High-pile wharf structure damage identification method and system

Abstract

The invention provides a high-pile wharf structure damage identification method and system, and relates to the technical field of wharf damage identification. According to the invention, through collaborative acquisition of multi-source data (strain, deformation and defect), a monitoring blind area of a single sensor is broken through, the multi-source data is integrated into multi-source heterogeneous data with time-space alignment, and high-dimensional input is provided for a deep learning model; and a damage feature vector output by the model is matched with a dynamic threshold library, so that a full-link closed loop from data acquisition to feature fusion to damage decision is realized. Specifically, the detection coverage is improved through multi-dimensional data collection, the anti-interference capability is enhanced through standardization and fusion processing, the damage analysis precision is improved through a deep learning model, and self-adaptive decision making is achieved through a dynamic threshold library.

Description

Technical Field

[0001] The present invention relates to the technical field of wharf damage identification, and particularly to a method and system for identifying structural damage of high-piled wharfs. Background Art

[0002] As a core facility of port engineering, high-piled wharfs are subject to complex actions such as ship loads, water flow impacts, and environmental corrosion for a long time, which are prone to cause structural damage. Existing damage detection technologies have the following problems: First, the detection dimension is single: Traditional methods rely on a single sensor (such as a strain gauge or a vibration sensor) to collect local data, and it is difficult to comprehensively reflect the internal and external damage states of the structure. For example, vibrating wire strain gauges can only monitor surface strains and cannot capture internal cracks or hidden defects in underwater pile foundations. Second, the real-time performance and anti-interference ability are insufficient: Existing technologies rely on manual inspections or offline analyses, cannot achieve real-time monitoring, and changes in environmental temperature and humidity are likely to lead to misjudgments. For example, instantaneous damage caused by overloading may be missed due to detection delays. Third, the multi-source data fusion ability is weak: Deformation, strain, and internal defect data are analyzed in isolation, lacking spatio-temporal correlation, resulting in fuzzy damage localization. For example, surface deformation data and internal defect data are not analyzed collaboratively, making it difficult to distinguish damage causes (such as ship collisions or material deterioration). Fourth, the threshold library is static: Traditional threshold libraries are statically set based on historical data and cannot adapt to the long-term performance degradation of wharfs and dynamic changes in the environment, resulting in an increased false alarm rate. Summary of the Invention

[0003] In view of the above problems existing in the prior art, the first aspect of the present invention proposes a method for identifying structural damage of a high-piled wharf, including: Step S1: Based on a distributed optical fiber sensing network, collect the structural strain data and temperature data of the high-piled wharf in real time to generate a first data set; Step S2: Obtain the surface deformation data of the high-piled wharf through a three-dimensional laser scanner to generate a second data set; Step S3: Use an ultrasonic tomography scanner to perform non-destructive testing on the internal structure of the high-piled wharf, obtain the original ultrasonic tomography signal data, and generate a third data set; Step S4: Perform feature extraction and standardization processing on the first data set, the second data set, and the third data set, including dimensionality reduction through principal component analysis, using a normalization algorithm to eliminate dimension differences, and aligning feature vectors based on geographic information system coordinates to generate multi-source heterogeneous data; Step S5: Input the multi-source heterogeneous data into a pre-trained deep learning model, extract spatial features through a convolutional neural network, and analyze temporal correlations through a long short-term memory network to generate a fused damage feature vector; Step S6: Based on the damage feature vector, similarity matching is performed in combination with the dynamically updated damage threshold library, and the damage location and damage level are output.

[0004] In combination with the first aspect, in some implementations of the first aspect, step S1 includes: Step S1-1: collecting strain distribution data and ambient temperature change data on the surface of the pile foundation through distributed optical fiber sensors arranged symmetrically along the axial direction of the pile foundation to generate an original strain data set; Step S1-2: inputting the original strain data set into the Kalman filter module, removing the noise interference data, and generating a first data set; Step S2 includes: Step S2-1: Scan the pile cap, pile body and panel of the high-pile wharf at multiple angles using a three-dimensional laser scanner to obtain surface deformation point cloud data; Step S2-2: Input the surface deformation point cloud data into the coordinate conversion module of the geographic information system, align it to the unified spatial reference, and generate a second data set.

[0005] In combination with the first aspect, in some implementations of the first aspect, step S4 includes: Step S4-1: Input the structural strain data in the first data set into the Hilbert-Huang transform module, decompose it into multiple intrinsic mode functions, extract the main frequency component and transient energy characteristics, and generate a frequency domain feature vector; Step S4-2: input the surface deformation data in the second data set into the iterative closest point algorithm module, remove noise points and perform registration based on geographic information system coordinates, extract the pile cap displacement and pile body inclination angle, and generate a deformation gradient vector; Step S4-3: inputting the ultrasonic tomography raw signal data in the third data set into the back-projection reconstruction algorithm module, analyzing the geometric parameters and position information of the internal defects, and generating a tomographic imaging map; Step S4-4: Input the frequency domain feature vector, deformation gradient vector and tomographic imaging spectrum into the normalization module, eliminate the dimension difference, and then perform dimensionality reduction through principal component analysis to generate standardized multi-source heterogeneous data.

