Bridge health monitoring method and system based on artificial intelligence

Abstract

The invention discloses a bridge health monitoring method and system based on artificial intelligence, and relates to the technical field of bridge monitoring, and the method comprises the steps: building a bridge finite element physical model based on a bridge geometric model; and performing modal analysis and strain energy analysis based on a finite element physical model, determining key nodes based on an analysis result to generate sensor candidate nodes, and optimizing the sensor layout by using a genetic algorithm. According to the method, on the basis of bridge design parameters, a finite element analysis model is used for building a physical model for a bridge, sensitive areas in the bridge are converted into key monitoring nodes according to modal analysis and strain energy analysis, and sensor layout is optimized according to the key monitoring nodes in combination with a genetic algorithm, so that high-quality data acquisition is achieved; the monitoring efficiency is maximized, the cost is reduced, and dynamic damage characteristic indexes in the bridge are comprehensively extracted by calculating parameters such as the frequency spectrum amplitude, the damping ratio and the strain energy change.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge monitoring, and in particular to a bridge health monitoring method and system based on artificial intelligence. Background Art

[0002] With the rapid development of modern transportation, bridges, as important transportation infrastructure, their safety and reliability have received extensive attention. Bridge health monitoring systems have emerged. By real-time monitoring data and combining data analysis and processing technologies, the health status of bridges is evaluated and warned. In terms of data processing and analysis, with the introduction of artificial intelligence and big data technologies, bridge health monitoring systems can more accurately identify potential problems, provide timely warnings and maintenance plans, and achieve intelligent monitoring and management.

[0003] However, there are still some deficiencies in existing bridge health monitoring technologies. First of all, traditional sensor layout methods often rely on experience or simple rules, lacking scientific optimization means, resulting in too many sensors or unreasonable distributions, increasing costs and the complexity of data processing. Secondly, for the detection of internal damage of bridges, existing methods mostly rely on single frequency domain analysis or strain energy calculation, and it is difficult to comprehensively and accurately extract damage characteristics. In addition, in terms of image data processing, traditional image analysis methods cannot effectively identify subtle defects on the bridge surface and have poor adaptability to images in complex environments. These deficiencies limit the accuracy and reliability of bridge health monitoring systems and are difficult to meet the high requirements of modern bridge safety management. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a bridge health monitoring method based on artificial intelligence, which solves the problems of unreasonable sensor layout, incomplete feature extraction, and low monitoring accuracy in existing bridge health monitoring.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a bridge health monitoring method based on artificial intelligence, which includes,

[0008] Constructing a physical model of the bridge through finite element analysis software based on the geometric model, material properties, and boundary conditions of the bridge;

[0009] Performing modal analysis and strain energy analysis based on the physical model of the bridge, determining key nodes based on the analysis results to generate sensor candidate nodes, and using a genetic algorithm to optimize the sensor layout;

[0010] Start the sensor to collect the original data of the bridge, deploy the drone inspection system to synchronously collect the image data of the bridge, and preprocess the original data and image data of the bridge;

[0011] Extract the internal damage characteristics of the bridge based on the preprocessed original data through frequency domain analysis and strain energy calculation;

[0012] Extract the appearance defect characteristics of the bridge surface based on the preprocessed image data through a convolutional neural network;

[0013] Build a bridge health monitoring model, and generate the health status monitoring results and warning information of the bridge based on the internal damage characteristics and appearance defect characteristics.

[0014] As a preferred solution of the bridge health monitoring method based on artificial intelligence according to the present invention, wherein: construct the physical model of the bridge through finite element analysis software based on the geometric model, material properties and boundary conditions of the bridge, including,

[0015] Precisely draw the three-dimensional geometric model of the bridge through CAD software according to the design drawings of the bridge;

[0016] Obtain the material properties from the material database according to the actual materials of the bridge;

[0017] Determine the boundary conditions of the bridge according to the actual support conditions of the bridge;

[0018] Based on the finite element analysis software, construct the finite element physical model of the bridge according to the three-dimensional geometric model, material property parameters and boundary conditions of the bridge, and select the appropriate mesh type and mesh size to divide the finite element mesh according to the structural characteristics of the bridge, and record the number of nodes and elements of the mesh.

[0019] As a preferred solution of the bridge health monitoring method based on artificial intelligence according to the present invention, wherein: perform modal analysis and strain energy analysis based on the physical model of the bridge, determine the key nodes based on the analysis results to generate sensor candidate nodes, and use the genetic algorithm to optimize the sensor layout, including,

[0020] Calculate the stiffness matrix K and mass matrix M of the bridge through the constructed physical model;

[0021] Construct the solution formula for the modal frequency and the corresponding modal vibration mode according to the stiffness matrix and mass matrix;

[0022] Use the QR decomposition algorithm to solve the formula, and output the modal frequency f of each order in the healthy state of the bridge n and the corresponding modal vibration mode φ(x);

[0023] Perform second-order partial derivative operations on the modal vibration modes, calculate the modal vibration mode curvature, and generate a distribution map of the modal vibration mode curvature;

