Bridge cable force prediction method and device based on time series data

Abstract

The present invention discloses a bridge cable force prediction method and device based on time series data, which belongs to the technical field of data processing and analysis. The method collects surface images and original radiographic images of multiple sampling points of a single cable, and calculates the surface state data and cross-sectional state data of the cable according to the surface image and the original radiographic image respectively. The surface state data and the cross-sectional state data comprehensively predict the state prediction data of the cable, which more comprehensively reflects the health status of the cable. Furthermore, the present invention calculates defect state data according to the flux change and current magnetic induction intensity of multiple sampling points within the monitoring period, generates upper and lower control limits of the cable damage according to the defect state data, and analyzes the state prediction data based on the upper and lower control limits to avoid the artificially set health evaluation standards from affecting the prediction results.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and analysis, and in particular to a method and device for predicting bridge cable force based on time series data. Background Art

[0002] As a key load-bearing component of a bridge, the cable has a profound impact on the safety and durability of the bridge. Since the bridge was put into use, the cable has been continuously subjected to environmental effects such as corrosion, fatigue and vibration. In particular, due to its external layout and small cross-sectional characteristics, it is extremely sensitive to corrosion. Surface damage directly weakens the mechanical properties of the inclined cable and reduces its bearing capacity. When the damage accumulates to a certain extent, it seriously threatens the safety of the bridge structure and may cause major accidents. Cracks in the steel wires inside the cable will also lead to a reduction in the cross-sectional area of ​​a single steel wire, thereby reducing the total bearing capacity of the entire cable. Therefore, the prediction of the health of the bridge is crucial. A Chinese patent application with publication number CN117688647A discloses a method for establishing a bridge structure monitoring data prediction model and its application. This method uses the incremental vector of the monitoring data to construct an incremental influence matrix, and establishes a prediction model based on this. However, the state of the bridge structure may show non-stationarity over time, and the construction of the incremental influence matrix requires that the monitoring data is stable over a period of time. The Chinese patent application with publication number CN107895014A discloses a time series bridge monitoring data analysis method based on the MapReduce framework. This method uses the MapReduce framework and the improved ARIMA model to process the time series data of the bridge to predict the overall health of the bridge and obtain the structural response of the bridge in a timely manner in a changing environment. However, the above method lacks the collection of detailed status data of a single cable, and the scheme does not disclose the evaluation criteria for bridge health, making it difficult to form an objective evaluation result. Summary of the invention

[0003] The present invention provides a bridge cable force prediction method and device based on time series data to solve the problems existing in the above-mentioned prior art. The present invention collects surface and cross-sectional information to improve the prediction accuracy. At the same time, upper and lower control limits are generated according to the cable structure state to avoid the influence of artificial health evaluation standards on the prediction results.

[0004] The invention objectives of this application can be achieved through the following technical means:

[0005] A bridge cable force prediction method based on time series data comprises the following steps:

[0006] Step 1: Multiple cables connect the bridge tower and the bridge, and the load index of each cable is determined according to the load model of the bridge;

[0007] Step 2: In each monitoring cycle, multiple surface images, original radiographic images, and magnetic flux changes are collected along the length direction of the cable, and surface state data are calculated based on the surface images, cross-sectional state data are calculated based on the original radiographic images, and defect state data are calculated based on the magnetic flux changes;

[0008] Step 3: Generate a surface state sequence from the surface state data of multiple sampling points, generate a cross-section state sequence from the cross-section state data of multiple sampling points, and generate a defect state sequence from the defect state data of multiple sampling points;

[0009] Step 4: Combining the plurality of surface state sequences, cross-section state sequences and defect state sequences into a surface state matrix, a cross-section state matrix and a defect state matrix respectively along the order of the monitoring period;

[0010] Step 5: Correct the surface state matrix into a surface variation matrix and correct the cross-section state matrix into a cross-section variation matrix according to the natural period;

[0011] Step 6: Calculate the mean and standard deviation of the defect state matrix of the cable, and then generate the upper control limit and lower control limit of the cable;

[0012] Step 7: Process the surface change matrix into the first observation sample, generate the surface prediction sequence according to the first observation sample, process the cross-section change matrix into the second observation sample, and generate the cross-section prediction sequence according to the second observation sample;

[0013] Step 8: Merge the surface prediction sequence and the cross-section prediction sequence into a state prediction sequence, extract the abnormal sampling points in the state prediction sequence where the state prediction data exceeds the upper control limit and the lower control limit, and predict the bridge cable tension state according to the number of continuous abnormal sampling points and the load index.

