CN117216844A

CN117216844A - Bridge structure damage detection method, system and storage medium

Info

Publication number: CN117216844A
Application number: CN202311169949.XA
Authority: CN
Inventors: 黄杰忠; 元思杰; 李东升
Original assignee: Shantou University
Current assignee: Shantou University
Priority date: 2023-09-12
Filing date: 2023-09-12
Publication date: 2023-12-12
Anticipated expiration: 2043-09-12
Also published as: CN117216844B

Abstract

The invention discloses a bridge structure damage detection method, a bridge structure damage detection system and a bridge structure damage detection storage medium, wherein the bridge structure damage detection method comprises the steps of collecting historical frequency signals of a plurality of sample bridges, preprocessing the historical frequency signals to obtain a plurality of first eigenfrequency characteristics and corresponding principal component damage coefficients of the first eigenfrequency characteristics, and further generating a model training set and a non-damage verification set; training an SVR model by using a model training set, verifying the trained model by using a non-damage verification set, and constructing a control chart according to a verification result; and collecting a current frequency signal of the bridge to be detected, and detecting structural damage to the current frequency signal by using a control chart and a model to obtain a structural detection result of the bridge to be detected. The method can effectively denoise and remove seasonal modes in the frequency time sequence, eliminates negative environmental impact factors from frequency data, improves the accuracy of structural damage detection and the detection efficiency of the structural damage, and has a certain engineering application prospect. The invention is applied to the technical field of structural damage detection.

Description

Bridge structure damage detection method, system and storage medium

Technical Field

The invention relates to the technical field of structural damage detection, in particular to a bridge structural damage detection method, a bridge structural damage detection system and a storage medium.

Background

Civil infrastructure systems may be subject to varying degrees of damage from factors such as earthquakes and excessive traffic loads. In order to find structural damage problems of a facility at as early a stage as possible, it is extremely important to monitor the structure of the facility in real time or periodically using structural health monitoring techniques. Since structural damage typically causes a change in structural stiffness or mass, the dynamic characteristics of the structure depend on the physical properties of the structure, such as mass and stiffness, and thus structural health monitoring techniques based on vibrational response can detect damage to the structure by identifying changes in the dynamic characteristics.

In the vibration-based structural health monitoring technology, the frequency is the most common and very effective structural damage characteristic, and the natural frequency of the structure can be automatically estimated from the response data by using a small number of sensors without any manual intervention, so that the structural health monitoring is realized by using the natural frequency. However, frequency is a structural damage characteristic, which is extremely susceptible to changes in environmental conditions, such as temperature, traffic load, humidity, wind speed, and the like. It is not uncommon for the frequency to vary by 10% relative due to environmental factors, such that the adverse environmental impact on frequency will completely mask the frequency variation due to structural damage, thereby rendering the actual structural damage effectively unrecognizable. Therefore, in order to accurately and reliably identify lesions, these negative environmental effects must be separated from the frequency data prior to structural lesion detection. In this regard, the relevant scholars propose damage recognition methods for separating the influence of environmental factors, which can be generally classified into an explicit method and an implicit method, wherein the explicit method needs to measure environmental variable parameters, and by establishing an explicit relation model between the environmental factors and frequencies, the influence of the environmental factors on the frequencies can be quantified, such as support vector-based regression, neural networks and the like; while the implicit method has the advantage that it does not require measuring environmental variables, which are regarded as embedded variables, it assumes that the frequency changes caused by environmental changes are different from those caused by structural damage, and can distinguish between two sources of change, such as principal component analysis, factor analysis, etc., by mapping the raw data to another subspace.

Although the related art has been widely used to separate the influence of environmental factors on damage recognition, it has the following drawbacks:

firstly, in the related art, the explicit technology requires that multiple types of sensors are deployed at different positions of a structure to measure environmental impact factors, so as to establish an association relationship between the environmental impact factors and damage features. However, this approach undoubtedly results in an increase in the cost of detecting structural damage, as well as an increase in the effort required to be put into the data processing stage. In addition, as not all positions are suitable for sensor deployment, once the data of the key positions are not available, the damage identification difficulty of the explicit technology is further improved.

Second, most implicit methods in the related art can only potentially reveal linear correlations between two sets of variables. However, when the monitored data exhibits a nonlinear relationship, implicit methods often become difficult to establish a nonlinear relationship. In practical applications, a nonlinear relation exists between the monitored variables, but related technologies cannot accurately monitor the nonlinear relation, which causes the problem that the linear implicit method is difficult to accurately identify structural damage.

Disclosure of Invention

The present invention aims to solve at least one of the technical problems existing in the prior art to a certain extent.

Therefore, the invention aims to provide a bridge structure damage detection method, a bridge structure damage detection system and a storage medium.

In order to achieve the technical purpose, the technical scheme adopted by the embodiment of the invention comprises the following steps:

in one aspect, the embodiment of the invention provides a bridge structure damage detection method, which comprises the following steps:

acquiring a plurality of historical frequency signals of a sample bridge, and preprocessing the historical frequency signals to generate a model training set and a nondestructive testing set;

training an SVR model by using the model training set to generate a structural damage identification model, verifying the structural damage identification model by using the nondestructive testing set, and constructing a control chart according to the verification result of the structural damage identification model;

and acquiring a current frequency signal of the bridge to be detected, preprocessing the current frequency signal, and detecting structural damage to the preprocessed current frequency signal by using the control diagram and the structural damage identification model to obtain a structural detection result of the bridge to be detected.

