Highway bridge safety state assessment method and system

Abstract

The invention discloses a highway bridge safety state assessment method and system, and relates to the technical field of highway bridge state assessment, and the method comprises the steps: collecting the time sequence vibration data of a plurality of key nodes of a highway bridge and the overall surface image data; inputting each time sequence vibration data into the improved generative adversarial network, generating denoised time sequence vibration data, and performing initial state evaluation according to the denoised data to obtain an initial evaluation value; wherein a generator in the network adopts an encoder-decoder structure, and a channel attention mechanism is additionally arranged between feature encoding and feature decoding; identifying all cracks in the surface image data, performing region division on the cracks, calculating according to the crack of each region to obtain a secondary evaluation value, and using the secondary evaluation value to assist in adjusting the initial evaluation value; and for each adjusted evaluation value, integration is carried out based on the support degree between the data, and an overall evaluation result is obtained. The accuracy and efficiency of highway bridge safety state evaluation can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of highway bridge status assessment, and in particular to a highway bridge safety status assessment method and system. Background Art

[0002] During the use of highway bridges, they are often affected by traffic loads, environmental factors and natural disasters, which may cause structural damage or performance degradation such as cracks, deformation and corrosion, resulting in reduced stability and safety. Therefore, in order to ensure the safety of highway bridges, it is necessary to conduct safety status assessments on highway bridges and carry out effective maintenance based on the assessment results to extend the service life of highway bridges and reduce traffic interruptions and accident risks.

[0003] Traditional assessment methods for the safety status of highway bridges mainly rely on regular inspections, embedded sensor detection, and manual inspections. However, this method is costly, time-consuming, and labor-intensive, and it is easy to overlook hidden defects. In order to avoid the drawbacks of traditional methods, in the prior art, considering that when the stability of the highway bridge structure changes, there will be abnormal vibrations, a method of obtaining highway bridge vibration data through detection and evaluating the safety status based on the vibration data is proposed. However, this method still has certain problems, namely:

[0004] (1) In general, the data obtained by vibration sensor detection contains noise, which leads to inaccurate assessment of highway bridge safety directly based on vibration data. In order to ensure the accuracy of the final assessment, the vibration data needs to be denoised. However, existing denoising methods such as mean filtering and wavelet transform have certain drawbacks and cannot meet the current demand for effective denoising of vibration data. For example, the selection of the filter window in the mean filter affects the final filtering denoising effect, and the filter window is usually selected manually and cannot guarantee the filtering effect; the wavelet transform requires multiple decomposition and analysis of the data, which has a high computational cost and low efficiency.

[0005] (2) Vibration data is often obtained based on vibration sensors distributed at key nodes on key cross-sections of bridges. However, these nodes are discrete. If the damaged part is not equipped with corresponding sensors, it is impossible to determine whether local damage has occurred, which will affect the accuracy of the overall safety status of the highway bridge. If sensors are installed on a large area of ​​highway bridges, it will not only be costly and time-consuming and labor-intensive to install, but the processing of a large amount of vibration data will also further affect the efficiency of data analysis and processing. Summary of the invention

[0006] In order to address the deficiencies of the above-mentioned prior art, the present invention provides a method and system for assessing the safety status of a highway bridge. By collecting multi-dimensional monitoring data of the highway bridge: vibration data and surface image data, and combining the multi-dimensional monitoring data for overall analysis, a more accurate and comprehensive assessment of the safety status of the highway bridge can be achieved, thereby addressing the problem of poor accuracy in existing assessments based only on single data.

[0007] In a first aspect, the present invention provides a method for assessing the safety status of a highway bridge.

[0008] A safety status assessment method for a highway bridge, comprising:

[0009] Collect time-series vibration data and overall surface image data at multiple key nodes of highway bridges;

[0010] Each time series vibration data is input into the trained improved generative adversarial network to generate denoised time series vibration data, and then a preliminary state evaluation is performed based on the denoised data to obtain an initial evaluation value; in the improved generative adversarial network, the generator adopts an encoder-decoder structure, and a channel attention mechanism is added between feature encoding and decoding;

[0011] Identify all cracks in the surface image data and divide the cracks into regions, calculate a secondary evaluation value based on the cracks in each region, and use the secondary evaluation value to assist in adjusting the initial evaluation value;

[0012] For each adjusted evaluation value, the data is integrated based on the support between the data to obtain the overall evaluation result of the safety status of the highway bridge.