[0014] In the present invention, in step 1, a finite element analysis is performed on the bridge according to the load model, the maximum tension of the cable in a natural period is extracted, the maximum tension of each cable is normalized, and the load index k of the cable is generated, k∈[0,1].

[0015] In the present invention, in step 2, the first acquisition device acquires at least one surface image of the cable, processes the surface image into a surface correction image, extracts the surface damage area of ​​the surface correction image based on the grayscale histogram and threshold segmentation method, and generates surface state data of multiple sampling points according to the number of pixels in the surface damage area.

[0016] In the present invention, in step 2, the second acquisition device acquires original X-ray images of the cable at different rotation angles, generates a cross-sectional correction image based on multiple original X-ray images, extracts the cross-sectional damage area of ​​the cross-sectional correction image, and generates cross-sectional state data of multiple sampling points based on the number of pixels in the cross-sectional damage area.

[0017] In the present invention, in step 2, the third acquisition device acquires the magnetic flux change and current magnetic induction intensity of the cable, and calculates the defect state data of multiple sampling points according to the magnetic flux change, current magnetic induction intensity and defect-free magnetic induction intensity of the cable.

[0018] In the present invention, in step 5, the surface change matrix A' is composed of surface change data of multiple sampling points in multiple monitoring periods, wherein the surface change data a' of sampling point j in monitoring period t is t×j =a t×j -a (t-τ)×j , a t×j is the surface state data of sampling point j in monitoring period t, a (t-τ)×j is the surface state data of sampling point j in the monitoring period t-τ, τ is the number of monitoring periods included in the natural period, t>τ.

[0019] In the present invention, in step 6, the mean of the defect state matrix is ​​μ, the standard deviation of the defect state matrix is ​​σ, the upper control limit of the cable is S1 = (μ+3σ) / μ, and the lower control limit is S2 = (μ-3σ) / μ.

[0020] In the present invention, in step 7, multiple surface change data of the first observation sample are input into the gated cycle prediction model along the order of the monitoring cycle, at least one model parameter of the gated cycle prediction model is trained, the surface change data of the next monitoring cycle is predicted according to the gated cycle prediction model, the surface state data of the next monitoring cycle is obtained according to the surface change data of the next monitoring cycle and the surface state data of the current monitoring cycle, and the multiple surface state data of the next monitoring cycle constitute a surface prediction sequence.

[0021] In the present invention, in step 8, the cable force weights w1 and w2 are constructed, and the state prediction data d of each sampling point in the state prediction sequence is t×j =(w1a t×j +w2b t×j ), where a t×j is the surface state data of sampling point j in monitoring period t, b t×j To monitor the cross-sectional state data of sampling point j in period t, calculate the t×j [d0S2, d0S1] abnormal sampling points, where the maximum number of consecutive abnormal sampling points is n, the load index of the cable is k, d0 is the reference value of cable damage, if n×k of at least one cable exceeds the allowable upper limit, the current bridge cable force state is an alarm state.

[0022] A prediction device for implementing the bridge cable force prediction method based on time series data, comprising:

[0023] A crawling device, the crawling device is configured to move to a plurality of sampling points along the length direction of the cable;

[0024] A first acquisition device, the first acquisition device is configured to acquire a plurality of sets of surface images;

[0025] A second acquisition device, the second acquisition device is configured to acquire multiple groups of original radiographic images;

[0026] A third acquisition device, the third acquisition device is configured to acquire multiple sets of magnetic flux variation;

[0027] A data processing device, the data processing device is configured to generate a surface state matrix, a cross-sectional state matrix, and a defect state matrix according to the surface image, the original radiographic image, and the magnetic flux variation;

[0028] A data storage device, the data storage device being configured to store a surface state matrix, a cross-sectional state matrix, and a defect state matrix for a plurality of natural periods;

[0029] A data analysis device, the data analysis device is configured to generate a state prediction sequence according to the surface state matrix, the cross-sectional state matrix, and the defect state matrix;

[0030] The cable force prediction device is configured to predict the cable force state of the bridge.