[0024] Calculate the stress and strain based on the modal vibration mode and material properties of each node;

[0025] Calculate the strain energy density based on the node stress and strain, perform volume integration based on the strain energy density, calculate the total strain energy of the node, and generate a distribution map of the strain energy in the healthy state of the bridge;

[0026] Comprehensively evaluate the sensitivity of each node by weighting based on the distribution map of the modal vibration mode curvature and the distribution map of the strain energy;

[0027] Determine the weighting weights and evaluation thresholds through cross-validation, and select the nodes with comprehensive evaluation values greater than the evaluation threshold as key nodes;

[0028] Based on the key nodes and historical cases, select nodes from the finite element physical model as candidate nodes for acceleration and strain sensors;

[0029] Generate an initial population using the genetic algorithm based on the sensor candidate nodes. Each individual in the population represents a set of sensor layout schemes. Design a fitness function according to the coverage range and cost of the sensors, and perform iterations through selection, crossover, and mutation operations. When the fitness function value no longer changes significantly, output the individual with the highest fitness function value as the optimal sensor layout scheme.

[0030] As a preferred solution of the bridge health monitoring method based on artificial intelligence according to the present invention, wherein: start the sensors to collect the original data of the bridge, and deploy an unmanned aerial vehicle inspection system to synchronously collect the image data of the bridge, and preprocess the original data and image data of the bridge, including,

[0031] Collect the real-time acceleration and strain data of the bridge through a wireless synchronous sampling system, and at the same time deploy an unmanned aerial vehicle inspection system to synchronously collect the image data on the surface of the bridge;

[0032] Perform cleaning, denoising, and signal smoothing operations on the acceleration and strain data of the bridge;

[0033] Perform image enhancement and denoising operations on the image data on the surface of the bridge.

[0034] As a preferred solution of the bridge health monitoring method based on artificial intelligence according to the present invention, wherein: extract the internal damage characteristics of the bridge through frequency domain analysis and strain energy calculation based on the preprocessed original data, including,

[0035] Convert the preprocessed real-time acceleration data into frequency domain data through fast Fourier transform;

[0036] Calculate the spectral amplitude |A i (f)| according to the conversion result;

[0037] Use the peak detection algorithm to detect the peak points in the spectral amplitude, extract the modal frequencies, and output the set of modal frequencies F = {f1, f2, …, f n} of the i-th acceleration sensor;

[0038] Perform weighted average on the modal frequencies extracted from all sensors to obtain the modal frequency of the entire bridge;

[0039] Use the preprocessed real-time acceleration data to decompose it into k intrinsic mode functions through empirical mode decomposition;

[0040] Perform Hilbert transform on each intrinsic mode function IMF i (t) to extract the instantaneous frequency f i (t) and the instantaneous amplitude A i (t);

[0041] By squaring each intrinsic mode function IMF i (t) and integrating the squared result over time to obtain the energy E i ;

[0042] Calculate the standard deviation σ i and the average value f fi of the instantaneous frequency sequence f i extracted from each intrinsic mode function;

[0043] Use the least squares method to fit an exponential decay curve to the instantaneous amplitude A i (t) extracted from each intrinsic mode function, determine the best initial amplitude and decay slope, calculate the damping ratio according to the decay slope, and output the set of damping ratios C = {C1, C2, …, C k};

[0044] Calculate the stress at the node where the strain sensor is located using the preprocessed strain data;

[0045] Calculate the strain energy at the node based on the stress and strain through volume integration;

[0046] Compare the strain energy at the node where each strain sensor is located with the strain energy at the corresponding node in the strain energy distribution map of the bridge in the healthy state, calculate the change in strain energy at the node, and synthesize the changes in strain energy at all nodes where the strain sensors are located to form the strain energy change matrix ΔU;

[0047] Directly splice the modal frequency, the energy of the intrinsic mode function, the instantaneous frequency, the instantaneous amplitude, the standard deviation and average value of the instantaneous frequency, the damping ratio, and the strain energy change matrix to form the internal damage characteristics.

[0048] As a preferred solution of the bridge health monitoring method based on artificial intelligence according to the present invention, wherein: the preprocessed image data is subjected to feature extraction through a convolutional neural network to extract the appearance defect features on the bridge surface, including

[0049] Select a convolutional neural network with the ResNet-50 architecture to construct an external defect feature extraction model, including an input layer, a convolutional layer, a pooling layer, and a fully connected layer;

[0050] Initialize the model using pre-trained weights, train the model using the training set, optimize the model parameters, and output the model after training is completed;

[0051] Process the preprocessed image data according to the input requirements of the model and input it into the model. After convolutional operations and pooling operations, the fully connected layer performs object detection on the pooled feature map to output the bounding box and class label of each defect area in the image;

[0052] Statistically analyze the number, category, coordinates, and area of the defects according to the output results to generate appearance defect features.