[0013] In a second aspect, the present invention provides a safety status assessment system for a highway bridge.

[0014] A highway bridge safety status assessment system, comprising:

[0015] A data acquisition module is used to collect time-series vibration data and overall surface image data at multiple key nodes of highway bridges;

[0016] An initial evaluation module is used to input each time series vibration data into the trained improved generative adversarial network to generate denoised time series vibration data, and then perform a preliminary state evaluation based on the denoised data to obtain an initial evaluation value; in the improved generative adversarial network, the generator adopts an encoder-decoder structure, and a channel attention mechanism is added between feature encoding and decoding;

[0017] A secondary adjustment module is used to identify all cracks in the surface image data and divide the cracks into regions, calculate a secondary evaluation value based on the cracks in each region, and use the secondary evaluation value to assist in adjusting the initial evaluation value;

[0018] The overall assessment module is used to integrate each adjusted assessment value based on the support between data to obtain the overall assessment result of the safety status of the highway bridge.

[0019] In a third aspect, the present invention further provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps of the method described in the first aspect are completed.

[0020] In a fourth aspect, the present invention further provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the method described in the first aspect.

[0021] One or more of the above technical solutions have the following beneficial effects:

[0022] 1. The present invention provides a method and system for assessing the safety status of a highway bridge. The method collects multi-dimensional monitoring data of the highway bridge, namely, one-dimensional time-series vibration data at each key node and two-dimensional surface image data of the whole bridge, and combines the two-dimensional monitoring data for overall analysis to solve the problem of poor assessment accuracy when the safety assessment is performed based on single-dimensional data. In the above scheme, on the one hand, time-series vibration data is used for preliminary monitoring and assessment, and on the other hand, an image recognition algorithm is used to identify surface cracks of the highway bridge, and the preliminary assessment results are adjusted according to the crack conditions in the local area, thereby compensating for the uneven distribution of vibration data and the inability to fully reflect local losses, thereby avoiding the problem that the assessment value with local errors affects the accuracy of the final global assessment, and avoiding the problem of low accuracy when single-dimensional assessment is performed using single vibration data or crack conditions.

[0023] 2. The present invention adopts an improved generative adversarial network to filter and denoise the time series vibration data. The improved generative adversarial network includes a generator and a discriminator. The generator adopts a U-NET structure (encoder-decoder structure), and a channel attention mechanism is added between each corresponding encoding layer and decoding layer to enhance the effect of feature fusion and improve the convergence speed of the network model. The time series vibration data is input into the trained network, and the generator extracts deep features and generates vibration data that is closer to the pure signal after denoising. Compared with existing signal denoising methods such as time domain-based, frequency domain-based and time-frequency analysis-based methods, more efficient denoising can be achieved and the denoising effect is better.

[0024] 3. Considering the high complexity of one-dimensional time series vibration data, some important features such as time correlation are often ignored during feature extraction and analysis. For this reason, the present invention adopts a data conversion method to convert the one-dimensional time series vibration data into two-dimensional image data, and then analyzes and evaluates based on the two-dimensional image data, which can effectively increase the time correlation. The neural network can also evaluate and diagnose based on the distribution of each pixel block of the image. Compared with one-dimensional time series data, two-dimensional images can highlight the intrinsic characteristics of the data, and the pixel block distribution rules of various data images are clear, which is more conducive to neural network recognition and improves the accuracy and efficiency of the final judgment.

[0025] 4. In the present invention, a secondary evaluation value based on crack identification is used to assist in adjusting the initial evaluation value, wherein a data consistency analysis method is used to analyze the consistency relationship between the initial evaluation and the secondary evaluation, and the initial evaluation result is assisted in adjusting according to the relationship, thereby taking into account various aspects and multi-dimensional influencing factors, ensuring that the final evaluation result is closer to the actual situation, and ensuring the accuracy of the evaluation; in addition, by analyzing the support between data, each evaluation value is integrated to obtain a total evaluation result that can cover the overall situation of the highway bridge, thereby accurately dividing the safety status of the highway bridge into safety levels, and providing data support for subsequent highway bridge maintenance measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0027] Figure 1 It is an overall flow chart of the safety status assessment method of a highway bridge according to an embodiment of the present invention;

[0028] Figure 2 Schematic diagram of the structure of an improved adversarial generative network in an embodiment of the present invention;

[0029] Figure 3 Schematic diagram of the attention module in the improved adversarial generative network in an embodiment of the present invention;

[0030] Figure 4 It is a schematic diagram of the process of crack target detection in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] It should be noted that the following detailed descriptions are exemplary only, are intended to describe specific embodiments, are intended to provide further explanation of the present invention, and are not intended to limit exemplary embodiments according to the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those of ordinary skill in the art to which the present invention belongs. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0032] Embodiment 1

[0033] This embodiment provides a method for evaluating the safety status of a highway bridge. Figure 1 As shown, the following steps are included:

[0034] Step S1, collecting time-series vibration data and overall surface image data at multiple key nodes of a highway bridge.

