阅读说明：本技术 基于时空关联模型的桥梁集群结构损伤定位方法 (Bridge cluster structure damage positioning method based on space-time correlation model ) 是由 刘洋 曹建新 郭成军 于 2021-07-12 设计创作，主要内容包括：本发明公开了一种基于时空关联模型的桥梁集群结构损伤定位方法,所述方法如下：一、在集群内各结构形式相似桥梁的关键断面布设应变传感器,实现桥梁集群结构应变测点群监测数据的实时采集；二、对桥梁集群结构采集的长期应变测点群监测数据进行预处理；三、建立参考桥梁应变测点群监测数据与集群内其它桥梁应变测点群监测数据的时空关联模型；四、利用时空关联模型,对集群内桥梁的应变响应进行预测,并利用应变测点群预测应变值与实测应变值构建损伤定位因子和阈值,完成集群内桥梁结构的损伤定位。本发明建立了集群内桥梁之间的时空关联模型,解决了复杂环境耦合作用下集群内全部结构形式相似桥梁的损伤定位难题。(The invention discloses a method for positioning damage of a bridge cluster structure based on a space-time correlation model, which comprises the following steps: firstly, strain sensors are distributed on key sections of bridges with similar structural forms in a cluster, so that real-time acquisition of monitoring data of strain measuring point groups of a bridge cluster structure is realized; secondly, preprocessing long-term strain measuring point group monitoring data collected by the bridge cluster structure; establishing a space-time correlation model of the reference bridge strain measuring point group monitoring data and other bridge strain measuring point group monitoring data in the cluster; and fourthly, predicting the strain response of the bridge in the cluster by using a space-time correlation model, and constructing a damage positioning factor and a threshold value by using the predicted strain value and the actually measured strain value of the strain measurement point cluster to complete the damage positioning of the bridge structure in the cluster. The invention establishes a space-time correlation model between the bridges in the cluster, and solves the problem of damage positioning of all bridges with similar structural forms in the cluster under the coupling action of complex environments.)

1. A method for positioning damage of a bridge cluster structure based on a space-time correlation model is characterized by comprising the following steps:

step one, strain sensors are distributed on key sections of bridges with similar structural forms in a cluster, a bridge cluster structure strain measurement point group monitoring system is established, and real-time collection of bridge cluster structure strain measurement point group monitoring data is achieved;

secondly, preprocessing long-term strain measuring point group monitoring data collected by the bridge cluster structure;

selecting one bridge in the cluster as a reference bridge, and establishing a space-time correlation model of the reference bridge strain measuring point group monitoring data and other bridge strain measuring point group monitoring data in the cluster by using the strain long-term monitoring data in the structural health state;

and step four, predicting the strain response of the bridge in the cluster by using the space-time correlation model constructed in the step three, and constructing a damage positioning factor and a threshold value by using the predicted strain value and the measured strain value of the strain measurement point group to complete the damage positioning of the bridge structure in the cluster.

2. The method for positioning damage to a bridge cluster structure based on a spatio-temporal correlation model according to claim 1, characterized in that the concrete steps of the second step are as follows:

step two, denoising and eliminating random vehicle-mounted influence processing are carried out on the strain monitoring data of all bridges in the cluster:

wherein n represents the total number of sampling points; k represents any sampling point of the strain of the bridge structure, and k belongs to (1,2, …, n); i represents any measuring point of the bridge structure strain, i belongs to (1,2, …, m), and m is the number of the bridge structure strain sensors;for any strain measurement point i, kthStrain data values of the sampling points;representing a strain data value of a kth sampling point of any strain measuring point i after noise removal and random vehicle-mounted influence elimination processing;

step two, standardizing the strain monitoring data after the noise removal and random vehicle-mounted influence elimination treatment, wherein the standardized treatment formula is as follows:

in the formula (I), the compound is shown in the specification,representing the strain value of the kth sampling point of the strain measuring point i after the standardization treatment; e (x)ⁱ) Representing the mean value of the strain measuring point i monitoring data set; d (x)ⁱ) Representing the variance of the strain measurement point i monitoring data set;

step two and step three, forming a strain monitoring matrix by using all the strain monitoring data after standardized processing of each bridge in the cluster

In the formula (I), the compound is shown in the specification,representing a vector formed by all the measuring point strain data of the single bridge at any kth sampling point after the normalization treatment,

