A method and device for cable-stayed bridge damage identification based on massive data

Abstract

The present invention relates to a method and device for damage identification of cable-stayed bridges based on massive data. The method comprises: constructing a benchmark finite element model of a cable-stayed bridge, collecting structural dynamic response data under a variety of damage conditions; proposing an improved fully adaptive noise ensemble empirical mode decomposition algorithm (ICEEMDAN) to decompose the data, dynamically adjusting the noise level through a standard deviation feedback mechanism to obtain IMF components; innovatively utilizing the Hurst index to screen target IMF components that satisfy H>0.5 and reconstruction error MSE<0.0003; constructing and training a one-dimensional convolutional neural network model, inputting the target IMF components, and outputting identification results of the damage location and degree. The model comprises a leading convolutional layer group, a mid-level feature layer group, a deep abstraction layer group, and a classification output layer. In addition, a damage identification device, an electronic device, a computer-readable storage medium, and a computer program product are provided to implement the method. The present invention can effectively identify damage to cable-stayed bridges and provide technical support for bridge health monitoring.

Description

Technical Field

[0001] The present invention relates to the field of monitoring technology, and in particular to a method and device for identifying damage of a cable-stayed bridge based on massive data. Background Art

[0002] With the widespread adoption of structural health monitoring systems for bridge engineering, a vast amount of long-term monitoring data has been accumulated for cable-stayed bridges and other long-span bridges. This data contains rich structural state information, providing an important foundation for damage identification. However, cable-stayed bridge monitoring data is subject to strong noise interference, non-stationarity, and high dimensionality, posing significant challenges to traditional damage identification methods.

[0003] Machine learning-based damage identification methods are currently gaining research attention, but they still face significant drawbacks in practical applications. These include noise sensitivity: When traditional one-dimensional convolutional neural networks directly process noisy signals, the model is easily affected by ambient noise, resulting in inadequate damage feature extraction and a sharp decline in identification accuracy with increasing noise intensity. Existing signal processing methods rely on fixed parameter settings, which can easily generate modal aliasing and false components during the decomposition process, leading to the loss of key feature information. Most studies directly use all intrinsic mode function components as input, without considering the information validity of different components. Redundant features increase model complexity and reduce generalization ability. Existing methods are often limited to single damage location or quantification, lacking the ability to collaboratively identify complex damage at multiple locations and degrees, and resulting in quantitative errors often exceeding 5%, making them difficult to meet engineering accuracy requirements. Traditional noise suppression algorithms use static thresholds and are unable to dynamically adjust the noise level based on signal characteristics, resulting in significantly increased signal reconstruction errors in high-noise environments. Although some studies have attempted to combine signal decomposition with deep learning, their noise adjustment mechanisms are rigid and feature quality assessment metrics are not introduced, resulting in insufficient robustness and computational efficiency for the massive data sets of cable-stayed bridges. Summary of the Invention

[0004] The present application provides a method and device for cable-stayed bridge damage identification based on massive data to solve the problems of insufficient noise sensitivity, limited signal decomposition, insufficient feature screening, and lack of dynamic adaptability in the prior art.

[0005] The first aspect of the present application provides a damage identification method for a cable-stayed bridge based on massive data, comprising the following steps: S1: constructing a benchmark finite element model of a cable-stayed bridge; S2: collecting structural dynamic response data under a variety of damage conditions according to the benchmark finite element model; S3: proposing an improved fully adaptive noise ensemble empirical mode decomposition algorithm ICEEMDAN to decompose the structural dynamic response data to obtain IMF components, wherein the decomposition algorithm ICEEMDAN is embodied in dynamically adjusting the noise level through a standard deviation feedback mechanism; S4: innovatively performing Hurst index screening on the IMF components, and selecting target IMF components that meet the Hurst index H>0.5H>0.5 and the reconstruction error MSE<0.0003MSE<0.0003; S5: constructing and training a one-dimensional convolutional neural network model, inputting the target IMF component into the trained one-dimensional convolutional neural network model, and outputting the identification results of the damage location and damage degree, wherein the one-dimensional convolutional neural network model includes a leading convolution layer group, a mid-level feature layer group, a deep abstraction layer group, and a classification output layer.

[0006] Preferably, the decomposition algorithm ICEEMDAN comprises: S3.1.1 adding the initial noise level to the original data to obtain noisy data, and the formula is as follows:

[0007]

[0008] Where: x(t) is the original signal; is the original noise level; N0(t) is the Gaussian white noise sequence;

[0009] S3.1.2: Use EMD to decompose the noisy data and obtain the first residual signal. The formula is as follows:

[0010] R1= <M(x i )> (2)

[0011] Where: <·> is the overall mean; M(·) is the local mean of the input signal;

[0012] S3.1.3: Calculate the error between the first and the original data using the following formula:

[0013]

[0014] in: is the standard deviation of the IMF1 component; σ x is the standard deviation of the original data; E1 is the error;

[0015] S3.1.4: Based on the error E1, the noise level formula is updated using a nonlinear strategy as follows:

[0016]

[0017] in: is the noise level after the first update; β is the control coefficient of noise adjustment;

[0018] S3.1.5: The formula for the residual signal obtained by removing the first IMF component from the original signal is as follows:

[0019] x1(t)=x(t)-IMF1(t) (7)

[0020] S3.1.6: Add the updated noise to the remaining signal, calculated as follows:

[0021]

[0022] Where: N1(t) is a new set of Gaussian white noise;

[0023] S3.1.7: Perform EMD decomposition on the new noisy signal to extract the second-order intrinsic mode function;

[0024] S3.1.8: Repeat the above calculations to add noise to the remaining signal, perform EMD decomposition, extract subsequent IMF components, and calculate the error and adjust the noise level after each iteration:

[0025]

[0026] in: is the noise level after the (n+1)th update; E n is the error at the nth time; β is the control coefficient of noise adjustment.

[0027] Preferably, the Hurst exponent calculation includes: S4.1.1: preprocessing a given time series x(t) to obtain a processed sequence, the formula is as follows:

[0028]

[0029] in: is the mean of the time series;

[0030] S4.1.2: Use the least squares method to obtain the local trend term of each series using the nonlinear fitting method. The formula is as follows:

[0031]

[0032] Among them: a n are the coefficients of the fitted polynomial;

[0033] S4.1.3: Use the local trend term y m(k) Calculate the second-order fluctuation function of the time series, which is as follows:

[0034]

[0035] F q (n)∝n α (13)

[0036] Where: α is the Hurst exponent;

[0037] S4.1.4: Calculate the slope of the fitted curve to obtain the Hurst exponent, as follows:

[0038]

[0039] S4.1.5: The IMF components can be effectively selected by calculating the Hurst exponent of each IMF component. The MSE calculation formula is:

[0040]

[0041] Where: x(t) is the original signal; To reconstruct the signal.