[0035] Step S1.1, collecting time-series vibration data at multiple key nodes of highway bridges.

[0036] Specifically, vibration sensors are arranged at key nodes of key sections of highway bridges. The vibration sensors may be highly sensitive seismographs or accelerometers. The vibration sensors are used for real-time monitoring, that is, monitoring the time series vibration data at multiple key nodes of the highway bridge under a set traffic load within a set time period, wherein the key nodes refer to the connection nodes between the bridge deck and each pier of the highway bridge; each vibration sensor is electrically connected to a data collector, which has a high sampling rate and can collect and capture instantaneous vibration data of the bridge under a set traffic load such as vehicle passing, so as to obtain time series vibration data.

[0037] Preferably, after acquiring a plurality of time-series vibration data, each time-series vibration data is processed using a linear interpolation method to avoid data loss due to fluctuations in collection frequency during the vibration data collection process, thereby ensuring the accuracy of data collection.

[0038] Step S1.2: Collect the surface image data of the entire highway bridge.

[0039] Specifically, a surface image of the entire outer surface of the highway bridge is obtained by using a drone or other equipment. In fact, the overall structure of the highway bridge includes a variety of components, such as a bridge span structure (such as a bridge deck, etc.), bearings, piers, abutments, main beams, cantilevers, pier foundations, etc. In this step, a surface image of each component of the highway bridge is obtained.

[0040] Step S2: input each time series vibration data into the trained improved generative adversarial network to generate denoised time series vibration data, and then perform a preliminary state evaluation based on the denoised data to obtain an initial evaluation value; wherein, in the improved generative adversarial network, the generator adopts an encoder-decoder structure, and adds a channel attention mechanism between feature encoding and decoding.

[0041] Specifically, the time series vibration data of each key node is denoised using an improved adversarial generative network, and the denoised time series vibration data is converted into a two-dimensional image, which is input into the trained state assessment model to output the preliminary assessment value of each key node. In this embodiment, the preliminary assessment value sequence of multiple key nodes is obtained according to the arrangement order of n vibration sensors from left to right (or from right to left) on the highway bridge. Where n represents the number of vibration sensors. Represents the preliminary evaluation value corresponding to the i-th key node.

[0042] Step S2.1: For the time series vibration data of each key node, denoising is performed using an improved generative adversarial network to generate denoised time series vibration data.

[0043] In the process of using sensors to measure and obtain time-series vibration data, noise influence is inevitable. For example, environmental anomalies may cause similar abnormal increases or drifts in the data collected by multiple sensors, defects in the sensor equipment itself may cause abnormal data collection, and large differences in accuracy between the deployed sensors may cause data collected by some sensors to have data jumps, trends, missing and drift. Therefore, when evaluating the safety status of highway bridges based on time-series vibration data, it is first necessary to eliminate the noise interference caused by environmental anomalies and sensor abnormalities through data cleaning and denoising to avoid errors or anomalies in subsequent analysis.

[0044] Since the existing denoising methods have low processing efficiency and poor accuracy, this embodiment adopts a machine learning method and proposes a vibration data denoising method based on an improved adversarial generative network model (GAN). The network is used to identify noise data (or noise signals) in the collected data and effectively suppress the noise signals. The noise in the original time series vibration data can be quickly and effectively removed, and then an accurate safety status assessment can be performed based on the denoised time series vibration data.

[0045] Specifically, for time series vibration data (also called vibration signals), in order to better retain useful signals and suppress noise signals, this embodiment adopts a denoising method based on an attention mechanism to improve the generative adversarial network. Figure 2As shown, the improved generative adversarial network includes a generator and a discriminator. First, the generator is used to remove the noise component in the noisy signal (i.e., the original time series vibration signal) and establish a mapping relationship between the noisy signal and the denoised signal; secondly, the denoised signal generated by the generator and the theoretically pure signal are input into the discriminator for classification. If the output label of the discriminator is closer to 1, it is considered that the input data signal is closer to the pure signal, otherwise the output label is closer to 0, that is, the input data signal is considered to be closer to the denoised signal.