3. the method for positioning damage to a bridge cluster structure based on a spatio-temporal correlation model according to claim 1, characterized in that the concrete steps of the third step are as follows:

step three, establishing a space-time correlation model training network combining a deep convolutional neural network and a long-time and short-time memory network;

step three, taking any bridge A in the cluster as a reference bridge, and monitoring the strain of the bridge A in a healthy state by using a strain monitoring matrixAs network input, simultaneously monitoring the strain of any other bridge B in the cluster and the bridge A at the same sampling timeAs a network output;

step three, training the time-space correlation model training network until the network loss function is converged, wherein the formula of the network loss function is as follows:

wherein Γ (Θ) is a loss function; theta is a hidden layer parameter set of the training network; kappa (-) is a strain monitoring data prediction function based on a space-time correlation model between bridges.

4. The method for positioning damage to a bridge cluster structure based on a spatio-temporal correlation model according to claim 3, wherein the structure of the spatio-temporal correlation model training network comprises: the system comprises a data input layer, a time sequence data folding layer, a convolution layer, a maximum pooling layer, a batch normalization layer, an activation function layer, a time sequence data expansion layer, a data flattening layer, a long-short time memory network layer I, a long-short time memory network layer II, a full connection layer and a data output regression layer, wherein the structural layers in the network are sequentially connected, and the time sequence data folding layer is connected with the time sequence data expansion layer.

5. The method for positioning damage to a bridge cluster structure based on a spatio-temporal correlation model according to claim 1, characterized in that the concrete steps of the fourth step are as follows:

and step four, based on a verification data set under the bridge health state, constructing a damage positioning factor under the health state by utilizing a training network of a time-space correlation model which is completely trained in the step three:

in the formula (I), the compound is shown in the specification,representing a residual vector of a predicted value and an actual measured value of any ith strain measuring point in the bridge health state;represents the k th strain measuring point of any ith strain measuring point under the bridge health state₁Residual values of the predicted values and the measured values of the sampling points;represents the k th strain measuring point of any ith strain measuring point under the bridge health state₁A damage localization factor of each sampling point; k is a radical of_hRepresenting the total number of verification data set samples in the bridge health state; e (-) represents the mean of the calculated vectors; d (-) represents the variance of the calculation vector;

on the basis, an arbitrary ith strain measuring point damage positioning factor set under the bridge health state is constructed

Step four, establishing a damage positioning threshold gamma of any ith strain measurement point of the bridgeⁱ：

In the formula, gamma_0.95(. cndot.) represents a 95% confidence median function of the extraction set;representing a guarantee coefficient of the damage positioning of the bridge structure;

step four, constructing a damage positioning factor of the bridge in a state to be diagnosed:

in the formula, d represents the state to be diagnosed of the bridge structure;

step four: judging the damage position of the bridge structure, and if the damage positioning factor set is the median in the state to be diagnosedIndicating that the structure at the measuring point is damaged; otherwise, the bridge structure is in a safe state.

Technical Field

The invention belongs to the field of monitoring of operation safety of a bridge cluster structure, and relates to a method for positioning damage of the bridge cluster structure based on a space-time correlation model.

Background

The bridge is used as an important component of road traffic infrastructure, and the coupling effect of various factors such as environmental erosion, material aging, long-term effect of load and the like during operation inevitably leads to structural damage accumulation and resistance attenuation. With the continuous update of advanced sensing technology and data processing method, the damage identification theory of the single bridge structure has been developed greatly, and the problem of monitoring the operation safety of the actual bridge structure is solved to a certain extent. Along with the development of large-scale, large-scale and clustering of bridge construction in China, the technical requirement for solving the operation safety monitoring of a bridge structure of multiple groups is more and more urgent. However, at present, the direct research results in the aspect of bridge cluster structure operation safety monitoring are almost blank, and the successful experience and cases which can be used for reference at home and abroad are almost zero. Therefore, the bridge cluster structure damage identification method becomes a difficult point which needs to be broken through urgently in the field of structural health monitoring.

There are often numerous bridges with similar structural forms within a bridge cluster, for example: a box type beam bridge group, a T-shaped beam bridge group, a hollow slab beam bridge group and the like. The external loads to which these bridges are subjected are similar, such as similar overall temperature loads, similar vertical temperature differential loads, similar vehicle loads, and the like. The similarity of the load borne by the bridges enables strain monitoring data among the bridges in the cluster to have a complex space-time mapping relation, and the traditional machine learning methods such as an artificial neural network cannot accurately mine the space-time correlation of the strain monitoring data among the bridges due to insufficient calculation depth. In recent years, with the continuous upgrade of computer hardware and the rapid development of big data analysis algorithms, deep learning algorithms are deeply researched.