[0006] In combination with the first aspect, in some implementations of the first aspect, the dimensionality reduction by principal component analysis in step S4-4 includes: Step S4-4-1: inputting the frequency domain feature vector, deformation gradient vector and tomographic imaging spectrum into the covariance matrix calculation module to generate the covariance matrix of multi-source data; Step S4-4-2: Perform eigenvalue decomposition on the covariance matrix, select the top three principal components with the highest contribution rate, and generate the eigenvector after dimensionality reduction.

[0007] Combined with the first aspect, in some implementation manners of the first aspect, in step S4-3, parsing the geometric parameters and position information of internal defects includes: Reconstruct the three-dimensional space coordinates and equivalent aperture of internal defects according to the relationship between acoustic wave propagation time and amplitude attenuation.

[0008] Combined with the first aspect, in some implementation manners of the first aspect, step S5 includes: Step S5-1: Input multi-source heterogeneous data into a convolutional neural network, extract spatial local features through multiple convolutional kernels, and generate a damage-sensitive feature map; Step S5-2: Input the damage-sensitive feature map into a long short-term memory network, analyze the temporal variation law, and generate a damage feature vector that fuses spatio-temporal correlations.

[0009] Combined with the first aspect, in some implementation manners of the first aspect, in step S6, the generation of a dynamically updated damage threshold library includes: Step S6-1: Train an initial damage threshold library with sample data in the historical damage database to generate benchmark matching parameters; Step S6-2: Input the first data set, the second data set, and the third data set into an adaptive update module, and dynamically adjust the benchmark matching parameters based on the structure strain data, surface deformation data, and ultrasonic tomography raw signal data collected in real time to generate a dynamically updated damage threshold library.

[0010] Combined with the first aspect, in some implementation manners of the first aspect, the deployment of the distributed optical fiber sensing network in step S1 includes: Mark two measuring lines symmetrically along the axial direction on the surface of the pile foundation, one measuring line is located on the side facing the water area, and the other measuring line is located on the side facing the slope, to generate measuring line positioning information; Lay the distributed optical fiber along the measuring line positioning information, and fix it on the pile surface in three sections through underwater epoxy resin, which are located in the pile top, pile middle, and pile bottom areas respectively, to form an optical fiber sensing network for continuous monitoring of the entire pile length.

[0011] Combined with the first aspect, in some implementation manners of the first aspect, the training process of the deep learning model includes: Generate virtual training samples by finite element simulation of the structure strain data, surface deformation data, and ultrasonic tomography raw signal data of the high-pile wharf under different damage conditions; Integrate the virtual training samples with historical monitoring data and on-site loading test data to construct a training sample set covering multiple scenarios; Input the training sample set into a generative adversarial network to generate an enhanced training set; Train a convolutional neural network and a long short-term memory network with the enhanced training set to generate a deep learning model.

[0012] In a second aspect, the present invention provides a structural damage identification system for high-piled wharves. The system adopts the method provided in any of the above embodiments, and the system includes: A data acquisition module, including a distributed optical fiber sensing network, a three-dimensional laser scanner, and an ultrasonic tomography scanner, for real-time collecting structural strain data, temperature data, surface deformation data, and ultrasonic tomography original signal data of the high-piled wharf, and respectively generating a first data set, a second data set, and a third data set; A data processing module, connected to the data acquisition module, for performing feature extraction and normalization processing on the first data set, the second data set, and the third data set, including dimensionality reduction through principal component analysis, using a normalization algorithm to eliminate the dimension difference, and aligning the feature vectors based on geographic information system coordinates to generate multi-source heterogeneous data; A model processing module, connected to the data processing module, including a pre-trained deep learning model, which is composed of a convolutional neural network and a long short-term memory network, for performing spatial feature extraction and temporal correlation analysis on the multi-source heterogeneous data to generate a fused damage feature vector; A damage identification module, connected to the model processing module, including a dynamically updated damage threshold library and a matching calculation unit, for performing similarity matching based on the damage feature vector and outputting the damage location and damage level.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: The detection coverage is improved through the multi-dimensional data acquisition in steps S1 to S3. Specifically, in step S1, the distributed optical fiber sensing network real-time collects structural strain data and temperature data, and covers the underwater and above-water areas through continuous monitoring of the entire pile length, solving the problem of blind spots in traditional local monitoring; in step S2, the three-dimensional laser scanner obtains surface deformation data, accurately captures macroscopic deformation characteristics such as pile cap displacement and pile body inclination, making up for the deficiency of single strain data; in step S3, the ultrasonic tomography scanner analyzes internal defect data through acoustic signals, realizing visual detection of hidden damages (such as cracks and cavities), and breaking through the limitations of surface monitoring.

[0014] In addition, in step S4, principal component analysis reduces the dimension of multi-source data and eliminates redundant features; the normalization algorithm unifies the dimensions of strain, deformation, and defect data, reducing the magnitude deviation caused by environmental temperature fluctuations; the geographic information system coordinates align the feature vectors to ensure the consistency of spatio-temporal data and avoid misjudgment.

