Bridge structure stability intelligent detection method and system based on DAS
By combining distributed optical fiber sensing devices and random subspace recognition methods with support vector machine models, the problems of low accuracy and insufficient intelligent recognition of bridge structural modal parameters were solved, realizing automated and intelligent detection of bridge structural stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAIBEI HANGRUI MECHANICAL & ELECTRICAL EQUIPMENT CO LTD
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-09
AI Technical Summary
Among existing bridge monitoring technologies, distributed optical fiber acoustic sensing (DAS) technology has low accuracy in identifying bridge structural modal parameters and lacks intelligent identification and early warning capabilities, making it difficult to achieve full-bridge monitoring and structural stability assessment.
Distributed fiber optic sensing devices are used to acquire bridge vibration signals. Modal analysis is performed through noise reduction and random subspace identification. Combined with a pre-trained support vector machine model, intelligent evaluation is carried out to achieve automated identification of bridge structural stability.
It significantly improves the accuracy of modal parameter identification, realizes automated and intelligent identification of bridge structural stability, reduces manual interpretation, and improves efficiency.
Smart Images
Figure CN122171137A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of bridge health monitoring technology, specifically to a method and system for intelligent detection of bridge structural stability based on DAS. Background Technology
[0002] With the rapid development of transportation networks, large bridges play a crucial role in urban traffic and regional economies, and their structural safety is directly related to public safety and socio-economic stability. As bridges age, fatigue, and damage become increasingly prominent issues. To prevent sudden structural failures, bridge health monitoring technology has gradually become an important research direction in the field of infrastructure operation and maintenance. Existing bridge monitoring technologies mainly rely on traditional electrical devices such as accelerometers, strain gauges, or displacement sensors. These systems are mostly deployed at discrete points, only acquiring structural response information at limited locations, making it difficult to achieve full-bridge monitoring. Furthermore, the limited number of sensors results in insufficient spatial resolution, failing to comprehensively capture the dynamic changes of the structure in different areas.
[0003] In recent years, the development of Distributed Acoustic Sensing (DAS) technology has provided a new solution for structural monitoring. DAS utilizes optical fiber as the sensing medium, detecting changes in backscattered signals caused by external vibrations affecting light within the fiber. This enables continuous distributed vibration sensing along the fiber path, offering significant advantages such as high spatial resolution, long-distance coverage, resistance to electromagnetic interference, and ease of deployment. It can simultaneously acquire data from thousands of equivalent monitoring points on a single fiber, making it ideal for dynamic response monitoring of large structures such as bridges.
[0004] Although DAS technology provides a new means for bridge health monitoring, it still faces some challenges in practical applications. For example, vibration signals acquired by DAS are often mixed with a large amount of environmental noise and non-structural interference, making it difficult to extract effective signals and resulting in low accuracy in modal parameter identification. In addition, existing bridge condition assessment methods mostly rely on empirical analysis or single characteristic indicators, lacking the ability to comprehensively judge multi-dimensional modal information, making it difficult to achieve intelligent and automated identification and early warning of structural stability. Summary of the Invention
[0005] The purpose of this invention is to solve the problems of low accuracy in identifying bridge structural modal parameters and lack of intelligent identification of structural state mentioned in the background art, and to propose a bridge structural stability intelligent detection method and system based on DAS.
[0006] A first aspect of this invention provides a method for intelligent detection of bridge structural stability based on DAS, the method comprising: Vibration signals from multiple monitoring locations on the bridge are acquired using distributed fiber optic sensing devices. The first vibration signal is denoised to obtain the second vibration signal; the first vibration signal is the vibration signal of any one of the multiple monitoring positions. A multi-channel vibration data matrix is established based on the second vibration signals from multiple monitoring locations. Based on the multi-channel vibration data matrix, modal analysis is performed using the random subspace identification method to obtain the real-time modal parameters of the bridge structure; Modal characteristic data are calculated based on the reference modal parameters of the bridge structure and the real-time modal parameters. The modal feature data is input into a pre-trained detection model to obtain the bridge state detection results.
