can be used as a supplementary means to collect image data on the top of the bridge or other hard-to-reach locations.

[0056] Data transmission and management:

[0057] In one implementation, all sensors transmit the collected data via a wired or wireless network. The wireless transmission method can be based on a low-power wide area network (such as LoRa or NB-IoT) to meet the monitoring needs of remote bridges. For bridge locations with less environmental interference, the wired method can ensure the stability of data transmission.

[0058] In order to ensure the synchronization of time series data, the data collected by the sensor is timestamped. For example, the sampling period of vibration and strain sensors can be set to 1 millisecond or less to capture fast-changing dynamic responses, while the sampling frequency of temperature, humidity and image data is usually lower, which may be 1 second or longer.

[0059] Data model and formula:

[0060] The data acquisition model in this module can be described by the following mathematical expression:

[0061] For vibration signals:

[0062] X v (t) = A v sin(2πf v t+φ v )

[0063] in:

[0064] X v (t) represents the vibration signal at time t;

[0065] A v is the amplitude, indicating the vibration intensity;

[0066] f v is the vibration frequency, indicating the main frequency of the vibration response;

[0067] φ v is the phase, indicating the initial state of vibration.

[0068] For strain data, the strain signal X s (t) can be expressed as:

[0069]

[0070] in:

[0071] X s (t) is the strain value at time t;

[0072] ΔL(t) is the change in beam length at sampling time t;

[0073] L0 is the original length of the beam.

[0074] The results of environmental parameter collection (such as temperature and humidity) are usually given in scalar form, for example, temperature T(t) and humidity H(t) reflect the thermal and humidity conditions of the environment respectively.

[0075] Image data can be represented as a two-dimensional matrix I(x,y), where each pixel I(x,y) reflects the grayscale value or color intensity at the position (x,y).

[0076] Extended technical content:

[0077] In some possible extended embodiments, the data acquisition module can integrate edge computing functions to enable the sensor to perform preliminary data processing locally. For example, the edge device can calculate the power spectral density (PSD) of the vibration signal in real time:

[0078]

[0079] in:

[0080] F(f) is the Fourier transform of the vibration signal;

[0081] T is the length of the sampling time window.

[0082] This method can reduce the amount of data before transmission and improve transmission efficiency.

[0083] In summary, the data acquisition module not only realizes comprehensive monitoring of the bridge operation status through multi-source sensors, but also ensures the accuracy and reliability of the data through scientific and reasonable layout and transmission methods. All collected data are time-stamped and output in a standardized format, providing consistency and high-quality data input for subsequent modules. Through the implementation of this module, the real-time bridge monitoring system has comprehensive and efficient input capabilities, fully supporting the subsequent processing and analysis process.

[0084] The data preprocessing module is used to clean, synchronize and extract features of the collected multi-dimensional data to obtain dynamic features, environmental features and visual features;

[0085] In this embodiment, the specific implementation of the data preprocessing module is as follows:

[0086] Generally, the data preprocessing module includes three submodules: data cleaning, time synchronization, and feature extraction. These submodules can be run in parallel or in sequence according to specific needs.

[0087] Specifically, the data cleaning submodule is responsible for removing outliers and noise from the raw data. As an option, statistical methods can be used to clean outliers in vibration signals and strain signals. For example, set upper and lower limits of the data range and mark data outside the range as outliers. For noise removal of image data, image processing algorithms such as Gaussian filtering or median filtering can be used.

[0088] In one possible implementation, missing values ​​are processed using interpolation methods. For vibration signals and strain signals, linear interpolation can effectively fill in missing data in short time intervals; for long-term missing data, a prediction model based on historical data is used to complete the missing data. For example, missing values ​​in temperature and humidity data can be predicted using time series analysis methods.

[0089] The time synchronization submodule is used to solve the problem of inconsistent sampling frequencies of different sensors. Vibration and strain signals are usually recorded at a high sampling rate (such as 1kHz), while the sampling rate of temperature, humidity and image data is lower (such as once per second or lower). In general, in order to achieve time alignment of multi-source data, the Dynamic Time Warping (DTW) algorithm is used to align data of different time scales. This algorithm can adjust the sampling time points by calculating the minimum alignment distance between data, so that different data sources are consistent under the same time base.