[0046] By pre-training the above-mentioned improved generative adversarial network, the denoised signal generated by the generator is closer to the pure signal, making it impossible for the discriminator to determine whether the input is a pure signal. On this basis, the trained generator is used to generate a denoised vibration signal for the input original vibration signal with noise.

[0047] Among them, in the above-mentioned improved adversarial generative network model, the generator adopts a U-NET structure (i.e., an encoder-decoder structure), including a feature encoding unit and a feature decoding unit, wherein the feature encoding unit includes multiple encoding layers, each encoding layer includes a downsampling unit, and the input noisy signal is subjected to multiple one-dimensional convolutions through multiple downsampling units to continuously compress and extract features; the feature decoding unit includes multiple decoding layers, each decoding layer includes an upsampling unit, and the compressed deep features are gradually restored to features of the same scale as the input signal through the one-dimensional deconvolution of the upsampling unit; and the feature encoding unit corresponds to the feature map size extracted in the feature decoding unit one by one, and is spliced ​​through jump connections to fuse low-dimensional features (i.e., encoding feature maps) and high-dimensional features (i.e., decoding feature maps); finally, a denoised signal is generated based on the fused feature output.

[0048] Furthermore, in order to enhance the fusion effect of low-dimensional features and high-dimensional features, the U-NET structure of the generator is improved, and a channel attention mechanism is added between feature encoding and feature decoding, such as Figure 3 As shown, for each corresponding encoding layer and decoding layer, the fusion method of the extracted encoding feature map and decoding feature map is:

[0049] (1) Encoded feature map x output by the encoding layer e And the decoded feature map x output by the decoding layer d Through the 1×1 convolution kernel, convolution with a step size of 1 is performed to obtain two weight vectors W e and W d ; Among them, the encoding feature map x e And decoded feature map x d The size of is (B, H, C), where B, H, and C represent length, width, and dimension respectively. In this embodiment, H=256;

[0050] (2) The weight vector W e and W d After adding, the LeakyRelu activation function is used to activate it and the activation result W is obtained. a1 ;

[0051] (3) The activation result is convolved with a 1×1 convolution kernel and activated by Sigmoid to obtain the attention weight layer W. a2 , the attention weight layer W a2 After resampling, it is combined with the decoding feature map x output by the current decoding layer d Multiply to complete the decoding feature map x d attention activation.

[0052] By adding a channel attention mechanism to the U-NET structure of the generator, the effect of feature fusion can be further enhanced, the convergence speed of the network model can be improved, and the training loss value can be lowered. Based on the improved generative adversarial network completed by the above training, the generator is used to process the input time series vibration data to generate vibration data that is closer to the pure signal after denoising. This data denoising method avoids the complex data analysis and processing process, and is more efficient and has better denoising effect.

[0053] Step S2.2: convert the denoised time series vibration data into a two-dimensional image, input the two-dimensional image into a state assessment model, and obtain an initial assessment value.

[0054] Since vibration data is one-dimensional time series data, it has high complexity. It is difficult for a simple neural network to fully extract the features of the time series data, and it often ignores important features such as time correlation, which affects the accuracy of the final output results. Correspondingly, if a complex model is used for learning and training, the training and time costs will increase. Therefore, the effect of abnormal diagnosis and evaluation directly based on one-dimensional time series vibration data is not ideal. For this reason, in this embodiment, the one-dimensional time series vibration data is converted into a two-dimensional image, which can effectively enhance the time correlation. The neural network can evaluate and diagnose based on the distribution of each pixel block of the image. Compared with one-dimensional time series data, two-dimensional images can highlight the intrinsic characteristics of the data. The pixel block distribution rules of various data images are clear, which is more conducive to neural network recognition and improves the accuracy of the final judgment.

[0055] In this embodiment, the Gramian Summation Angular Field (GAF) method is used to convert the one-dimensional time series data into a two-dimensional grayscale image, so as to better reflect the potential features between the time series data and extract richer feature representations. As another implementation method, the Markov transition field (MTF), recursive graph (RP) and other methods can also be used to convert the one-dimensional time series data into a two-dimensional image.