[0031] The implementation of the bridge cable force prediction method and device based on time series data of the present invention has the following beneficial effects: The present invention combines the surface damage and cross-sectional damage of each cable during the monitoring period to more comprehensively reflect the structural state of a single cable. The present invention independently analyzes the damage data of different sampling points of the cable, and at the same time eliminates seasonal changes that may affect the prediction results before data prediction, thereby improving the prediction accuracy. Furthermore, the present invention also combines the flux changes of multiple sampling periods to predict the aging trend of the cable, thereby generating upper and lower control limits for evaluating cable damage, thereby avoiding the influence of artificially set health evaluation standards on the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a schematic diagram of a cable-stayed bridge applicable to the present invention;

[0033] Figure 2 It is a flow chart of the bridge cable force prediction method based on time series data of the present invention;

[0034] Figure 3 A schematic diagram of collecting surface images according to the present invention;

[0035] Figure 4 A schematic diagram of generating a surface correction image for the present invention;

[0036] Figure 5A schematic diagram of collecting original radiographic images according to the present invention;

[0037] Figure 6 It is a schematic diagram of the current magnetic induction intensity when the cable is damaged inside the present invention;

[0038] Figure 7 A sampling schematic diagram of the surface state matrix of the present invention;

[0039] Figure 8 A schematic diagram of state prediction data of multiple sampling points of the present invention;

[0040] Fig. 9 It is a schematic diagram of mapping the pixel points of the original radiographic image to the cross-sectional coordinates of the present invention;

[0041] Fig.10 A schematic diagram of generating a cross-sectional correction image for the present invention;

[0042] Fig.11 A block diagram of a bridge cable force prediction device according to the present invention;

[0043] Fig.12 It is a structural diagram of the crawling device and the first collecting device, the second collecting device and the third collecting device of the present invention;

[0044] Fig.13 It is a cross-sectional schematic diagram of a first collection device of the present invention;

[0045] Fig.14 is a cross-sectional schematic diagram of a second collection device of the present invention;

[0046] Fig.15 It is a cross-sectional schematic diagram of the third collection device of the present invention.

[0047] Reference numerals in the drawings: imaging unit 102 , annular light source 103 , ray generator 202 , ray detector 203 , magnetic induction detection device 302 , magnetic yoke 303 , permanent magnet 304 , Hall sensor 305 , induction coil 306 . DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0049] like Figure 1, multiple cables connect the bridge tower and the bridge. The cables are cylindrical rod-shaped components, which are usually twisted together by multiple high-strength steel wire ropes. In view of the fact that the existing technology is not comprehensive enough in cable detection, the present invention combines the surface state data and cross-sectional state data of each cable, removes the periodic item, generates the state prediction data of the cable, and uses the defect state data of the natural period of the cable to obtain the upper and lower control limits of the cable, and then analyzes the state prediction data according to the upper and lower control limits, and then evaluates the cable tension condition of the entire bridge. Embodiment 1

[0050] Reference Figure 2 The bridge cable force prediction method based on time series data of the present invention described in detail in this embodiment comprises the following steps:

[0051] Step 1: Multiple cables connect the bridge tower and the bridge, and the load index of each cable is determined according to the load model of the bridge. Finite element analysis of the bridge is performed according to the load model to simulate the various loads that the bridge may bear under different environments, and the tension of the cable under different load conditions is obtained. The load model includes permanent loads, time-related variable loads, etc. Combined with the stress conditions of the cable, computer-aided engineering (CAE) software, such as finite element analysis (FEA) tools, is used to extract the maximum tension of the cable in a natural cycle, and the maximum tension of each cable is normalized to generate the load index k of the cable, k∈[0,1].

[0052] Step 2: Collect multiple surface images, original radiographic images, and magnetic flux changes along the length direction of the cable in each monitoring cycle, calculate surface state data based on the surface images, calculate cross-sectional state data based on the original radiographic images, and calculate defect state data based on the magnetic flux changes. In the present invention, the sampling points are arranged along the length direction of the cable, and each sampling point corresponds to a cable section. The number of pixels in the surface damage area of ​​the cable section is the surface state data, the number of pixels in the cross-sectional damage area of ​​the cable section is the cross-sectional state data, and the loss amount of the cable section is the defect state data.