[0007] Optionally, the step of denoising the first vibration signal to obtain the second vibration signal includes: The first vibration signal is modally decomposed using a fully ensemble empirical mode decomposition adaptive noise algorithm to obtain multiple intrinsic mode function components and a residual term; Calculate the sample entropy and permutation entropy for each intrinsic mode function component; The evaluation score is calculated based on the sample entropy and permutation entropy of each intrinsic mode function component. Based on the evaluation score, each intrinsic mode function component is classified into two categories: effective components and noisy components, thus obtaining the effective component set and the first noisy component set. Calculate the peak cross-correlation coefficient of any two noise components in the first noise component set, and merge the two noise components whose peak cross-correlation coefficient is greater than a preset merging threshold to obtain the second noise component set. Wavelet denoising is performed on each noise component in the second noise component set to obtain the denoised noise component. Based on the effective component, the noise-mixed component after noise reduction, and the residual term, the signal is reconstructed to obtain the vibration signal of the second target.
[0008] Optionally, the step of calculating the evaluation score based on the sample entropy and permutation entropy of each intrinsic mode function component, and classifying each intrinsic mode function component into two categories—effective components and noisy components—based on the evaluation score includes: The evaluation score is calculated based on the sample entropy and permutation entropy of each intrinsic mode function component: Among them, C i SE i and PE i These are the evaluation score, sample entropy, and permutation entropy of the i-th intrinsic mode function component, respectively; SE max and PE max These are the maximum values of the sample entropy and permutation entropy of each intrinsic mode function component, respectively; Calculate the average of the evaluation scores for each intrinsic mode function component. and standard deviation ; The screening threshold thr is calculated based on the mean and standard deviation of the evaluation scores: ; Intrinsic mode function components with evaluation scores lower than the screening threshold are classified as valid components; otherwise, they are classified as noisy components.
[0009] Optionally, the step of performing wavelet denoising on each noise component in the second noise component set to obtain the denoised noise component includes: Wavelet decomposition is performed on the target noise-mixing component to obtain detail coefficients and approximation coefficients; the target noise-mixing component is any noise-mixing component in the second noise-mixing component set. An improved thresholding function is used to threshold the detail coefficients: Where w is the detail coefficient; These are the detail coefficients after thresholding. It is the threshold corresponding to the detail coefficient; sgn() is the sign function; exp() is an exponential function with the natural constant e as the base; These are preset adjustment parameters; The approximation coefficients and the detail coefficients after thresholding are subjected to inverse wavelet transform to obtain the denoised target noise component.
[0010] Optionally, the detection model includes a two-level support vector machine; wherein: The first-level support vector machine is a single-class support vector machine, used to determine whether the bridge structure is normal or abnormal; during its training process, it is trained using only normal sample data. The second-level support vector machine is a multi-class support vector machine, used to determine the damage state of the bridge when the first-level support vector machine classifies the bridge structure as abnormal. During its training, it uses abnormal sample data with various damage states.
[0011] A second aspect of this invention provides a DAS-based intelligent detection system for bridge structural stability, the system comprising: The data acquisition module is used to acquire vibration signals from multiple monitoring locations on the bridge through a distributed optical fiber sensing device. The preprocessing module is used to denoise the first vibration signal to obtain the second vibration signal; the first vibration signal is the vibration signal of any one of the multiple monitoring positions. The data fusion module is used to establish a multi-channel vibration data matrix based on the second vibration signals from multiple monitoring locations; The modal analysis module is used to perform modal analysis based on the multi-channel vibration data matrix using the random subspace identification method to obtain the real-time modal parameters of the bridge structure. The feature extraction module is used to calculate modal feature data based on the reference modal parameters of the bridge structure and the real-time modal parameters; The state detection module is used to input the modal feature data into a pre-trained detection model to obtain the bridge state detection results.