In another aspect, an embodiment of the present invention provides a bridge structure damage detection system, including:

The data acquisition module is used for acquiring a plurality of historical frequency signals of the sample bridge and a current frequency signal of the bridge to be detected;

the data preprocessing module is used for preprocessing a plurality of historical frequency signals, generating a model training set and a nondestructive testing set, and preprocessing the current frequency signals;

the model processing module is used for training the SVR model by utilizing the model training set and generating a structural damage recognition model;

the control diagram processing module is used for verifying the structural damage recognition model by utilizing the nondestructive testing set, and constructing a control diagram according to the verification result of the structural damage recognition model;

and the detection module is used for detecting the structural damage of the preprocessed current frequency signal by utilizing the control diagram and the structural damage identification model to obtain a structural detection result of the bridge to be detected.

In yet another aspect, an embodiment of the present invention provides a storage medium in which a program executable by a processor is stored, where the program executable by the processor is configured to implement the above-mentioned bridge structure damage detection method when executed by the processor.

The beneficial effects of the invention are as follows: the method comprises the steps of taking a frequency signal of a structure as a main detection object, decomposing an identified frequency time sequence of the structure into two oscillation modes through a variation mode decomposition algorithm, utilizing principal component analysis and frequency characteristic combination to remove environmental influence factors, generating a Euclidean distance of principal component matrix residual errors, constructing a mapping relation between the frequency signal of the structure and the principal component damage coefficients through an SVR model according to principal component damage coefficients corresponding to the frequency characteristic, constructing a control chart for real-time detection according to a trained SVR model and a verification set, preprocessing and predicting current frequency data through the SVR model in actual detection, judging whether a predicted result accords with the condition of structural damage through the control chart, and further obtaining a final structural damage detection result. The method can effectively denoise and remove seasonal modes in the frequency time sequence, well eliminates negative environmental impact factors from frequency data, overcomes the negative impact of the environmental impact factors on the frequency data, improves the characterization of the frequency data on the structural damage condition, improves the precision of structural damage detection and the efficiency of structural damage detection, reduces the detection cost of the structural damage, the data operation amount and the processing load of real-time detection, and has a certain engineering application prospect.

Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the application. The objectives and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Drawings

FIG. 1 is a flow chart of a method for detecting damage to a bridge structure provided by the application;

FIG. 2 is a flow chart of a variation modal decomposition provided by the present application;

FIG. 3 is a flow chart of principal component analysis provided by the present application;

FIG. 4 is a flow chart of the generation control chart provided by the present application;

FIG. 5 is a schematic diagram of a method for detecting damage to a bridge structure according to the present application;

FIG. 6A is a schematic diagram showing the evolution of air temperature of a three-span prestressed highway bridge over time for a certain period of time according to the present application;

FIG. 6B is a schematic diagram showing the evolution of four natural frequencies of a three-span prestressed highway bridge over time in a certain period of time;

FIG. 7 is an exploded view of the natural frequency and its center frequency of the three-span prestressed highway bridge according to the present application;

FIG. 8A is a plot of the correlation between the original frequency signals provided by the present application;

FIG. 8B is a graph showing the correlation between VMD processed frequency signatures provided in the present application;

FIG. 9A is a graph of the correlation between Euclidean distances of sample points and residuals of training and validation sets provided by the present application;

FIG. 9B is a graph of the Euclidean distance between sample points and residuals of a training set, a validation set, and a test set provided by the present application;

FIG. 10A is a graph of results on a training set provided by the present application;

FIG. 10B is a graph of results on a training set and a validation set provided by the present application;

fig. 10C is a graph of results on a training set, a validation set, and a test set provided by the present application.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

The application will be further described with reference to the drawings and specific examples. The described embodiments should not be taken as limitations of the present application, and all other embodiments that would be obvious to one of ordinary skill in the art without making any inventive effort are intended to be within the scope of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the application only and is not intended to be limiting of the application.

Civil infrastructure systems may be subject to varying degrees of damage from factors such as earthquakes and excessive traffic loads. In order to find structural damage problems of a facility at as early a stage as possible, it is extremely important to monitor the structure of the facility in real time or periodically using structural health monitoring (Structural Health Monitoring, SHM) techniques. Since structural damage typically causes a change in structural stiffness or mass, the dynamic characteristics of the structure depend on the physical properties of the structure, such as mass and stiffness, and thus structural health monitoring techniques based on vibrational response can detect damage to the structure by identifying changes in the dynamic characteristics.

In the vibration-based structural health monitoring technology, the frequency is the most common and very effective structural damage characteristic, and the natural frequency of the structure can be automatically estimated from the response data by using a small number of sensors without any manual intervention, so that the structural health monitoring is realized by using the natural frequency. However, frequency is a structural damage characteristic, which is extremely susceptible to changes in environmental conditions, such as temperature, traffic load, humidity, wind speed, and the like. It is not uncommon for the frequency to vary by 10% relative due to environmental factors, such that the adverse environmental impact on frequency will completely mask the frequency variation due to structural damage, thereby rendering the actual structural damage effectively unrecognizable. Therefore, in order to accurately and reliably identify lesions, these negative environmental effects must be separated from the frequency data prior to structural lesion detection.

In this regard, relevant scholars propose methods for identifying lesions that separate the influence of environmental factors, which methods can be generally classified into explicit methods and implicit methods. The explicit method needs to measure environmental variable parameters, and can quantify the influence of environmental factors on frequency by establishing an explicit relation model between the environmental factors and the frequency. Common methods are: linear regression, polynomial regression, autoregressive models with inputs, bilinear regression, support vector-based regression, neural networks, and the like. The explicit method takes the environment variable as the input variable of the model, the physical meaning of the model is more definite, and the built model is easier to explain. While the implicit method has the advantage that it does not require measuring the environment variable, which is regarded as an embedded variable, it can distinguish between two sources of variation by mapping the original data to another subspace, assuming that the frequency variation caused by the environment variation is different from the variation caused by the structural damage. Common methods are: principal component analysis, factor analysis, coordination, nuclear typical correlation analysis, local principal component, nuclear principal component, gaussian mixture model and the like.