[0090] Feature extraction submodule

[0091] The function of the feature extraction submodule is to extract representative information about the health status of the bridge from the cleaned and synchronized data.

[0092] Specifically, for the vibration signal, time domain and frequency domain features can be extracted. In a possible implementation, the time domain features include mean, standard deviation and peak amplitude, and the frequency domain features include power spectrum density (PSD) and main frequency.

[0093] For strain signals, the main extraction methods are the mean strain value and the maximum strain change, which are used to reflect the long-term deformation trend and instantaneous response capability of the bridge structure. The characteristics of temperature and humidity data are usually the mean value and change rate of their time series, which are used to describe the potential impact of the environment on the performance of bridge materials.

[0094] For image data, the feature extraction submodule uses a convolutional neural network (CNN) to automatically extract crack width, length, and distribution features. c and length L c It can be initially obtained through the image edge detection algorithm, and the detection results can be further optimized using the deep learning network.

[0095] Data standardization and output

[0096] In order to ensure the scale consistency of multi-source data, all extracted features need to be standardized. As an implementation method, normalization or Z-score normalization method can be used. For example, the normalization formula is:

[0097]

[0098] in:

[0099] X is the original eigenvalue;

[0100] X min and X max are the minimum and maximum values ​​of the feature, respectively.

[0101] After standardization, the data is packaged and output in a standardized format. The features of vibration, strain, environment, and image data are passed to the data fusion module in the form of vectors or matrices. For example, the features of a vibration signal can be represented as a time series matrix V, where each row represents a feature value within a sampling time window.

[0102] Extended technical content

[0103] In some embodiments, the data preprocessing module can integrate real-time monitoring and processing functions. For example, by embedding edge computing nodes, some data cleaning and feature extraction can be completed locally on the sensor. This design can significantly reduce the pressure of data transmission and improve real-time response capabilities.

[0104] In addition, the cleaning algorithm can be further optimized for data collection scenarios in high-noise environments. For example, an adaptive noise removal algorithm can be used to dynamically suppress background noise in vibration signals.

[0105] The data fusion module is used to extract the coupling relationship between multi-dimensional data based on the constructed multi-source data relationship graph and graph neural network model, and generate fused global features;

[0106] In this embodiment, the specific implementation of the data fusion module is as follows:

[0107] In general, the data fusion module includes three submodules: data relationship graph construction, graph neural network modeling, and global feature generation. The functions of each submodule are interrelated, and together they realize the comprehensive analysis and representation of multi-source data.

[0108] Specifically, the data relationship graph construction submodule is used to analyze the correlation characteristics of multi-source data and construct a weighted graph structure containing nodes and edges. As an option, vibration signals, strain signals, environmental parameters, and image features are defined as nodes in the graph, and the edge weights between nodes are obtained by calculating the correlation or mutual information between the data. For example, in some implementations, the edge weights can be defined as:

[0109]

[0110] in:

[0111] w ij Represents node X i and X j The edge weights between

[0112] Cov(X i ,X j ) indicates X i and X j The covariance of

[0113] and X i and X j The standard deviation of .

[0114] In some embodiments, the data relationship graph may be dynamically adjusted. For example, in an environment with large temperature and humidity changes, the weight between the temperature and humidity node and the strain signal node may be increased to reflect the effect of thermal expansion on structural deformation.

[0115] Graph Neural Network Modeling Submodule

[0116] In one possible implementation, the data relationship graph is input into the graph neural network model, and the network updates and aggregates the features of each node through a message passing mechanism. In general, the update of node features can be expressed as:

[0117]

[0118] in:

[0119] Represents the feature representation of node i at the k+1th layer;

[0120] Represents the set of neighbor nodes of node i;

[0121] w ij represents the edge weight between node i and its neighbor node j;

[0122] W k and b k are the learnable weights and biases of the graph neural network at the kth layer respectively;

[0123] σ represents a non-linear activation function, such as ReLU.

[0124] As an option, the number of layers K of the graph neural network can be set according to the complexity of the data. For example, when the nonlinear coupling between the data is strong, the depth of the network can be increased to extract higher-level features. In another implementation, the attention mechanism can be combined to give different weights to the neighbor information of different nodes, thereby enhancing the modeling ability of specific relationships.