[0015] In step S5, the convolutional neural network extracts spatial local features (such as crack propagation patterns) of strain and deformation data, and the long short-term memory network analyzes the temporal correlation law (such as the cumulative effect of overloading and stacking), and fuses to generate a damage feature vector, solving the problem of poor adaptability of traditional threshold methods to complex damage patterns and improving the damage analysis accuracy.

[0016] In step S6, the dynamically updated damage threshold library adaptively adjusts the matching parameters based on real-time monitoring data, overcomes the problem of benchmark failure caused by material aging and environmental corrosion in the static threshold library, and improves the reliability of long-term monitoring.

[0017] The synergistic effect of steps S1 to S6 is reflected in that multi-source data (strain, deformation, defect) are integrated into spatio-temporally aligned multi-source heterogeneous data in step S4, providing high-dimensional input for the deep learning model in step S5; the damage feature vectors output by the model are matched with the dynamic threshold library to achieve a full-link closed loop from "data acquisition - feature fusion - damage decision". For example, fiber optic data captures instantaneous overload strain, laser scanning data reflects the cumulative effect of deformation, and ultrasonic data locates internal defects. The three work together to determine the damage level (such as minor cracks or severe fractures), reducing the misjudgment risk of a single data source. Brief Description of the Drawings

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are 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.

[0019] Figure 1 The figure shows a schematic flow chart of a method for identifying damage to a high-piled wharf structure provided by an embodiment of the present invention.

[0020] Figure 2a The figure shows a schematic diagram of a distributed fiber optic sensing network provided by an embodiment of the present invention.

[0021] Figure 2b The figure shows a schematic diagram of a distributed fiber optic sensing network provided by another embodiment of the present invention.

[0022] Figure 3 The figure shows a schematic diagram of a scanned image provided by an embodiment of the present invention.

[0023] Figure 4 The figure shows a schematic structural diagram of a system for identifying damage to a high-piled wharf structure provided by an embodiment of the present invention. Detailed Embodiments

[0024] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0025] The following describes the specific embodiments of the present invention.

[0026] Embodiment 1 As Figure 1 shown, the present invention proposes a method for identifying structural damage of a high-pile wharf, including: Step S1: Based on a distributed optical fiber sensing network (as Figure 2a and Figure 2b shown), the structural strain data and temperature data of the high-pile wharf are collected in real time to generate a first data set; Step S2: The surface deformation data of the high-pile wharf is obtained by a three-dimensional laser scanner to generate a second data set; Wherein, Figure 3 shown is the scanned image of the high-pile wharf obtained by the three-dimensional laser scanner, and the surface deformation data of the high-pile wharf is obtained through this scanned image; Step S3: The internal structure of the high-pile wharf is nondestructively detected by an ultrasonic tomography scanner to obtain the original ultrasonic tomography signal data and generate a third data set; Step S4: Feature extraction and standardization processing are performed on the first data set, the second data set and the third data set, including dimensionality reduction through principal component analysis, using a normalization algorithm to eliminate the dimension difference, and aligning the feature vectors based on the geographical information system coordinates to generate multi-source heterogeneous data; Step S5: The multi-source heterogeneous data is input into a pre-trained deep learning model, spatial features are extracted through a convolutional neural network, and temporal correlations are analyzed through a long short-term memory network to generate a fused damage feature vector; Step S6: Based on the damage feature vector, similarity matching is performed in combination with a dynamically updated damage threshold library, and the damage location and damage level are output.

[0027] The structural damage identification method for high-piled wharves realizes accurate damage positioning through multi-source data acquisition, fusion processing, and intelligent analysis. The distributed optical fiber sensing network is arranged along the surface of the pile foundation to collect structural strain data and temperature data in real time, forming the first data set covering the entire pile length; the three-dimensional laser scanner scans the pile cap, pile body, and panel from multiple angles to generate surface deformation point cloud data, which forms the second data set after geospatial information system coordinate registration; the ultrasonic tomography scanner detects internal defects through acoustic reflection signals to generate the third data set. In the feature extraction and standardization processing stage, principal component analysis reduces the dimensionality of the multi-source data, eliminates redundant features, the normalization algorithm unifies the dimensional differences of strain, deformation, and defect data, and the geospatial information system coordinate alignment ensures spatio-temporal consistency, generating multi-source heterogeneous data. In the deep learning model, the convolutional neural network extracts spatial local features (such as crack propagation patterns), and the long short-term memory network analyzes temporal correlations (such as overloading cumulative effects) to generate damage feature vectors integrating spatio-temporal characteristics. The dynamically updated damage threshold library adjusts matching parameters based on real-time monitoring data, and performs similarity matching in combination with the damage feature vectors to output the damage location and damage level.

[0028] In the embodiments of the present invention, the monitoring blind area of a single sensor is broken through by the collaborative acquisition of multi-source data (strain, deformation, defect), environmental interference is eliminated through standardization processing, the analytical ability of complex damage patterns is improved by the deep learning model, and the dynamic threshold library adapts to the long-term performance degradation of the wharf. For example, the distributed optical fiber captures instantaneous overloading strain, the laser scanning data reflects the cumulative trend of deformation, and the ultrasonic tomography data locates internal cracks. The fusion of the three can distinguish ship impact (instantaneous high-frequency strain) from material deterioration (long-term deformation accumulation), avoiding misjudgment caused by isolated data in traditional methods.