[0053] like Figure 3 The first acquisition device includes a camera unit, which acquires at least one surface image of the cable. Figure 4 , the surface image needs to be processed into a surface correction image. For any point p(x p ,y p ), and convert it to the corresponding point p' (x' p ,y' p ), x' p The mathematical mapping of the coordinates is x' p =Rarcsin(x p / R), where arcsin(x p / R) is the radian value of the angle θ between the line connecting point p and the center of the cable cross section and the central axis of the cable, and R is the radius of the cable cross section. In this embodiment, the camera unit is completely perpendicular to the central axis of the cable, then y' p Approximately equal to y p .

[0054] The grayscale level of the surface correction image is counted to obtain a grayscale histogram. In this embodiment, the artificial lighting conditions are relatively uniform, and the contrast between the undamaged area and the damaged area on the cable surface is good, so a simple single threshold segmentation method is used. The surface image threshold is set, and the grayscale histogram is segmented by a single threshold according to the surface image threshold. The surface state data of multiple sampling points are generated according to the number of pixels in the surface damaged area. Furthermore, if the camera unit is not completely perpendicular to the central axis of the cable, then y' p = y p cos(β). β is the angle between the optical axis of the camera unit and the central axis of the cable.

[0055] Reference Figure 5 The second acquisition device is composed of a ray generator and a ray detector. The ray generator emits X-rays to the cable at different rotation angles, and the ray detector receives the X-rays and generates an original ray image. A cross-sectional correction image is generated based on multiple original ray images, and a cross-sectional damage area of ​​the cross-sectional correction image is extracted. The cross-sectional state data of multiple sampling points is generated based on the number of pixels in the cross-sectional damage area. Embodiment 3 further discloses a method for obtaining cross-sectional state data based on the original ray image.

[0056] As the cable ages, the current magnetic induction intensity of the cable gradually decreases. The third acquisition device consists of a yoke, a permanent magnet and a magnetic induction detection device. The magnetic induction detection device collects the magnetic flux change and current magnetic induction intensity of the cable, and calculates the defect state data of multiple sampling points based on the magnetic flux change, current magnetic induction intensity and defect-free magnetic induction intensity of the cable. Defect state data c=[ΔΦ+κ2(z1+ z2)] / κ1, where ΔΦ is the magnetic flux change and κ1 is the defect-free magnetic induction intensity. Figure 6 , κ2 is the current magnetic induction intensity of the cable, z1 is the defect-free internal cross-sectional area of ​​the cable, and z2 is the defect-free air cross-sectional area between the magnetic induction detection device and the cable.

[0057] Step 3: Generate a surface state sequence from the surface state data of multiple sampling points, generate a cross-section state sequence from the cross-section state data of multiple sampling points, and generate a defect state sequence from the defect state data of multiple sampling points. Surface state sequence A of monitoring period t t ={a t×1 ,a t×2 ,...,a t×J}、Section state sequence Bt ={b t×1 ,b t×2 ,...,b t×J}、Defective state sequence C t ={c t×1 ,c t×2 ,...,c t×J}, J is the number of sampling points. t×1 is the surface state data of the first sampling point in monitoring period t, b t×1 is the cross-sectional state data of the first sampling point in monitoring period t, c t×1 It is the defect status data of the first sampling point in the monitoring period t.

[0058] Step 4: Combine multiple surface state sequences, cross-section state sequences and defect state sequences into a surface state matrix, a cross-section state matrix and a defect state matrix respectively along the order of the monitoring period. , after expansion , cross-section state matrix , after expansion , defect state matrix , after expansion , T is the number of monitoring cycles.

[0059] Step 5: Correct the surface state matrix to the surface change matrix according to the natural period, and correct the cross-section state matrix to the cross-section change matrix. The surface state matrix contains periodic terms, which will reduce the accuracy of data prediction. The present invention further eliminates the periodic terms. The surface change matrix A' after elimination is composed of surface change data of multiple sampling points in multiple monitoring periods. The surface change data a' of sampling point j in monitoring period t t×j =a t×j -a (t-τ)×j , refer to Figure 7 , a t×j is the surface state data of sampling point j in monitoring period t, a (t-τ)×j is the surface state data of sampling point j in the monitoring period t-τ, τ is the number of monitoring periods included in the natural period, t>τ. The acquisition method of the cross-sectional change matrix refers to the acquisition method of the surface change matrix. In the invention, the natural period is one year, the monitoring period is 1 day, and τ=360.