Disclosure of Invention

The invention provides a bridge cluster structure damage positioning method based on a space-time correlation model, and aims to solve the problems that at present, bridge cluster monitoring data are subjected to the coupling effect of various operating environment factors, decoupling of the monitoring data under the coupling effect of the various environment factors is often difficult to directly realize, and damage cannot be effectively positioned. The method provides a novel deep learning algorithm, establishes a space-time correlation model between bridges, and further realizes damage positioning of bridge structures with similar structural forms in a cluster based on the model.

The purpose of the invention is realized by the following technical scheme:

a method for positioning damage of a bridge cluster structure based on a space-time correlation model comprises the following steps:

step one, strain sensors are distributed on key sections of bridges with similar structural forms in a cluster, a bridge cluster structure strain measurement point group monitoring system is established, and real-time collection of bridge cluster structure strain measurement point group monitoring data is achieved;

step two, preprocessing long-term strain measurement point group monitoring data acquired by the bridge cluster structure, and mainly comprising the following steps of: denoising, eliminating random vehicle-mounted influence processing and strain data standardization processing;

selecting one bridge in the cluster as a reference bridge, and establishing a space-time correlation model of the reference bridge strain measuring point group monitoring data and other bridge strain measuring point group monitoring data in the cluster by using the strain long-term monitoring data in the structural health state;

and step four, predicting the strain response of the bridge in the cluster by using the space-time correlation model constructed in the step three, and constructing a damage positioning factor and a threshold value by using the predicted strain value and the measured strain value of the strain measurement point group to complete the damage positioning of the bridge structure in the cluster.

Compared with the prior art, the invention has the following advantages:

the invention establishes a space-time correlation model between the bridges in the cluster, and solves the problem of damage positioning of all bridges with similar structural forms in the cluster under the coupling action of complex environments.

Drawings

FIG. 1 is a flowchart of a method for positioning damage to a bridge cluster structure based on a spatio-temporal correlation model.

FIG. 2 is a diagram of a spatiotemporal correlation model training network.

FIG. 3 is a photograph of an embodiment of an overpass.

FIG. 4 is a diagram showing the positions of sensors of a bridge A according to the embodiment.

FIG. 5 is a diagram showing the positions of sensors of a bridge B according to an embodiment.

Fig. 6 is a photograph of the sensor mounting in the example.

Fig. 7 is a photograph of a sensor data acquisition in an embodiment.

FIG. 8 is a graph showing the result of strain prediction of the #4 measuring point of the bridge A bottom plate by using the method of the present invention in the examples.

FIG. 9 shows the strain prediction results of the #5 measuring point of the top plate of the bridge A using the method of the present invention in the examples.

FIG. 10 is a diagram illustrating strain prediction results of a #4 measuring point of a bottom plate of a bridge A by using a conventional support vector machine method in the embodiment.

FIG. 11 is a strain prediction result of the #5 measuring point of the top plate of the bridge A by using the conventional support vector machine method in the embodiment.

Fig. 12 shows the damage localization result of the bridge a according to the method of the present invention in the embodiment.

Fig. 13 is a result of positioning the damage to the bridge a by using the conventional support vector machine method in the embodiment.

Detailed Description

The technical solution of the present invention is further described below with reference to the accompanying drawings, but not limited thereto, and any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention shall be covered by the protection scope of the present invention.

The invention provides a method for positioning damage of a bridge cluster structure based on a space-time correlation model, which comprises the steps of firstly, arranging strain sensors on bridge key sections with similar structural forms in a cluster to realize real-time acquisition of strain measurement point group monitoring data of the bridge cluster structure; secondly, denoising and eliminating random vehicle-mounted influence processing are carried out on long-term strain measuring point group monitoring data collected by a bridge cluster structure, and strain data are processed in a standardized mode; on the basis, a space-time correlation model of the strain monitoring data among the bridges in the cluster is established; and finally, constructing a damage positioning factor and a threshold value by using the predicted strain value and the measured strain value of the strain measurement point group, and completing the damage positioning of the bridge structure in the cluster. As shown in fig. 1, the specific implementation steps are as follows:

step one, strain sensors are distributed on key sections of bridges with similar structural forms in a cluster, a bridge cluster structure strain measurement point group monitoring system is built, and real-time collection of bridge cluster structure strain measurement point group monitoring data is achieved.