[0125] Global feature generation submodule

[0126] After all graph neural network layers are completed, node features are further aggregated into global features. Specifically, the aggregation method can be global average pooling or global maximum pooling. For example, for global average pooling, the feature generation formula is:

[0127]

[0128] in:

[0129] H represents the global feature;

[0130] Represents the final feature representation of node i at layer K;

[0131] n is the total number of nodes in the graph.

[0132] The global feature H represents a unified representation of multi-source data, integrating multi-dimensional information of vibration, strain, environment, and image features. In some embodiments, the global feature can directly reflect the overall health status of the bridge. For example, the presence of abnormal frequencies in the vibration signal may significantly affect the value of H, and the crack features in the image will also be reflected in the global feature through the modeling of the relationship graph.

[0133] Extended technical content

[0134] In some embodiments, the data fusion module can further optimize the construction method of the graph relationship. For example, by introducing a dynamic edge weight adjustment mechanism based on time series analysis, the edge weight is updated in real time to reflect the latest association between nodes. For example, when the vehicle load on the bridge decreases at night, the association between the vibration signal node and the environmental parameter node may be weakened.

[0135] In addition, in some special environments (such as bridges under extreme temperature difference conditions), specific adjustment factors can be introduced for environmental parameter nodes. For example, when the temperature difference exceeds a certain threshold, additional edges can be added to the temperature and humidity nodes and the strain nodes to enhance their influence on the global features.

[0136] The physical constraint optimization module is used to impose physical constraints based on the vibration modal characteristics and thermal expansion characteristics of the bridge on the fused global features to optimize the global features;

[0137] In this embodiment, the specific implementation of the physical constraint optimization module is as follows:

[0138] Generally speaking, the physical constraint optimization module includes vibration mode constraints, thermal expansion constraints and comprehensive optimization sub-modules. These sub-modules cooperate with each other and work together on the optimization process of the global feature H.

[0139] Vibration modal constraints

[0140] Specifically, the vibration modal characteristics of a bridge are important physical parameters for evaluating the health of the bridge structure. The vibration modal frequency f i It has a direct physical relationship with the stiffness K and mass M of the bridge. In a possible implementation, the present invention is based on modal analysis theory and defines the physical relationship of vibration frequency as follows:

[0141]

[0142] in:

[0143] f i is the i-th order modal frequency;

[0144] α i is the modal coefficient, which is related to the geometry and boundary conditions of the bridge;

[0145] K is the equivalent stiffness of the bridge;

[0146] M is the equivalent mass of the bridge.

[0147] In the present invention, the vibration mode constraint submodule compares the vibration frequency predicted in the global feature The actual modal characteristics Generate modal constraint loss:

[0148]

[0149] As an option, when the stiffness or mass of the bridge is significantly affected by vehicle loads or environmental factors, the estimated values ​​of K and M can be dynamically adjusted to ensure the accuracy of the vibration modal constraints.

[0150] Thermal expansion restraint

[0151] The bridge will expand or contract when the ambient temperature changes, and the strain signal ∈ is positively correlated with the temperature change ΔT. In general, the present invention describes the thermal expansion phenomenon by the following formula:

[0152] ∈=α·ΔT

[0153] in:

[0154] ∈ is the thermal expansion strain of the bridge structure;

[0155] α is the coefficient of thermal expansion, which is related to the physical properties of the bridge material;

[0156] ΔT is the temperature change.

[0157] In one possible implementation, the thermal expansion constraint submodule dynamically adjusts the thermal expansion coefficient α according to the real-time changes of the ambient temperature data T(t) and the strain data ∈(t). For example, when the bridge is in a high temperature area, the system will increase the weight of α to increase the influence of the ambient data in the optimization target.

[0158] The loss function for thermal expansion constraint is defined as:

[0159]

[0160] Through this constraint, the present invention can effectively calibrate the impact of environmental parameters on the global feature H.