[0029] In alternative solutions, the distributed optical fiber sensing network can adopt fiber Bragg grating sensors or distributed acoustic sensing technology; the normalization algorithm can be replaced by Z-score standardization or decimal scaling standardization; the deep learning model can select a graph convolutional network to replace the convolutional neural network to better process non-Euclidean space data.

[0030] Combined with the first aspect, in some implementation manners of the first aspect, step S1 includes: Step S1-1: Through distributed optical fiber sensors symmetrically arranged along the axial direction of the pile foundation, collect the strain distribution data on the surface of the pile foundation and the environmental temperature change data, and generate the original strain data set; Step S1-2: Input the original strain data set into the Kalman filter module to eliminate noise interference data and generate the first data set; Step S2 includes: Step S2-1: Through a three-dimensional laser scanner, scan the pile cap, pile body, and panel of the high-piled wharf from multiple angles to obtain surface deformation point cloud data; Step S2-2: Input the surface deformation point cloud data into the geographic information system coordinate transformation module, register it to a unified spatial reference, and generate a second dataset.

[0031] In step S1-1, distributed fiber optic sensors are symmetrically arranged along the axial direction of the pile foundation. One survey line is located on the side facing the water area to monitor the action of ship loads, and the other is located on the side facing the slope to detect the influence of slope earth pressure. The sensor network covers the pile top, pile middle, and pile bottom areas and is fixed in segments by underwater epoxy resin to ensure continuous monitoring of the entire pile length. The original strain dataset contains spatio-temporally correlated strain distribution and temperature change data. The Kalman filter module eliminates noise interference caused by wave impact or temperature fluctuations based on the state space model and generates a first dataset with high signal-to-noise ratio. In step S2-1, the three-dimensional laser scanner acquires surface deformation point cloud data by multi-view scanning. The geographic information system coordinate transformation module registers the point cloud data to the global coordinate system of the wharf, eliminates the equipment pose error, and generates a second dataset with a unified spatial reference.

[0032] The symmetrically arranged fiber optic sensors enhance the representativeness of the data, the Kalman filter improves the reliability of the strain data, and the coordinate registration ensures the spatial consistency of the deformation data. For example, the survey line on the side facing the water area can preferentially capture the strain mutation caused by ship impact, while the Kalman filter can suppress the influence of daily temperature fluctuations on long-term monitoring; after the laser scan data is coordinate-registered, the calculation error of the pile cap displacement and the pile body inclination angle is reduced, avoiding deformation misjudgment caused by coordinate system offset.

[0033] In an alternative solution, the Kalman filter can be replaced by wavelet transform denoising; the geographic information system coordinate transformation can use the ICP (Iterative Closest Point) algorithm or feature matching algorithm; the three-dimensional laser scanner can be replaced by a structured light scanner, but the resolution and scanning speed need to be balanced.

[0034] Combined with the first aspect, in some implementation manners of the first aspect, step S4 includes: Step S4-1: Input the structural strain data in the first dataset into the Hilbert-Huang transform module, decompose it into multiple intrinsic mode functions, extract the main frequency component and transient energy characteristics, and generate a frequency domain feature vector; Step S4-2: Input the surface deformation data in the second dataset into the Iterative Closest Point algorithm module, remove noise points and register based on the geographic information system coordinates, extract the pile cap displacement and the pile body inclination angle, and generate a deformation gradient vector; Step S4-3: Input the ultrasonic tomography original signal data in the third dataset into the back-projection reconstruction algorithm module, analyze the geometric parameters and position information of internal defects, and generate a tomographic imaging map; Step S4-4: Input the frequency-domain feature vector, deformation gradient vector, and tomographic imaging map into the normalization module. After eliminating the dimensional differences, perform dimensionality reduction through principal component analysis to generate standardized multi-source heterogeneous data.

[0035] In step S4-1, the Hilbert-Huang transform module performs empirical mode decomposition on the structural strain data in the first dataset, decomposes the non-stationary signal into multiple intrinsic mode functions (IMFs), extracts the main frequency components and transient energy characteristics of the first three IMF components with the highest energy proportion, and generates a frequency-domain feature vector. In step S4-2, the iterative closest point algorithm module registers the surface deformation point cloud data in the second dataset. After removing the outlier noise points, calculate the pile cap displacement vector (including horizontal displacement and vertical settlement) and the pile body inclination angle based on the geographic information system coordinates to generate a deformation gradient vector. In step S4-3, the back-projection reconstruction algorithm module analyzes the original ultrasonic tomographic signal data in the third dataset, and reconstructs the three-dimensional spatial coordinates and equivalent aperture of the internal defect according to the acoustic wave propagation time difference and amplitude attenuation coefficient to generate a tomographic imaging map. In step S4-4, the normalization module unifies the dimensions of the frequency-domain feature vector, deformation gradient vector, and tomographic imaging map, and the principal component analysis selects the first three principal components with the highest contribution rate to generate standardized multi-source heterogeneous data.