[0042] Preferably, the one-dimensional convolutional neural network model includes: S5.1.1: the input layer receives a 6000-dimensional time series signal; S5.1.2: the first convolution layer uses 16 convolution kernels of size 8 and a step size of 2; S5.1.3: the second pooling layer adopts maximum pooling with a pooling size of 2; S5.1.4: the deep convolution layer uses 512 convolution kernels of size 2; S5.1.5: the global average pooling layer is connected to a 512-node fully connected layer; S5.1.6: the Dropout layer sets a dropout rate of 0.3; S5.1.7: the output layer adopts the Softmax activation function.

[0043] Preferably, an improved fully adaptive noise ensemble empirical mode decomposition algorithm ICEEMDAN is proposed, which further includes the following steps before decomposing the structural dynamic response data to obtain IMF components: S3.2.1: adding Gaussian white noise to the original stress response signal x(t) to construct a noisy signal:

[0044] x_noise(t)=x(t)+ησ_xξ(t)(16)

[0045] Where σ_x is the signal standard deviation, η∈{5%, 10%, 15%} is the noise intensity coefficient, and ξ(t) is a standard normal distribution random number.

[0046] Preferably, the multiple damage conditions adopt a multi-label classification strategy, wherein the multi-label classification strategy includes lossless state encoding, single damage condition and multiple damage conditions, including: S5.2.1: the lossless state is encoded as category 0, corresponding to the lossless condition; S5.2.2: the single damage condition is graded and encoded according to the damage factor: the 0.05 / 0.10 / 0.15 damage factors of measuring point 3 correspond to categories 1-3 respectively, and the same-level damage factors of measuring point 8 correspond to categories 4-6; S5.2.3: Multiple damage condition encoding rules: when measuring points 3 and 11 are damaged in combination, the 0.05 / 0.10 / 0.15 damage factors correspond to categories 7-9 respectively; S5.2.4: the one-dimensional convolutional neural network output layer is set with 10 neurons, corresponding to the probability distribution of damage conditions of categories 0-9 respectively.

[0047] The second aspect of the present application provides a massive data damage identification device for a cable-stayed bridge, including: a modeling module, which constructs a benchmark finite element model of the cable-stayed bridge; a data acquisition module, which collects structural dynamic response data under various damage conditions based on the benchmark finite element model; a signal decomposition module, which proposes an improved fully adaptive noise set empirical mode decomposition algorithm ICEEMDAN to decompose the structural dynamic response data to obtain IMF components, wherein the decomposition algorithm ICEEMDAN is embodied in dynamically adjusting the noise level through a standard deviation feedback mechanism; a feature screening module, which innovatively performs Hurst index screening on the IMF components, and selects target IMF components that meet the Hurst index H>0.5H>0.5 and the reconstruction error MSE<0.0003MSE<0.0003; a network construction module, which constructs and trains a one-dimensional convolutional neural network model, inputs the target IMF component into the trained one-dimensional convolutional neural network model, and outputs the identification results of the damage location and damage degree, wherein the one-dimensional convolutional neural network model includes a leading convolution layer group, a mid-level feature layer group, a deep abstraction layer group and a classification output layer.

[0048] The third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and runnable on the processor. The processor executes the program to implement a damage identification method for a cable-stayed bridge based on massive data as in the above embodiment.

[0049] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a method and device for damage identification of a cable-stayed bridge based on massive data as in the above-mentioned embodiment.

[0050] The fifth embodiment of the present application provides a computer program product, including a computer program or instructions, for implementing a damage identification method for a cable-stayed bridge based on massive data as described in the above embodiment.

[0051] Therefore, this application has the following beneficial effects:

[0052] The embodiment of the present application dynamically adjusts the noise suppression level through the improved ICEEMDAN algorithm, combines the Hurst exponent to screen key feature components, and constructs a deeply optimized 1D-CNN model, achieving high-precision identification of cable-stayed bridge damage and a breakthrough in noise resistance. In a low-noise environment, the model can accurately extract damage-sensitive features; under 15% high-intensity noise interference, it can still ensure the quality of signal reconstruction through a dynamic compensation mechanism. At the same time, based on the multi-label classification strategy and the fusion of finite element simulation data, the system supports the collaborative identification of composite damage across measurement points and realizes quantitative grading of the degree of damage. As a result, the present application solves the problems of high noise sensitivity, high feature redundancy, insufficient multi-damage identification capabilities, and low computational efficiency in the prior art, and improves the accuracy of cable-stayed bridge damage identification to more than 99%, with the quantitative error controlled within 3%, realizing the optimization of the entire process from data acquisition, feature extraction to damage diagnosis, and providing a highly reliable intelligent solution for bridge health monitoring.

[0053] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0055] Figure 1 A flowchart of a method for damage identification of a cable-stayed bridge based on massive data according to an embodiment of the present application;

[0056] Figure 2 This is a flowchart of a method for damage identification of cable-stayed bridges based on massive data according to one embodiment of the present application;

[0057] Figure 3 A schematic diagram of a finite element model of a cable-stayed bridge provided according to an embodiment of the present application;

[0058] Figure 4 A schematic diagram of the damaged area, measuring points, and stress sensor arrangement of the main beam of a cable-stayed bridge provided according to an embodiment of the present application;

[0059] Figure 5 Schematic diagram of stress signal response before and after damage of measuring point 3 provided in an embodiment of the present application;

[0060] Figure 6 A schematic diagram of the position of a damaged area on a unit according to an embodiment of the present application;

[0061] Figure 7A schematic diagram of reconstruction error provided according to an embodiment of the present application;

[0062] Figure 8 The loss value and accuracy curve of the training set and test set provided according to the embodiment of the present application

[0063] Schematic diagram of;

[0064] Figure 9 A schematic diagram of stress response data of a real bridge provided according to an embodiment of the present application;

[0065] Figure 10 This is a schematic diagram of the structure of a cable-stayed bridge massive data damage identification device provided according to an embodiment of the present application;

[0066] Figure 11 A schematic diagram of the structure of an electronic device provided according to an embodiment of the present application;

[0067] Figure 12 This is an introduction diagram of a method and device for identifying damage to a cable-stayed bridge based on massive data according to an embodiment of the present application. DETAILED DESCRIPTION

[0068] The following describes in detail embodiments of the present application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0069] The following describes a method and device for cable-stayed bridge damage identification based on massive data in accordance with an embodiment of the present application with reference to the accompanying drawings. In response to the problems of strong noise sensitivity, high feature redundancy, and insufficient accuracy in multi-damage identification raised in the above-mentioned background technology, the present application provides a method for cable-stayed bridge damage identification based on massive data. In this method, the noise suppression level is dynamically adjusted by an improved ICEEMDAN algorithm, key feature components are screened in combination with the Hurst exponent, and a deeply optimized 1D-CNN model is constructed, thereby achieving high-precision identification of cable-stayed bridge damage and a breakthrough in noise resistance. The noise compensation factor γ (γ=0.3) is optimized in real time through a standard deviation feedback mechanism, and a damage location accuracy of 98.59% can be maintained under 15% high-intensity noise interference (traditional methods ≤58.44%), and the reconstruction error is reduced to 0.0002 (compared to VMD's 0.0015). At the same time, based on a multi-label classification strategy and finite element simulation data fusion, the system supports the collaborative identification of composite damage across measurement points, with a single damage location error of ≤0.5m (measurement point spacing 2m) and a multi-damage quantification error of ≤2.94% (traditional methods ≥8.2%). This method significantly improves the robustness and computational efficiency of damage identification in complex noise environments through the collaborative optimization of dynamic noise suppression, feature validity evaluation, and lightweight network design. It also addresses the problems of strong noise sensitivity, high feature redundancy, insufficient multi-damage identification capabilities, and high computing resource consumption in traditional methods. It provides a highly reliable, low-latency intelligent solution for cable-stayed bridge health monitoring, achieving a breakthrough in the entire process from data acquisition, feature extraction, to damage diagnosis.