[0161] Comprehensive optimization submodule

[0162] In order to combine the vibration mode constraint and the thermal expansion constraint to optimize the global features, the present invention designs a comprehensive optimization objective function, which combines the data-driven loss function with the two physical constraint loss functions:

[0163] L=L data +λ1L modal +λ2L thermal

[0164] in:

[0165] L data It is a data-driven loss function, usually the error between the global feature H and the actual output health state;

[0166] λ1 and λ2 are weight coefficients used to balance the effects of modal constraints and thermal expansion constraints on the optimization objective.

[0167] In general, the choice of weight coefficient is related to the operating environment of the bridge and the specific evaluation requirements. For example, when the bridge is located in an area affected by heavy traffic, the value of λ1 can be increased to emphasize the role of vibration modal constraints.

[0168] Extended technical content

[0169] In some embodiments, the physical constraint optimization module can also add other constraints according to the special design conditions of the bridge. For example, for offshore bridges, the constraints of wind load on the structural dynamic response can be introduced; for mountain bridges, the correction term for the effect of humidity on the mechanical properties of materials can be added.

[0170] In addition, in order to improve the efficiency of the optimization process, the present invention can adopt an optimization algorithm based on gradient descent to iteratively adjust the model parameters. As an option, an adaptive optimization method (such as Adam optimizer) can also be used to dynamically adjust the learning rate to improve the convergence speed.

[0171] A health status assessment module, used to generate a health status score based on the optimized global features;

[0172] In this embodiment, the specific implementation of the health status assessment module is as follows:

[0173] In general, the health status assessment module receives the global feature H output from the physical constraint optimization module. opt This feature contains multi-dimensional information such as vibration frequency, strain amplitude, temperature and humidity changes, and crack width, representing the overall characteristics of the current state of the bridge. To facilitate evaluation and practical application, the system maps these complex features into a single health score ranging from 0 to 1. The closer the score is to 1, the healthier the bridge is; the closer it is to 0, the greater the risk of the bridge.

[0174] Specifically, the generation process of the health score is based on a multi-layer perceptron (MLP) model. In one possible implementation, the MLP model receives the optimized feature H opt As input, it generates a health score through a fully connected layer and an activation function. The core calculation process of MLP can be expressed as:

[0175] HI=σ(W (2) ReLU(W (1) ·H opt +b (1) )+b (2) )

[0176] in:

[0177] HI is the health status score;

[0178] H opt is the optimized global feature;

[0179] W (1) and W (2) are the weight matrices of the first and second layers respectively;

[0180] b (1) and b (2) are the biases of the first and second layers, respectively;

[0181] ReLU(x)=max(0,x) is an activation function used to introduce nonlinearity;

[0182] The Sigmoid function maps the output to the range of 0 to 1.

[0183] Feature weight design

[0184] Alternatively, the weights W and biases b of the MLP model can be trained by supervised learning. During the training process, the system optimizes the model parameters based on a large number of historical bridge datasets, including health status labels (such as intact, slightly damaged, moderately damaged, and severely damaged).

[0185] In some embodiments, a feature weight adjustment mechanism may be introduced. For example, when a significant increase in crack width is detected, the system increases the weight of the crack feature in the global feature, thereby more sensitively reflecting the severity of structural damage.

[0186] Hierarchical classification of health status scores

[0187] In order to facilitate engineering applications and risk management, health status scores are usually divided into multiple levels. For example:

[0188] HI>0.8: The structure is in good condition and no intervention is required.

[0189] 0.5≤HI≤0.8: The structure is in general condition and regular inspection is recommended.

[0190] 0.2≤HI<0.5: The structure has potential risks and requires intensive monitoring.

[0191] HI<0.2: The structure is in a serious condition and maintenance measures should be taken immediately.

[0192] This division method can provide managers with intuitive evaluation results and assist in making quick decisions.

[0193] Visualization of health index

[0194] In this embodiment, the health status score can also be displayed graphically through the visualization module. In some embodiments, the system generates a real-time health status trend chart to show the time evolution of the score. For example, the system can use a line chart to show the change of the score in the past week to help users identify potential structural degradation trends.

[0195] In another possible implementation, the health status score can also be combined with the structural model of the bridge to generate a three-dimensional visualization image. For example, the system can mark different parts of the bridge with different colors according to the score, with green indicating health, yellow indicating medium risk, and red indicating high risk. This intuitive visualization method can improve monitoring efficiency.

