Beam bridge structure damage diagnosis method based on line-surface information fusion

Abstract

The application discloses a kind of beam bridge structure damage diagnosis methods based on line-surface information fusion, the method utilizes distributed optical fiber to collect beam bridge sparse measuring point information, and the strain data set of non-encryption and encryption area is obtained by encrypting key section;Using the method of sparse optimization, the data collected is denoised, the way of introducing Gaussian process regression combined with space kernel is used, the strain data of non-encryption area is used to extend and reconstruct the monitoring data of encryption area, and the strain data of the whole monitoring area is obtained;After obtaining the strain data, the strain data correlation coefficient in different time domain is calculated, and the damage feature matrix is constructed;By solving the Frobenius norm of correlation matrix, the structure damage threshold is established by using kernel density estimation method, and the effective early warning of bridge structure risk is realized.The method can improve the monitoring efficiency of sensor, realize the lightweight structure health monitoring, and provide effective protection for the safe operation and maintenance of bridge structure in operation period.

Description

Technical Field

[0001] This invention relates to a method for diagnosing damage to beam bridge structures, specifically a method for diagnosing damage to beam bridge structures based on line-surface information fusion. Background Technology

[0002] Bridges are not only vital links in regional economic development but also critical infrastructure for ensuring public safety. Among them, beam bridges dominate due to their simple structure and convenient construction, and are widely used in highway, railway, and urban transportation systems. However, during long-term service, bridge structures are subjected to the coupled effects of various complex loads, making them prone to various defects that seriously affect their service performance and safety. Currently, bridge structural damage monitoring mainly relies on point sensor technology and distributed sensing technology, such as strain gauges, accelerometers, and distributed optical fibers. These sensors are typically installed at key locations on the bridge in a point-like arrangement, collecting strain, vibration, or displacement data in localized areas to analyze the structural health status.

[0003] However, these monitoring technologies are characterized by high costs, limited coverage, and low sensor utilization. Furthermore, the massive data volume, complex processing, and heavy computational requirements are also major challenges and pain points in current monitoring technologies, hindering the achievement of lightweight health monitoring and severely limiting real-time performance and cost-effectiveness. Improving sensor utilization efficiency, eliminating the influence of complex environmental factors on monitoring data, and achieving more efficient and lightweight bridge damage diagnosis technology with fewer sensors to monitor the entire structural range while improving diagnostic accuracy and robustness remain challenging issues. Summary of the Invention

[0004] To address the current problem of low utilization of distributed optical fibers and their monitoring data, this invention provides a beam bridge structure damage diagnosis method based on line-surface information fusion. This method can improve sensor monitoring efficiency and achieve health monitoring of lightweight structures.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A method for diagnosing damage to beam bridge structures based on line-surface information fusion includes the following steps:

[0007] Step 1: Based on locally encrypted fiber optic sensors, obtain strain monitoring values ​​within a certain area under the key cross section of the structure, and obtain strain monitoring values ​​at other measuring points along the longitudinal direction of the beam through distributed optical fibers to construct a strain dataset based on locally encrypted distributed optical fibers.

[0008] Step 2: Clean the collected data using sparse optimization methods to obtain the strain monitoring values ​​after noise removal;

[0009] Step 3: Using Gaussian process regression combined with a spatial kernel, the strain monitoring values ​​of the local densified area are expanded and reconstructed using the monitoring data of a single optical fiber, thus obtaining the strain data of the entire monitoring area.

[0010] Step 4: Based on Step 3, construct a damage feature matrix by calculating the correlation coefficient between strain response data in different time domains, obtain the Frobenius norm for the damage feature matrix, and establish a structural damage threshold using the kernel density estimation method.

[0011] Compared with the prior art, the present invention has the following advantages:

[0012] This invention extracts structural damage information from historical bridge structure inspection data, improves the efficiency of monitoring data utilization by optimizing the fiber optic deployment scheme, and establishes structural risk early warning limits to achieve effective early warning of bridge structural risks. This enables lightweight monitoring of bridge structures and provides effective protection for the safe operation and maintenance of bridge structures during their operational cycle. Attached Figure Description

[0013] Figure 1 This is a flowchart of a beam bridge structural damage diagnosis method based on line-surface information fusion.

[0014] Figure 2 The image shows the selected continuous beam bridge for monitoring.