[0070] Specifically, Figure 1 A flowchart of a damage identification method for a cable-stayed bridge based on massive data provided in an embodiment of the present application.

[0071] like Figure 1 As shown in FIG, the damage identification method of the cable-stayed bridge based on massive data includes the following steps:

[0072] In step S1, a reference finite element model of a cable-stayed bridge is constructed.

[0073] The benchmark finite element model was established using ABAQUS software, the main beam used C3D8R solid elements (grid size 0.5m), the cables used T3D2 truss elements, and consolidation constraints were set at the bottom of the tower.

[0074] It can be understood that a refined model containing 500,000 units (main beam C3D8R / cable T3D2) was established through ABAQUS, 10 damage conditions were set (damage factor 0.05-0.15) and the model accuracy was verified (modal frequency error <1.2%), providing a highly reliable data foundation for subsequent simulations.

[0075] It should be noted that the model was calibrated using both measured modal parameters (frequency error <1.2%, mode MAC value >0.95) and static load testing (displacement deviation <2 mm) to ensure engineering applicability of the damage simulation. The main beam element size was optimized to 0.5 m to capture localized damage effects, cable prestressing was applied with an accuracy of 0.1 kN, and the tower constraints closely replicated the actual bridge bearing configuration.

[0076] In step S2, structural dynamic response data under various damage conditions are collected based on the baseline finite element model.

[0077] The dynamic response data include the stress time history signals of 15 measuring points on the main beam (sampling rate 200 Hz) and the cable force vibration spectrum data (frequency resolution 0.1 Hz).

[0078] It can be understood that the stress time history (sampling at 200 Hz) and cable force spectrum data covering 15 measuring points (spacing 2 m) on the main beam were used to generate 4,752,000 sets of samples containing 5%-15% noise, simulating the data characteristics under actual environmental interference.

[0079] It should be noted that the data acquisition module integrates a temperature compensation algorithm to eliminate the effects of ambient temperature drift on the strain signal (temperature drift error <0.5με / °C). The white noise excitation bandwidth covers the 0.1-50Hz main bridge vibration frequency range, and damage conditions include typical damage types such as mid-span cracks in main beams and loose cable end anchorages, ensuring the completeness and project representativeness of the dataset.

[0080] In step S3, an improved fully adaptive noise ensemble empirical mode decomposition algorithm ICEEMDAN is proposed to decompose the structural dynamic response data to obtain IMF components.

[0081] Among them, the decomposition algorithm ICEEMDAN introduces complementary white noise to reduce modal aliasing and adds a dynamic noise adjustment mechanism to significantly improve signal decomposition accuracy and reconstruction effect.

[0082] It can be understood that through dynamic noise adjustment (β iteration formula) and standard deviation feedback, modal aliasing is suppressed and 8-10 order IMF components are decomposed, which reduces 63% of false components compared with traditional methods and improves the signal decomposition purity.

[0083] It should be noted that the initial noise intensity β_0 is set to 20% of the standard deviation of the input signal (β_0 = 0.2σ_x). During the decomposition process, the amount of noise injection is dynamically adjusted using the residual energy ratio E_k. Compared with the traditional CEEMDAN, the correlation coefficient of the IMF components in a 15% noise environment is improved to 0.92 (compared to 0.68 for the traditional method), and the proportion of false modal energy is reduced to 3.7% (compared to 9.8% for the traditional method).

[0084] In the embodiment of the present application, the decomposition algorithm ICEEMDA includes: S3.1.1 adding the initial noise level to the original data to obtain noisy data, and the formula is as follows:

[0085]

[0086] Where: x(t) is the original signal; is the original noise level; N0(t) is the Gaussian white noise sequence;

[0087] S3.1.2: Use EMD to decompose the noisy data and obtain the first residual signal. The formula is as follows:

[0088] R1= <M(x i )> (2)

[0089] Where: <·> is the overall mean; M(·) is the local mean of the input signal;

[0090] S3.1.3: Calculate the error between the first and the original data using the following formula:

[0091]

[0092]

[0093] in: is the standard deviation of the IMF1 component; σ x is the standard deviation of the original data; E1 is the error;

[0094] S3.1.4: Based on the error E1, the noise level formula is updated using a nonlinear strategy as follows:

[0095]

[0096] in: is the noise level after the first update; β is the control coefficient of noise adjustment;

[0097] S3.1.5: The formula for the residual signal obtained by removing the first IMF component from the original signal is as follows:

[0098] x1(t)=x(t)-IMF1(t) (7)

[0099] S3.1.6: Add the updated noise to the remaining signal, calculated as follows:

[0100]

[0101] Where: N1(t) is a new set of Gaussian white noise;

[0102] S3.1.7: Perform EMD decomposition on the new noisy signal to extract the second-order intrinsic mode function;

[0103] S3.1.8: Repeat the above calculations to add noise to the remaining signal, perform EMD decomposition, extract subsequent IMF components, and calculate the error and adjust the noise level after each iteration:

[0104]

[0105] in: is the noise level after the (n+1)th update; E n is the error at the nth time; β is the control coefficient of noise adjustment.

[0106] As can be understood, by controlling noise intensity through decomposition error feedback, this method actively injects high-intensity noise in the initial stage to separate high-frequency interference components (such as ambient vibration noise). As the number of iterations increases, the noise intensity is exponentially reduced to focus on mid- and low-frequency damage features. This strategy reduces the modal aliasing of the IMF component to 0.12 (compared to 0.54 with traditional methods) and improves the correlation coefficient to 0.92, ensuring the reliability of subsequent feature screening.

[0107] In step S4, the Hurst index of the IMF component innovation is screened, and the target IMF component that meets the Hurst index H>0.5H>0.5 and the reconstruction error MSE<0.0003MSE<0.0003 is selected.

[0108] The Hurst index is calculated using the detrended fluctuation analysis (DFA) method. After eliminating the trend term through cubic polynomial fitting, a logF(n)-logn linear regression model is established, and 1 / 2 of the slope value is H.