Step two, preprocessing long-term strain measurement point group monitoring data acquired by the bridge cluster structure, and mainly comprising the following steps of: denoising, eliminating random vehicle-mounted influence processing and strain data standardization processing. The method comprises the following specific steps:

and step two, performing denoising and random vehicle-mounted influence elimination processing on the strain monitoring data of all bridges in the cluster, wherein the formula (1) is as follows:

wherein n represents the total number of sampling points; k represents any sampling point of the strain of the bridge structure, and k belongs to (1,2, …, n); i represents any measuring point of the bridge structure strain, i belongs to (1,2, …, m), and m is the number of the bridge structure strain sensors;strain data values of the kth sampling point of any strain measuring point i;and expressing the strain data value of the kth sampling point of any strain measuring point i after the noise is removed and the random vehicle-mounted influence is eliminated.

Step two, standardizing the strain monitoring data after the noise removal and random vehicle-mounted influence elimination processing to improve the precision of the space-time correlation model, wherein the standardization processing is shown as a formula (2):

in the formula (I), the compound is shown in the specification,representing the strain value of the kth sampling point of the strain measuring point i after the standardization treatment; e (x)ⁱ) Representing the mean value of the strain measuring point i monitoring data set; d (x)ⁱ) The variance of the monitored dataset of strain points i is represented.

Step two and step three, forming a strain monitoring matrix by using all the strain monitoring data after standardized processing of each bridge in the clusterAs shown in formula (3):

in the formula (I), the compound is shown in the specification,representing a vector formed by all the measuring point strain data of the single bridge at any kth sampling point after the normalization treatment,

and step three, selecting one bridge in the cluster as a reference bridge, and establishing a space-time correlation model of the reference bridge strain measuring point group monitoring data and other bridge strain measuring point group monitoring data in the cluster by using the strain long-term monitoring data in the structural health state. The method comprises the following specific steps:

establishing a space-time correlation model training network combining a deep convolutional neural network and a long-time and short-time memory network, wherein the structure of the space-time correlation model training network comprises the following steps: the system comprises a data input layer, a time sequence data folding layer, a convolution layer, a maximum pooling layer, a batch normalization layer, an activation function layer, a time sequence data expansion layer, a data flattening layer, a long and short time memory network layer I, a long and short time memory network layer II, a full connection layer and a data output regression layer. And sequentially connecting the structural layers in the network, and connecting the time sequence data folding layer with the time sequence data unfolding layer. The spatio-temporal correlation model training network structure is shown in FIG. 2.

Step three, taking any bridge A in the cluster as a reference bridge, and monitoring the strain of the bridge A in a healthy state by using a strain monitoring matrixAs network input, simultaneously monitoring the strain of any other bridge B in the cluster and the bridge A at the same sampling timeAs a network output.

Step three, training the time-space correlation model training network until the network loss function is converged, as shown in formula (4):

wherein Γ (Θ) is a loss function; theta is a hidden layer parameter set of the training network; kappa (-) is a strain monitoring data prediction function based on a space-time correlation model between bridges.

And step four, predicting the strain response of the bridge in the cluster by using the space-time correlation model constructed in the step three, and constructing a damage positioning factor and a threshold value by using the predicted strain value and the measured strain value of the strain measurement point group to complete the damage positioning of the bridge structure in the cluster. The method comprises the following specific steps:

step four, based on a verification data set under the bridge health state, utilizing a training network of a space-time correlation model which is completely trained in the step three to construct a damage positioning factor under the health state, as shown in a formula (5):

in the formula (I), the compound is shown in the specification,representing a residual vector of a predicted value and an actual measured value of any ith strain measuring point in the bridge health state;represents the k th strain measuring point of any ith strain measuring point under the bridge health state₁Residual values of the predicted values and the measured values of the sampling points;represents the k th strain measuring point of any ith strain measuring point under the bridge health state₁A damage localization factor of each sampling point; k is a radical of_hRepresenting the total number of verification data set samples in the bridge health state; e (-) represents the mean of the calculated vectors; d (-) denotes the variance of the calculated vector.