[0012] Optionally, the preprocessing module includes: The decomposition module is used to perform mode decomposition on the first vibration signal using a fully ensemble empirical mode decomposition adaptive noise algorithm to obtain multiple intrinsic mode function components and a residual term; The entropy calculation module is used to calculate the sample entropy and permutation entropy of each intrinsic mode function component; The classification module is used to calculate the evaluation score of each intrinsic mode function component based on the sample entropy and permutation entropy, and classify each intrinsic mode function component into two categories, effective components and noisy components, based on the evaluation score, thus obtaining the effective component set and the first noisy component set. The merging module is used to calculate the peak cross-correlation coefficient of any two noise components in the first noise component set, and merge two noise components whose peak cross-correlation coefficient is greater than a preset merging threshold to obtain a second noise component set. The component noise reduction module is used to perform wavelet noise reduction on each noise component in the second noise component set to obtain the noise-reduced noise component. The reconstruction module is used to reconstruct the signal based on the effective component, the noise-reduced mixed component, and the residual term to obtain the second target vibration signal.
[0013] Optionally, the classification module includes: The comprehensive evaluation module is used to calculate the evaluation score based on the sample entropy and permutation entropy of each intrinsic mode function component: Among them, C i SE i and PE i These are the evaluation score, sample entropy, and permutation entropy of the i-th intrinsic mode function component, respectively; SE max and PE max These are the maximum values of the sample entropy and permutation entropy of each intrinsic mode function component, respectively; The threshold determination module is used to calculate the average evaluation score of each intrinsic mode function component. and standard deviation The screening threshold thr is calculated based on the average and standard deviation of the evaluation scores. ; The filtering module is used to classify intrinsic mode function components with evaluation scores less than the filtering threshold as valid components, and otherwise as noisy components.
[0014] Optionally, the component noise reduction module includes: The wavelet decomposition module is used to perform wavelet decomposition on the target noise-mixing component to obtain detail coefficients and approximation coefficients; the target noise-mixing component is any noise-mixing component in the second noise-mixing component set. The thresholding module is used to perform thresholding on detail coefficients using an improved thresholding function. Where w is the detail coefficient; These are the detail coefficients after thresholding. It is the threshold corresponding to the detail coefficient; sgn() is the sign function; exp() is an exponential function with the natural constant e as the base; These are preset adjustment parameters; The inverse transform module is used to perform wavelet inverse transform on the approximation coefficients and the detail coefficients after thresholding to obtain the denoised target noise-mixing component.
[0015] Optionally, the detection model includes a two-level support vector machine; wherein: The first-level support vector machine is a single-class support vector machine, used to determine whether the bridge structure is normal or abnormal; during its training process, it is trained using only normal sample data. The second-level support vector machine is a multi-class support vector machine, used to determine the damage state of the bridge when the first-level support vector machine determines that the bridge structure is abnormal; during its training process, it is trained using abnormal sample data of various damage states.
[0016] The beneficial effects of this invention are: By using distributed fiber optic sensing technology, high-density vibration signals from multiple locations on the bridge are collected. High-quality data is obtained through noise reduction processing, and modal analysis is performed using the random subspace identification method, which significantly improves the accuracy of modal parameter identification. Combined with a pre-trained detection model, the modal features are intelligently evaluated, realizing automated and intelligent identification of bridge structural stability, reducing manual interpretation and improving efficiency. Attached Figure Description
[0017] Figure 1 A flowchart illustrating a bridge structure stability intelligent detection method based on DAS provided in this embodiment of the invention; Figure 2 This is an architecture diagram of a bridge structure stability intelligent detection system based on DAS provided in an embodiment of the present invention. Detailed Implementation
[0018] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0019] This invention provides a method for intelligent detection of bridge structural stability based on DAS. See also... Figure 1 , Figure 1 A flowchart illustrating a DAS-based intelligent detection method for bridge structural stability, provided as an embodiment of the present invention. The method includes the following steps: S101 acquires vibration signals from multiple monitoring locations on the bridge through a distributed optical fiber sensing device.
[0020] S102, the first vibration signal is denoised to obtain the second vibration signal.
[0021] S103, establish a multi-channel vibration data matrix based on the second vibration signals from multiple monitoring locations.
[0022] S104. Based on the multi-channel vibration data matrix, modal analysis is performed using the random subspace identification method to obtain the real-time modal parameters of the bridge structure.