[0109] It can be understood that the Hurst index (H>0.5) screens and retains the damage-sensitive components, and combined with the MSE<0.0003 constraint, 90% of redundant modes are eliminated, the feature dimensions are compressed to 3-4, 95% of the effective information is retained, and the reconstruction error is <0.03%.

[0110] It should be noted that the Hurst exponent is calculated using detrended fluctuation analysis (DFA) with a window length covering 10-1000 sampling points, and trend terms are eliminated through cubic polynomial fitting. The Damage Sensitivity Index (DSI) of the filtered target IMF component reaches 8.7, a 4.3-fold improvement over the unfiltered component, effectively enhancing feature separability.

[0111] In the embodiment of the present application, the Hurst exponent calculation includes: S4.1.1: preprocessing a given time series x(t) to obtain a processed sequence, and the formula is as follows:

[0112]

[0113] Where: x is the mean of the time series;

[0114] S4.1.2: Use the least squares method to obtain the local trend term of each series using the nonlinear fitting method. The formula is as follows:

[0115]

[0116] Among them: a n are the coefficients of the fitted polynomial;

[0117] S4.1.3: Use the local trend term y m (k) Calculate the second-order fluctuation function of the time series, which is as follows:

[0118]

[0119] F q (n)∝n α (13)

[0120] Where: α is the Hurst exponent;

[0121] S4.1.4: Calculate the slope of the fitted curve to obtain the Hurst exponent, as follows:

[0122]

[0123] S4.1.5: The IMF components can be effectively selected by calculating the Hurst exponent of each IMF component. The MSE calculation formula is:

[0124]

[0125] Where: x(t) is the original signal; To reconstruct the signal.

[0126] Among them, the screening conditions are set as follows: Hurst exponent threshold H>0.5 (retaining damage-sensitive components with long-term correlation), reconstruction error threshold MSE<0.0003 (ensuring that the signal reconstruction error is lower than the engineering allowable threshold of 0.03%), and window length n∈[10,1000] (covering short-term fluctuations and long-term trend analysis).

[0127] As can be seen, the cubic polynomial fit eliminates 98.5% of baseline drift compared to linear / quadratic fits, and the multi-scale analysis of window length ensures the robustness of H-value calculation (standard deviation < 0.03). The damage sensitivity index (DSI = 8.7) of the filtered target IMF component increases 4.3 times compared to the unfiltered component (DSI = 2.0), effectively enhancing feature separability.

[0128] In step S5, a one-dimensional convolutional neural network model is constructed and trained, the target IMF component is input into the trained one-dimensional convolutional neural network model, and the recognition results of the damage location and damage degree are output.

[0129] Among them, the one-dimensional convolutional neural network model includes: leading convolution layer group: dual-channel 16 cores (size 8, stride 2), intermediate feature layer group: 64 cores (size 4) + maximum pooling layer, deep abstraction layer group: 256 cores (size 2) + global average pooling, classification output layer: 10-neuron fully connected layer with Softmax activation.

[0130] It can be understood that the stress time history (sampling at 200 Hz) and cable force spectrum data covering 15 measuring points (spacing 2 m) on the main beam were used to generate 4,752,000 sets of samples containing 5%-15% noise, simulating the data characteristics under actual environmental interference.

[0131] It should be noted that the data acquisition module integrates a temperature compensation algorithm to eliminate the effects of ambient temperature drift on the strain signal (temperature drift error <0.5με / °C). The white noise excitation bandwidth covers the main bridge vibration frequency range of 0.1-50Hz, and damage conditions include typical damage types such as mid-span cracks in the main beam and loose cable end anchorage, ensuring the completeness and project representativeness of the dataset.

[0132] In an embodiment of the present application, the one-dimensional convolutional neural network model includes: S5.1.1: the input layer receives a 6000-dimensional time series signal; S5.1.2: the first convolutional layer uses 16 convolution kernels of size 8 and a step size of 2; S5.1.3: the second pooling layer adopts maximum pooling with a pooling size of 2; S5.1.4: the deep convolutional layer uses 512 convolution kernels of size 2; S5.1.5: the global average pooling layer is connected to a 512-node fully connected layer; S5.1.6: the Dropout layer sets a dropout rate of 0.3; S5.1.7: the output layer adopts the Softmax activation function.

[0133] Among them, the network optimization strategies include: designing the convolution kernel size in a decreasing manner (8→4→2) to achieve multi-scale feature abstraction, replacing the fully connected layer with global average pooling to reduce the number of parameters by 83%, and a Dropout rate of 0.3 to suppress overfitting (the loss fluctuation of the validation set is <0.5%).

[0134] It is understandable that the deep small core (size 2) focuses on weak damage signals under high-frequency noise (such as 5με-level strain changes with a damage factor of 0.05), and combined with lightweight design (parameter quantity 1.2M) to achieve 50ms-level real-time diagnosis (traditional method 220ms), meeting the embedded deployment requirements of bridge monitoring.

[0135] In the embodiment of the present application, an improved fully adaptive noise ensemble empirical mode decomposition algorithm ICEEMDAN is proposed. Before decomposing the structural dynamic response data to obtain the IMF components, the following steps are also included: S3.2.1: Adding Gaussian white noise to the original stress response signal x(t) to construct a noisy signal:

[0136] x_noise(t)=x(t)+ησ_xξ(t)(16)

[0137] Where σ_x is the signal standard deviation, η∈{5%, 10%, 15%} is the noise intensity coefficient, and ξ(t) is a standard normal distribution random number.

[0138] It can be understood that this encoding strategy reduces the number of network parameters by 40% (10 vs 18 output nodes) by mapping multi-dimensional information (position + degree) with a single label, and supports diagnosis of three composite damages (such as the combination of measurement points 3+8+11), with a quantitative error of ≤2.94% (traditional method ≥8.2%) and a misjudgment rate of <0.5%.

[0139] In an embodiment of the present application, the quantitative analysis of damage adopts a multi-label classification strategy, including: S5.2.1: the damageless state is encoded as category 0, corresponding to the damageless working condition; S5.2.2: the single damage working condition is graded and encoded according to the damage factor: the 0.05 / 0.10 / 0.15 damage factors of measuring point 3 correspond to categories 1-3 respectively, and the same-level damage factors of measuring point 8 correspond to categories 4-6; S5.2.3: Coding rules for multiple damage working conditions: when measuring points 3 and 11 are damaged in combination, the 0.05 / 0.10 / 0.15 damage factors correspond to categories 7-9 respectively; S5.2.4: the output layer of the one-dimensional convolutional neural network is set with 10 neurons, corresponding to the probability distribution of damage working conditions of categories 0-9 respectively.

[0140] The noise parameters are set based on the following principles: η covers the statistical range of actual bridge environmental noise (mean 10.3% ± 4.7%), and σ_x is dynamically calculated using a sliding window (length 1 s, step size 0.1 s) to achieve adaptive adjustment of noise intensity.