Although the related art has been widely used to separate the influence of environmental factors on damage recognition, it has the following drawbacks: in the related art, the explicit technology requires deployment of multiple types of sensors at different positions of a structure to measure environmental impact factors, so as to establish an association relationship between the environmental impact factors and damage characteristics. However, this approach undoubtedly results in an increase in the cost of detecting structural damage, as well as an increase in the effort required to be put into the data processing stage. In addition, as not all positions are suitable for sensor deployment, once the data of the key positions are not available, the damage identification difficulty of the explicit technology is further improved. Second, most implicit methods in the related art can only potentially reveal linear correlations between two sets of variables. However, when the monitored data exhibits a nonlinear relationship, implicit methods often become difficult to establish a nonlinear relationship. In practical applications, a nonlinear relation exists between the monitored variables, but related technologies cannot accurately monitor the nonlinear relation, which causes the problem that the linear implicit method is difficult to accurately identify structural damage.

Aiming at the defects and problems of the related art, the embodiment of the invention provides a bridge structure damage detection method, a bridge structure damage detection system and a storage medium, which are realized based on variation modal decomposition (Variational Mode Decomposition, VMD), principal component analysis (Principal Components Analysis, PCA) and support vector machine regression (Support Vector Regression, SVR), and the performance of the bridge structure damage detection method, the system and the storage medium in the aspects of actual environment influence separation and damage identification is verified by utilizing vibration data of three-span prestressed highway bridge continuous structure health monitoring.

The method for detecting damage to the bridge structure provided by the embodiment of the invention is described in detail below with reference to the accompanying drawings.

The method in the embodiment of the invention can be applied to the terminal, the server, software running in the terminal or the server and the like. The terminal may be, but is not limited to, a tablet computer, a notebook computer, a desktop computer, etc. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligent platforms.

Referring to fig. 1, the method provided by the embodiment of the invention mainly includes the following steps:

s101, acquiring a plurality of historical frequency signals of a sample bridge, and preprocessing the historical frequency signals to generate a model training set and a nondestructive testing set.

Further, the step of obtaining the frequency signal includes: and collecting vibration acceleration signals of the bridge structure, and carrying out modal parameter identification on the acceleration signals to obtain frequency signals of the bridge.

The method comprises the steps of preprocessing historical frequency signals based on a variation modal decomposition algorithm and principal component analysis, and further obtaining a model training set and a non-destructive testing set, wherein the model training set is mainly used for training an SVR model, and the non-destructive testing set is mainly used for generating a control chart for structural damage detection.

S102, training the SVR model by using the model training set to generate a structural damage recognition model.

And S103, verifying the structural damage identification model by using the non-damage verification set, and constructing a control chart according to the verification result of the structural damage identification model.

S104, obtaining and preprocessing the current frequency signal of the bridge to be detected, and detecting structural damage to the preprocessed current frequency signal by using the control diagram and the structural damage identification model to obtain a structural detection result of the bridge to be detected.

Optionally, the structure detection result of the bridge to be detected includes any one of a result that the bridge to be detected has structural damage or a result that the bridge to be detected does not have structural damage.

The implementation process of preprocessing the frequency signal in the method according to the embodiment of the present invention will be further described below with reference to the accompanying drawings.

The embodiment of the invention realizes the preprocessing of a plurality of historical frequency signals based on a variation modal decomposition algorithm and principal component analysis, and the process of preprocessing the historical frequency signals mainly comprises the following steps: firstly, denoising the historical frequency signal by using a variation modal decomposition method, removing features belonging to short-term seasonal modes, and generating a first eigenfrequency feature. Wherein the features belonging to short-term seasonal patterns are defined as IMFs ₂ The signal and the first eigenfrequency characteristic is defined as IMF ₁ A signal. Then, for IMF ₁ The signal is subjected to principal component analysis to obtain IMF ₁ The principal component damage coefficient corresponding to the signal.

1. IMF is obtained by using a variational modal decomposition method ₁ The signal is realized as follows:

environmental impact factors can be categorized into short-term seasonal modes and long-term non-seasonal modes, with damage detection algorithms often affected by variance changes in heteroscedastic data caused by the former. The variational modal decomposition method is a self-adaptive, completely non-recursive modal variational and signal processing method for decomposing complex nonlinear and non-stationary signals into a series of local vibration modes. In order to eliminate the heteroscedastic effect of short-term seasonal patterns, embodiments of the present invention utilize advanced signal decomposition techniques of variational modal decomposition to decompose frequency signature data into signatures that belong to long-term non-seasonal patterns, i.e., IMFs ₁ Signals and features belonging to short-term seasonal modes, i.e. IMF ₂ The signal is removed from the frequency signal to obtain the characteristic of long-term non-seasonal pattern, namely the first eigenvalueFrequency characteristics.

Referring to fig. 2, the implementation steps of the variant modal decomposition are as follows:

the first step: defining the historical frequency signal to be composed of a plurality of modal components is shown as follows:

，

in the method, in the process of the invention,is the original signal, which is the signal to be decomposed as a function of time t, K is the number of modes to be decomposed, +.>Is a modal function, i.e. a modal component.

Then, the historical frequency signal is decomposed into a plurality of modal componentsThe modal components are expressed as follows:

，

in the method, in the process of the invention,representing the amplitude variation of each modal component over time as an amplitude function; />As a function of phase, i.e. the phase of the mode over time.

And a second step of: each modal component has a center frequencyIs provided, while the sum of the estimated bandwidths of the modal components is minimal. The embodiment of the invention constructs the constrained variation problem according to a plurality of modal components, namely the constrained variation problem, and defines constraint conditions as the sum of all modal components and the original constraint conditions The starting signals are equal, and the corresponding expression of the constraint variation problem is as follows:

，

wherein k is the number of modal components to be decomposed, which is a positive integer;for +.>Partial derivative of>And->Corresponds to the k-th modal component and center frequency after decomposition, respectively,>is a dirac function, j is an imaginary unit,/, and>is a convolution operator.