[0036] The technical effects are as follows: The Hilbert-Huang transform effectively processes non-stationary strain signals, the iterative closest point algorithm improves the registration accuracy of deformation data, the back-projection reconstruction realizes three-dimensional visualization of defects, and the principal component analysis reduces the data dimension. For example, the transient energy characteristics can identify instantaneous damage caused by overloading, the deformation gradient vector quantifies the inclination trend of the pile body, and the tomographic imaging map locates internal cracks. After the three are normalized, the dimensional differences are eliminated, providing high-quality input for the deep learning model.

[0037] In an alternative solution, the Hilbert-Huang transform can be replaced by the wavelet transform; the iterative closest point algorithm can be replaced by the NDT (Normal Distribution Transform) registration; the back-projection reconstruction algorithm can be replaced by the filtered back-projection or iterative reconstruction algorithm.

[0038] Combined with the first aspect, in some implementation manners of the first aspect, the dimensionality reduction through principal component analysis in step S4-4 includes: Step S4-4-1: Input the frequency-domain feature vector, deformation gradient vector, and tomographic imaging map into the covariance matrix calculation module to generate the covariance matrix of the multi-source data; Step S4-4-2: Perform eigenvalue decomposition on the covariance matrix, select the first three principal components with the highest contribution rate, and generate the reduced-dimensional feature vector.

[0039] In step S4-4-1, the covariance matrix calculation module receives the frequency-domain feature vector, the deformation gradient vector, and the tomography map, calculates the covariance matrix between the multi-source data, and quantifies the correlation between each feature dimension. In step S4-4-2, the eigenvalue decomposition module decomposes the covariance matrix, sorts the eigenvalues, and selects the top three principal components with a cumulative contribution rate of more than 85% to generate the reduced-dimensional feature vector. For example, the main frequency component in the frequency-domain feature vector and the pile inclination angle in the deformation gradient vector may show a high correlation. Principal component analysis integrates them into a single-dimensional feature to reduce data redundancy.

[0040] The technical effects are as follows: The covariance matrix reveals the internal relationship of multi-source data, eigenvalue decomposition retains key information, and the reduced-dimensional feature vector improves the model training efficiency. For example, if the response frequency domain feature is highly correlated with the deformation gradient, principal component analysis combines them into a comprehensive damage index to avoid model overfitting; setting the contribution rate threshold balances the data compression rate and information integrity to meet the monitoring requirements of different wharves.

[0041] In an alternative solution, the contribution rate threshold can be adjusted to 80% or 90%; the number of principal components can be dynamically adjusted according to the actual data characteristics.

[0042] Combined with the first aspect, in some implementation manners of the first aspect, in step S4-3, analyzing the geometric parameters and position information of internal defects includes: According to the relationship between the acoustic wave propagation time and the amplitude attenuation, reconstruct the three-dimensional spatial coordinates and the equivalent aperture of the internal defect.

[0043] In step S4-3, the back-projection reconstruction algorithm module calculates the acoustic wave propagation time between the transmitter and the receiver according to the difference in the acoustic wave propagation speed in the concrete, and combines the amplitude attenuation coefficient (negatively correlated with the defect size) to reconstruct the three-dimensional spatial coordinates of the internal defect. The equivalent aperture is calculated by the acoustic wave diffraction effect and reflects the equivalent size of the defect. For example, cracks will cause the acoustic wave propagation path to extend and the amplitude attenuation to increase. Through cross-verification of multiple groups of sensor signals, the crack depth and width can be accurately calculated; the hole defect is manifested as the absence of the acoustic wave reflection signal, and its spatial position is determined by triangulation.

[0044] The technical effects are as follows: The two-parameter analysis of the acoustic wave propagation time and the amplitude attenuation improves the defect detection accuracy, and the equivalent aperture quantifies the severity of the defect. For example, the difference in propagation time can distinguish surface cracks from deep cavities, and the amplitude attenuation coefficient combined with the equivalent aperture evaluates the crack propagation risk, providing a quantitative basis for maintenance decisions.

[0045] In an alternative solution, the acoustic wave parameters can be increased by frequency component analysis; the equivalent aperture calculation can adopt the geometric diffraction theory or numerical simulation methods.

[0046] In combination with the first aspect, in some implementation manners of the first aspect, step S5 includes: Step S5-1: Input the multi-source heterogeneous data into a convolutional neural network, extract spatial local features through multiple layers of convolutional kernels, and generate a damage-sensitive feature map; Step S5-2: Input the damage-sensitive feature map into a long short-term memory network, analyze the time series change rule, and generate a damage feature vector that fuses spatio-temporal correlations.

[0047] In step S5-1, the convolutional neural network receives the multi-source heterogeneous data as input and extracts spatial local features layer by layer through multiple layers of convolutional kernels. For example, the first layer of convolutional kernels can identify local mutations in strain data (such as stress concentration at the crack tip), the second layer of convolutional kernels can capture macroscopic displacement patterns in deformation data (such as the inclination trend of the pile cap), and the third layer of convolutional kernels can analyze the defect geometry in the tomography map (such as the crack length and direction). In step S5-2, the long short-term memory network receives the damage-sensitive feature map output by the convolutional neural network and analyzes the time series correlation rule through time step expansion. For example, the strain accumulation effect caused by overloading and stacking shows an increasing gradient in the feature map, while ship impact presents an instantaneous pulse feature. The long short-term memory network selects key time series nodes through a gating mechanism and generates a damage feature vector that fuses spatio-temporal correlations.