[0056] Specifically, the principle of the Gram angular field is to convert the scaled one-dimensional sequence data from the rectangular coordinate system to the polar coordinate system, and then identify the correlation of different time points by considering the angle sum or difference between different points. This embodiment adopts the Gramian Difference Angular Field (GADF) method to convert the one-dimensional time series data into a two-dimensional grayscale image by angle difference calculation, which specifically includes the following steps:

[0057] Step S2.2.1: for each denoised time series vibration data, such as the i-th time series vibration data x i ={x i1 , x i2 , ..., x ij ,...x im}, where i = 1, 2, ..., n, j = 1, 2, ..., m, m represents the number of time series data, and the data is normalized using the following formula to scale the sequence to the range of [-1, 1]. The normalization formula is:

[0058]

[0059] Step S2.2.2: Encode the normalized data as the cosine of the vector angle The corresponding timestamp is encoded as radius r j , then the time series in the Cartesian coordinate system can be transformed into the time series in the polar coordinate system as follows:

[0060]

[0061]

[0062] In the above formula, t j is the timestamp, j represents the jth data, and m is the sample size of the data.

[0063] Under the above encoding method, as time increases, the corresponding values ​​are distorted between different corners on the span circle. By converting to the polar coordinate system, the polar coordinates are divided by the radius r j The time dependency of the sequences is preserved and each sequence produces a unique polar coordinate map.

[0064] Step S2.2.3, for time series data with a length of n, the converted GADF size is a matrix of [n, n]. The vector angle between the two values ​​is obtained by the following formula, and the sine value of the angle difference between the two vectors is calculated to obtain the Gram matrix. The image is drawn according to the matrix to obtain a two-dimensional grayscale image.

[0065]

[0066] In the above formula, Represents the cosine of the angle between the vectors.

[0067] Step S2.2.4, input the two-dimensional grayscale image into the trained BP network-based state assessment model, output the confidence of the safety state of the highway bridge, and output the confidence as the assessment value, which is the value obtained after the preliminary state assessment. In this embodiment, the state assessment model is constructed using a BP neural network, with time series vibration data of several pre-labeled state values ​​as a training set, and the time series vibration data is converted into a two-dimensional grayscale image, and the BP neural network is trained using the image training set to obtain a trained state assessment model.

[0068] Step S3, identifying all cracks in the surface image data and dividing the cracks into regions, calculating a secondary evaluation value based on the cracks in each region, and using the secondary evaluation value to assist in adjusting the initial evaluation value.

[0069] Step S3.1: Input the surface image of the entire outer surface of the highway bridge into the target detection model, identify the cracks in the image and detect the length and width of the cracks.

[0070] In this embodiment, a target detection model is constructed, which includes a crack recognition module and a crack extraction module. The model is used to recognize cracks in the surface image and extract crack targets with a single pixel width.

[0071] like Figure 4 As shown, first, for the crack recognition module: since the crack target to be detected has the characteristics of a small-size structure, its detail information is easily lost during the feature extraction process, resulting in difficulty in crack recognition and low recognition accuracy. For this reason, this embodiment uses the VGG16 network as a feature extraction network, and uses the network to extract the global feature map of the input image. The network only uses a 3×3 convolution kernel. Under the condition of the same receptive domain, it can improve the network depth while reducing the number of parameters, and reduce the probability of loss of crack target details due to convolution operations; the feature map extracted by the VGG16 feature extraction network is input into the RPN network (Region Proposal Network), and a series of candidate boxes for identifying cracks are output through the network; the output candidate box is combined with the feature map and input into the RoIAlign layer, and the part of the candidate box corresponding to the feature map is cropped, that is, the candidate box feature map; finally, the candidate box feature map is classified and regressed by the fully connected layer, and the identified crack is output.

[0072] Secondly, for the crack extraction module: the identified crack area image is binarized, denoised and refined in turn to extract the crack target from the crack area. Specifically, the adaptive threshold method is used to segment the crack target from the background of the crack area image to generate a crack target binary image; considering that the binarization process usually produces isolated foreground noise and target fractures, a clean and complete crack target binary image is obtained through fracture connection and denoising, in which the closed operation is used to bridge the fractured part of the crack target, the crack target boundary is smoothed while keeping the original crack target area unchanged, and the connected component labeling method is used to remove background noise; the crack target binary image is refined to obtain a single-pixel wide crack target.

[0073] Furthermore, based on the proposed crack target, the length and width values ​​of the crack target are obtained by calculating the number of pixels of the crack target.

[0074] Step S3.2: Divide all identified cracks into regions. Specifically, according to the locations of n vibration sensors on the highway bridge, the entire highway bridge is divided into n regions, and each region includes at least one component; then, based on the divided regions, all identified cracks are divided into regions to determine all cracks in each region.