[0141] As can be understood, dynamic standard deviation calculation adds weak noise (η = 5%) to the high-energy range (σ_x > 10 MPa) to prevent damage signatures from being obscured; and strong noise (η = 15%) is added to the low-energy range (σ_x < 5 MPa) to simulate signal attenuation at the sensor's far end. This strategy keeps the damage signal attenuation rate below 3% (compared to 8-12% attenuation with traditional fixed threshold methods), and the residual noise is ≤ 12% (compared to ≥ 20% with traditional methods).

[0142] The embodiment of the present application provides a damage identification method for cable-stayed bridges based on massive data, which can significantly improve the damage identification accuracy and computational efficiency in complex noise environments. This method achieves enhanced anti-interference ability and full-process performance breakthroughs in damage identification through dynamic noise suppression mechanism, damage-sensitive feature screening, lightweight convolutional network design and multi-damage collaborative diagnosis optimization. This method establishes a dynamic update model of the noise level parameter β_k (β_{k+1}=β_k·exp(-γE_k)) through the improved ICEEMDAN algorithm, and combines it with the standard deviation feedback mechanism to maintain a damage location accuracy of 98.59% under 15% Gaussian noise interference (traditional methods ≤58.44%). The dual thresholds of Hurst index (H>0.5) and reconstruction error (MSE<0.0003) are used to screen the IMF components, eliminate 90% of redundant modes, and achieve a feature dimension compression ratio of 8:1, retaining more than 95% of effective damage information. By building a 1D-CNN model with a deep separable convolutional architecture, the number of parameters is reduced to 1.2M (compared to 5.7M for traditional CNNs), supporting 50ms-level real-time diagnosis, with a single damage location error of ≤0.5m (measurement point spacing 2m), and a composite damage quantification error of ≤2.94%. This solves problems such as high noise sensitivity, high feature redundancy, insufficient multi-damage recognition, and high resource consumption.

[0143] The following will describe the damage identification method of cable-stayed bridge based on massive data through a specific embodiment. Figure 2 Shown, including:

[0144] Step 1: Construct a baseline finite element model of the cable-stayed bridge.

[0145] like Figure 3 As shown in the figure, the benchmark finite element model was established using ABAQUS software, the main beam used C3D8R solid elements (grid size 0.5m), the cable used T3D2 truss elements, and the consolidation constraint conditions were set at the bottom of the tower. Figure 4 As shown in the figure, the damaged area covers the middle span of the main beam (measurement point 3), 1 / 4 span (measurement point 8) and the cable end anchorage area (measurement point 11), and the distance between the measurement points is 2m. Figure 5 As shown in the figure, the positioning accuracy of the damaged area on the finite element unit reaches 0.1m, and the prestressed steel bars of the main beam are connected to the inclined cables through the built-in area.

[0146] Specifically, a single-span, single-tower cable-stayed bridge was used as an example. The towers were A-shaped vase-shaped towers, and the cables were parallel steel cables, 11 pairs per side. A benchmark finite element model of the cable-stayed bridge was established using ABAQUS software. Three-dimensional, eight-node linear hexahedral elements were used for the main beams and towers, while three-dimensional, two-node truss elements were used for the prestressed steel bars in the main beams and the cables. A full fixation was applied at the base of the towers, with vertical and lateral constraints applied at both ends of the cable-stayed bridge. The main beams and towers were connected by binding, and the cables, prestressed steel bars, main beams, and towers were connected by built-in regions. A benchmark finite element model was established using the model update method. A transient dynamic analysis was performed by applying an environmental excitation to the main beams with the same duration as the analysis step increment.

[0147] Step 2: Collect structural dynamic response data under various damage conditions based on the baseline finite element model.

[0148] like Figure 6 As shown in the figure, the data acquisition module includes a distributed fiber optic sensor array (sampling rate 200Hz), covering 15 measuring points (spacing 2m) on the main beam and the anchor end of the cable. The dynamic response data includes: stress time history signal: recording the dynamic strain response of each measuring point on the main beam (accuracy ±0.5με); cable force vibration spectrum: analyzing the frequency offset of the cable force (resolution 0.1Hz); environmental noise data: collecting 5%-15% Gaussian white noise to simulate the actual monitoring environment interference. Figure 7 As shown in Figure 1, at a damage factor of 0.15, the peak stress at measuring point 3 dropped from 15.2 MPa to 12.9 MPa (a decrease of 14.8%), demonstrating significant damage characteristics. Finite element simulation generated 4,752,000 sets of samples, covering 10 operating conditions, including single damage, combined damage, and no damage. The data classification is shown in Table 1.

[0149] Table 1 Working condition results

[0150]

[0151] Among them, the damage type and damage factor of each working condition are shown in Table 2.

[0152] Table 2 Damage types and damage factors of working conditions

[0153]

[0154] Specifically, for the 0.10 damage factor condition at main beam measuring point 3 (mid-span): Signal acquisition: The fiber optic sensor records the stress time history signal (duration 30s, sampling rate 200Hz), adding 10% Gaussian noise; Feature extraction: The cable tension vibration spectrum shows that the first-order frequency drops from 2.35Hz to 2.28Hz (offset 0.07Hz); Dataset construction: Generate 120,000 sets of samples for this condition, including variants with noise intensity of 5%-15%.

[0155] Step 3: Use the improved ICEEMDAN algorithm to decompose the dynamic response data to obtain the IMF components.

[0156] like Figure 8 As shown, the improved ICEEMDAN decomposition process includes: Adding it to the original data, the formula for obtaining noisy data is as follows:

[0157]

[0158] Where: x(t) is the original signal; is the noise level; N0(t) is the Gaussian white noise sequence; use EMD to decompose the noisy data and get the first residual signal:

[0159] R1= <M(x i )>

[0160] The modal component formula is as follows:

[0161] IMF1(t)=x(t)-R1.

[0162] Where: <·> is the overall mean; M(·) is the local mean of the input signal; calculate the error between the first IMF1(t) and the original data x(t) as follows:

[0163]

[0164]

[0165] in: is the standard deviation of the IMF1 component; σ x is the standard deviation of the original data; E1 is the error; according to the error E1, the noise level formula is updated using a nonlinear strategy as follows:

[0166]

[0167] in, is the updated noise level; β is the control coefficient of noise adjustment; the formula for the residual signal obtained by removing the first IMF component from the original signal is as follows:

[0168] x1(t)=x(t)-IMF1(t)

[0169] Add the updated noise to the residual signal x1(t) The formula is as follows:

[0170]

[0171] Where: N1(t) is a new set of Gaussian white noise; for the new noisy signal Perform EMD decomposition to extract the second-order intrinsic mode function; repeat the above calculation to continue adding noise to the remaining signal, perform EMD decomposition, extract subsequent IMF components, and calculate the error and adjust the noise level after each round of iteration:

[0172]

[0173] in: is the updated noise level; En is the error; β is the noise adjustment control coefficient. The latest IMF component is subtracted from each decomposed signal, and the updated noise is added to the remaining signal. Decomposition and noise adjustment are repeated iteratively until the error En is below the preset error tolerance. Results of the decomposition of cable-stayed bridge monitoring data using the ICEEMDAN algorithm. By analyzing different IMF components, key components containing damage information can be effectively selected for subsequent CNN model training. The IMF component reconstruction errors are shown in Table 3.