And a third step of: aiming at the constraint variation problem, a quadratic punishment term and a Lagrange multiplier are introduced, and the constraint variation problem is converted into an unconstrained variation problem based on the constraint variation problem, and the unconstrained variation problem is an extended Lagrange equation, which is shown as follows:

in the method, in the process of the invention,and->For Lagrangian multiplier and penalty factor, < ->Representation->Andan inner product of the two.

Fourth step: aiming at the solution of the extended Lagrangian equation, the embodiment of the invention adopts a multiplication operator alternating direction method to solve the variation problem, and the minimum problem of the solution, namely the saddle point of the extended Lagrangian expression, is solved through alternating updating calculation, so that the first eigenfrequency characteristic after the characteristic belonging to the short-term seasonal mode is separated is obtained.

Further, the above-described solution to the variational problem applies to a relatively large number of mathematical tools, including quadratic penalty terms, lagrangian, and augmented lagrangian functions, among others. In this regard, embodiments of the present invention require specifying some parameters prior to solving:

1) k: the original signal is decomposed into the number of IMFs. The embodiment of the invention enablesTo remove environmental impact factors in the frequency time series.

2）: and a second punishment term. In the invention, the value of the embodiment of the three-span prestressed highway bridge is 10, and the value of the other embodiments is 100, so that the interference of Gaussian noise is reduced. It should be noted that larger values allow less noise to enter the decomposition process.

3）: denoising factor to be validated. The embodiment of the invention is->. When the denoising factor is any non-zero value, a forced perfect reconstruction is required, i.e., no denoising occurs. Thus, the ginsengThe value of the number becomes insignificant.

4）: tolerance parameters for control algorithm convergence. The embodiment of the invention is->Select smaller +.>The time required for convergence of the variant mode decomposition algorithm is longer at this value.

5) init: center frequency of modal component. The embodiment of the invention enablesTo initialize the center frequency of the modal component. In other embodiments of the present invention, init may be set to 0, 1 or 2 to achieve the effect of zero initialization, uniform initialization or random initialization, respectively.

6) DC: a modal component is determined that maintains the center frequency at zero. The embodiment of the invention makes dc=1 so that the center frequency of the first modal component remains zero, which is done in order to preserve the non-stationary behaviour of the first modal component. In other embodiments of the invention, dc=0 may be made, which means that the center frequency of the first modal component may be non-zero.

2. The principal component damage coefficient corresponding to the first eigenvalue is obtained by principal component analysis, and the implementation process is as follows:

principal component analysis is a method in multivariate statistics, theoretically a coordinate-based double transformation. In the present invention, the original data is first projected into the vector space generated by the principal components, then only a portion of the principal components are retained, and then remapped back to the original space. The principal components are statistical independent variables, and the contribution rates of different principal components to the variance of the original data are different. The retained principal components represent environmental factors (such as temperature, temperature gradient, humidity, wind, etc.) that contribute significantly to the data change, which in the embodiments of the present invention are also referred to as environmental impact factors, and subtracting the principal components represents separating the environmental impact factors from the data.

Referring to fig. 3, the principal component analysis is implemented as follows:

the first step: and constructing a damage characteristic matrix in the original space by utilizing the data of the characteristics belonging to the short-term seasonal modes, namely the first eigenfrequency characteristics.

In this step, the damage characteristic matrix is composed of IMF (IMF) which is a characteristic of long-term non-seasonal pattern ₁ A matrix of signals. For the damage characteristic matrix in the original space, defining the damage characteristic matrix asN represents the number of samples and N is the dimension of the impairment feature matrix.

And a second step of: singular value decomposition is carried out on a covariance matrix of the damage characteristic matrix to generate a first matrix, and the damage characteristic matrix is projected into a principal component space through the first matrix to generate a second matrix located in the principal component space.

In this step, if natural frequency data is used as the lesion feature, the basic idea of principal component analysis is to matrix the lesion feature with original dimension nAnd mapping the environment factor characteristic space to the environment factor characteristic space with the lower dimension m, wherein the environment factor characteristic space is a vector space or a principal component space generated by principal components, and m is the number of environment influence factors of the first eigenfrequency characteristic.

More specifically, in the second step, the mapping manner adopted in the embodiment of the present invention is:

firstly, singular value decomposition is carried out on a covariance matrix of a damage characteristic matrix to obtain a principal component matrix, wherein the principal component matrix is shown in the following formula:

，/>，/>，

in the method, in the process of the invention,is a damage characteristic matrix; />Is a principal component matrix, and its column vector is the principal component; />Is a unit matrix; />Is a diagonal matrix >Included are singular values of the covariance matrix, the diagonal terms of which are the singular values of the covariance matrix, representing the contribution of each principal component to the variance; />Is a diagonal matrix->Left upper submatrix of>Is a diagonal matrix->Is a right lower sub-matrix of (c).

Then, a first matrix is constructed using part of the principal component elements of the principal component matrix. Specifically, by retaining only the principal component matrixIs the first m columns of the matrix of principal components +.>To construct a first matrix +.>。

Finally, the first matrix is utilized to project the damage characteristic matrix into the principal component space, and a second matrix is generated, wherein the second matrix is located in the principal component space and is shown in the following formula:，

in the method, in the process of the invention,called the score matrix, and also the secondary matrix. />Is a first matrix, also called loading matrix, comprising a principal component matrix +.>The first m principal component elements of (a) principal component matrix +.>The first m principal constituent elements of (1) correspond to the principal constituent matrix +.>Is the first m columns of (c). m is the dimension of the second matrix and is expressed as the number of environmental impact factors that have a major impact on the natural frequency.

And a third step of: and performing inverse transformation on the second matrix to obtain a third matrix in the original space.