[0023] S105. Modal characteristic data are calculated based on the reference modal parameters and real-time modal parameters of the bridge structure.
[0024] S106. Input the modal feature data into the pre-trained detection model to obtain the bridge state detection results.
[0025] The first vibration signal is the vibration signal from any one of the multiple monitoring locations. Modal parameters include natural frequency, damping ratio, and modal shape. Modal characteristic data include frequency variation, damping ratio variation, and modal shape similarity. The reference modal parameters are those measured under healthy bridge conditions.
[0026] The present invention provides an intelligent bridge structural stability detection method based on DAS, which uses distributed optical fiber sensing technology to collect high-density vibration signals from multiple locations on the bridge, obtains high-quality data through noise reduction processing, and performs modal analysis using the random subspace identification method, significantly improving the accuracy of modal parameter identification. Combined with a pre-trained detection model, the modal features are intelligently evaluated, realizing automated and intelligent identification of bridge structural stability, reducing manual interpretation and improving efficiency.
[0027] In one implementation, optical fibers are continuously laid along the bottom / side of the main beam, covering the entire span, with a spatial resolution of 1m. At the connection point between the support and the main beam, the optical fiber is locally densified (spatial resolution can be set to 0.5m) to capture sudden changes in vibration frequency caused by support loosening or aging. Using optical fiber + DAS equipment deployed along the bridge, vibration waveforms along the line are collected in real time, and the signals are intelligently analyzed at the edge or in the cloud to achieve graded alarms for bridge structural stability.
[0028] In one implementation, corresponding baseline modal parameters are established for different temperature scenarios. When calculating modal feature data in real time, these baseline modal parameters at the corresponding temperature are used. Bridge modal parameters naturally drift with temperature changes. By establishing corresponding baseline modal parameters for different temperature scenarios, the appropriate baseline can be automatically matched during real-time analysis, effectively eliminating temperature-induced frequency drift or damping fluctuations, and improving the stability and accuracy of damage identification.
[0029] In one embodiment, step S102, which involves denoising the first vibration signal to obtain the second vibration signal, includes: Step 1: The first vibration signal is modally decomposed using the Complete Ensemble Empirical Mode Decomposition Adaptive Noise Algorithm (CEEMDAN), yielding multiple intrinsic mode function components and a residual term. Compared to traditional EMD and EEMD, CEEMD introduces adaptive noise ensemble averaging during the decomposition process, effectively reducing mode aliasing and making each intrinsic mode function component correspond to a clearer frequency band.
[0030] Step two involves calculating the sample entropy and permutation entropy for each intrinsic mode function component. Sample entropy reflects the complexity and regularity of the time series; permutation entropy measures the randomness of local sequence variations in the signal. Combining the two can more accurately distinguish between structural vibrations (which have strong regularity) and noise (which has high randomness).
[0031] Step 3: Calculate the evaluation score of each intrinsic mode function component based on the sample entropy and permutation entropy. Based on the evaluation score, classify each intrinsic mode function component into two categories: effective components and noisy components, and obtain the effective component set and the noisy component set (denoted as the first noisy component set).
[0032] Step four: Calculate the peak cross-correlation coefficient of any two noise components in the first noise component set. Merge the two noise components whose peak cross-correlation coefficient is greater than a preset merging threshold to obtain the second noise component set. Noise components often exhibit correlation; correlation analysis can merge redundant components, reduce redundant noise interference, and decrease computational complexity.
[0033] Step 5: Perform wavelet denoising on each noise component in the second noise component set to obtain the denoised noise component.
[0034] Step 6: Based on the effective component, the noise-reduced mixed component, and the residual term, the signal is reconstructed to obtain the vibration signal of the second target.
[0035] This embodiment achieves adaptive and refined denoising of vibration signals through a multi-stage processing approach, including a fully ensemble empirical mode decomposition adaptive noise algorithm, entropy feature selection, correlation analysis, and wavelet denoising. This method automatically distinguishes between effective vibration-dominant components and noise-dominant components based on signal complexity, effectively suppressing mode aliasing and information loss during denoising. The reconstructed vibration signal not only preserves the true dynamic characteristics of the structure to the greatest extent but also significantly improves the signal-to-noise ratio and stability, providing high-quality input data for subsequent modal analysis and bridge condition identification.