[0075] Step S3.3: For all cracks in each area, calculate the secondary evaluation value of each area of ​​the highway bridge according to the length and width of the cracks, wherein the corresponding secondary evaluation value sequence is obtained corresponding to the area order from left to right (or from right to left).

[0076] Specifically, according to the cracks identified in the current area and their length and width, the secondary assessment value of the area is calculated. include:

[0077] Step S3.3.1, calculate the crack area s of all cracks in each area of ​​the highway bridge q , the crack area of ​​the i-th region is:

[0078]

[0079] Among them, l q represents the crack length, h q represents the crack width, q represents the qth crack identified in the area, and i = 1, 2, ... n.

[0080] Step S3.3.2: Based on the crack area of ​​each region and the ratio of the crack area to the total crack area, determine the secondary assessment value of each region. The formula is:

[0081]

[0082] Step S3.4: Based on the data consistency analysis method, the secondary evaluation value is used to assist in adjusting the initial evaluation value, specifically:

[0083] Step S3.4.1: normalize the initial evaluation value sequences and the secondary evaluation value sequences of the multiple regions respectively.

[0084] Step S3.4.2: Based on the normalized initial evaluation value sequence and the secondary evaluation value sequence, perform data consistency analysis on each pair of evaluation values, calculate the weight coefficient of each pair of evaluation values, and obtain the adjusted evaluation value sequence through weighted fusion, which is recorded as y1, y2, ..., y i ,...,y n .

[0085] Specifically, based on the initial evaluation value sequence and the second evaluation value sequence The support function is used to analyze the consistency between each pair of evaluation values. The formula is:

[0086]

[0087] In the above formula, It represents the support between the two evaluation values ​​at the ith node (or the ith region), and the support is recorded as the consistency coefficient; G[*] represents the support function, K∈[0,1] is the amplitude of the support function, and β≥0 is the decay rate of the support function.

[0088] Through the above method, the consistency coefficient sequence is calculated and recorded as z1,z2,...,z i ,...,z n .

[0089] Furthermore, a threshold Z of the consistency coefficient is set. In this embodiment, the threshold Z is set to 0.8. When the consistency coefficient is greater than or equal to the threshold, it means that the initial evaluation value and the secondary evaluation value are close, and the results of the two evaluation methods are not much different. The initial evaluation value at this time can more accurately reflect the actual situation. The consistency coefficient is used as the weight coefficient of the two evaluation values, and weighted fusion is performed to obtain the adjusted evaluation value y n , and its calculation formula is:

[0090]

[0091] Correspondingly, when the consistency coefficient is less than the set threshold value Z, it means that the initial evaluation value and the secondary evaluation value are quite different at this time, and the results of the two evaluation methods are quite different. At this time, the initial evaluation value may be difficult to accurately reflect the actual state of the highway bridge due to problems such as unreasonable sensor layout. Therefore, it is necessary to use the secondary evaluation value for auxiliary adjustment at this time: on the one hand, considering that the initial evaluation value reflects the safety status of different aspects of the highway bridge, directly using the secondary evaluation value for replacement is likely to lead to the loss of some factors affecting the safety status. On the other hand, the secondary evaluation value can better reflect the real safety status at this time. Therefore, directly using the above-mentioned weighted fusion method will lead to biased final results. For this reason, considering that the various regions and components of the highway bridge are continuous and related, their consistency also has certain continuity and correlation. In this embodiment, the current consistency coefficient is used as the weight coefficient of the initial evaluation value, and the maximum consistency coefficient within the set scale before and after the current consistency coefficient (in this embodiment, the set scale is 1, that is, the current consistency coefficient and its two coefficients before and after) is used as the weight coefficient of the secondary evaluation value. Weighted fusion is performed to obtain the adjusted evaluation value y n , and its calculation formula is:

[0092]

[0093] Through the above method, the adjusted evaluation value sequence y1, y2, ..., y i ,...,y n .

[0094] Step S4: for each adjusted evaluation value, integrate the data based on the support between the data to obtain the overall evaluation result of the highway bridge safety status. Specifically, based on the adjusted evaluation value sequence y1, y2, ..., y i ,...,y n , analyze the support between data, integrate them according to the support, and get the final overall evaluation value y of the safety status of highway bridges. This method can effectively avoid the problem of artificially setting weights in traditional weighted fusion methods that affects the accuracy of the final results.