In this step, the original damage characteristic matrix Projected into the vector space generated by the principal component, and transformed by inversion +.>Estimated data in the original space will be obtained, i.e. the third matrix +.>The following formula is shown: />。

The principal component included in the third matrix is an environmental impact factor.

Fourth step: and generating a principal component damage coefficient corresponding to the first eigen frequency characteristic according to the third matrix and the damage characteristic matrix.

In this step, due to the third matrixThe first r principal component elements are retained, which are similar to the environmental impact factors, contributing mainly to the variance of the data, thus according to the third matrix ∈ ->And the original injury feature matrix->The damage matrix after the environmental impact factors are separated can be obtained. Then, based on the damage matrix and Euclidean norms, corresponding principal component damage coefficients can be generated.

More specifically, in the fourth step, first, the original lesion characterization matrix is utilizedSubtracting the third matrix->The residual error obtained can be used as a damage matrix after the environmental impact factors are separated, and the damage matrix is shown as the following formula: />Wherein->The residual matrix which is the main component is the damage matrix. Then, use residual matrix- >To define the principal component damage coefficient: />So that the first eigenfrequency features have corresponding principal component damage coefficients.

The implementation process of constructing a model training set and a nondestructive verification set, performing model training by using the model training set, and constructing a control chart by using the verification set in the method provided by the embodiment of the invention will be further described below.

1. The implementation process of constructing the model training set and the nondestructive verification set is as follows:

the model training set of the embodiment of the invention is a data set for training the SVR model, the nondestructive testing set is a data set for testing the SVR model, and the construction process of the model training set and the testing set is as follows:

and taking the first eigenvalues and the corresponding principal component damage coefficients as an initial data set, extracting one part of data from the initial data set as a model training set, and the other part of data as a non-damage verification set.

It should be noted that the model training set and the non-destructive testing set each include a plurality of first eigenfrequency features and their corresponding principal component damage coefficients, where the first eigenfrequency features are IMFs ₁ The signal will be the input feature and the principal component damage coefficient will be the IMF ₁ The signal corresponds to the marking result, which is the output characteristic.

2. The implementation process for constructing the structural damage recognition model is as follows:

the embodiment of the invention trains the SVR model by using a model training set to obtain the structural damage identification model for constructing the control diagram. SVR is a machine learning method based on statistical theory, has good effect on solving nonlinear problems, has strong generalization capability and obvious advantages, and is widely applied to various fields in industry.

For a given training set sample as shownWherein, the method comprises the steps of, wherein,for the first eigenfrequency characteristic,/>Is the corresponding principal component damage coefficient. For the sampleA regression model is usually obtained such that +.>And->As close as possible, assuming tolerance +.>And->There is at most->Deviation of only->The loss is calculated. When->And if so, the prediction is considered to be accurate. The SVR problem can be translated into the following formula: />，

Wherein the weight vector w and the bias term b are regression parameters to be determined,the prediction output of the model; />The first eigenvector is the input eigenvector; />The corresponding target output is the damage coefficient of the main component. m is the number of samples of the model training set, and C is the regularization parameter. / >Is a insensitive loss function, as shown in the following equation:

，

wherein,is a threshold parameter.

In introducing relaxation variablesAnd->The SVR problem described above can then be converted into the following formula:

constrained by: />。

Similarly, the lagrangian multiplier is introduced, defining the lagrangian function as follows:

wherein,and->Is Lagrangian multiplier, +.>And->Are respectively about->And->Lagrangian multiplier of (c). Let L pair w, b, < >>And->The partial derivative of (a) is 0, and the weight vector w and the bias term b can be determined.

According to the duality of the Lagrangian equation, the duality problem of the SVR problem is a maximum and minimum problem, and after the dual problem is converted into the duality problem, a final SVR solution form can be obtained through the duality of the Lagrangian equation, and the final SVR solution form is shown in the following formula:

，/>is the predicted output of the model, and b is the bias term, intercept in linear regression.

More specifically, the steps for training the SVR model according to the embodiment of the invention are as follows:

characterised by a first eigenfrequency, i.e. IMF ₁ The signal is used as a characteristic, the main component damage coefficient is used as a target, and the model training set is used for carrying out model training on the SVR model, so that the SVR model learns the first eigenfrequency characteristic, namely IMF ₁ And taking the current trained SVR model as a structural damage identification model and outputting the structural damage identification model according to the association relation between the signals and the main component damage coefficients.

3. The implementation process for constructing the control chart is as follows:

the error signal of the model of the embodiment of the invention can be regarded as a control chart based on the difference between the real principal component damage coefficient and the predicted principal component damage coefficient. Referring to fig. 4, the step of constructing a control map may be divided into two steps:

the first step: first, each IMF in the set of non-destructive authentications ₁ The signals are input into a structural damage recognition model, and each IMF is obtained through the output of the structural damage recognition model ₁ And predicting the damage coefficient of the principal component corresponding to the signal. Then, each IMF is collected using non-invasive verification ₁ The marking result of the signal, namely the real principal component damage coefficient, is combined with the predicted principal component damage coefficient to calculate and obtain a predicted error signal of the model, and the predicted error signal of the model is used as a verification result of the model.

More specifically, in the first step, the first eigenfrequency feature, i.e., IMF, in the set is verified by the non-destructive verification ₁ And carrying out prediction verification on the model by the signal to obtain a predicted principal component damage coefficient predicted by the model. Because the nondestructive testing set also comprises a label result corresponding to each first eigenfrequency characteristic, namely a real principal component damage coefficient, the prediction error signal of the model can be calculated based on the obtained prediction principal component damage coefficient and the real principal component damage coefficient to serve as a testing result.

And a second step of: and calculating the average value and standard deviation of a plurality of prediction error signals of the model, determining the central line and the upper limit value of the control chart according to the average value and the standard deviation of the damage coefficients of the plurality of prediction principal components, and further constructing the control chart.