[0036] In one implementation, an evaluation score is calculated based on the sample entropy and permutation entropy. Based on the evaluation score, each intrinsic mode function component is categorized into two types: effective components and noisy components. Step 1: Calculate the evaluation score based on the sample entropy and permutation entropy of each intrinsic mode function component: Among them, C i SE i and PE i These are the evaluation score, sample entropy, and permutation entropy of the i-th intrinsic mode function component, respectively; SE max and PE max These are the maximum values of the sample entropy and permutation entropy of each intrinsic mode function component, respectively.
[0037] Step 2: Calculate the average evaluation score of each intrinsic mode function component. and standard deviation .
[0038] Step 3: Calculate the screening threshold thr based on the average and standard deviation of the evaluation scores. .
[0039] Step four: eigenmode function components with evaluation scores less than the screening threshold are classified as valid components; otherwise, they are classified as noise components.
[0040] This implementation constructs an evaluation score by jointly using sample entropy and permutation entropy, avoiding the subjectivity of relying solely on a single indicator or human experience for classification. Through normalization and adaptive threshold calculation, the classification process has good universality and robustness, thus more accurately distinguishing the effective components of structural vibration from noise-mixed components.
[0041] In one implementation, the preset merging threshold can be set to 0.8. Let r be the peak value of the cross-correlation coefficient between component i and component j. ijIf the similarity is greater than the merging threshold, then component i and component j are said to be similar. If there is chain similarity, such as r... ab and r bc If all values are greater than the merging threshold, then components a, b, and c are grouped together, summed, and merged into a single noisy component to avoid unnecessary cross-correlation calculations caused by multiple pairwise merging.
[0042] In one implementation, wavelet denoising is performed on each noise component in the second noise component set to obtain the denoised noise components, including: Step 1: Perform wavelet decomposition on the target noise component to obtain detail coefficients and approximation coefficients; the target noise component is any noise component in the second noise component set. Specifically, the wavelet basis function can be sym4, and the decomposition is performed at 5 levels.
[0043] Step two: Apply a modified threshold function to the detail coefficients for thresholding. Where w is the detail coefficient; These are the detail coefficients after thresholding. This is the threshold corresponding to the detail coefficients, and a fixed threshold (sqtwolog threshold) can be selected; sgn() is the sign function; exp() is an exponential function with the natural constant e as the base; This is a preset adjustment parameter, which can be set to 0.3.
[0044] Step 3: Perform inverse wavelet transform on the approximation coefficients and the detail coefficients after thresholding to obtain the denoised target noise component.
[0045] Traditional hard thresholding functions are discontinuous at the threshold point, leading to oscillations and pseudo-Gibbs phenomena in the denoised signal. While soft thresholding functions are continuous, they excessively shrink wavelet coefficients, easily causing edge signal distortion. Compared to traditional hard / soft thresholding methods, the improved thresholding function in this embodiment achieves a better balance between edge preservation and continuity, reducing the risk of signal distortion.
[0046] In one embodiment, the detection model comprises a two-level support vector machine. Wherein: The first-level support vector machine is a single-class support vector machine used to determine whether the bridge structure is normal or abnormal. During its training, it is trained using only normal sample data. The second-level support vector machine is a multi-class support vector machine, used to determine the damage state of the bridge when the first-level support vector machine classifies the bridge structure as abnormal, into minor, moderate, and severe damage. During its training, it uses abnormal sample data of various damage states.
[0047] The first-level single-class SVM is trained using only normal samples, effectively establishing the health feature boundaries of bridge structures. Even when abnormal samples are scarce or difficult to obtain comprehensively, it can still accurately identify anomalies, reducing the false alarm rate. The second-level multi-class SVM further subdivides different degrees of damage based on abnormal samples, achieving multi-level classifications such as mild, moderate, and severe damage. This makes the detection results more engineering-guided and facilitates the development of graded maintenance strategies. The two-level structure separates "normal / abnormal judgment" from "damage grading," reducing class confusion from a single model and improving the model's adaptability in complex environments.