[0048] The technical effect is reflected in that the spatial feature extraction ability of the convolutional neural network is combined with the time series modeling ability of the long short-term memory network to achieve accurate analysis of complex damage patterns. For example, spatial features can locate the crack propagation area, and time series features can distinguish instantaneous impacts from long-term deterioration. After the two are fused, the damage development stage (such as initial micro-cracks or severe fractures) can be accurately determined.

[0049] In an alternative solution, the convolutional neural network can be replaced by a visual model based on the Transformer architecture, and the long short-term memory network can be replaced by a temporal convolutional network; the generation of the damage-sensitive feature map can use an attention mechanism to enhance the feature weights of key regions.

[0050] In combination with the first aspect, in some implementation manners of the first aspect, in step S6, the generation of the dynamically updated damage threshold library includes: Step S6-1: Train the initial damage threshold library with the sample data in the historical damage database to generate benchmark matching parameters; Step S6-2: Input the first data set, the second data set, and the third data set into the adaptive update module, and dynamically adjust the benchmark matching parameters based on the real-time collected structural strain data, surface deformation data, and ultrasonic tomography raw signal data to generate a dynamically updated damage threshold library.

[0051] In step S6-1, the historical damage database contains sample data of various typical damage conditions, such as ship impact, overloading and stacking, material corrosion, etc. The initial damage threshold library is generated through supervised learning training, and the benchmark matching parameters include the confidence threshold of the damage position, the classification boundary of the damage level, etc. In step S6-2, the adaptive update module receives the first data set (strain data), the second data set (deformation data), and the third data set (defect data) collected in real time, and dynamically adjusts the benchmark matching parameters through an online learning algorithm (such as an incremental support vector machine). For example, in long-term monitoring, it is found that the elastic modulus of the material decreases due to corrosion, and the adaptive module will reduce the absolute value of the strain threshold to match the structural state after performance degradation.

[0052] The technical effect is reflected in that the dynamic threshold library adapts to the performance evolution and environmental changes of the wharf structure through online learning, avoiding false alarms or missed alarms caused by the failure of the static threshold due to the benchmark. For example, when the strain baseline drifts due to material aging, the dynamically adjusted threshold can still accurately identify abnormal strains; new damage modes (such as new ship impacts) can be incorporated into the threshold library through incremental learning, improving the generalization ability of the system.

[0053] In an alternative solution, the online learning algorithm can adopt random forest incremental training or Bayesian probability model; the benchmark matching parameters can be extended to distance metric parameters in a multi-dimensional feature space.

[0054] Combined with the first aspect, in some implementation manners of the first aspect, the deployment of the distributed optical fiber sensing network in step S1 includes: Mark two measuring lines symmetrically along the axial direction on the surface of the pile foundation, one measuring line is located on the side facing the water area, and the other measuring line is located on the side facing the slope, generating measuring line positioning information; Lay the distributed optical fiber along the measuring line positioning information and fix it on the surface of the pile body in three sections through underwater epoxy resin, which are located in the pile top, pile middle and pile bottom areas respectively, forming an optical fiber sensing network for continuous monitoring of the whole pile length.

[0055] In the measuring line positioning stage, the measuring line on the side facing the water area is arranged on the water-facing side of the pile foundation to monitor the dynamic strain caused by ship berthing and wave impact; the measuring line on the side facing the slope is arranged on the back water side of the pile foundation to detect the static strain offset caused by the settlement or slip of the slope soil body. After the distributed optical fiber is laid along the measuring line, it is fixed in three sections by underwater epoxy resin: high-viscosity epoxy resin is used in the pile top area to resist water flow scouring, flexible epoxy resin is used in the pile middle area to adapt to the micro-bending deformation of the pile body, and corrosion-resistant epoxy resin is used in the pile bottom area to prevent marine organisms from attaching. The optical fiber sensing network for continuous monitoring of the whole pile length covers the area from above water to below water, and realizes multi-point parallel data acquisition through wavelength division multiplexing technology.

[0056] The technical effects are as follows: The symmetric layout of survey lines enhances data comparability, and the segmented fixation strategy improves the environmental adaptability of the sensor network. For example, the survey line facing the water area can preferentially capture ship impact signals, and the segmented fixation design avoids the fiber optic cable detachment caused by pile deformation; the wavelength division multiplexing technology supports a high-density measuring point layout and realizes high-resolution monitoring of strain distribution.

[0057] In an alternative solution, the number of survey lines can be extended to four (cross-symmetric layout), but the hardware cost needs to be increased; the underwater epoxy resin can be replaced by mechanical clamps for fixation, but stress concentration interference may be introduced.