[0196] Extended technical content

[0197] In some embodiments, the health status assessment module can also introduce expert rules for correction. For example, in the case where the crack width increases rapidly but the health score changes little, the system can appropriately adjust the score in combination with expert rules. In addition, the module can dynamically update the parameters of the scoring model based on the historical operation data of the bridge to adapt it to the long-term operation characteristics of the bridge.

[0198] To further improve the reliability of the score, the health status score generation process can also be combined with uncertainty assessment. For example, the confidence interval of the score can be estimated through a Bayesian neural network to provide additional reference information for managers.

[0199] The health trend prediction module is used to predict the future health status of the bridge based on the time series model and generate maintenance recommendations.

[0200] In this embodiment, the specific implementation of the health trend prediction module is as follows:

[0201] Generally, the health trend prediction module receives the health score sequence {HI t ,HI t-1 ,HI t-2 ,…,HI t-n}. Here, HI t Indicates the current health status score, HI t-1 Indicates the score of the previous moment, and so on. In order to predict future health status, the module uses a time series prediction model to train and analyze the score sequence.

[0202] Time Series Modeling

[0203] Specifically, the temporal evolution of the bridge health score has significant time dependence. For example, the downward trend of the score may be affected by the long-term fatigue of the bridge, environmental changes, or sudden loads. In one possible implementation, the module uses a long short-term memory network (LSTM) as a time series modeling tool, and its structure can effectively capture short-term changes and long-term trends in the health score.

[0204] The core formula of the LSTM network is as follows:

[0205]

[0206] f t =σ(W f x t +U f h t-1 +b f )

[0207] i t=σ(W i x t +U i h t-1 +b i )

[0208] in:

[0209] h t : The hidden state at the current moment;

[0210] x t : Input health score sequence;

[0211] f t : Forget gate, controls the influence of the hidden state at the previous moment;

[0212] i t : Input gate, which controls the update of the hidden state by the current input;

[0213] Candidate hidden states;

[0214] W and U: trainable weights of the network;

[0215] b: bias;

[0216] σ: Sigmoid activation function;

[0217] ⊙: Element-wise multiplication.

[0218] As an option, the length n of the input sequence can be dynamically adjusted according to the frequency and amount of bridge monitoring data. For example, for short-term prediction tasks, a smaller n can be selected to focus on recent health score changes; while for long-term prediction tasks, n can be increased to capture long-term trends.

[0219] Prediction of future health status

[0220] After completing the model training, the health trend prediction module can generate the predicted health status score for the future period. For example, for the future m moments, the prediction sequence can be expressed as:

[0221] {HI t+1 ,HI t+2 ,…,HI t+m}

[0222] Here, HI t+k Represents the health status score at the kth moment in the future.

[0223] As a possible implementation method, the prediction results can be corrected according to the actual operating environment and historical trends of the bridge. For example, when the system detects abnormal changes in ambient temperature and humidity, the weight parameters of the model output can be adjusted to increase the impact of environmental characteristics on the prediction results.

[0224] Abnormal trend identification and early warning

[0225] In general, the health trend prediction module also has the function of identifying abnormal trends. For example, when the predicted health score drops faster than the set threshold, the system can trigger an early warning message to prompt managers to conduct key inspections and maintenance on the bridge.

[0226] Specifically, the module calculates the predicted rate of change of the health score:

[0227]

[0228] When ΔHI is less than a preset threshold (e.g., -0.1), the system will mark the bridge as being in a high-risk state and generate maintenance recommendations.

[0229] Visualization of prediction results

[0230] The health trend prediction module can also present the prediction results in a graphical form. In some embodiments, the module displays the predicted changes in the health score through a line graph. For example, the system can draw a curve of the health score change from the current moment to the next week to intuitively reflect the health evolution trend of the bridge.

[0231] In another implementation, the prediction results can also be dynamically displayed in combination with the three-dimensional model of the bridge. For example, the prediction scores of different areas can be marked on the bridge model in a color gradient, with green indicating good health and red indicating potential risks. This visualization method can help managers quickly locate high-risk areas.