More specifically, in the second step, the upper limit value of the control map is determined based on the average value and the standard deviation of the plurality of prediction error signals obtained in the first step, as shown in the following formula:

，

where UCL corresponds to the upper control limit of the control map.And->Is the mean and standard deviation of the freshness index NI in the formula under the structural health state. />Typically take 3, corresponding to 99.7% confidence.

The embodiment of the invention sets the upper limit of the control chart, when the data is abnormal, the data exceeds the upper limit of the control chart, and in this case, an alarm signal is generated to indicate that the current structure has damage.

The real-time detection process of structural damage in the method according to the embodiment of the present invention will be further described below.

The first step: acquiring and preprocessing a current frequency signal of the bridge to be detected to obtain a current IMF ₁ The signal and its corresponding current real principal component damage coefficient.

In the step, the method for preprocessing real-time data is the same as the method for preprocessing training sets, and the method firstly preprocesses the bridge current frequency signal by using a variation mode decomposition method to generate current frequency characteristics, namely the current IMF ₁ A signal; thereafter, for the current IMF ₁ The signal is subjected to principal component analysis to obtain the current IMF ₁ The current principal component damage coefficient corresponding to the signal is the current real principal component damage coefficient.

And a second step of: current IMF ₁ Inputting the signals into a structural damage recognition model to obtain the current IMF ₁ The current predicted principal component injury coefficient corresponding to the signal.

In this step, during training, the structural damage recognition model learns the IMF without damage to the structure ₁ Correlation between signal and principal component damage coefficient. According to the relevance, the current IMF can be obtained through the structural damage identification model ₁ The current predicted principal component injury coefficient corresponding to the signal.

And a third step of: and calculating the difference between the current prediction principal component damage coefficient and the current real principal component damage coefficient to obtain a current prediction error signal.

Fourth step: and judging whether the current prediction error signal exceeds the upper limit of the control diagram according to the upper limit value and the central line of the control diagram, and obtaining the structure detection result of the bridge to be detected according to the judgment result.

The following describes the principle of a bridge structure damage detection method according to an embodiment of the present invention by way of an example, and verifies the performance thereof. Referring to fig. 5, the overall implementation of the embodiment of the present invention can be divided into: data preprocessing, construction of a training set and a verification set, model training and generation, control diagram generation and real-time detection; the performance of the method of the embodiment of the invention in the aspects of actual environmental impact separation and damage identification is verified by using vibration data of continuous monitoring of the three-span prestressed highway bridge.

The data set adopted for verifying the performance is data of a three-span prestressed highway bridge, and the data set comprises environmental data of temperature and humidity, bridge vibration data and real damage data, and also comprises four inherent frequencies of the bridge calculated from acceleration measurement by using a random subspace method. Referring to fig. 6A and 6B, fig. 6A shows the evolution of air temperature over time for a three-span prestressed highway bridge over a period of time, with temperature on the vertical axis and sample data points on the horizontal axis; FIG. 6B shows the evolution of four natural frequencies of a three-span prestressed highway bridge over time, with the vertical axis being frequency and the horizontal axis being sample data points, over a certain period of time; the broken line indicates the time of occurrence of the injury. It can be seen that a very cold temperature at around 2000 sample points results in a non-linear relationship between natural frequency and temperature. The frequency change is therefore significant due to environmental effects, with a much larger amplitude than the relative change in frequency during the injury phase. It is difficult to identify the structural damage moment only from the frequency variation, and the influence of the environment on the frequency data should be eliminated.

For data preprocessing, the traditional damage detection algorithm is influenced by variance change of heteroscedastic data, so that the damage detection process is more complex. Thus, the first step of the embodiment of the present invention is to denoise and remove seasonal features from the collected frequency data using a variational modal decomposition algorithm. After removing seasonal features and denoising, principal component analysis is used to remove environmental impact factors and construct the residual euclidean distance of the principal component matrix, i.e. the principal component damage coefficients.

More specifically, the present invention utilizes VMD to decompose natural frequency time series into two IMFs (IMFs ₁ And IMF (inertial measurement unit) ₂ ) I.e. modal component, removing IMF corresponding to environmental influence factors ₂ . The same parameters as in the previous embodiment are used in the VMD setting here, where the quadratic penalty termThe obtained decomposition result and its center frequency are shown in FIG. 7, FIG. 7 shows a decomposition diagram of the natural frequency and its center frequency of a three-span prestressed highway bridge, the vertical axis is frequency, the horizontal axis is sample data point, (a) is IMF ₁ (b) is IMF ₂ ，IMF ₁ Maintained at DC (zero center frequency), embodiments of the present invention suggest using only IMF ₁ As a frequency characteristic.

Referring to fig. 8A and 8B, fig. 8A shows a correlation distribution diagram between original frequency signals, it can be understood that each natural frequency correlation scatter diagram can observe nonlinear relationship between different frequencies, in particularBilinear relationship with one of the other three natural frequencies. Fig. 8B shows a correlation distribution diagram between frequency characteristics after VMD processing. By comparing the distribution shapes of the main diagonal graphs before and after the VMD is applied, the data discrete degree is reduced after the VMD is processed, and the VMD can effectively reduce the noise content of the co-distributed scatter graph of the off-diagonal items. In the embodiment of the invention, the main principle of using the VMD is to remove the environmental influence factors in the frequency signal and denoise the signal, which further contributes to the training process of the SVR model.

For constructing the training set and the verification set, the embodiment of the invention takes q% of the first eigenfrequency characteristic and the corresponding principal component damage coefficient thereof as a model training set, and takes 1-q% of the first eigenfrequency characteristic and the corresponding principal component damage coefficient thereof as a nondestructive verification set.

Alternatively, q% is 80%, i.e., 80% of the lossless data is selected as training samples and 20% is selected as validation samples. In addition, lossy data is selected as test data.