[0015] Figure 3 The following are schematic diagrams of the fiber optic encryption scheme: (a) a cross-sectional view of the fiber optic arrangement, (b) a longitudinal view of the fiber optic arrangement, and (c) a plan view of the fiber optic arrangement.

[0016] Figure 4 This is historical strain monitoring data for a single bridge structure.

[0017] Figure 5 This is a finite element model of the bridge structure.

[0018] Figure 6 The diagram shows the strain results of the beam under random traffic flow.

[0019] Figure 7 This is a denoised image of data based on a sparse optimization algorithm.

[0020] Figure 8 This is an early warning result for bridge structural damage. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0022] This invention provides a method for diagnosing beam bridge structural damage based on line-surface information fusion. The method utilizes distributed optical fibers to collect sparse measurement point information of the beam bridge and densifies key sections to obtain strain data sets for both undensed and densified areas. A sparse optimization method is employed to denoise the collected data. Gaussian process regression combined with a spatial kernel is introduced to expand and reconstruct the monitoring data in the densified area using the fiber strain data from the undensed area, yielding strain data for the entire monitoring area. After obtaining the strain data, a damage feature matrix is ​​constructed by calculating the correlation coefficients of the strain data in different time domains. By calculating the Frobenius norm of the correlation matrix and using kernel density estimation, a structural damage threshold is established, enabling effective early warning of bridge structural risks. Figure 1 As shown, the specific steps include the following:

[0023] Step 1: Based on locally encrypted fiber optic sensors, acquire strain monitoring values ​​within a certain area (the encrypted measuring points are rectangular) under the key cross-section of the structure (usually at mid-span or 1 / 4 of the span). Acquire strain monitoring values ​​at the remaining measuring points along the longitudinal direction of the beam through distributed optical fibers. The specific steps are as follows:

[0024] Step 11: Based on the distributed fiber optic monitoring data of the unencrypted area of ​​the bridge structure, construct the monitoring data set f of the bridge structure strain:

[0025] f = [ε1, ε2, ..., ε f ,…,ε m ] Τ (1)

[0026] In the formula, (·) T ε is the transpose of the matrix; m is the total number of sampling points for the monitoring data; f For a specific sampling point of the monitoring data, each measuring point simultaneously covers multiple measurement data, and the sample set ε of the monitoring data of the f-th measuring point is... f It consists of the following formula,

[0027]

[0028] In the formula, v is the total number of samples taken at the measuring point within a certain period of time; j∈(1,2,…,v); This is the v-th sampling point under this measurement point.

[0029] Steps 1 and 2: Based on the monitoring data from the distributed fiber optic encrypted zone, construct a monitoring dataset g of the bridge structural strain.

[0030]

[0031] In the formula, n is the number of columns of measuring points along the longitudinal direction of the bridge in the encrypted area; k is the number of rows of measuring points along the transverse direction of the bridge; εki Let k be the sampling point in the k-th row and i-th column of the monitoring data; the sample set of this sampling point is composed of the following formula:

[0032]

[0033] In the formula, w is the total number of samples taken at a certain measurement point in the encrypted area; u∈(1,2,…,w); Let be the u-th sampling point under this measurement point.

[0034] Step 2: Clean the collected data using sparse optimization methods to obtain noise-removed strain monitoring values. The specific steps are as follows:

[0035] Step 21: Filter the original data using a low-pass filter to enhance data sparsity:

[0036]

[0037] In the formula, mask is a low-pass filter; This is the filtered data set.

[0038] Step 22: Use a sparse optimization algorithm to denoise the filtered data, and obtain the denoised dataset:

[0039] min||z||1 (6)

[0040] In the formula, ||||1 represents the minimum value of the L1 norm of z, which is used to eliminate noise.

[0041] Reconstruct the signal, calculate the vector residual, and then denoise the data:

[0042]

[0043] In the formula, A is the transformation matrix. This represents projecting the sparse coefficients z into the signal space; ∈ represents the noise tolerance, filtering out random disturbances; and s represents the strain field after denoising.

[0044] The noise-removed distributed fiber optic data is represented as follows:

[0045] f'=Ax * (8)

[0046] In the formula, f' is the distributed optical fiber monitoring dataset of the unencrypted area after denoising, and x * The sparsity coefficients are used to obtain the minimum L1 norm.

[0047] Similarly, the denoised fiber optic encrypted zone monitoring data set can be obtained as follows:

[0048] g'=A1x1 * (9)

[0049] In the formula, g' is the fiber optic monitoring dataset of the encrypted area after denoising, and A1 is the transformation matrix; x1 * The sparsity coefficients are used to obtain the minimum L1 norm.