[0060] Step 6: Calculate the mean and standard deviation of the defect state matrix of the cable, and then generate the upper and lower control limits of the cable. , the standard deviation of the defect state matrix is , the upper control limit of the cable is S1 = (μ+3σ) / μ, and the lower control limit is S2 = (μ-3σ) / μ. The defect state measured by the magnetic flux mainly reflects the aging and defect growth of the cable with the natural cycle. The present invention generates the control limit by means of the 3σ criterion of the normal distribution.

[0061] Step 7: Process the surface change matrix into a first observation sample, generate a surface prediction sequence based on the first observation sample, process the cross-section change matrix into a second observation sample, and generate a cross-section prediction sequence based on the second observation sample. In the present invention, the matrix data can be discretized and denoised to obtain the first observation sample to improve the comparability of the data. At the same time, the surface prediction sequence is obtained with the help of a gated recurrent prediction model. Multiple surface change data of the first observation sample are input into a gated recurrent prediction model (GRU) in the order of the monitoring cycle, and at least one model parameter of the gated recurrent prediction model is trained. Model parameters include but are not limited to activation functions or weight vectors of update gates, reset gates, candidate states, and hidden states.

[0062] After the training of the gated cycle prediction model is completed, the surface change data of the next monitoring cycle is predicted according to the gated cycle prediction model. The sum of the surface change data of the next monitoring cycle and the surface state data of the current monitoring cycle is equal to the surface state data of the next monitoring cycle. The multiple surface state data of the next monitoring cycle are arranged along the sampling points to form a surface prediction sequence. In the same way, the cross-sectional change data of the next monitoring cycle is predicted according to the gated cycle prediction model. The sum of the cross-sectional change data of the next monitoring cycle and the cross-sectional state data of the current monitoring cycle is equal to the cross-sectional state data of the next monitoring cycle. The multiple cross-sectional state data of the next monitoring cycle are arranged along the sampling points to form a cross-sectional prediction sequence.

[0063] Step 8: Combine the surface prediction sequence and the cross-section prediction sequence into a state prediction sequence, extract the abnormal sampling points in the state prediction sequence where the state prediction data exceeds the upper control limit and the lower control limit, and predict the bridge cable state according to the number of consecutive abnormal sampling points and the load index. Construct the cable weights w1 and w2, and the state prediction data d of each sampling point in the state prediction sequence t×j =(w1a t×j +w2b t×j ), where a t×j is the surface state data of sampling point j in monitoring period t, b t×j To monitor the cross-sectional state data of sampling point j in period t, w1 and w2 respectively express the influence of surface state and cross-sectional state on cable force. t×j [d0S2, d0S1] abnormal sampling point, refer to Figure 8, abnormal sampling points appear in different areas. The higher the concentration of abnormal sampling points, the larger the damage surface or the greater the damage depth. The maximum number of continuous abnormal sampling points is n, the load index of the cable is k, d0 is the baseline value of cable damage, and n×k of the cable is calculated. If the n×k of at least one cable exceeds the allowable upper limit, the cable is difficult to provide the design tension, and the current bridge cable tension state is in an alarm state. The allowable upper limit is the upper limit value of damage provided by the bridge structure design. The allowable upper limits of cables of different sizes and design loads are different, which can be determined in combination with the bridge design specifications. Embodiment 2

[0064] The cable force weights w1 and w2 of the present invention affect the prediction of the cable force state of the bridge. In a more simplified embodiment, w1=w2=0.5. This embodiment further discloses a method for determining the cable force weights w1 and w2.

[0065] Step 201: Construct a judgment matrix H = (h ij ) L×L ,h ij =[e ij ,f ij ],h ij represents the interval of relative importance between the i-th evaluation index and the j-th evaluation index, e ij and f ij Respectively represent h ij The lower and upper limits of , L is the number of evaluation indicators, i=1,2,…,L, j=1,2,…,L.

[0066] Step 202: Construct the consistency digital judgment matrix M of the judgment matrix H = (m ij ) L×L , , the weight vector of the consistency digital judgment matrix M is W'=(w'1,w'2,…,w' l ,…,w' L ),in .

[0067] Step 203: Determine the range matrix ΔM=(Δm ij ) L×L , Δm ij =MAX((m ij -e ij ), (f ij -m ij )) Determine the weight error Δw based on the extreme difference matrix ΔM l , , then the adjusted weight vector is W''=(w''1,w''2,…,w'' l ,…,w'' L)Where w'' l =[w' l –Δw l , w' l +Δw l ].