And training and generating a model, training the SVR model by using a training set, and learning a relation rule between the first eigenfrequency characteristic and the principal component damage coefficient when the structure is lossless so as to obtain a structure damage identification model. Referring to fig. 9A, which is a graph of correlations between sample points of a model training set and a validation set and euclidean distances of residuals, which are frequency features; referring to fig. 9B, which is a graph of correlations between the euclidean distances between the sample points of the model training set, the validation set, and the test set and the residuals, the residual euclidean distances are frequency features, and it is known through the validation of the model that when a damage occurs, the prediction error will deviate significantly from the upper control graph limit (UCL).

More specifically, referring to fig. 10A, a result diagram on a training set is shown, the abscissa is a sample point, the ordinate is the euclidean distance and the prediction error, respectively, 80% of training data (the first IMFs of the decomposed frequency signal) is used as training features of the SVR model, the frequency feature corresponding to each data point is used as a training target of the SVR model, and the SVR model is trained through the training features and the training targets. The embodiment of the invention provides the following optional SVR model hyper-parameters: the kernel function adopts a radial basis kernel function; SVR loss function The method comprises the steps of carrying out a first treatment on the surface of the Radial basis function parameter +.>The method comprises the steps of carrying out a first treatment on the surface of the Loss function in SVR->Values of (2)。

For the generation of the control chart, the error signal of the obtained model can be regarded as the difference between the real principal component damage coefficient in the verification set and the predicted principal component damage coefficient outputted through the model prediction, and referring to fig. 10B, the result chart on the training set and the verification set is shown, the abscissa is the sample point, the ordinate is the euclidean distance and the prediction error respectively, and the mean value of the error signal in the verification set is taken as the upper limit of the control chart, so as to construct the control chart.

For real-time detection, referring again to fig. 9B, the current frequency signal is collected and preprocessed into the current frequency characteristic and the corresponding current principal component damage coefficient, the current prediction principal component damage coefficient is predicted and output according to the current frequency characteristic by using the trained SVR model, the current prediction error signal of the model is obtained according to the difference between the current prediction principal component damage coefficient and the current principal component damage coefficient, the result diagram on the training set, the test set and the verification set is shown with reference to fig. 10C, the abscissa is the sample point, the ordinate is the euclidean distance and the prediction error respectively, and the structural damage detection result is obtained by using the correlation between the current prediction error signal and the control diagram.

The above examples demonstrate the feasibility, effectiveness and high availability of the proposed method in using noisy data for structural state monitoring.

In summary, the embodiment of the invention provides the following technical effects:

firstly, the embodiment of the invention takes the frequency signal of the structure as a main detection object, can realize rapid and convenient structural damage detection without collecting a plurality of environmental data and constructing the relation between the environmental data and damage characteristics, reduces the detection cost of structural damage, and simultaneously can reduce the data operand and processing load of real-time detection of the structural damage in practical application because the data source only has one data of the frequency signal of the structure.

Secondly, the embodiment of the invention decomposes the identified structural frequency time sequence into two eigenmodes through a variation mode decomposition algorithm, can effectively denoise and remove seasonal modes in the frequency time sequence, and utilizes principal component analysis to combine frequency characteristics to generate Euclidean distance of a principal component matrix, so that negative environmental impact factors are well removed from frequency data by using principal component damage coefficients corresponding to the frequency characteristics, the negative impact of the environmental impact factors on the frequency data is overcome, the characterization of the frequency data on structural damage conditions is improved, and the training efficiency of a subsequent SVR model is further accelerated.

Thirdly, according to the embodiment of the invention, the basic rule of Euclidean distance of each point is determined from frequency distribution through SVR model learning, namely, the mapping relation between the frequency signal of the structure and the damage coefficient of the main component is constructed, the SVR model is made to learn, and a control chart for real-time detection is constructed according to the trained SVR model and a verification set. Finally, in actual detection, the current frequency data is predicted by utilizing the SVR model, and whether the predicted result accords with the condition of structural damage is judged by using the control diagram, so that a final structural damage detection result is obtained. The embodiment of the invention takes the fact that the data monitored during actual structural damage detection are usually nonlinear, and for the embodiment of the invention, the mapping relation between the frequency signal of the structure and the damage coefficient of the main component is learned by utilizing the SVR algorithm, so that the nonlinear relation among variables is constructed, the structural damage detection precision and the structural damage detection efficiency are improved, and the method has a certain engineering application prospect.

In addition, the embodiment of the invention also provides a bridge structure damage detection system, which comprises:

The content in the method embodiment is applicable to the system embodiment, the functions specifically realized by the system embodiment are the same as those of the method embodiment, and the achieved beneficial effects are the same as those of the method embodiment.

The embodiment of the invention also provides a computer readable storage medium, in which a program executable by a processor is stored, the program executable by the processor is used for executing the bridge structure damage detection method.

Similarly, the content in the above method embodiment is applicable to the present storage medium embodiment, and the specific functions of the present storage medium embodiment are the same as those of the above method embodiment, and the achieved beneficial effects are the same as those of the above method embodiment.

In some alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may in fact be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more thorough understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed, and in which sub-operations described as part of a larger operation are performed independently.

Furthermore, while the invention is described in the context of functional modules, it should be appreciated that, unless otherwise indicated, one or more of the functions and/or features may be integrated in a single physical device and/or software module or may be implemented in separate physical devices or software modules. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary to an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be apparent to those skilled in the art from consideration of their attributes, functions and internal relationships. Accordingly, one of ordinary skill in the art can implement the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative and are not intended to be limiting upon the scope of the invention, which is to be defined in the appended claims and their full scope of equivalents.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in the form of a software product stored in a storage medium, including several programs for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable programs for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with a program execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the programs from the program execution system, apparatus, or device and execute the programs. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the program execution system, apparatus, or device.