[0050] Step 3: Using Gaussian process regression combined with a spatial kernel, the strain monitoring values ​​of the local densified area are expanded and reconstructed using monitoring data from a single optical fiber to obtain strain data for the entire monitoring area. The specific steps are as follows:

[0051] Step 31: Establish a coordinate system and extract the denoised strain data and the corresponding distributed fiber optic measuring point coordinate set M:

[0052] M1={x1,y1} (10)

[0053] M2={x2,y2} (11)

[0054] In the formula, M1 is the set of coordinates of strain measurement points in the unencrypted zone, and M2 is the set of coordinates of strain measurement points in the encrypted zone.

[0055] Step 32: Select p groups of unencrypted area data and q groups of encrypted area data to construct training sets for the measurement point coordinates and strain data respectively:

[0056]

[0057] In the formula, M train M represents the training set of known measurement point coordinates. p and M q The training data sets represent the coordinates of the unencrypted and encrypted areas, respectively.

[0058]

[0059] In the formula, Y train Given a known strain training set at measurement points, f p The unencrypted strain data training set represents the vectorization of the two-dimensional matrix; g q This is a training set of strain data for dense regions.

[0060] Step 33: Construct the full beam strain prediction network. Assume the number of transverse measurement points to be predicted is *a*, the number of longitudinal measurement points is *b*, and *d* = *a* × *b*. The measurement points are distributed in a rectangular array. We have:

[0061]

[0062] In the formula, M test This is the set of coordinates of the measurement points to be predicted.

[0063] Steps 3 and 4: Construct the spatial coordinates and strain relationship to predict unmeasured points:

[0064]

[0065] Y test =K(M test M train )[K(M train M train )+σ n 2 I] -1 Y train (16)

[0066] In the formula, Y is a Gaussian process, representing a random function distribution, where k(M,M') is the spatial kernel, and ψ is noise, which can be ignored after denoising; test Let I be the set of strain points to be predicted, where I is the identity matrix and n is the number of training points.

[0067] Step 4: Based on Step 3, a damage characteristic matrix is ​​constructed by calculating the correlation coefficients between strain response data in different time domains. The Frobenius norm is then obtained from the damage characteristic matrix, and a structural damage threshold is established using kernel density estimation. The specific steps are as follows:

[0068] Step 41: Based on the data predicted in Step 3, at a certain moment, select a certain measuring point and use that point as a reference to calculate the correlation coefficient between that reference measuring point and the other measuring points at different moments:

[0069]

[0070] In the formula, To predict the strain at point q at time p.

[0071] Step 42: Construct the correlation matrix of strain data at different times between this measuring point and other measuring points. And this matrix is ​​used as the characteristic matrix:

[0072]

[0073] In the formula, Strain at a reference prediction point at a certain moment; This is to predict the strain at point j at time n.

[0074] Step 43: Calculate the Frobenius norm of each characteristic matrix:

[0075]

[0076] In the formula, ||H|| F This represents finding the sum of squares and the square root of all elements in H.

[0077] Step 44: Construct the Frobenius norm set of the feature matrix at different time points under the healthy state.

[0078] Steps four and five: For the dataset of each prediction point, construct the damage threshold using kernel density estimation.

[0079]

[0080] In the formula, Let K be the estimated probability density function, and let HI denote the Gaussian kernel function; B (Z j (k) is the damage index, representing the internal variability of the data.

[0081] Calculate the cumulative distribution function

[0082]

[0083] Calculate the 98.5th percentile to determine the damage threshold θ. j :

[0084] θ j =F -1 (0.985) (22)

[0085] This invention achieves lightweight monitoring of bridge structures by densifying key cross-sections, optimizing fiber optic cable layout, denoising data using sparse optimization methods, constructing a structural damage matrix through data expansion and prediction, and finally using kernel density estimation to select values ​​that meet certain confidence levels to construct a structural risk threshold. Existing bridge structural damage monitoring and early warning systems typically require deploying a large number of point sensors or distributed fiber optic sensors, which suffers from challenges such as large data volume, low data utilization, and high computational cost. This invention employs an optimized distributed fiber optic cable layout and densifies key cross-sections, significantly improving the utilization rate of distributed fiber optics. Simultaneously, it removes interference terms from the data using sparse optimization, expands the strain data of the entire monitoring area using Gaussian regression combined with spatial kernel methods, constructs a correlation coefficient matrix of strain data at different measuring points, and uses kernel density estimation to select values ​​that meet certain confidence levels to construct a structural risk threshold, thus achieving effective early warning and lightweight monitoring of bridge structural damage.