[0068] Step 204: Convert the scores made by the G experts into G row vectors , , …, , …, ,in =(α g1 , α g2 , …, α gL ), α gL is the score of expert g on the evaluation index L. Construct a similarity matrix O=(o gv ) G×G ,in , g=1,2,…,G, v=1,2,…,G,o gv represents the similarity of the scores between expert g and expert v, o gv The closer it is to 1, the higher the similarity between the scores of expert g and expert v.

[0069] Step 205: Calculate the credibility of each expert based on the similarity of the scores. For expert g, its credibility is ,g=1,2,…,G, v=1,2,…,G,the credibility U g Standardized to the interval [0,1], the standardized credibility of expert g is U' g =(U g -U min ) / (U max -U min ), U max is the maximum value of the credibility of all experts, U min is the minimum value of the credibility of all experts.

[0070] Step 206: Calculate the final weights of the L indicators W=(w1,w2,…,w l ,…,w L ), , where α is the maximum score. If w l Not in w'' l If the expert with the lowest credibility is within the range of , the expert with the lowest credibility is removed, and the scores of the remaining experts are recalculated to perform steps 204 and 205, and the final weight is recalculated. In this embodiment, α is 10, L=2, and the L indicators are cross-sectional state data and surface state data. Embodiment 3

[0071] This embodiment further discloses a method for calculating the cross-sectional state data according to the original radiographic image in step 2. In one detection cycle, the ray generator and the ray detector rotate along the cable for one circle to obtain λ original radiographic images corresponding to the sampling points.

[0072] Step 301: Divide the original radiographic image into Q rows of pixel strips, each pixel strip corresponding to a cross-sectional position. Fig. 9 When the X-ray passes through the cross-sectional coordinate (x, y) of the cable, the X-ray maps the cross-sectional coordinate (x, y) of the cable to a pixel position on the pixel strip. The pixel position represents the pixel position on the ray detector corresponding to the cross-sectional coordinate (x, y). The pixel value of the pixel position is affected by factors such as the density and attenuation coefficient of the cross-sectional coordinate (x, y).

[0073] Step 302: Extract the coordinates of the ray generator (x0, y0) of the current rotation angle, and determine the distance γ1 from the ray generator coordinates to the center of the cable and the distance γ2 from the ray detector to the center of the inclined cable. The expansion angle of the X-ray passing through the cross-sectional coordinates (x, y) is ω, and the ray projection distance of the cross-sectional coordinates (x, y) is γ3, γ3=γ1sinω. And ω=arctan[(y0-y) / (x0-x)]-arctan(y0 / x0), the ray projection distance γ3=γ1sin{arctan[(y0-y) / (x0-x)]-arctan(y0 / x0)}. And ω also satisfies ω=arctan[γ4 / (γ1+γ2)], so the pixel position γ4=tan(arcsinγ3 / γ1)×(γ1+γ2). In this way, the mapping relationship between the pixel strip and the cross-sectional coordinates can be determined. The pixel bit of the present invention is a length value, which refers to the distance between the projection position of the X-ray in the pixel bar and the midpoint of the pixel bar. The positive and negative signs of the pixel bit can be determined according to the projection direction of the X-ray.

[0074] Step 303: Reference Fig.10 , extract the qth row of pixel strips of the λ original ray images to obtain λ groups of pixel strips. According to the method of step 302, the cross-sectional coordinates corresponding to each pixel position of the λ groups of pixel strips can be determined.

[0075] Step 304: Extract the initial intensity of the X-ray, measure the projection intensity of each pixel of the pixel strip, and the projection intensity of the pixel is determined by the initial intensity and the attenuation coefficient of all cross-sectional coordinates through which the X-ray corresponding to the pixel passes. Calculate the intensity attenuation logarithm of the pixel, which reflects the attenuation coefficient of multiple cross-sectional coordinates through which the X-ray corresponding to the pixel passes.