[0095] Step S4.1: Calculate the support r between the two evaluation values ​​based on the support function calculation formula ij , construct the support matrix, which can be expressed as:

[0096]

[0097] The values ​​in the above matrix represent the mutual support between the two evaluation values. Since it cannot explain the total support of a certain value relative to the whole, the weight of each evaluation value is calculated in the following way to complete the integrated calculation.

[0098] Step S4.2: Based on the support matrix, calculate the weight coefficient of each evaluation value in all evaluation values, specifically:

[0099] Randomly set a set of non-negative numbers (a1, a2, ..., a i , ..., a n ), such that:

[0100] w i =a1r i1 +a2r i2 +…+a n r in , i=1,2,...n;

[0101] The matrix form can be expressed as:

[0102] W = R * A;

[0103] Where W = [w1, w2, ..., w i , ..., w n ] T , A=[a1,a2,...,a i , ..., a n ] T .

[0104] Since the support matrix is ​​a non-negative matrix, there is a maximum modulus eigenvalue λ≥0. The eigenvector A of the maximum modulus eigenvalue can be obtained by solving 1*A=R*A, and then the weight of each data can be obtained, which can be expressed as:

[0105]

[0106] Step S4.3: Perform integration calculation based on the weight coefficients to obtain the overall evaluation result of the safety status of the highway bridge, that is, y = w1y1+w2y2+…+w n y n .

[0107] As another implementation method, based on the overall evaluation results obtained by the above calculations and combined with preset evaluation levels (different evaluation values ​​correspond to different levels), the safety status of the highway bridge is accurately divided into safety levels to provide data support for subsequent maintenance measures for the highway bridge.

[0108] Embodiment 2

[0109] This embodiment provides a highway bridge safety status assessment system, including:

[0110] A data acquisition module is used to collect time-series vibration data and overall surface image data at multiple key nodes of highway bridges;

[0111] An initial evaluation module is used to input each time series vibration data into the trained improved generative adversarial network to generate denoised time series vibration data, and then perform a preliminary state evaluation based on the denoised data to obtain an initial evaluation value; in the improved generative adversarial network, the generator adopts an encoder-decoder structure, and a channel attention mechanism is added between feature encoding and decoding;

[0112] A secondary adjustment module is used to identify all cracks in the surface image data and divide the cracks into regions, calculate a secondary evaluation value based on the cracks in each region, and use the secondary evaluation value to assist in adjusting the initial evaluation value;

[0113] The overall assessment module is used to integrate each adjusted assessment value based on the support between data to obtain the overall assessment result of the safety status of the highway bridge.

[0114] Embodiment 3

[0115] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above-mentioned method for assessing the safety status of a highway bridge are completed.

[0116] Embodiment 4

[0117] This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned method for assessing the safety status of a highway bridge are completed.

[0118] The steps involved in the above embodiments 2 to 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0119] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0120] The above description is only a preferred embodiment of the present invention. Although the specific implementation mode of the present invention is described in conjunction with the accompanying drawings, it is not a limitation of the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the protection scope of the present invention.

Claims

1. A method for evaluating the safety status of a highway bridge, characterized in that: include: Collect time-series vibration data and overall surface image data at multiple key nodes of highway bridges; Each time series vibration data is input into the trained improved generative adversarial network to generate denoised time series vibration data, and then a preliminary state evaluation is performed based on the denoised data to obtain an initial evaluation value; In the improved generative adversarial network, the generator adopts an encoder-decoder structure, and a channel attention mechanism is added between feature encoding and decoding; Identify all cracks in the surface image data and divide the cracks into regions, calculate a secondary evaluation value based on the cracks in each region, and use the secondary evaluation value to assist in adjusting the initial evaluation value; For each adjusted evaluation value, the data is integrated based on the support between the data to obtain the overall evaluation result of the safety status of the highway bridge.

2. A method for evaluating the safety status of a highway bridge according to claim 1, characterized in that: The improved adversarial generative network includes a generator and a discriminator. The time series vibration data is used as a noisy signal. The generator is used to remove the noise component in the noisy signal and establish a mapping relationship between the noisy signal and the denoised signal; the discriminator is used to classify and judge the denoised signal generated by the generator; The generator adopts an encoder-decoder structure, including a feature encoding unit and a feature decoding unit; The feature encoding unit includes multiple encoding layers, each encoding layer includes a downsampling unit, and the input noisy signal is subjected to multiple one-dimensional convolutions through multiple downsampling units to extract deep features; The feature decoding unit includes multiple layers of decoding layers, each decoding layer includes an upsampling unit, and the extracted deep features are gradually restored to feature vectors with the same scale as the input signal through multiple one-dimensional deconvolutions of multiple upsampling units; wherein the encoding feature map output by each encoding layer and the decoding feature map output by the corresponding decoding layer are jump-joined through a channel attention mechanism.