[0086] The following experiments were conducted to verify the effectiveness of the present invention:

[0087] This experiment uses historical monitoring and testing data and a finite element model of a specific simply supported beam bridge as an example to verify the effectiveness of the method.

[0088] The specific details of this experiment are as follows:

[0089] 1. Based on the historical monitoring data of the selected specific bridge structure, the strain data is denoised using a sparse optimization algorithm to remove environmental interference terms from the strain data;

[0090] 2. Construct a finite element model, add random traffic flow, and extract strain information from key cross-sectional areas;

[0091] 3. Based on Gaussian process regression combined with spatial kernel, the data of key sections (densified areas) are expanded and predicted to obtain the strain data set of the monitoring area;

[0092] 4. Cross-validation is used to compare the expanded dataset with the data in the finite element model;

[0093] 5. Based on the reconstructed bridge structure monitoring data set, establish a correlation coefficient feature matrix;

[0094] 6. Calculate the Frobenius norm, establish the early warning feature vector using the kernel density estimation method, fit the probability density function of the early warning feature vector, and construct the initial value of the monitoring data for early warning.

[0095] Figure 2 For the selected actual monitoring structure, distributed optical fibers are arranged inside the box girder, and the specific cross-sectional arrangement scheme is as follows: Figure 3 As shown in -a, the distributed optical fibers form a loop along the longitudinal direction, as shown in 3-b. Figure 3 -c is a schematic diagram of distributed fiber optic local encryption. Local encryption of fiber optics is usually set at key sections, such as the mid-span or quarter-span. At the encryption point, the fiber optics form a dense rectangular area. The specific encryption method needs to be determined according to the actual site conditions.

[0096] Figure 4 The image shows the result of denoising a measurement point in the time domain using a sparse optimization algorithm. Figure 5 This is a finite element model established based on the actual structure. Figure 6 This is a diagram showing the stress results of the structure under random traffic flow. Figure 7 The plot shows the time-domain strain results at a certain prediction point, with time on the horizontal axis and strain value on the vertical axis. Figure 8 The diagnostic results graph based on the kernel density function early warning algorithm shows the measurement points on the horizontal axis and the degree of damage and damage threshold on the vertical axis.

Claims

1. A method for diagnosing damage to beam bridge structures based on line-surface information fusion, characterized in that... The method includes the following steps: Step 1: Based on locally encrypted fiber optic sensors, obtain strain monitoring values ​​within a certain area under the key cross-section of the structure. The key cross-section is at the mid-span and 1 / 4 of the span. Obtain strain monitoring values ​​at other measuring points along the longitudinal direction of the beam through distributed optical fibers to construct a strain dataset based on locally encrypted distributed optical fibers. Step 2: Clean the collected data using sparse optimization methods to obtain the strain monitoring values ​​after noise removal; Step 3: Using Gaussian process regression combined with a spatial kernel, the strain monitoring values ​​of the local densified area are expanded and reconstructed using monitoring data from a single optical fiber to obtain strain data for the entire monitoring area. The specific steps are as follows: Step 31: Establish a coordinate system and extract the denoised strain data and the corresponding set of distributed fiber optic measurement point coordinates. : （10） （11） In the formula, This is the set of coordinates of strain measurement points in the unencrypted area. This is the set of coordinates of strain measurement points in the encrypted zone; Step 32: Select p groups of unencrypted area data and q groups of encrypted area data to construct training sets for the measurement point coordinates and strain data respectively: （12） In the formula, This represents a training set with known coordinates of measurement points. and The training datasets represent the coordinates of the unencrypted and encrypted regions, respectively. （13） In the formula, Given a training set of strain measurements at known measurement points, This represents the strain data training set in the unencrypted area. This represents the vectorization of a two-dimensional matrix; This is a training set of strain data for dense regions; Step 33: Construct the full beam strain prediction network, assuming the number of transverse measurement points of the full beam to be predicted is... The number of longitudinal measurement points is ,and The predicted measurement points are distributed in a rectangular array, as follows: （14） In the formula, This is the set of coordinates of the measurement points to be predicted. Steps 3 and 4: Construct the spatial coordinates and strain relationship to predict unmeasured points: （15） （16） In the formula, Let be a Gaussian process, representing the distribution of a random function. For space core, This is noise, which can be ignored after noise reduction; For the set of measurement points to be predicted, The identity matrix is ​​n; the number of training points is n. Step 4: Based on Step 3, construct a damage feature matrix by calculating the correlation coefficient between strain response data in different time domains, obtain the Frobenius norm for the damage feature matrix, and establish a structural damage threshold using the kernel density estimation method.