[0048] This invention provides a method for intelligent detection of bridge structural stability based on DAS. See also... Figure 2 , Figure 2 This is an architecture diagram of a bridge structural stability intelligent detection system based on DAS, provided as an embodiment of the present invention. The system includes: The data acquisition module is used to acquire vibration signals from multiple monitoring locations on the bridge through a distributed optical fiber sensing device.
[0049] The preprocessing module is used to denoise the first vibration signal to obtain the second vibration signal.
[0050] The data fusion module is used to establish a multi-channel vibration data matrix based on the second vibration signals from multiple monitoring locations.
[0051] The modal analysis module is used to perform modal analysis based on the multi-channel vibration data matrix using the random subspace identification method to obtain the real-time modal parameters of the bridge structure.
[0052] The feature extraction module is used to calculate modal feature data based on the reference modal parameters and real-time modal parameters of the bridge structure.
[0053] The state detection module is used to input modal feature data into a pre-trained detection model to obtain bridge state detection results.
[0054] The first vibration signal is the vibration signal from any one of the multiple monitoring locations.
[0055] The present invention provides a bridge structural stability intelligent detection system based on DAS, which realizes high-density vibration signal acquisition at multiple locations of the bridge through distributed optical fiber sensing technology, obtains high-quality data through noise reduction processing, and performs modal analysis using the random subspace identification method, which significantly improves the accuracy of modal parameter identification. Combined with a pre-trained detection model, the system performs intelligent evaluation of modal features, thereby realizing automated and intelligent identification of bridge structural stability.
[0056] In one embodiment, the preprocessing module includes: The decomposition module is used to perform modal decomposition on the first vibration signal using a fully ensemble empirical mode decomposition adaptive noise algorithm, obtaining multiple intrinsic mode function components and a residual term.
[0057] The entropy calculation module is used to calculate the sample entropy and permutation entropy of each intrinsic mode function component.
[0058] The classification module is used to calculate the evaluation score of each intrinsic mode function component based on the sample entropy and permutation entropy of each component, and classify each component into two categories, effective components and noisy components, based on the evaluation score, thus obtaining the effective component set and the first noisy component set.
[0059] The merging module is used to calculate the peak cross-correlation coefficient of any two noise components in the first noise component set, and merge the two noise components whose peak cross-correlation coefficient is greater than a preset merging threshold to obtain the second noise component set.
[0060] The component noise reduction module is used to perform wavelet noise reduction on each noise component in the second noise component set to obtain the noise-reduced noise component.
[0061] The reconstruction module is used to reconstruct the signal based on the effective component, the noise-reduced mixed component, and the residual term to obtain the second target vibration signal.
[0062] In one implementation, the classification module includes: The comprehensive evaluation module is used to calculate the evaluation score based on the sample entropy and permutation entropy of each intrinsic mode function component: Among them, C i SE i and PE i These are the evaluation score, sample entropy, and permutation entropy of the i-th intrinsic mode function component, respectively; SE max and PE max These are the maximum values of the sample entropy and permutation entropy of each intrinsic mode function component, respectively.
[0063] The threshold determination module is used to calculate the average evaluation score of each intrinsic mode function component. and standard deviation The screening threshold thr is calculated based on the average and standard deviation of the evaluation scores. .
[0064] The filtering module is used to classify intrinsic mode function components with evaluation scores less than the filtering threshold as valid components, and otherwise as noisy components.
[0065] In one implementation, the component noise reduction module includes: The wavelet decomposition module is used to perform wavelet decomposition on the target noise component to obtain detail coefficients and approximation coefficients; the target noise component is any noise component in the second noise component set.
[0066] The thresholding module is used to perform thresholding on detail coefficients using an improved thresholding function. Where w is the detail coefficient; These are the detail coefficients after thresholding. It is the threshold corresponding to the detail coefficient; sgn() is the sign function; exp() is an exponential function with the natural constant e as the base; These are preset adjustment parameters.