On the basis, an arbitrary ith strain measuring point damage positioning factor set under the bridge health state is constructedAs shown in formula (6):

step four, establishing a damage positioning threshold gamma of any ith strain measurement point of the bridgeⁱAs shown in formula (7):

in the formula, gamma_0.95(. cndot.) represents a 95% confidence median function of the extraction set;representing structural damage of a bridgeThe value range of the guarantee coefficient of the injury location is 1.1-1.2.

Step four, constructing a damage positioning factor of the bridge in the state to be diagnosed, as shown in formula (8):

in the formula, d represents the state to be diagnosed of the bridge structure; the other parameters have the same meanings as the fourth step and the first step.

Fourthly, judging the damage position of the bridge structure, and if the damage position is the median of the damage positioning factor set in the state to be diagnosedIndicating that the structure at the measuring point is damaged; otherwise, the bridge structure is in a safe state.

According to the method, aiming at the problem of how to establish a space-time correlation model between bridges in a cluster and realize the damage positioning of the bridges with similar structural forms in the cluster under the coupling action of a complex environment, the real-time acquisition of the monitoring data of the strain measuring point group of the bridge cluster structure is realized by establishing a strain measuring point group monitoring system of the bridge cluster structure; denoising and eliminating random vehicle-mounted influence processing are carried out on long-term strain measuring point group monitoring data collected by a bridge cluster structure, and strain data are processed in a standardized manner; establishing a time-space correlation model of the reference bridge strain measuring point group monitoring data and other bridge strain measuring point group monitoring data in the cluster; and finally, constructing a damage positioning factor and a threshold value by the predicted strain value and the actually measured strain value of the variable measuring point group, and completing the damage positioning of the bridge structure in the group. According to the method, the influence of a complex operation environment on the monitoring data is effectively considered, the space-time correlation model between the bridges in the cluster is accurately established, and the damage positioning precision of the bridge cluster structure is greatly improved.

Example (b):

in the embodiment, 2 single-box multi-chamber bridge clusters in a large overpass in a certain city in China are selected as an example, and the effectiveness of the bridge cluster structure damage positioning method based on the space-time correlation model provided by the invention is verified. The photo of the overpass cluster structure is shown in fig. 3, strain sensors are arranged on 2 bridge cluster structures of the south and north main lines, and a bridge cluster structure strain measurement point group monitoring system is established. Sensor locations are shown in fig. 4 and 5, and in-situ sensor deployment and acquisition are shown in fig. 6 and 7. The strain sensors arranged in the bridge span are positioned on a bridge bottom plate, and the strain sensors arranged near the bridge support are positioned on a bridge top plate. The SP4 to SP7 serve as a bridge a in the present embodiment, and the SP7 to SP10 serve as a bridge B in the present embodiment.

Strain monitoring data are collected under the bridge health state, and all strain data are subjected to noise removal, random vehicle-mounted influence elimination and data standardization processing. On the basis, the strain monitoring data after the strain measuring point group of the bridge B is subjected to standardization processing is used as network input, the strain monitoring data after the strain measuring point group of the bridge A is subjected to standardization processing is used as network output, and the space-time correlation model network of the bridge A and the bridge B is trained until convergence.

And performing model precision verification by using strain monitoring data in a state to be diagnosed, wherein the strain data prediction results of the measuring point #4 of the bottom plate and the measuring point #5 of the top plate are shown in fig. 8 and 9. In order to embody the advantages of the present invention, compared with the traditional support vector machine learning algorithm, the prediction results based on the strain data of the #4 measuring point of the bottom plate and the #5 measuring point of the top plate of the support vector machine are shown in fig. 10 and fig. 11. According to the prediction result, the method provided by the invention has more accurate prediction result on the strain monitoring data.

And applying 10 mu epsilon of simulated damage to the strain monitoring data to be diagnosed at the measuring point #1 and the measuring point #15 of the bridge A, and positioning the damage of the bridge A based on the established space-time correlation model, wherein the positioning result is shown in figure 12, and the damage of the measuring point #1 and the measuring point #15 can be effectively positioned due to the accuracy of the space-time correlation model. Similarly, as shown in fig. 13, the damage positioning result based on the support vector machine is that the accuracy of the strain prediction result based on the support vector machine is not sufficient, so that the simulated damage cannot be effectively positioned finally, and other measuring points which are not damaged are prone to misjudgment.