[0053] As a preferred solution of the bridge health monitoring method based on artificial intelligence according to the present invention, wherein: the construction of the bridge health monitoring model generates a bridge health status evaluation result and a warning message based on internal damage features and appearance defect features, including

[0054] Concatenate the internal damage features and the appearance defect features to form a comprehensive feature F f ;

[0055] Select the radial basis function as the kernel function based on SVM to construct a bridge health monitoring model to evaluate the bridge health status;

[0056] Train the bridge health monitoring model using the labeled comprehensive feature training set, select the cross-entropy loss function to calculate the difference between the evaluation result and the true label, use the Adam optimizer for gradient descent optimization, fine-tune the parameters of the model, and stop the iteration and output the model when the loss of the model no longer decreases significantly during continuous iteration;

[0057] Input the new comprehensive feature F into the trained bridge health monitoring model f Generate a bridge health status evaluation result;

[0058] Generate corresponding warning messages according to the evaluation results and send them to the maintenance personnel.

[0059] In a second aspect, the present invention provides a bridge health monitoring system based on artificial intelligence, including

[0060] A bridge analysis module, based on bridge design parameters, constructs a physical model of the bridge through finite element analysis software, and determines the key areas and physical characteristics of the bridge according to modal analysis and strain energy analysis;

[0061] A layout optimization module, based on the results of modal analysis and strain energy analysis, optimizes the sensor layout through a genetic algorithm to maximize the monitoring coverage and sensor layout efficiency;

[0062] A data acquisition module activates the arranged sensors to collect the original internal data of the bridge, and synchronously collects the external image data of the bridge in combination with the UAV inspection system. After preprocessing, it is used as the input for subsequent analysis;

[0063] A feature extraction module uses frequency domain analysis and empirical mode decomposition to extract the internal damage features of the bridge, and uses a convolutional neural network to extract the external defect features of the bridge;

[0064] A monitoring and warning module constructs a bridge health monitoring model based on SVM, integrates the internal damage features and external defect features, inputs them into the model, and generates warning information about the health status monitoring results of the bridge.

[0065] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the bridge health monitoring method based on artificial intelligence as described in the first aspect of the present invention is implemented.

[0066] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the bridge health monitoring method based on artificial intelligence as described in the first aspect of the present invention is implemented.

[0067] The beneficial effects of the present invention are as follows: Based on bridge design parameters, a finite element analysis model is used to construct a physical model for the bridge. According to modal analysis and strain energy analysis, sensitive areas in the bridge are transformed into key monitoring nodes, and the sensor layout is optimized according to the key monitoring nodes in combination with a genetic algorithm to achieve high-quality data acquisition, maximize the monitoring efficiency, and reduce costs. By calculating parameters such as spectral amplitude, damping ratio, and strain energy change, internal dynamic damage feature indicators of the bridge are comprehensively extracted. A ResNet-50 is used to construct an external defect feature extraction model, which can automatically detect subtle external defects in images without manual intervention and provide parametric output for external damage analysis. By integrating internal damage and external defect data, the comprehensiveness of the overall state description of the bridge structure is improved. An SVM model is used to generate health status evaluation results and warning information, realizing multi-angle prediction and real-time alarm of the bridge health status. Description of the Drawings

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0069] Figure 1 It is a flowchart of the bridge health monitoring method based on artificial intelligence in Embodiment 1.

[0070] Figure 2 It is a structural diagram of the bridge health monitoring system based on artificial intelligence in Embodiment 1. Specific Embodiments

[0071] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the drawings in the specification.

[0072] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0073] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0074] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a bridge health monitoring method based on artificial intelligence, including the following steps:

[0075] S1. Construct a physical model of the bridge through finite element analysis software based on the geometric model, material properties, and boundary conditions of the bridge.

[0076] Specifically, accurately draw a three-dimensional geometric model of the bridge through CAD software according to the design drawings of the bridge;

[0077] Obtain the material properties from the material database according to the actual materials of the bridge, including elastic modulus, Poisson's ratio, and density;

[0078] Determine the boundary conditions of the bridge according to its actual support conditions, including fixed supports, sliding supports, elastic supports, and free boundaries, simulate the actual stress conditions of the bridge, and ensure that the finite element analysis is consistent with the actual bridge state;

[0079] Based on the finite element analysis software, construct the finite element physical model of the bridge according to the three-dimensional geometric model, material property parameters, and boundary conditions of the bridge, select an appropriate mesh type and mesh size according to the structural characteristics of the bridge to divide the finite element mesh, and record the number of nodes and elements of the mesh.

[0080] Through precise three-dimensional geometric modeling and input of real material parameters, eliminate the structural deviation of the traditional simplified model, ensure that the basic data for modal analysis and strain energy calculation are consistent with the mechanical characteristics of the real bridge, dynamically adjust the mesh density according to the geometric complexity of different regions of the bridge, achieve a balance between accuracy and efficiency, determine the boundary conditions of the bridge according to the actual support constraints, accurately restore the physical characteristics and structural state of the bridge, generate a bridge finite element physical model that conforms to the actual situation, and provide a benchmark model that highly matches the real bridge for subsequent structural dynamic analysis such as modal analysis and strain energy calculation.