[0067] The inverse transform module is used to perform wavelet inverse transform on the approximation coefficients and the detail coefficients after thresholding to obtain the denoised target noise component.
[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall fall within the scope of the present invention.
Claims
1. A smart detection method for bridge structural stability based on DAS, characterized in that, The method includes: Vibration signals from multiple monitoring locations on the bridge are acquired using distributed fiber optic sensing devices. The first vibration signal is denoised to obtain the second vibration signal; the first vibration signal is the vibration signal of any one of the multiple monitoring positions. A multi-channel vibration data matrix is established based on the second vibration signals from multiple monitoring locations. Based on the multi-channel vibration data matrix, modal analysis is performed using the random subspace identification method to obtain the real-time modal parameters of the bridge structure; Modal characteristic data are calculated based on the reference modal parameters of the bridge structure and the real-time modal parameters. The modal feature data is input into a pre-trained detection model to obtain the bridge state detection results.
2. The intelligent detection method for bridge structural stability based on DAS according to claim 1, characterized in that, The step of denoising the first vibration signal to obtain the second vibration signal includes: The first vibration signal is modally decomposed using a fully ensemble empirical mode decomposition adaptive noise algorithm to obtain multiple intrinsic mode function components and a residual term; Calculate the sample entropy and permutation entropy for each intrinsic mode function component; The evaluation score is calculated based on the sample entropy and permutation entropy of each intrinsic mode function component. Based on the evaluation score, each intrinsic mode function component is classified into two categories: effective components and noisy components, thus obtaining the effective component set and the first noisy component set. Calculate the peak cross-correlation coefficient of any two noise components in the first noise component set, and merge the two noise components whose peak cross-correlation coefficient is greater than a preset merging threshold to obtain the second noise component set. Wavelet denoising is performed on each noise component in the second noise component set to obtain the denoised noise component. Based on the effective component, the noise-mixed component after noise reduction, and the residual term, the signal is reconstructed to obtain the vibration signal of the second target.
3. The intelligent detection method for bridge structural stability based on DAS according to claim 2, characterized in that, The process of calculating the evaluation score based on the sample entropy and permutation entropy of each intrinsic mode function component, and classifying each intrinsic mode function component into two categories—effective components and noisy components—based on the evaluation scores includes: The evaluation score is calculated based on the sample entropy and permutation entropy of each intrinsic mode function component: Among them, C i SE i and PE i These are the evaluation score, sample entropy, and permutation entropy of the i-th intrinsic mode function component, respectively; SE max and PE max These are the maximum values of the sample entropy and permutation entropy of each intrinsic mode function component, respectively; Calculate the average of the evaluation scores for each intrinsic mode function component. and standard deviation ; The screening threshold thr is calculated based on the mean and standard deviation of the evaluation scores: ; Intrinsic mode function components with evaluation scores lower than the screening threshold are classified as valid components; otherwise, they are classified as noisy components.
4. The intelligent detection method for bridge structural stability based on DAS according to claim 2, characterized in that, The step of performing wavelet denoising on each noise component in the second noise component set to obtain the denoised noise components includes: Wavelet decomposition is performed on the target noise-mixing component to obtain detail coefficients and approximation coefficients; the target noise-mixing component is any noise-mixing component in the second noise-mixing component set. An improved thresholding function is used to threshold the detail coefficients: Where w is the detail coefficient; These are the detail coefficients after thresholding. It is the threshold corresponding to the detail coefficient; sgn() is the sign function; exp() is an exponential function with the natural constant e as the base; These are preset adjustment parameters; The approximation coefficients and the detail coefficients after thresholding are subjected to inverse wavelet transform to obtain the denoised target noise component.
5. The intelligent detection method for bridge structural stability based on DAS according to claim 1, characterized in that, The detection model comprises a two-level support vector machine; wherein: The first-level support vector machine is a single-class support vector machine, used to determine whether the bridge structure is normal or abnormal; during its training process, it is trained using only normal sample data. The second-level support vector machine is a multi-class support vector machine, used to determine the damage state of the bridge when the first-level support vector machine determines that the bridge structure is abnormal; during its training process, it is trained using abnormal sample data of various damage states.