[0076] Step 305: Given the intensity attenuation logarithm of each pixel and the cross-sectional coordinates corresponding to the pixel, a multivariate equation of the intensity attenuation logarithm and the cross-sectional coordinates can be established to obtain the attenuation coefficient of each cross-sectional coordinate. The attenuation coefficient reflects the depth of the original cross-sectional image. Affected by factors such as damage and corrosion, the attenuation coefficient of the cross-sectional coordinate may be less than the reference attenuation coefficient. The cross-sectional correction image can be reconstructed based on the attenuation coefficient of each cross-sectional coordinate. The cross-sectional damage area of ​​the cross-sectional correction image is extracted (the pixel value does not match the cable material), and the cross-sectional state data of multiple sampling points is generated based on the number of pixels in the cross-sectional damage area. Embodiment 4

[0077] Reference Figures 11 to 15 This embodiment further discloses a prediction device for implementing the bridge cable force prediction method based on time series data, including: a crawling device, a first acquisition device, a second acquisition device, a third acquisition device, a data processing device, a data storage device, a data analysis device, a cable force prediction device, and a controller. The first acquisition device and the second acquisition device can rotate around the crawling device, and a light shield is set outside the first acquisition device and the second acquisition device. It can be expected that when the first acquisition device, the second acquisition device, and the third acquisition device work simultaneously, the surface images, original radiographic images, and magnetic flux changes of three adjacent sampling points are respectively collected. The controller is configured to control the operation of the first acquisition device, the second acquisition device, and the third acquisition device.

[0078] The crawling device is configured to move to multiple sampling points along the length direction of the cable. The first acquisition device is configured to acquire multiple sets of surface images. The first acquisition device is composed of a camera unit 102 and an annular light source 103, and the annular light source 103 makes the lighting conditions uniform. The second acquisition device is configured to acquire multiple sets of original radiographic images. The second acquisition device is composed of a ray generator 202 and a ray detector 203. The ray generator 202 emits X-rays to the cable, and the ray detector 203 receives the X-rays passing through the cable and generates original radiographic images. The third acquisition device is configured to acquire multiple sets of magnetic flux changes. The third acquisition device is composed of a magnetic induction detection device 302, a yoke 303 and a permanent magnet 304. The yoke 303 and the permanent magnet 304 are used to magnetize the inside of the cable, wherein the magnetic induction detection device 302 is composed of a Hall sensor 305 and an induction coil 306. The Hall sensor 305 is used to detect the current magnetic induction intensity, and the induction coil 306 is used to detect the magnetic flux change. The data processing device is configured to generate a surface state matrix, a cross-section state matrix, and a defect state matrix according to the surface image, the original radiographic image, and the magnetic flux variation. The data storage device is configured to store the surface state matrix, the cross-section state matrix, and the defect state matrix of multiple natural periods. The data analysis device is configured to generate a state prediction sequence according to the surface state matrix, the cross-section state matrix, and the defect state matrix. The cable force prediction device is configured to predict the cable force state of the bridge.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A bridge cable force prediction method based on time series data, characterized in that: The following steps are involved: Step 1: Multiple cables connect the bridge tower and the bridge, and the load index of each cable is determined according to the load model of the bridge; Step 2: In each monitoring cycle, multiple surface images, original radiographic images, and magnetic flux changes are collected along the length direction of the cable, and surface state data are calculated based on the surface images, cross-sectional state data are calculated based on the original radiographic images, and defect state data are calculated based on the magnetic flux changes; Step 3: Generate a surface state sequence from the surface state data of multiple sampling points, generate a cross-section state sequence from the cross-section state data of multiple sampling points, and generate a defect state sequence from the defect state data of multiple sampling points; Step 4: Combining the plurality of surface state sequences, cross-section state sequences and defect state sequences into a surface state matrix, a cross-section state matrix and a defect state matrix respectively along the order of the monitoring period; Step 5: Correct the surface state matrix into a surface variation matrix and correct the cross-section state matrix into a cross-section variation matrix according to the natural period; Step 6: Calculate the mean and standard deviation of the defect state matrix of the cable, and then generate the upper control limit and lower control limit of the cable; Step 7: Process the surface change matrix into the first observation sample, generate the surface prediction sequence according to the first observation sample, process the cross-section change matrix into the second observation sample, and generate the cross-section prediction sequence according to the second observation sample; Step 8: Merge the surface prediction sequence and the cross-section prediction sequence into a state prediction sequence, extract the abnormal sampling points in the state prediction sequence where the state prediction data exceeds the upper control limit and the lower control limit, and predict the bridge cable tension state according to the number of continuous abnormal sampling points and the load index.