[0081] S2. Conduct modal analysis and strain energy analysis based on the physical model of the bridge, determine the key nodes based on the analysis results to generate candidate sensor nodes, and use the genetic algorithm to optimize the sensor layout.

[0082] Specifically, calculate the stiffness matrix K and mass matrix M of the bridge through the constructed physical model, expressed as:

[0083] K = ∫ V B T DBdV;

[0084] M = ρ∫ V N T NdV;

[0085] Where, K represents the stiffness matrix, M represents the mass matrix, B represents the element strain-displacement matrix, which is automatically generated by the finite element software when constructing the physical model, D represents the material rigidity matrix, which is obtained from the expressions of elastic modulus and Poisson's ratio, V represents the element volume, ρ represents the material density, and N represents the shape function matrix, which is generated when the finite element software divides the mesh;

[0086] Construct the solution formula for the modal frequency and the corresponding modal vibration mode according to the stiffness matrix and mass matrix, expressed as:

[0087] (K - λM)φ = 0;

[0088]

[0089] Where, λ represents the eigenvalue, fn represents the nth - order modal frequency, and φ represents the modal shape;

[0090] Use the QR decomposition algorithm to solve the formula and output the modal frequency f of each order under the healthy state of the bridge n and the corresponding modal shape φ(x). Each modal frequency corresponds to different vibration modes of the overall bridge structure, and the modal shape describes the relative displacement of each node under different vibration modes;

[0091] Perform the second - order partial derivative operation on the modal shape to calculate the modal shape curvature and generate the modal shape curvature distribution diagram, which is expressed as:

[0092]

[0093] where MC(x) represents the modal shape curvature at node position x, and φ(t) represents the modal shape at node position x;

[0094] Calculate the stress and strain according to the modal shape and material properties of each node, which is expressed as:

[0095] σ(x) = D·B·u(x);

[0096] ∈(x) = B·u(x);

[0097] where σ(x) represents the stress at node position x, ∈(x) represents the strain at node position x, u(x) represents the node displacement vector, and in finite - element analysis, u(x) = φ(x);

[0098] Calculate the strain - energy density according to the node stress and strain, perform volume integration based on the strain - energy density, calculate the total strain energy of the node, and generate the strain - energy distribution diagram under the healthy state of the bridge, which is expressed as:

[0099]

[0100] where SE(x) represents the strain - energy density at node position x, V x represents the total volume of all elements directly connected to node x, and U x represents the total strain energy at node x;

[0101] Comprehensively evaluate the sensitivity of each node by a weighted method based on the modal - shape curvature distribution diagram and the strain - energy distribution diagram;

[0102] Determine the weighted weights and evaluation thresholds through cross - validation, and select the nodes with comprehensive evaluation values greater than the evaluation threshold as key nodes;

[0103] Based on the key nodes and historical cases, select nodes from the finite - element physical model as candidate nodes for acceleration and strain sensors;

[0104] Based on the sensor candidate nodes, use the genetic algorithm to generate an initial population. Each individual in the population represents a set of sensor layout schemes. Design a fitness function according to the coverage range and cost of the sensors, and perform iterations through selection, crossover, and mutation operations. When the value of the fitness function no longer changes significantly, output the individual with the highest fitness function value as the optimal sensor layout scheme.

[0105] Precisely calculate the stiffness matrix and mass matrix through the constructed bridge physical model, use the QR algorithm to solve for the modal frequencies and modal vibration modes, describe the dynamic response of the bridge structure, and calculate the modal curvature distribution and strain energy distribution based on the modal vibration modes, which respectively reflect the stiffness mutation and energy concentration of the structure. The calculated modal frequencies and strain energy distribution provide the benchmark modal frequencies and strain energy distribution in the healthy state for subsequent bridge health monitoring. Identify potential damages by monitoring the changes in the modal frequencies and strain energy distribution, and evaluate the severity of the damages.

[0106] Through the comprehensive screening of the extreme points of the modal curvature and the high-density areas of the strain energy, combined with the genetic algorithm for sensor layout optimization, ensure that the sensors cover all potential vulnerable positions, which can improve the efficiency of signal acquisition and the accuracy of damage identification, reduce the number of sensors, and lower the cost of the monitoring system.

[0107] S3. Start the sensors to collect the original bridge data, and deploy the UAV inspection system to synchronously collect the bridge image data, and preprocess the original bridge data and image data.

[0108] Specifically, collect the real-time acceleration and strain data of the bridge through the wireless synchronous sampling system, and at the same time deploy the UAV inspection system to synchronously collect the image data on the bridge surface;

[0109] Perform cleaning, denoising, and signal smoothing operations on the acceleration and strain data of the bridge;

[0110] Perform image enhancement and denoising operations on the image data on the bridge surface.