[0174] Table 3 IMF component reconstruction error

[0175]

[0176]

[0177] Specifically, the noisy signal at measurement point 3 (σ_x = 15.2 MPa) was decomposed as follows: Initial noise: β0 = 0.2 × 15.2 = 3.04; Iteration process: First iteration: E1 = 0.15 → β1 = 3.04 × exp(-0.3 × 0.15) = 2.89; Third iteration: E3 = 0.03 → β3 = 2.41 × exp(-0.3 × 0.03) = 2.38. Output: The decomposition yields nine IMF components, with the correlation coefficient between IMF3 and IMF5 exceeding 0.90, and the false modal energy accounting for 3.5%.

[0178] Step 4: Screen the Hurst index of IMF component innovation and select the target component.

[0179] like Figure 9 To further verify the noise robustness of the proposed method, as shown in Table 4

[0180] Table 4 Noise levels

[0181]

[0182] Gaussian white noise with a degree of 5%, 10% and 15% is added to the extracted main beam stress data. The noisy stress response calculation formula is:

[0183] s k =sk,u +s k,u σχ

[0184] Where: s k is the stress response signal with noise added; s k,u is a noise-free stress response signal; σ is a standard normal distribution with mean 0 and variance 1; χ is the intensity of the Gaussian white noise level; the Hurst exponent of each IMF component is calculated, and the screening criteria are based on the Hurst exponent threshold. The formula derivation process includes: preprocessing the given time series to obtain the processed series. The formula is as follows:

[0185]

[0186] in: is the mean of the time series, x(t) is the time series; the local trend term formula of each series is obtained by least squares nonlinear fitting as follows:

[0187]

[0188] Among them: a n To fit the polynomial coefficients, use the local trend term to calculate the second-order fluctuation function of the time series, which is as follows:

[0189]

[0190] F q (n)∝n α

[0191] Where: α is the Hurst exponent, y m (k) is the local trend term. By changing the subinterval length of formula (10), the fluctuation function is derived from formulas (11) to (13). The slope of the fitting curve is calculated to obtain the Hurst index. The index formula is as follows:

[0192]

[0193] Among them, the smaller the α value, the greater the signal fluctuation and the rougher the signal. When 0<α<0.5, the time series is negatively correlated and the noise accounts for a large proportion; when α=0.5, the signal is uncorrelated; when 0.5<α<1, the time series is positively correlated and the useful information accounts for a large proportion. Therefore, by calculating the Hurst exponent of each IMF component, the IMF component can be effectively selected; the component with 0.5<α<1 is obtained and the mean square error (MSE) is used to quantify the reconstruction error. And the optimal unique component is selected, and the MSE calculation formula is as follows:

[0194]

[0195] Where: x(t) is the original signal; To reconstruct the signal;

[0196] The comparison of model performance is shown in Table 5 and Table 6. The lower the MSE, the greater the contribution of the IMF component to the reconstruction of the signal and the better the quality, which is the target feature.

[0197] Table 5 Model performance comparison

[0198]

[0199] Table 6 Model scores

[0200]

[0201] Specifically, the quantitative damage analysis of the IMF3 component (H = 0.68, MSE = 0.00018) is shown in Table 7.

[0202] Table 7 Quantitative damage analysis

[0203]

[0204] Specific analysis includes: Hurst calculation: when window n = 500, logF(n) = 1.82, logn = 2.70 → slope = 0.67 → H = 0.335; trend elimination: cubic polynomial fitting R 2 =0.992, baseline drift reduced by 98.7%; screening results: IMF3 (H=0.68>0.5, MSE=0.00018<0.0003) was retained, and its DSI=9.1, which was 4.6 times higher than that of the unscreened component.

[0205] Step 5: Construct a one-dimensional convolutional neural network model to output damage identification results.

[0206] Among them, the architecture of the target model is shown in Table 8.

[0207] Table 8 Architecture of the target model

[0208]

[0209]

[0210] To validate the model's effectiveness in real-world projects, this study conducted field tests on an operational cable-stayed bridge. As shown in Table 9, the training conditions covered key vulnerable areas of the main girder (measurement points 3 and 8) and typical damage levels (0.05-0.15), ensuring the model's sensitivity to actual damage patterns.

[0211] Table 9 Test results of cable-stayed bridge

[0212]

[0213] Distributed fiber optic sensors (sampling rate 200 Hz) were deployed on the main beams of the actual bridge to collect stress time history data over a 300-second period during operation. After ICEEMDAN decomposition and Hurst exponent filtering, the measured data were input into a trained 1D-CNN model for diagnosis. As shown in Table 10, all four sets of measured signals were judged to be damage-free (labeled 0), consistent with the conclusions of manual inspection.

[0214] Table 10 Measured signal results

[0215]

[0216] Specifically, the 0.15 damage factor sample of measuring point 3 was diagnosed: input data: 6000-dimensional time series signal of IMF3 component; network reasoning: the first layer convolution outputs a 16-channel feature map (size 2996×16); after global average pooling, a 512-dimensional feature vector is generated; the Softmax layer outputs the probability distribution: the probability of category 3 (measuring point 3-0.15 damage) is 99.3%; the results verify: the actual damage location error is 0.32m (<0.5m), and the calculated damage factor is 0.147 (error 1.93%).

[0217] In summary, the present invention implements a highly accurate and robust intelligent damage identification method for cable-stayed bridges. Dynamic noise suppression is achieved through an improved ICEEMDAN algorithm. Combined with the Hurst exponent feature screening mechanism and a lightweight one-dimensional convolutional neural network, this method overcomes the identification bottleneck of traditional methods in complex noise environments. By dynamically adjusting the noise injection intensity (β_{k+1}=β_k·exp(-γE_k)) through a standard deviation feedback mechanism, high-purity IMF components can still be extracted under 15% Gaussian noise interference, reducing the modal aliasing to 0.12 (compared to 0.54 using traditional methods) and increasing the IMF correlation coefficient to 0.92. Using a dual-threshold screening mechanism of Hurst exponent (H>0.5) and reconstruction error (MSE<0.0003), the feature dimension compression ratio reaches 8:1, retaining 95% of effective damage information and improving the damage sensitivity index (DSI) by 4.3 times. A 1D-CNN model with a deep separable convolutional architecture was constructed, with the number of parameters compressed to 1.2M (traditional CNN 5.7M), an inference speed of 50ms / time, support for embedded device deployment, single damage location error ≤ 0.5m, and composite damage quantification error ≤ 2.94%. The compensation factor γ (0.1-0.5) was dynamically adjusted based on the noise intensity η (5%-15%), and the network weights were updated through backpropagation to ensure the robustness of the system in complex environments such as temperature drift (error < 0.5με / ℃) and vibration frequency band offset (0.1-50Hz). This solution provides a full-link solution for cable-stayed bridge health monitoring from signal processing to intelligent diagnosis, significantly improving the structural safety warning capability and operation and maintenance efficiency, and solving the technical bottlenecks of traditional methods in noise sensitivity, feature redundancy, multi-damage analysis and real-time performance.

