A multi-step identification method and system for railway station building structural damage

Abstract

A multi-step identification method and system for railway station building structural damage, the method comprising: step 1, dividing the railway station building structure into multiple substructures according to the different characteristics of each part of the railway station building structure; step 2, determining the damaged substructure in the railway station building structure, then determining the damaged components in the known damaged substructure, and finally determining the damage degree of the damaged components. The present invention first divides the station building into multiple substructures according to the different characteristics of each part of the structure, and further divides the components in each substructure into important components and general components according to sensitivity analysis, determines the damaged substructure in the station building, then determines the damaged components in the known damaged substructure, and finally determines the damage degree of the damaged components. This method can reduce the complexity of the network, greatly reduce the learning samples, and thus save learning time.

Description

Technical Field

[0001] The present invention relates to the field of building health monitoring, and in particular to a multiple-step identification method and system for railway station building structural damage. Background Art

[0002] Structural health monitoring uses sensors to obtain information related to the health status of the structure (such as stress, strain, temperature, natural frequency, mode, etc.) online and in real time. It combines advanced communication systems, information processing technology and finite element modeling methods to diagnose the location and extent of structural damage, and evaluate the service condition, reliability, durability and bearing capacity of the structure. In this way, structural damage can be controlled at an early stage to eliminate safety hazards or an early warning signal can be issued when the structural condition deteriorates seriously, thereby avoiding major safety accidents and ensuring the safety of people’s lives and property.

[0003] In recent decades, many typical engineering accidents have occurred in large-span space structures at home and abroad. The frequent engineering accidents have not only caused the loss of life and property, but also had a bad social impact. Therefore, how to diagnose the structure and predict the failure behavior of the structure in advance has become an urgent problem to be solved in current large-scale projects. Summary of the invention

[0004] In view of the technical defects and technical drawbacks in the prior art, the embodiments of the present invention provide a multi-step identification method and system for railway station building structural damage that overcomes the above problems or at least partially solves the above problems. The specific scheme is as follows:

[0005] As a first aspect of the present invention, a multi-step identification method for railway station building structural damage is provided, the method comprising:

[0006] Step 1, dividing the railway station building structure into multiple substructures according to the different characteristics of each part of the railway station building structure;

[0007] Step 2, determine the damaged substructure in the railway station structure, then determine the damaged components in the substructure known to be damaged, and finally determine the damage extent of the damaged components.

[0008] Furthermore, the method also includes: dividing the components in each substructure into important components and general components based on sensitivity analysis, and only performing damage identification on the important components.

[0009] Furthermore, based on sensitivity analysis, they are divided into important components and general components. Specifically, based on the Spearman rank correlation coefficient as the standard for distinguishing important components from general components, those with a Spearman rank correlation coefficient equal to or greater than 0.025 are important components, otherwise they are general components.

[0010] Furthermore, in step 2, determining the damaged substructure in the railway station building structure includes: determining a first damage indicator that affects the railway station building structure, using the first damage indicator as a training sample to train a first neural network; obtaining the first damage indicator of each substructure as an input parameter of the trained first neural network, and determining the damaged substructure in the railway station building structure based on the output result of the first neural network, wherein the first damage indicator is a damage indicator related to the damage location.

[0011] Furthermore, in step 2, determining a damaged component in a substructure known to be damaged includes: determining a second damage indicator that affects the components in the substructure, and using the second damage indicator as a training sample to train a second neural network; obtaining the second damage indicator of the substructure as an input parameter of the trained second neural network, and determining the damaged component in the substructure based on an output result of the second neural network, wherein the second damage indicator is a damage indicator related to the damage location.

[0012] Furthermore, in step 2, determining the degree of damage of the damaged component includes: determining a third damage indicator that affects the degree of damage of the component in the substructure, and using the third damage indicator as a training sample to train the third neural network; obtaining the third damage indicator of the damaged component in the substructure as an input parameter of the trained third neural network, and determining the degree of damage of the component based on the output result of the third neural network.