[0111] Through the wireless synchronous sampling system, ensure that the acquisition times of all acquired data are strictly aligned, providing a basis for subsequent data analysis. The UAV inspection can cover areas that are difficult for humans to reach, such as the bottom of the bridge pier and the inside of the box girder. Combining the wide-angle lens and zoom shooting, achieve a full-view detection of the bridge surface and hidden structures. Through the synchronous combination of the UAV inspection system and the sensor acquisition system, dynamic and visual information can jointly construct a bridge damage assessment model, avoiding the deficiency of a single data source. Preprocessing the data not only improves the quality of the original data but also reduces the computational burden during model training and analysis, thereby improving the overall system efficiency.

[0112] S4. Extract the internal damage characteristics of the bridge based on the preprocessed original data through frequency-domain analysis and strain energy calculation.

[0113] Specifically, the preprocessed real-time acceleration data is converted into frequency-domain data through fast Fourier transform, expressed as:

[0114]

[0115] where, A i (f) represents the complex spectrum of the time-series data of the i-th acceleration sensor at frequency f, a i (t) represents the time-series data of the i-th acceleration sensor, e -j2πft represents the kernel function, defined by Euler's formula;

[0116] Calculate the spectral amplitude |A i (f)| according to the conversion result, expressed as:

[0117]

[0118] where, Re[A i (f)] represents the real part of the complex spectrum, Im[A i (f)] represents the imaginary part of the complex spectrum, |A i (f)| represents the spectral amplitude;

[0119] Use the peak detection algorithm to detect the peak points in the spectral amplitude to extract the modal frequencies, and output the set of modal frequencies F = {f1, f2, …, f n} of the i-th acceleration sensor, expressed as:

[0120]

[0121] where, f n represents the n-th order modal frequency;

[0122] Perform weighted averaging on the modal frequencies extracted from all sensors to obtain the overall modal frequency of the bridge;

[0123] Use the preprocessed real-time acceleration data through empirical mode decomposition to decompose it into k intrinsic mode functions, expressed as:

[0124]

[0125] where, a(t) represents the preprocessed acceleration data, IMF i (t) represents the i-th intrinsic mode function decomposed, r(t) represents the residual term, and k represents the total number of intrinsic mode functions decomposed;

[0126] For each intrinsic mode function IMF i (t), perform the Hilbert transform to extract the instantaneous frequency f i (t) and the instantaneous amplitude A i (t), expressed as:

[0127]

[0128] where H[IMF i (t)] represents the result of the Hilbert transform, A i (t) represents the instantaneous amplitude, and f i (t) represents the instantaneous frequency;

[0129] By squaring each intrinsic mode function IMF i (t) and integrating the squared result over time, the energy E i ;

[0130] Calculate the standard deviation σ i and the average value f fi for the instantaneous frequency sequence f i extracted from each intrinsic mode function;

[0131] Use the least squares method to fit an exponential decay curve to the instantaneous amplitude A i (t) extracted from each intrinsic mode function, determine the optimal initial amplitude and decay slope, calculate the damping ratio based on the decay slope, and output the damping ratio set C = {C1, C2, …, C k}, expressed as:

[0132] A i (t) = A0e bt ;

[0133]

[0134] where A0 represents the initial amplitude, b represents the decay slope, f i represents the center frequency of the i-th intrinsic mode function, determined by the average value of f i (t), and C i represents the damping ratio of the i-th intrinsic mode function;

[0135] Calculate the stress at the node where the strain sensor is located using the preprocessed strain data, expressed as:

[0136] σ i = E·∈ i ;

[0137] where σ i represents the strain data of node i, and σ iLet $\sigma_i$ denote the stress of node $i$, and $E$ denote the elastic modulus of the bridge material, which is obtained from the material properties;

[0138] Based on the stress and strain, the strain energy at the node is calculated through volume integration, expressed as:

[0139]

[0140] where $U$ i denotes the strain energy of node $i$, $\sigma$ i denotes the stress of node $i$, $\varepsilon$ i denotes the strain data of node $i$, and $V$ i denotes the total volume of all elements directly connected to node $i$;

[0141] Compare the strain energy of each node where the strain sensor is located with the strain energy of the corresponding node in the strain energy distribution map under the healthy state of the bridge, calculate the change in strain energy of the node, and synthesize the changes in strain energy of all nodes where the strain sensors are located to form a strain energy change matrix $\Delta U$;

[0142] Directly splice the modal frequency, the energy of the intrinsic modal function, the instantaneous frequency, the instantaneous amplitude, the standard deviation and mean of the instantaneous frequency, the damping ratio, and the strain energy change matrix to form the internal damage characteristics.

[0143] Through fast Fourier transform for frequency-domain analysis to extract the modal frequency, which can be compared with the reference modal frequency, and the overall stiffness degradation of the bridge can be reflected according to the deviation degree. Through the empirical mode decomposition algorithm for time-domain analysis to separate the high-frequency components of IMF to identify weak damage, the detection sensitivity for initial damage is significantly improved. Through calculating the change in damping ratio and the fluctuation of instantaneous frequency to reveal local non-linear damage of the bridge, and by calculating the strain energy distribution change matrix, the energy redistribution caused by damage is quantified, the severity of damage is evaluated, and it can be associated with the finite element mesh node coordinates to accurately locate the damage position, realizing the leap from "overall abnormal alarm" to "damage type - location - degree quantitative diagnosis". Combine the frequency-domain characteristics, time-domain characteristics, and strain energy characteristics to form a multi-dimensional damage characteristics matrix, covering different aspects of bridge damage classification and quantification, and improving the accuracy of damage location.