6. A bridge structural stability intelligent detection system based on DAS, characterized in that, The system includes: The data acquisition module is used to acquire vibration signals from multiple monitoring locations on the bridge through a distributed optical fiber sensing device. The preprocessing module is used to denoise the first vibration signal to obtain the second vibration signal; the first vibration signal is the vibration signal of any one of the multiple monitoring positions. The data fusion module is used to establish a multi-channel vibration data matrix based on the second vibration signals from multiple monitoring locations; The modal analysis module is used to perform modal analysis based on the multi-channel vibration data matrix using the random subspace identification method to obtain the real-time modal parameters of the bridge structure. The feature extraction module is used to calculate modal feature data based on the reference modal parameters of the bridge structure and the real-time modal parameters; The state detection module is used to input the modal feature data into a pre-trained detection model to obtain the bridge state detection results.
7. The intelligent bridge structural stability detection system based on DAS according to claim 6, characterized in that, The preprocessing module includes: The decomposition module is used to perform mode decomposition on the first vibration signal using a fully ensemble empirical mode decomposition adaptive noise algorithm to obtain multiple intrinsic mode function components and a residual term; The entropy calculation module is used to calculate the sample entropy and permutation entropy of each intrinsic mode function component; The classification module is used to calculate the evaluation score of each intrinsic mode function component based on the sample entropy and permutation entropy, and classify each intrinsic mode function component into two categories, effective components and noisy components, based on the evaluation score, thus obtaining the effective component set and the first noisy component set. The merging module is used to calculate the peak cross-correlation coefficient of any two noise components in the first noise component set, and merge two noise components whose peak cross-correlation coefficient is greater than a preset merging threshold to obtain a second noise component set. The component noise reduction module is used to perform wavelet noise reduction on each noise component in the second noise component set to obtain the noise-reduced noise component. The reconstruction module is used to reconstruct the signal based on the effective component, the noise-reduced mixed component, and the residual term to obtain the second target vibration signal.
8. The intelligent bridge structural stability detection system based on DAS according to claim 7, characterized in that, The classification module includes: The comprehensive evaluation module is used to calculate the evaluation score based on the sample entropy and permutation entropy of each intrinsic mode function component: Among them, C i SE i and PE i These are the evaluation score, sample entropy, and permutation entropy of the i-th intrinsic mode function component, respectively; SE max and PE max These are the maximum values of the sample entropy and permutation entropy of each intrinsic mode function component, respectively; The threshold determination module is used to calculate the average evaluation score of each intrinsic mode function component. and standard deviation The screening threshold thr is calculated based on the average and standard deviation of the evaluation scores. ; The filtering module is used to classify intrinsic mode function components with evaluation scores less than the filtering threshold as valid components, and otherwise as noisy components.
9. The intelligent bridge structural stability detection system based on DAS according to claim 7, characterized in that, The component noise reduction module includes: The wavelet decomposition module is used to perform wavelet decomposition on the target noise-mixing component to obtain detail coefficients and approximation coefficients; the target noise-mixing component is any noise-mixing component in the second noise-mixing component set. The thresholding module is used to perform thresholding on detail coefficients using an improved thresholding function. Where w is the detail coefficient; These are the detail coefficients after thresholding. It is the threshold corresponding to the detail coefficient; sgn() is the sign function; exp() is an exponential function with the natural constant e as the base; These are preset adjustment parameters; The inverse transform module is used to perform wavelet inverse transform on the approximation coefficients and the detail coefficients after thresholding to obtain the denoised target noise-mixing component.
10. The intelligent bridge structural stability detection system based on DAS according to claim 6, characterized in that, The detection model comprises a two-level support vector machine; wherein: The first-level support vector machine is a single-class support vector machine, used to determine whether the bridge structure is normal or abnormal; during its training process, it is trained using only normal sample data. The second-level support vector machine is a multi-class support vector machine, used to determine the damage state of the bridge when the first-level support vector machine determines that the bridge structure is abnormal; during its training process, it is trained using abnormal sample data of various damage states.