2. The method for diagnosing beam bridge structural damage based on line-surface information fusion according to claim 1, characterized in that... The specific steps of step one are as follows: Step 11: Based on the distributed fiber optic monitoring data of the unencrypted area of ​​the bridge structure, construct the monitoring data set f of the bridge structure strain: （1） In the formula, is the transpose of the matrix; m is the total number of sampling points for the monitoring data; For a specific sampling point of the monitoring data, each measuring point simultaneously covers multiple measurement data, and the sample set of the monitoring data for the f-th measuring point is... It consists of the following formula, : （2） In the formula, This represents the total number of samples collected at this measuring point within a certain period of time. ; This is the v-th sampling point under this measurement point; Steps 1 and 2: Based on the monitoring data from the distributed fiber optic encrypted zone, construct a monitoring dataset g of the bridge structural strain. （3） In the formula, n is the number of columns of measuring points along the longitudinal direction of the bridge in the encrypted area; k is the number of rows of measuring points along the transverse direction of the bridge. Let k be the sampling point in the k-th row and i-th column of the monitoring data; the sample set of this sampling point is composed of the following formula: : （4） In the formula, w is the total number of samples taken at a certain measurement point in the encrypted area; ; Let be the u-th sampling point under this measurement point.

3. The method for diagnosing beam bridge structural damage based on line-surface information fusion according to claim 2, characterized in that... In steps one and two, the measurement points in the encryption area are rectangular.

4. The method for diagnosing beam bridge structural damage based on line-surface information fusion according to claim 2, characterized in that... The specific steps of step two are as follows: Step 21: Filter the original data using a low-pass filter to enhance data sparsity: （5） In the formula, mask is a low-pass filter; This is the filtered data set; Step 22: Use a sparse optimization algorithm to denoise the filtered data, and obtain the denoised dataset: （6） In the formula, To obtain Find the minimum L1 norm value to remove noise; Reconstruct the signal, calculate the vector residual, and then denoise the data: （7） In the formula, A is the transformation matrix. , which means projecting the sparse coefficients z into the signal space; is the noise tolerance, used to filter out random disturbances; s is the strain field after noise reduction. The noise-removed distributed fiber optic data is represented as follows: （8） In the formula, This is the denoised, unencrypted area distributed fiber optic monitoring dataset. To obtain the sparsity coefficients with the minimum L1 norm; Similarly, the denoised fiber optic encrypted zone monitoring data set can be obtained as follows: （9） In the formula, This is the dataset for fiber optic monitoring in the encrypted area after denoising. The transformation matrix; The sparsity coefficients are used to obtain the minimum L1 norm.

5. The method for diagnosing beam bridge structural damage based on line-surface information fusion according to claim 1, characterized in that... The specific steps of step four are as follows: Step 41: Based on the data predicted in Step 3, at a certain moment, select a certain measuring point and use that point as a reference to calculate the correlation coefficient between that reference measuring point and the other measuring points at different moments: （17） In the formula, To predict the strain at point q at time p; Step 42: Construct the correlation matrix of strain data at different times between this measuring point and other measuring points. And use this matrix as the characteristic matrix: （18） In the formula, Strain at a reference prediction point at a certain moment; To predict the strain at point j at time n; Step 43: Calculate the Frobenius norm of each characteristic matrix: （19） In the formula, Indicates the request The sum of squares of all elements and the square root; Step 44: Construct the Frobenius norm set of the feature matrices at different time points under the healthy state; Steps four and five: For the dataset of each prediction point, construct the damage threshold using kernel density estimation. （20） In the formula, Let K be the estimated probability density function, and let K denote the Gaussian kernel function. The damage index represents the internal variability of the data. h represents bandwidth; Calculate the cumulative distribution function : （21） Calculate the 98.5th percentile to determine the damage threshold. : （22）。

Images