[0058] Combined with the first aspect, in some implementation manners of the first aspect, the training process of the deep learning model includes: Generating virtual training samples by finite element simulation of the structural strain data, surface deformation data, and ultrasonic tomography original signal data of the high-pile wharf under different damage conditions; Integrating the virtual training samples with historical monitoring data and on-site loading test data to construct a training sample set covering multiple scenarios; Inputting the training sample set into the generative adversarial network to generate an enhanced training set; Training a convolutional neural network and a long short-term memory network through the enhanced training set to generate a deep learning model.

[0059] In the finite element simulation stage, a parametric model of the high-pile wharf is established, and virtual training samples are generated by changing the load types (concentrated force, distributed force), damage positions (pile top, pile middle, pile bottom), and damage sizes (crack length, aperture). The historical monitoring data is used to screen typical damage events (such as the caisson displacement data after a typhoon), and the on-site loading test applies a stepped load and records the structural response to construct a training sample set covering multiple scenarios. The generative adversarial network synthesizes rare damage patterns (such as pile bottom explosion and silt replacement damage) through the generator, and the discriminator distinguishes real and synthetic data to generate a diversity-enhanced training set. The convolutional neural network and the long short-term memory network adopt end-to-end joint training, and the loss function combines the spatial reconstruction error and the temporal prediction error to optimize the model parameters.

[0060] The technical effects are as follows: The virtual training samples fill the shortage of measured data, and the generative adversarial enhancement improves the model's recognition ability for rare damage. For example, the finite element simulation can generate damage data in the underwater area at the pile bottom (a traditional detection blind spot), and the extreme condition data (such as seismic action) synthesized by the generative adversarial network enhances the model's robustness.

[0061] In an alternative solution, the finite element simulation can be replaced by a discrete element model; the generative adversarial network can be replaced by a variational autoencoder; the joint training strategy can be changed to staged pre-training and fine-tuning.

[0062] Embodiment 2

[0063] As Figure 4 shown, in a second aspect, the present invention provides a high-piled wharf structure damage identification system. The system adopts the method provided in any of the above embodiments. The system includes: A data acquisition module, including a distributed optical fiber sensing network, a three-dimensional laser scanner, and an ultrasonic tomography scanner, is used to collect the structural strain data, temperature data, surface deformation data, and ultrasonic tomography original signal data of the high-piled wharf in real time, and generate a first data set, a second data set, and a third data set respectively; A data processing module, connected to the data acquisition module, is used to perform feature extraction and normalization processing on the first data set, the second data set, and the third data set, including dimensionality reduction through principal component analysis, using a normalization algorithm to eliminate the dimension difference, and aligning the feature vectors based on the geographic information system coordinates to generate multi-source heterogeneous data; A model processing module, connected to the data processing module, includes a pre-trained deep learning model. The deep learning model is composed of a convolutional neural network and a long short-term memory network, and is used to perform spatial feature extraction and temporal correlation analysis on the multi-source heterogeneous data to generate a fused damage feature vector; A damage identification module, connected to the model processing module, includes a dynamically updated damage threshold library and a matching calculation unit, and is used to perform similarity matching based on the damage feature vector and output the damage location and damage level.

[0064] The identification system provided by the embodiment of the present invention corresponds to the identification method provided in Embodiment 1. Multi-source data (strain, deformation, defect) are integrated into spatiotemporally aligned multi-source heterogeneous data, providing high-dimensional input for the deep learning model; the damage feature vector output by the model is matched with the dynamic threshold library to achieve a full-link closed loop from "data acquisition - feature fusion - damage decision".

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying the damage of a high-piled wharf structure, characterized in that, include: Step S1: collecting structural strain data and temperature data of the high-pile wharf in real time based on a distributed optical fiber sensing network to generate a first data set; Step S2: acquiring surface deformation data of the high-pile wharf by means of a three-dimensional laser scanner to generate a second data set; Step S3: using an ultrasonic tomography scanner to perform nondestructive testing on the internal structure of the high-pile wharf, obtaining ultrasonic tomography raw signal data, and generating a third data set; Step S4: extracting features and standardizing the first data set, the second data set, and the third data set, including dimensionality reduction through principal component analysis, eliminating dimensional differences using a normalization algorithm, and aligning feature vectors based on geographic information system coordinates to generate multi-source heterogeneous data; Step S5: inputting the multi-source heterogeneous data into a pre-trained deep learning model, extracting spatial features through a convolutional neural network, and analyzing temporal associations through a long short-term memory network to generate a fused damage feature vector; Step S6: Based on the damage feature vector, similarity matching is performed in combination with the dynamically updated damage threshold library to output the damage location and damage level.

2. The method for identifying structural damage of a high-piled wharf according to claim 1, characterized in that Step S1 includes: Step S1-1: collecting strain distribution data and ambient temperature change data on the surface of the pile foundation through distributed optical fiber sensors arranged symmetrically along the axial direction of the pile foundation to generate an original strain data set; Step S1-2: inputting the original strain data set into a Kalman filter module, removing noise interference data, and generating a first data set; Step S2 includes: Step S2-1: Scan the pile cap, pile body and panel of the high-pile wharf at multiple angles using a three-dimensional laser scanner to obtain surface deformation point cloud data; Step S2-2: input the surface deformation point cloud data into a geographic information system coordinate conversion module, align it to a unified spatial reference, and generate a second data set.