More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable program execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.

In the foregoing description of the present specification, reference has been made to the terms "one embodiment/example", "another embodiment/example", "certain embodiments/examples", and the like, means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

While the preferred embodiment of the present invention has been described in detail, the present invention is not limited to the embodiments described above, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the present invention, and these equivalent modifications and substitutions are intended to be included in the scope of the present invention as defined in the appended claims.

Claims

1. The bridge structure damage detection method is characterized by comprising the following steps of:

2. The method for detecting damage to a bridge structure according to claim 1, wherein the step of preprocessing the historical frequency signal comprises:

firstly, denoising the historical frequency signal by using a variation modal decomposition method, removing features belonging to short-term seasonal modes, and generating a first eigenfrequency feature; and then, carrying out principal component analysis on the first eigenfrequency characteristic to obtain a principal component damage coefficient corresponding to the first eigenfrequency characteristic.

3. The method for detecting damage to a bridge structure according to claim 2, wherein the step of denoising the historical frequency signal and removing features belonging to short-term seasonal patterns by using a variation modal decomposition method, and generating the first eigenfrequency feature comprises:

Defining that the historical frequency signal is composed of a plurality of modal components, and decomposing the historical frequency signal into a plurality of modal components;

constructing constrained variation problems according to a plurality of modal components, and converting the constrained variation problems into unconstrained variation problems by utilizing Lagrange multipliers and quadratic penalty items to obtain an extended Lagrange equation;

and solving the extended Lagrange equation by adopting a multiplication operator alternating direction method, so as to obtain a first eigenfrequency characteristic after separating the characteristic belonging to the short-term seasonal mode.

4. The method for detecting damage to a bridge structure according to claim 3, wherein the step of performing principal component analysis on the first eigenfrequency feature to obtain a principal component damage coefficient corresponding to the first eigenfrequency feature includes:

constructing a damage characteristic matrix in an original space by utilizing the first eigenfrequency characteristics, wherein the dimension of the damage characteristic matrix is obtained by constructing the number of modal components selected by the damage characteristic matrix;

singular value decomposition is carried out on a covariance matrix of the damage characteristic matrix to generate a first matrix, the damage characteristic matrix is projected into a principal component space through the first matrix to generate a second matrix positioned in the principal component space, and the dimension of the second matrix is obtained through the number of environmental influence factors of the first eigenfrequency characteristic;

And performing inverse transformation on the second matrix to obtain a third matrix in an original space, and generating a main component damage coefficient corresponding to the first eigenfrequency characteristic according to the third matrix and the damage characteristic matrix.

5. The method for detecting damage to a bridge structure according to claim 4, wherein the step of generating the model training set and the damage-free verification set using the preprocessed plurality of historical frequency signals comprises:

and extracting a part of the first eigenvalues and the corresponding principal component damage coefficients thereof from the plurality of first eigenvalues and the corresponding principal component damage coefficients thereof to serve as a model training set, and taking the rest of the data as a non-damage verification set, wherein the non-damage verification set and the model training set both comprise the first eigenvalues and the corresponding principal component damage coefficients thereof, and the principal component non-damage coefficients are used as marking results of the first eigenvalues.

6. The method for detecting damage to a bridge structure according to claim 5, wherein the step of training the SVR model using the model training set to generate the structural damage recognition model comprises:

and carrying out model training on the SVR model by taking the first eigenfrequency characteristic as a characteristic and the main component damage coefficient as a target and utilizing the model training set, so that the SVR model learns the association relation between the first eigenfrequency characteristic and the main component damage coefficient, and further obtaining a structural damage identification model.

7. The method for detecting damage to a bridge structure according to claim 6, wherein the step of verifying the structural damage recognition model using the non-destructive verification set and constructing a control chart according to the verification result of the structural damage recognition model comprises:

firstly, inputting each first eigenfrequency characteristic in the nondestructive testing set into the structural damage identification model, and outputting the predicted principal component damage coefficient corresponding to each first eigenfrequency characteristic through the structural damage identification model;

then, obtaining a principal component damage coefficient corresponding to each first eigenfrequency characteristic in the nondestructive testing set as a real principal component damage coefficient, calculating a difference value between a predicted principal component damage coefficient corresponding to each first eigenfrequency characteristic and the real principal component damage coefficient, and using the difference value as a prediction error signal of the structural damage identification model to further obtain a plurality of prediction error signals, namely a testing result of the structural damage identification model;

and finally, calculating the average value and the standard deviation of the plurality of prediction error signals, and determining the upper limit value of the control chart according to the average value and the standard deviation of the plurality of prediction error signals, thereby constructing the control chart.

8. The method for detecting structural damage of bridge according to claim 1, wherein the step of obtaining the current frequency signal of the bridge to be detected and preprocessing the current frequency signal, and performing structural damage detection on the preprocessed current frequency signal by using the control diagram and the structural damage recognition model, to obtain the structural detection result of the bridge to be detected comprises the following steps:

acquiring a current frequency signal of a bridge to be detected, denoising the current historical frequency signal by using a variation modal decomposition method, removing features belonging to a short-term seasonal mode, generating a current frequency feature, and then performing principal component analysis on the current frequency feature to obtain a current principal component damage coefficient corresponding to the current frequency feature;

inputting the current frequency characteristic into a structural damage identification model to obtain a current predicted principal component damage coefficient;

calculating the difference between the current principal component damage coefficient and the current predicted principal component damage coefficient, and taking the difference as a current prediction error signal of a structural damage model;

and obtaining a structure detection result of the bridge to be detected according to the current prediction error signal and the control diagram.

9. A bridge construction damage detection system, comprising:

10. A storage medium having stored therein processor-executable instructions which, when executed by a processor, are for performing a bridge construction damage detection method according to any one of claims 1-8.