[0013] Furthermore, the damage index related to the damage position is the regularized frequency change rate or the change ratio of the second-order frequency.

[0014] Furthermore, the third damage index determining the damage degree of the components in the substructure is the square ratio of the second-order frequency change.

[0015] As a second aspect of the present invention, a multiple-step identification system for railway station building structural damage is provided, the system comprising a classification module and a step-by-step identification module;

[0016] The classification module is used to divide the railway station building structure into a plurality of substructures according to the different characteristics of each part of the railway station building structure;

[0017] The step-by-step identification module is used to determine the damaged substructure in the railway station building structure, then determine the damaged components in the substructure known to be damaged, and finally determine the damage degree of the damaged components.

[0018] The present invention has the following beneficial effects:

[0019] The present invention first divides the station building into multiple substructures according to the different characteristics of each part of the structure, and further divides the components in each substructure into important components and general components according to sensitivity analysis, determines the damaged substructures in the station building, and then determines the damaged components in the substructures known to be damaged, and finally determines the damage degree of the damaged components. This method can reduce the complexity of the network, greatly reduce the learning samples, and thus save learning time. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a multiple-step identification method for railway station building structural damage provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, as a first embodiment of the present invention, a multi-step identification method for railway station building structural damage is provided, which mainly includes the following steps:

[0023] Step 1, dividing the railway station building structure into multiple substructures according to the different characteristics of each part of the railway station building structure;

[0024] Step 2, determine the damaged substructure in the railway station structure, then determine the damaged components in the substructure known to be damaged, and finally determine the damage extent of the damaged components.

[0025] The idea of ​​this method is described in detail below:

[0026] Among them: The specific classification of railway station building structures is shown in the following table. Since the use of substructures may still fall into the dimensionality disaster, the components in the substructure need to be further divided into important components and general components based on sensitivity analysis. Damage identification is only for important components. Among them, the Spearman rank correlation coefficient is used as the standard for distinguishing important components from general components. Components with a correlation coefficient of 0.025 or above are important components, otherwise they are general components.

[0027]

[0028] Preferably, determining the damaged substructure in the railway station building structure comprises: determining a first damage indicator affecting the railway station building structure, using the first damage indicator as a training sample to train a first neural network; obtaining the first damage indicator of each substructure as an input parameter of the trained first neural network, and determining the damaged substructure in the railway station building structure based on an output result of the first neural network, wherein the first damage indicator is a damage indicator related to the damage location;

[0029] In the above embodiment, since only the position of the damaged substructure needs to be determined, the Monte Carlo method is used to randomly change the stiffness of all important components in the substructure to generate the required data, i.e., the first damage index, and then the corresponding training samples are generated using these data. Here, only the position of the damaged substructure needs to be determined, so when selecting input parameters, parameters that are not related to the degree of damage but only to the damage position should be used as much as possible, and the number of output units is the number of substructures.

[0030] Preferably, determining a damaged component in a substructure known to be damaged includes: determining a second damage indicator that affects the component in the substructure, and using the second damage indicator as a training sample to train a second neural network; obtaining the second damage indicator of the substructure as an input parameter of the trained second neural network, and determining the damaged component in the substructure based on an output result of the second neural network, wherein the second damage indicator is a damage indicator related to the damage location.

[0031] In the above embodiment, since the damaged components are selected in the substructure instead of the entire railway station structure, the training samples can be greatly reduced and the network training can be simplified. At this time, the Monte Carlo method needs to be used to generate the required data, i.e., the second damage index, for all important components in the substructure, and then the corresponding training samples are generated using these data. Similarly, here, it is also necessary to try to select parameters related to the damage location as input, and the number of output units is the number of important components in the substructure.

[0032] Preferably, determining the degree of damage of the damaged component includes: determining a third damage indicator that affects the degree of damage of the component in the substructure, using the third damage indicator as a training sample to train the third neural network; obtaining the third damage indicator of the damaged component in the substructure as an input parameter of the trained third neural network, and determining the degree of damage of the component based on the output result of the third neural network.