[0144] S5. Based on the preprocessed image data, feature extraction is performed through a convolutional neural network to extract the appearance defect features on the bridge surface.

[0145] Specifically, select a convolutional neural network with the ResNet-50 architecture to construct an external defect feature extraction model, including an input layer, a convolutional layer, a pooling layer, and a fully connected layer;

[0146] Initialize the model using the pre-trained weights, train the model using the training set, optimize the model parameters, and output the model after training;

[0147] The preprocessed image data is processed according to the input requirements of the model and then input into the model. After convolutional operations and pooling operations, the fully connected layer performs object detection on the pooled feature map to output the bounding boxes and class labels of each defect region in the image.

[0148] Based on the output results, the number, category, coordinates, and area of the defects are statistically analyzed to generate appearance defect features.

[0149] Utilize the general feature extraction ability of the pre-trained ResNet-50 model to fine-tune the network parameters according to the characteristics of bridge appearance defects (cracks, spalling, corrosion). High-precision recognition can be achieved with only a small amount of labeled data, solving the pain point of traditional methods relying on a large amount of manual labeling. Through the feature pyramid structure of the convolutional neural network, millimeter-scale micro-cracks and centimeter-scale spalling regions can be synchronously recognized, avoiding missed detections caused by a single scale in traditional algorithms. Map the detected defect position coordinates to the finite element mesh to clarify the position coordinates to which the defects belong, provide engineering semantic information for subsequent maintenance decisions, replace the subjective and inefficient manual visual inspection, and achieve automatic and standardized evaluation of surface defects, providing surface defect analysis results for the overall bridge health status assessment.

[0150] S6. Construct a bridge health monitoring model to generate the bridge health status assessment results and early warning information based on the internal damage features and appearance defect features.

[0151] Specifically, the internal damage features and appearance defect features are concatenated to form the comprehensive feature F f ;

[0152] Based on SVM, select the radial basis function as the kernel function to construct a bridge health monitoring model to evaluate the bridge health status;

[0153] Use the labeled comprehensive feature training set to train the bridge health monitoring model. Select the cross-entropy loss function to calculate the difference between the evaluation result and the true label, use the Adam optimizer for gradient descent optimization, and fine-tune the parameters of the model. Stop the iteration and output the model when the loss of the model no longer significantly decreases during consecutive iterations.

[0154] Input the new comprehensive feature F into the trained bridge health monitoring model f Generate the bridge health status assessment results;

[0155] Generate corresponding early warning information according to the evaluation results and send it to the maintenance personnel.

[0156] The SVM radial basis kernel function can characterize the complex non - linear relationship between damage features and health states. Compared with linear models, the classification accuracy is significantly improved. Internal damage features and appearance defect features respectively reflect the structural mechanical state and apparent integrity of the bridge. Their fusion can cross - verify the authenticity of damage, reduce the false alarm rate, and enhance the comprehensiveness of the overall structural state description, realizing the comprehensive monitoring of the bridge health state. Through the trained bridge health monitoring model, the health state assessment results can be generated in real - time according to the fusion features, and warning information can be generated according to the assessment results, improving the real - time performance and reliability of bridge health monitoring, being able to detect potential problems in time and take corresponding maintenance measures to ensure the safe operation of the bridge.

[0157] This embodiment also provides an artificial - intelligence - based bridge health monitoring system, including: a bridge analysis module, which constructs a physical model of the bridge through finite - element analysis software based on bridge design parameters, and determines the key areas and physical characteristics of the bridge according to modal analysis and strain - energy analysis;

[0158] A layout optimization module, which optimizes the sensor layout through a genetic algorithm based on the results of modal analysis and strain - energy analysis to maximize the monitoring coverage and sensor layout efficiency;

[0159] A data acquisition module, which starts the arranged sensors to collect the original internal data of the bridge and synchronously collects the external image data of the bridge in combination with the UAV inspection system, and after pre - processing, uses it as the input for subsequent analysis;

[0160] A feature extraction module, which uses frequency - domain analysis and empirical mode decomposition to extract the internal damage features of the bridge, and uses a convolutional neural network to extract the appearance defect features of the bridge;

[0161] A monitoring and warning module, which constructs a bridge health monitoring model based on SVM, integrates internal damage features and appearance defect features, inputs them into the model, and generates warning information about the health state monitoring results of the bridge.

[0162] This embodiment also provides a computer device applicable to the artificial - intelligence - based bridge health monitoring method, including: a memory and a processor; the memory is used to store computer - executable instructions, and the processor is used to execute the computer - executable instructions to implement the artificial - intelligence - based bridge health monitoring method as proposed in the above embodiment.