3. A method for evaluating the safety status of a highway bridge according to claim 2, characterized in that: A channel attention mechanism is added between the decoded feature map and the encoded feature map, including: The encoded feature map x output by the encoding layer e And the decoded feature map x output by the decoding layer d Convolution is performed through the same convolution kernel to obtain two weight vectors W e and W d ; The weight vector W e and W d After addition, the LeakyRelu activation function is used for activation to obtain the activation result W a1 ; Activation result W a1 After convolution and activation, we get the attention weight layer W a2 , the attention weight layer W a2 After resampling, it is combined with the decoding feature map x output by the current decoding layer d Multiply to complete the decoding feature map x d attention activation.

4. A method for evaluating the safety status of a highway bridge according to claim 1, characterized in that: Perform a preliminary state assessment based on the denoised data to obtain the initial assessment value, including: The Gram angular field method is used to convert the denoised time series vibration data into a two-dimensional grayscale image, which is then input into a state assessment model based on a BP neural network to output an initial assessment value.

5. A method for evaluating the safety status of a highway bridge according to claim 1, characterized in that: Identify all cracks in the surface image data and divide the cracks into regions. Calculate the secondary evaluation value based on the cracks in each region, including: The surface image of the entire outer surface of the highway bridge is input into the target detection model, all cracks in the image are identified and each crack target is extracted, and the length and width of each crack target are obtained by calculating the number of pixels of the crack target; All identified cracks are zoned; For all cracks in each area, the secondary assessment value of each area of ​​the highway bridge is calculated based on the length and width of the cracks.

6. A method for evaluating the safety status of a highway bridge according to claim 1, characterized in that: Based on the data consistency analysis method, the secondary evaluation value is used to assist in adjusting the initial evaluation value, including: Normalization processing is performed on the initial evaluation value sequences and the secondary evaluation value sequences of multiple regions respectively; Based on the normalized initial evaluation value sequence and the secondary evaluation value sequence, data consistency analysis is performed for each pair of evaluation values, and the weight coefficient of each pair of evaluation values ​​is calculated. Through weighted fusion, the adjusted evaluation value sequence is obtained; Among them, for each pair of evaluation values, if the calculated consistency coefficient is greater than or equal to the set threshold, the consistency coefficient is used as the weight coefficient of the two evaluation values; if the calculated consistency coefficient is less than the set threshold, the current consistency coefficient is used as the weight coefficient of the initial evaluation value, and the maximum consistency coefficient within the set scale before and after the current consistency coefficient is used as the weight coefficient of the secondary evaluation value.

7. A method for evaluating the safety status of a highway bridge according to claim 1, characterized in that: For each adjusted evaluation value, the support between the data is integrated to obtain the overall evaluation results of the highway bridge safety status, including: Based on the support function calculation formula, the support between the two evaluation values ​​is calculated to construct a support matrix; Based on the support matrix, calculate the weight coefficient of each evaluation value in all evaluation values; The integrated calculation is performed according to the weight coefficients to obtain the overall assessment result of the safety status of the highway bridge.

8. A highway bridge safety status assessment system, characterized in that: include: A data acquisition module is used to collect time-series vibration data and overall surface image data at multiple key nodes of highway bridges; An initial evaluation module is used to input each time series vibration data into the trained improved generative adversarial network to generate denoised time series vibration data, and then perform a preliminary state evaluation based on the denoised data to obtain an initial evaluation value; In the improved generative adversarial network, the generator adopts an encoder-decoder structure, and a channel attention mechanism is added between feature encoding and decoding; A secondary adjustment module is used to identify all cracks in the surface image data and divide the cracks into regions, calculate a secondary evaluation value based on the cracks in each region, and use the secondary evaluation value to assist in adjusting the initial evaluation value; The overall assessment module is used to integrate each adjusted assessment value based on the support between data to obtain the overall assessment result of the safety status of the highway bridge.

9. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of a method for evaluating the safety status of a highway bridge as described in any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a method for assessing the safety status of a highway bridge as described in any one of claims 1-7.

Images