2. The bridge cable force prediction method based on time series data according to claim 1 is characterized in that: In step 1, a finite element analysis is performed on the bridge according to the load model, the maximum tension of the cable in a natural period is extracted, the maximum tension of each cable is normalized, and the load index k of the cable is generated, k∈[0,1].

3. The bridge cable force prediction method based on time series data according to claim 1 is characterized in that: In step 2, the first acquisition device acquires at least one surface image of the cable, processes the surface image into a surface correction image, extracts the surface damage area of ​​the surface correction image based on the grayscale histogram and threshold segmentation method, and generates surface state data of multiple sampling points according to the number of pixels in the surface damage area.

4. The bridge cable force prediction method based on time series data according to claim 1 is characterized in that: In step 2, the second acquisition device acquires original X-ray images of the cable at different rotation angles, generates a cross-sectional correction image based on multiple original X-ray images, extracts the cross-sectional damage area of ​​the cross-sectional correction image, and generates cross-sectional state data of multiple sampling points based on the number of pixels in the cross-sectional damage area.

5. The bridge cable force prediction method based on time series data according to claim 1 is characterized in that: In step 2, the third acquisition device acquires the magnetic flux change and current magnetic induction intensity of the cable, and calculates defect status data of multiple sampling points according to the magnetic flux change, current magnetic induction intensity and defect-free magnetic induction intensity of the cable.

6. The bridge cable force prediction method based on time series data according to claim 1, characterized in that: In step 5, the surface change matrix A' is composed of surface change data of multiple sampling points in multiple monitoring periods, where the surface change data a' of sampling point j in monitoring period t is t×j =a t×j -a (t-τ)×j , a t×j is the surface state data of sampling point j in monitoring period t, a (t-τ)×j is the surface state data of sampling point j in the monitoring period t-τ, τ is the number of monitoring periods included in the natural period, t>τ.

7. The bridge cable force prediction method based on time series data according to claim 1 is characterized in that: In step 6, the mean of the defect state matrix is ​​μ, the standard deviation of the defect state matrix is ​​σ, the upper control limit of the cable is S1 = (μ+3σ) / μ, and the lower control limit is S2 = (μ-3σ) / μ.

8. The bridge cable force prediction method based on time series data according to claim 1 is characterized in that: In step 7, multiple surface change data of the first observation sample are input into the gated cycle prediction model along the order of the monitoring cycle, at least one model parameter of the gated cycle prediction model is trained, the surface change data of the next monitoring cycle is predicted according to the gated cycle prediction model, the surface state data of the next monitoring cycle is obtained according to the surface change data of the next monitoring cycle and the surface state data of the current monitoring cycle, and the multiple surface state data of the next monitoring cycle constitute a surface prediction sequence.

9. The bridge cable force prediction method based on time series data according to claim 7 is characterized in that: In step 8, the cable weights w1 and w2 are constructed, and the state prediction data d of each sampling point in the state prediction sequence is t×j =(w1a t×j +w2b t×j ), where a t×j is the surface state data of sampling point j in monitoring period t, b t×j To monitor the cross-sectional state data of sampling point j in period t, calculate the t×j [d0S2, d0S1] abnormal sampling points, where the maximum number of consecutive abnormal sampling points is n, the load index of the cable is k, d0 is the reference value of cable damage, if n×k of at least one cable exceeds the allowable upper limit, the current bridge cable force state is an alarm state.

10. A prediction device for implementing the bridge cable force prediction method based on time series data as claimed in claim 1, characterized in that: include: A crawling device, wherein the crawling device is configured to move to a plurality of sampling points along the length direction of the cable; A first acquisition device, the first acquisition device is configured to acquire a plurality of sets of surface images; A second acquisition device, the second acquisition device is configured to acquire multiple groups of original radiographic images; A third acquisition device, the third acquisition device is configured to acquire multiple sets of magnetic flux variation; A data processing device, the data processing device is configured to generate a surface state matrix, a cross-sectional state matrix, and a defect state matrix according to the surface image, the original radiographic image, and the magnetic flux variation; A data storage device, the data storage device being configured to store a surface state matrix, a cross-sectional state matrix, and a defect state matrix for a plurality of natural periods; A data analysis device, the data analysis device is configured to generate a state prediction sequence according to the surface state matrix, the cross-sectional state matrix, and the defect state matrix; The cable force prediction device is configured to predict the cable force state of the bridge.

Images