[0163] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be implemented through WIFI, carrier networks, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads provided on the outer shell of the computer device. It can also be an external keyboard, touchpad, or mouse, etc.

[0164] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for bridge health monitoring based on artificial intelligence proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, magnetic disks, or optical discs.

[0165] In summary, the present invention provides reliable data support for sensor layout optimization and damage identification by constructing a physical model; determines key nodes through analysis, optimizes the sensor layout using a genetic algorithm, improves data quality, and reduces monitoring costs; covers different perspectives of internal damage classification and quantification of bridges by combining frequency domain, time domain, and strain energy characteristics; automatically extracts features in images through a convolutional neural network, identifies appearance defects such as cracks and spalling on the bridge surface, and accurately locates the position and scope of the defects; the health monitoring model constructed by integrating internal damage characteristics and appearance defect characteristics can comprehensively evaluate the health status of the bridge.

[0166] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based bridge health monitoring method, characterized in that: Including Construct a physical model of the bridge through finite element analysis software based on the geometric model, material properties, and boundary conditions of the bridge Conduct modal analysis and strain energy analysis based on the physical model of the bridge. Determine key nodes based on the analysis results to generate candidate sensor nodes, and use the genetic algorithm to optimize the sensor layout Start the sensors to collect the original data of the bridge, and deploy an unmanned aerial vehicle inspection system to synchronously collect the image data of the bridge, and preprocess the original data and image data of the bridge Extract the internal damage characteristics of the bridge based on the preprocessed original data through frequency domain analysis and strain energy calculation Extract the appearance defect characteristics of the bridge surface based on the preprocessed image data through convolutional neural network Construct a bridge health monitoring model, and generate the health status monitoring results and warning information of the bridge based on the internal damage characteristics and appearance defect characteristics 2. The method for bridge health monitoring based on artificial intelligence according to claim 1, characterized in that: The step of constructing a physical model of the bridge through finite element analysis software based on the geometric model, material properties, and boundary conditions of the bridge includes Accurately draw the three-dimensional geometric model of the bridge according to the design drawings of the bridge through CAD software Obtain the material properties from the material database according to the actual materials of the bridge Determine the boundary conditions of the bridge according to the actual support conditions of the bridge Based on the finite element analysis software, construct the finite element physical model of the bridge according to the three-dimensional geometric model, material performance parameters, and boundary conditions of the bridge, and select the appropriate mesh type and mesh size to divide the finite element mesh according to the structural characteristics of the bridge, and record the number of nodes and elements of the mesh 3. The method for bridge health monitoring based on artificial intelligence according to claim 2, wherein: The step of conducting modal analysis and strain energy analysis based on the physical model of the bridge, determining key nodes based on the analysis results to generate candidate sensor nodes, and using the genetic algorithm to optimize the sensor layout includes Calculate the stiffness matrix K and mass matrix M of the bridge through the constructed physical model Construct the solution formula for modal frequency and the corresponding modal vibration mode according to the stiffness matrix and mass matrix Solve the formula using the QR decomposition algorithm and output the modal frequencies \(f\) of each order in the healthy state of the bridge n and the corresponding modal shapes \(\varphi(x)\); Perform second-order partial derivative operations on the modal vibration mode, calculate the modal vibration mode curvature, and generate the modal vibration mode curvature distribution map Calculate the stress and strain according to the modal vibration mode and material properties of each node Calculate the strain energy density according to the node stress and strain, perform volume integration based on the strain energy density, calculate the total strain energy of the node, and generate the strain energy distribution map under the healthy state of the bridge Comprehensively evaluate the sensitivity of each node through a weighted method based on the modal vibration mode curvature distribution map and the strain energy distribution map Determine the weighted weight and evaluation threshold through cross-validation, and select the nodes with a comprehensive evaluation value greater than the evaluation threshold as key nodes Based on the key nodes and historical cases, select nodes from the finite element physical model as candidate nodes for acceleration and strain sensors Generate an initial population based on the candidate sensor nodes using the genetic algorithm. Each individual in the population represents a set of sensor layout schemes. Design a fitness function according to the coverage range and cost of the sensors, and perform iteration through selection, crossover, and mutation operations. When the fitness function value no longer changes significantly, output the individual with the highest fitness function value as the optimal sensor layout scheme 4. The method for bridge health monitoring based on artificial intelligence according to claim 3, wherein: The startup sensor collects the original data of the bridge, and deploys an unmanned aerial vehicle (UAV) inspection system to synchronously collect the image data of the bridge. The original data and image data of the bridge are preprocessed, including collecting the real-time acceleration and strain data of the bridge through a wireless synchronous sampling system, and at the same time deploying a UAV inspection system to synchronously collect the image data on the surface of the bridge; performing cleaning, denoising, and signal smoothing operations on the acceleration and strain data of the bridge; performing image enhancement and denoising operations on the image data on the surface of the bridge.