[0218] Next, a device for identifying damage from massive data of a cable-stayed bridge according to an embodiment of the present application will be described with reference to the accompanying drawings.

[0219] Figure 10 It is a block diagram of a massive data damage identification device for a cable-stayed bridge according to an embodiment of the present application.

[0220] like Figure 10 As shown, the cable-stayed bridge massive data damage identification device 10 includes: a modeling module 100, a data acquisition module 200, a signal decomposition module 300, a feature screening module 400 and a network construction module 500.

[0221] Among them, the modeling module 100 is used to construct a benchmark finite element model of a cable-stayed bridge; the data acquisition module 200 is used to collect structural dynamic response data under various damage conditions based on the benchmark finite element model; the signal decomposition module 300 is used to propose an improved fully adaptive noise ensemble empirical mode decomposition algorithm ICEEMDAN to decompose the structural dynamic response data to obtain IMF components, wherein the decomposition algorithm ICEEMDAN is embodied in dynamically adjusting the noise level through a standard deviation feedback mechanism; the feature screening module 400 is used to innovatively perform Hurst index screening on the IMF components, and select target IMF components that meet the Hurst index H>0.5H>0.5 and the reconstruction error MSE<0.0003MSE<0.0003; the network construction module 500 is used to construct and train a one-dimensional convolutional neural network model, input the target IMF component into the trained one-dimensional convolutional neural network model, and output the identification results of the damage location and damage degree, wherein the one-dimensional convolutional neural network model includes: a leading convolution layer group, a mid-level feature layer group, a deep abstraction layer group, and a classification output layer.

[0222] It should be noted that the above explanation of the embodiment of the cable-stayed bridge damage identification method based on massive data is also applicable to the cable-stayed bridge damage identification device based on massive data of this embodiment, and will not be repeated here.

[0223] According to the cable-stayed bridge massive data damage identification device proposed in the embodiment of the present application, high-precision real-time identification of cable-stayed bridge damage in complex noise environments is achieved through dynamic noise suppression, damage-sensitive feature screening and lightweight convolutional network design. Based on the intelligent analysis of structural dynamic response data, the device overcomes the technical bottlenecks of modal aliasing, feature redundancy and low computational efficiency of traditional methods under noise interference. The integrated improved ICEEMDAN algorithm dynamically adjusts the noise injection intensity (β_{k+1}=β_k·exp(-γE_k), γ=0.3) through the standard deviation feedback mechanism. High-purity IMF components can still be extracted under 15% Gaussian noise, the modal aliasing degree is reduced to 0.12 (traditional methods ≥0.54), and the IMF component correlation coefficient is increased to 0.92. Using the dual threshold screening mechanism of Hurst index (H>0.5) and reconstruction error (MSE<0.0003), the feature dimension compression ratio reaches 8:1, retaining 95% of effective damage information and improving the damage sensitivity index (DSI) by 4.3 times. A 1D-CNN model with a deep separable convolutional architecture was constructed, reducing the number of parameters to 1.2M (compared to 5.7M for traditional CNNs), achieving an inference speed of 50ms / time, and supporting deployment on embedded devices. The error in localizing single damage is ≤0.5m (with a 2m spacing between measurement points), and the error in quantifying composite damage is ≤2.94%. The compensation factor γ (0.1-0.5) is dynamically adjusted based on the real-time noise intensity η (5%-15%), and network weights are updated through backpropagation, ensuring system robustness in complex environments such as temperature drift (error <0.5με / °C) and vibration frequency band shift (0.1-50Hz).

[0224] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0225] A memory 1101 , a processor 1102 , and a computer program stored in the memory 1101 and executable on the processor 1102 .

[0226] When the processor 1102 executes the program, the cable-stayed bridge damage identification method based on massive data provided in the above embodiment is implemented.

[0227] Furthermore, the electronic device further includes:

[0228] The communication interface 1103 is used for communication between the memory 1101 and the processor 1102 .

[0229] The memory 1101 is used to store computer programs that can be run on the processor 1102 .

[0230] The memory 1101 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0231] If the memory 1101, the processor 1102, and the communication interface 1103 are implemented independently, the communication interface 1103, the memory 1101, and the processor 1102 can be connected to each other via a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0232] Optionally, in a specific implementation, if the memory 1101, the processor 1102 and the communication interface 1103 are integrated on a chip, the memory 1101, the processor 1102 and the communication interface 1103 can communicate with each other through an internal interface.

[0233] The processor 1102 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0234] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for damage identification of cable-stayed bridges based on massive data.

[0235] In addition, an embodiment of the present application also provides a computer program product, including a computer program or instructions, which, when executed, implements the above-mentioned cable-stayed bridge damage identification method based on massive data.

[0236] In the description of this specification, reference to the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0237] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features being referred to. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0238] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0239] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0240] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0241] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A damage identification method for cable-stayed bridges based on massive data, characterized in that: The method includes: S1: Construct a benchmark finite element model of a cable-stayed bridge; S2: collecting structural dynamic response data under various damage conditions according to the benchmark finite element model; S3: An improved fully adaptive noise ensemble empirical mode decomposition algorithm (ICEEMDAN) is proposed to decompose the structural dynamic response data to obtain IMF components, wherein the decomposition algorithm ICEEMDAN dynamically adjusts the noise level through a standard deviation feedback mechanism, wherein the decomposition algorithm ICEEMDAN includes: S3.1.1 Add the initial noise level to the original data to obtain the noisy data, and the formula is as follows: Where: x(t) is the original signal; is the original noise level; N0(t) is the Gaussian white noise sequence; S3.1.2: Use EMD to decompose the noisy data and obtain the first residual signal. The formula is as follows: R1= <M(x i )> (2) Where: <·> is the overall mean; M(·) is the local mean of the input signal; S3.1.3: Calculate the error between the first and the original data using the following formula: in: is the standard deviation of the IMF1 component; σ x is the standard deviation of the original data; E1 is the error; S3.1.4: Based on the error E1, the noise level formula is updated using a nonlinear strategy as follows: in: is the noise level after the first update; β is the control coefficient of noise adjustment; S3.1.5: The formula for the residual signal obtained by removing the first IMF component from the original signal is as follows: x1(t)=x(t)-IMF1(t) (7) S3.1.6: Add the updated noise to the remaining signal, calculated as follows: Where: N1(t) is a new set of Gaussian white noise; S3.1.7: Perform EMD decomposition on the new noisy signal to extract the second-order intrinsic mode function; S3.1.8: Repeat the above calculations to add noise to the remaining signal, perform EMD decomposition, extract subsequent IMF components, and calculate the error and adjust the noise level after each iteration: in: is the noise level after the (n+1)th update; E n is the error at the nth time; β is the control coefficient of noise adjustment; S4: Perform Hurst index screening on the IMF components, and select target IMF components that meet the Hurst index H>0.5H>0.5 and the reconstruction error MSE<0.0003MSE<0.0003; S5: Construct and train a one-dimensional convolutional neural network model, input the target IMF component into the trained one-dimensional convolutional neural network model, and output the recognition results of the damage location and damage degree, wherein the one-dimensional convolutional neural network model includes a leading convolution layer group, a mid-level feature layer group, a deep abstraction layer group and a classification output layer.