[0232] Extended technical content

[0233] In some embodiments, the health trend prediction module can dynamically adjust the hyperparameters of the prediction model according to the structural characteristics and operating environment of different bridges. For example, for bridges with larger spans, the number of hidden layer units of the model can be appropriately increased to enhance the ability to capture the evolution of complex health states.

[0234] In addition, to improve the stability of the prediction results, the module can combine the outputs of multiple prediction models. For example, the LSTM model can be combined with a simple moving average model to smooth out short-term fluctuations in health scores.

[0235] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Real-time monitoring and maintenance system for roads and bridges based on intelligent perception, characterized by: include: A data acquisition module, used to collect multi-dimensional data of the bridge operation status through the deployed multi-source sensors, wherein the multi-source sensors include vibration sensors, strain sensors, temperature and humidity sensors, and image acquisition equipment; The data preprocessing module is used to clean, synchronize and extract features of the collected multi-dimensional data to obtain dynamic features, environmental features and visual features; The data fusion module is used to extract the coupling relationship between multi-dimensional data based on the constructed multi-source data relationship graph and graph neural network model, and generate fused global features; The physical constraint optimization module is used to impose physical constraints based on the vibration modal characteristics and thermal expansion characteristics of the bridge on the fused global features to optimize the global features; A health status assessment module, used to generate a health status score based on the optimized global features; The health trend prediction module is used to predict the future health status of the bridge based on the time series model and generate maintenance recommendations.

2. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The data acquisition module further comprises: The multi-source sensor deployment submodule is used to deploy vibration sensors, strain sensors, and temperature and humidity sensors according to the key stress positions of the bridge structure, and obtain images of bridge cracks through image acquisition equipment; The data transmission submodule is used to transmit the collected data to the data preprocessing module via a wireless network or a wired network.

3. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The data preprocessing module further comprises: The data cleaning submodule is used to remove outliers and fill in missing data during sensor acquisition; The data synchronization submodule is used to synchronize the time of data with different sampling frequencies; The feature extraction submodule is used to extract power spectral density features from dynamic features, extract normalized statistical features from environmental features, and extract the width and length of cracks from visual features.

4. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The data fusion module comprises: A data relationship graph construction submodule is used to construct a relationship graph including multi-dimensional data nodes and data association edges based on the correlation of multi-source data, and the weight of the edge is calculated according to the information sharing amount between the node data; A graph neural network submodule, used for performing multi-level message transmission and aggregation processing on node features in the relationship graph to extract nonlinear coupling features between multi-source data; The global feature generation submodule is used to generate global features describing the health status of the bridge through feature aggregation method.

5. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The physical constraint optimization module includes: The vibration modal constraint submodule is used to calibrate the global features based on the vibration modal characteristics of the bridge structure dynamics; Thermal expansion constraint submodule, used to correct the relationship between temperature and humidity characteristics and strain characteristics based on the thermal expansion formula of bridge materials; The comprehensive optimization submodule is used to adjust the weight parameters of the global features according to the optimization objective function, where the optimization objective function includes the data-driven loss term, the modal constraint loss term and the thermal expansion constraint loss term.

6. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The vibration modal characteristics are determined by the vibration frequency of the bridge structure. The vibration frequency is positively correlated with the stiffness and mass of the bridge. During the optimization process, the vibration modal characteristics are ensured to be consistent with the actual dynamic characteristics of the bridge.

7. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The health status assessment module includes: A health index generation submodule, used to generate a bridge health status score according to the global characteristics, wherein the health status score ranges from 0 to 1; The indicator visualization submodule is used to graphically display the current health status of the bridge.

8. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The health trend prediction module includes: The time series model submodule is used to predict the future trend of bridge health status scores through the long short-term memory network model; The early warning generation submodule is used to generate health early warning information according to the downward trend of the health status score.

9. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The maintenance recommendations are generated based on historical data and future predicted values ​​of health status scores, with priority given to repairing critical bridge parts with health status scores below a preset threshold.

10. The road and bridge real-time monitoring and maintenance system based on intelligent perception according to claim 1 is characterized in that: The global feature is generated by a fusion representation including all sensor collected data, and the fusion representation can simultaneously reflect the dynamic behavior, environmental impact and visual damage status of the bridge.

Images