5. The method for bridge health monitoring based on artificial intelligence according to claim 4, wherein: Based on the preprocessed original data, internal damage characteristics of the bridge are extracted through frequency domain analysis and strain energy calculation, including converting the preprocessed real-time acceleration data into frequency domain data through fast Fourier transform; Calculate the spectral amplitude |A i (f)| according to the conversion result; Use the peak detection algorithm to detect the peak points in the spectral amplitude, extract the modal frequencies, and output the set of modal frequencies F = {f1, f2, …, f n} for the i-th acceleration sensor; performing weighted averaging on the modal frequencies extracted by all sensors to obtain the overall modal frequency of the bridge; decomposing the preprocessed real-time acceleration data into k intrinsic mode functions through empirical mode decomposition; For each intrinsic mode function IMF i (t), perform the Hilbert transform to extract the instantaneous frequency f i (t) and the instantaneous amplitude A i (t); By squaring each Intrinsic Mode Function IMF i (t) and integrating the squared result over time to obtain the energy E i ; For each instantaneous frequency sequence f i (t) extracted from the intrinsic mode functions, calculate the standard deviation and the mean value f i ; For the instantaneous amplitude A i (t) extracted from each intrinsic mode function, the least squares method is used to fit an exponential decay curve to determine the optimal initial amplitude and decay slope. The damping ratio is calculated based on the decay slope, and the damping ratio set C = {C1, C2, …, C k}; calculating the stress of the node where the strain sensor is located using the preprocessed strain data; calculating the strain energy at the node based on stress and strain through volume integration; comparing the strain energy of each node where the strain sensor is located with the strain energy of the corresponding node in the strain energy distribution map under the healthy state of the bridge, calculating the change in strain energy of the node, and synthesizing the changes in strain energy of all nodes where the strain sensors are located to form a strain energy change matrix ΔU; directly splicing the modal frequency, the energy of the intrinsic mode function, the instantaneous frequency, the instantaneous amplitude, the standard deviation and average value of the instantaneous frequency, the damping ratio, and the strain energy change matrix to form the internal damage characteristics.

6. The method for bridge health monitoring based on artificial intelligence according to claim 5, characterized in that: Based on the preprocessed image data, feature extraction is performed through a convolutional neural network to extract the appearance defect characteristics on the surface of the bridge, including selecting a convolutional neural network with the ResNet-50 architecture to construct an external defect feature extraction model, including an input layer, a convolutional layer, a pooling layer, and a fully connected layer; initializing the model using the pre-trained weights, training the model using the training set, optimizing the model parameters, and outputting the model after training is completed; processing the preprocessed image data according to the input requirements of the model and then inputting it into the model. After convolutional operations and pooling operations, the fully connected layer performs object detection on the pooled feature map to output the bounding box and class label of each defect area in the image; counting the number, category, coordinates, and area of the defects according to the output results to generate the appearance defect characteristics.

7. The method for bridge health monitoring based on artificial intelligence according to claim 6, characterized in that: The bridge health monitoring model is constructed, and the health status evaluation result and early warning information of the bridge are generated based on the internal damage characteristics and appearance defect characteristics, including The internal damage features and appearance defect features are spliced to form a comprehensive feature F f ; selecting the radial basis function as the kernel function based on the support vector machine (SVM) to construct a bridge health monitoring model to evaluate the health status of the bridge; training the bridge health monitoring model using the labeled comprehensive feature training set, selecting the cross-entropy loss function to calculate the difference between the evaluation result and the true label, using the Adam optimizer for gradient descent optimization, and fine-tuning the parameters of the model. Stop iterating and output the model when the loss of the model no longer decreases significantly during consecutive iterations; Input the new comprehensive feature F into the trained bridge health monitoring model f Generate the evaluation result of the bridge health status; generating corresponding early warning information according to the evaluation result and sending it to the maintenance personnel.

8. An artificial intelligence-based bridge health monitoring system, based on the artificial intelligence-based bridge health monitoring method according to any one of claims 1 to 7, characterized in that: including The bridge analysis module constructs a physical model of the bridge based on the bridge design parameters through finite element analysis software, and determines the key areas and physical properties of the bridge according to modal analysis and strain energy analysis; The layout optimization module optimizes the sensor layout through the genetic algorithm based on the results of modal analysis and strain energy analysis to maximize the monitoring coverage and sensor layout efficiency; The data acquisition module starts the arranged sensors to collect the original internal data of the bridge, and synchronously collects the external image data of the bridge in combination with the UAV inspection system, and preprocesses it as the input for subsequent analysis; The feature extraction module uses frequency domain analysis and empirical mode decomposition to extract the internal damage features of the bridge, and uses convolutional neural network to extract the appearance defect features of the bridge; The monitoring and warning module constructs a bridge health monitoring model based on SVM, integrates the internal damage features and appearance defect features, inputs them into the model, and generates warning information on the health status monitoring results of the bridge.

9. A computer device, including a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based bridge health monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based bridge health monitoring method according to any one of claims 1 to 7.

Images