3. A method for identifying structural damage of a high-pile wharf according to claim 1, characterized in that Step S4 includes: Step S4-1: input the structural strain data in the first data set into the Hilbert-Huang transform module, decompose it into multiple intrinsic mode functions, extract the main frequency component and transient energy characteristics, and generate a frequency domain feature vector; Step S4-2: inputting the surface deformation data in the second data set into an iterative closest point algorithm module, removing noise points and performing registration based on the geographic information system coordinates, extracting the pile cap displacement and the pile body inclination angle, and generating a deformation gradient vector; Step S4-3: inputting the ultrasonic tomography raw signal data in the third data set into the back-projection reconstruction algorithm module, analyzing the geometric parameters and position information of the internal defects, and generating a tomographic imaging map; Step S4-4: Input the frequency domain feature vector, the deformation gradient vector and the tomographic imaging spectrum into a normalization module, eliminate the dimension difference, perform dimensionality reduction through principal component analysis, and generate standardized multi-source heterogeneous data.

4. A method for identifying structural damage of a high-piled wharf according to claim 3, characterized in that, The dimensionality reduction by principal component analysis in step S4-4 includes: Step S4-4-1: inputting the frequency domain feature vector, the deformation gradient vector and the tomographic imaging spectrum into a covariance matrix calculation module to generate a covariance matrix of multi-source data; Step S4-4-2: Perform eigenvalue decomposition on the covariance matrix, select the top three principal components with the highest contribution rate, and generate the eigenvectors after dimensionality reduction.

5. A method for identifying structural damage of a high-piled wharf according to claim 3, characterized in that, In step S4-3, the parsing of the geometric parameters and position information of internal defects includes: According to the relationship between the acoustic wave propagation time and amplitude attenuation, reconstruct the three-dimensional spatial coordinates and equivalent aperture of internal defects.

6. The method for identifying structural damage of a high-piled wharf according to claim 1, wherein, The said step S5 includes: Step S5-1: Input the multi-source heterogeneous data into a convolutional neural network, extract spatial local features through multiple convolutional kernels, and generate a damage-sensitive feature map. Step S5-2: Input the damage-sensitive feature map into a long short-term memory network, analyze the temporal variation law, and generate a damage feature vector that fuses spatio-temporal correlations.

7. A method for identifying structural damage of a high-piled wharf according to claim 1, characterized in that, In step S6, the generation of the dynamically updated damage threshold library includes: Step S6-1: Train the initial damage threshold library with the sample data in the historical damage database to generate benchmark matching parameters. Step S6-2: Input the first data set, the second data set, and the third data set into an adaptive update module, and dynamically adjust the benchmark matching parameters based on the real-time collected structural strain data, surface deformation data, and ultrasonic tomography raw signal data to generate a dynamically updated damage threshold library.

8. A method for identifying structural damage of a high-pile wharf according to claim 1, characterized in that, The deployment of the distributed optical fiber sensing network in step S1 includes: Mark two measurement lines symmetrically along the axial direction on the surface of the pile foundation, one measurement line is located on the side facing the water area, and the other measurement line is located on the side facing the slope, to generate measurement line positioning information. Lay the distributed optical fiber along the measurement line positioning information and fix it on the surface of the pile body in three sections through underwater epoxy resin, which are located in the pile top, pile middle, and pile bottom regions respectively, to form an optical fiber sensing network for continuous monitoring of the entire pile length.

9. A method for identifying structural damage of a high-piled wharf according to claim 1, characterized in that, The training process of the said deep learning model includes: Generate virtual training samples through finite element simulation of the structural strain data, surface deformation data, and ultrasonic tomography raw signal data of high-pile wharves under different damage conditions. Integrate the virtual training samples with historical monitoring data and on-site loading test data to construct a training sample set covering multiple scenarios. Input the training sample set into an adversarial generation network to generate an enhanced training set. Train the convolutional neural network and the long short-term memory network through the enhanced training set to generate the deep learning model.

10. A high-pile wharf structure damage identification system, characterized in that, The said system adopts the method described in any one of claims 1 to 9, and the system includes: A data acquisition module, including a distributed optical fiber sensing network, a three-dimensional laser scanner, and an ultrasonic tomography scanner, for real-time collecting the structural strain data, temperature data, surface deformation data, and ultrasonic tomography raw signal data of the high-pile wharf, and generating the first data set, the second data set, and the third data set respectively. A data processing module, connected to the data acquisition module, for performing feature extraction and normalization processing on the first data set, the second data set, and the third data set, including dimensionality reduction through principal component analysis, using a normalization algorithm to eliminate the dimension difference, and aligning the feature vectors based on the geographic information system coordinates to generate multi-source heterogeneous data. The model processing module, connected to the data processing module, includes a pre-trained deep learning model composed of a convolutional neural network and a long short-term memory network, and is used to extract spatial features and perform temporal correlation analysis on the multi-source heterogeneous data to generate a fused damage feature vector; The damage recognition module, connected to the model processing module, includes a dynamically updated damage threshold library and a matching calculation unit, and is used to perform similarity matching based on the damage feature vector and output the damage location and damage level.

Images