2. The cable-stayed bridge damage identification method based on massive data according to claim 1 is characterized in that: The Hurst exponent calculation includes: S4.1.1: Preprocess the given time series x(t) to obtain the processed sequence, as follows: Where: x is the mean of the time series; S4.1.2: Use the least squares method to obtain the local trend term of each series using the nonlinear fitting method. The formula is as follows: Among them: a n are the coefficients of the fitted polynomial; S4.1.3: Use the local trend term y m (k) Calculate the second-order fluctuation function of the time series, which is as follows: F q (n)∝n α (13) Where: α is the Hurst exponent; S4.1.4: Calculate the slope of the fitted curve to obtain the Hurst exponent, as follows: S4.1.5: The IMF components can be effectively selected by calculating the Hurst exponent of each IMF component. The MSE calculation formula is: Where: x(t) is the original signal; To reconstruct the signal.

3. The cable-stayed bridge damage identification method based on massive data according to claim 1 is characterized in that: The one-dimensional convolutional neural network model includes: S5.1.1: The input layer receives a 6000-dimensional time series signal; S5.1.2: The first convolutional layer uses 16 convolution kernels of size 8 with a stride of 2; S5.1.3: The second pooling layer uses maximum pooling with a pooling size of 2; S5.1.4: The deep convolutional layer uses 512 convolution kernels of size 2; S5.1.5: The global average pooling layer is connected to a 512-node fully connected layer; S5.1.6: The Dropout layer sets a dropout rate of 0.3; S5.1.7: The output layer uses the Softmax activation function.

4. The cable-stayed bridge damage identification method based on massive data according to claim 1 is characterized in that: An improved fully adaptive noise ensemble empirical mode decomposition algorithm ICEEMDAN is proposed. Before decomposing the structural dynamic response data to obtain IMF components, the algorithm also includes: S3.2.1: Add Gaussian white noise to the original stress response signal x(t) to construct a noisy signal: x_noise(t)=x(t)+ησ_xξ(t)(16) Where σ_x is the signal standard deviation, η∈{5%, 10%, 15%} is the noise intensity coefficient, and ξ(t) is a standard normal distribution random number.

5. The cable-stayed bridge damage identification method based on massive data according to claim 1 is characterized in that: The multiple damage conditions adopt a multi-label classification strategy, wherein the multi-label classification strategy includes lossless state coding, single damage condition and multiple damage conditions, including: S5.2.1: Encode the damage-free state as category 0, corresponding to the damage-free condition; S5.2.2: Single damage conditions are coded according to the damage factor classification: the 0.05 / 0.10 / 0.15 damage factors at measuring point 3 correspond to categories 1-3, respectively, and the same damage factors at measuring point 8 correspond to categories 4-6; S5.2.3: Coding rules for multiple damage conditions: When measuring points 3 and 11 are damaged in combination, damage factors of 0.05 / 0.10 / 0.15 correspond to categories 7-9 respectively; S5.2.4: The one-dimensional convolutional neural network output layer is set with 10 neurons, corresponding to the probability distribution of 0-9 types of damage conditions respectively.

6. A cable-stayed bridge damage identification device based on massive data, characterized in that: include: Modeling module, S1: constructing a benchmark finite element model of a cable-stayed bridge; Data acquisition module, S2: collecting structural dynamic response data under various damage conditions according to the benchmark finite element model; Signal decomposition module, S3: An improved fully adaptive noise ensemble empirical mode decomposition algorithm (ICEEMDAN) is proposed to decompose the structural dynamic response data to obtain IMF components. The decomposition algorithm ICEEMDAN dynamically adjusts the noise level through a standard deviation feedback mechanism. The decomposition algorithm ICEEMDAN includes: S3.1.1 Add the initial noise level to the original data to obtain the noisy data, and the formula is as follows: Where: x(t) is the original signal; is the original noise level; N0(t) is the Gaussian white noise sequence; S3.1.2: Use EMD to decompose the noisy data and obtain the first residual signal. The formula is as follows: R1= <M(x i )> (2) Where: <·> is the overall mean; M(·) is the local mean of the input signal; S3.1.3: Calculate the error between the first and the original data using the following formula: in: is the standard deviation of the IMF1 component; σ x is the standard deviation of the original data; E1 is the error; S3.1.4: Based on the error E1, the noise level formula is updated using a nonlinear strategy as follows: in: is the noise level after the first update; β is the control coefficient of noise adjustment; S3.1.5: The formula for the residual signal obtained by removing the first IMF component from the original signal is as follows: x1(t)=x(t)-IMF1(t) (7) S3.1.6: Add the updated noise to the remaining signal, calculated as follows: Where: N1(t) is a new set of Gaussian white noise; S3.1.7: Perform EMD decomposition on the new noisy signal to extract the second-order intrinsic mode function; S3.1.8: Repeat the above calculations to add noise to the remaining signal, perform EMD decomposition, extract subsequent IMF components, and calculate the error and adjust the noise level after each iteration: in: is the noise level after the (n+1)th update; E n is the error at the nth time; β is the control coefficient of noise adjustment; Feature screening module, S4: innovatively perform Hurst index screening on the IMF components, and select target IMF components that meet the Hurst index H>0.5H>0.5 and the reconstruction error MSE<0.0003MSE<0.0003; Network construction module, S5: construct and train a one-dimensional convolutional neural network model, input the target IMF component into the trained one-dimensional convolutional neural network model, and output the recognition results of the damage location and damage degree, wherein the one-dimensional convolutional neural network model includes a leading convolution layer group, a mid-level feature layer group, a deep abstraction layer group and a classification output layer.

7. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the damage identification method for a cable-stayed bridge based on massive data as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed, the method for cable-stayed bridge damage identification based on massive data as described in any one of claims 1 to 5 is implemented.

9. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed, the method for cable-stayed bridge damage identification based on massive data as described in any one of claims 1 to 5 is implemented.

Images