[0033] In the above embodiment, after the location of the damaged component is determined, the problem is no longer a classification problem, but a regression problem. Therefore, the label no longer uses a discrete value of one-hot encoding (a vector composed of 0 and 1), but a continuous value such as the degree of damage. At this time, it is necessary to use input parameters related to the degree of damage as much as possible, such as the change in frequency before and after damage, the square of the frequency change, the change in vibration mode before and after damage, etc.

[0034] Among them, the input parameters of the neural network are varied. Selecting different input parameters can achieve different damage identification goals. The cost and performance of selecting different input parameters are also different. Therefore, in order to better realize the damage identification of the structure, it is necessary to analyze the advantages and disadvantages of various damage indicators. Only in this way can the appropriate damage indicator be selected. The present invention mainly uses two types of damage indicators: frequency-based and modal vibration-based damage indicators;

[0035] Frequency-based dynamic damage indicators include:

[0036] (1) Regularized frequency change rate NFRN

[0037] Kaminski pointed out that the frequency change rate is related to both the damage location and the damage degree, while the regularized frequency change rate is only related to the damage location.

[0038] The natural frequency reduction (NFR) is expressed as:

[0039]

[0040] The normalization of natural frequency reduction (NFRN) is expressed as:

[0041]

[0042] Where q represents the number of all vibration modes involved in the calculation.

[0043] Regularized frequency change rate NFRN i The principle used for damage location is derived as follows: the change in the i-th order natural frequency of the structure caused by local damage is a function of the stiffness change and the damage location:

[0044] δω i =f(δK, r) (3)

[0045] Where r is the damage position vector, and the series expansion and ignoring the high-order terms yields:

[0046]

[0047] Since there is no damage, there is no frequency change, so f(0, r) = 0, then:

[0048] NFR i =δKg i (0, r) (5)

[0049] Therefore, when a single damage occurs in the structure, the regularized frequency change rate can be expressed as:

[0050]

[0051] Obviously, the regularized rate of change of frequency is only a function of the damage location:

[0052] (2) Change ratio of the second-order frequency

[0053] From formula (5), we can get:

[0054] NFR j =δKg j (0, r) (7)

[0055] Assuming that the change in stiffness is independent of frequency, we have:

[0056]

[0057] Obviously, the ratio of the changes in the two-order frequencies is only a function of the damage location.

[0058] (3) Square ratio of frequency change

[0059] The solution to the natural frequency comes down to the following generalized eigenvalue problem:

[0060]

[0061] Where M and K are the total mass matrix and total stiffness matrix of the structure, respectively, ω i is the i-th order circular frequency, φ i is the corresponding vibration mode vector. Assuming that the change of the stiffness matrix caused by damage is ΔK, while the mass remains unchanged, we have:

[0062]

[0063] In the formula, and Δφ i are the changes of frequency and vibration mode vector caused by damage respectively; multiply equation (10) by φ on the left i T , applying equation (9) and ignoring the higher-order terms, we get:

[0064]

[0065] The damage ΔK of the overall stiffness matrix is ​​composed of the stiffness matrices of each damaged unit, namely:

[0066]

[0067] Where n e is the total number of damaged units, ΔK e is the stiffness matrix of the e-th damaged unit in the expanded dimension. Substituting equation (12) into equation (11), we get:

[0068]

[0069] Considering the unit damage index α n The introduction of can be considered as:

[0070] ΔK e =α e K e (14)

[0071] So, we get:

[0072]

[0073] For a single damaged unit e, equation (15) can be simplified as:

[0074]

[0075] This formula shows that the rate of change of the frequency square is related to the degree of unit damage α e Related to the position (unit e), when using the i and j order frequency information, we can get:

[0076]

[0077] It can be seen that when there is only one damaged unit or the damage degree of each damaged unit is similar, the change ratio of the frequency square is only a function of the damage location; the commonly used parameters related to the damage degree include the frequency change rate and the frequency square change rate. The station building is a high-order hyperstatic structure. Generally, it can withstand the sudden break of a certain component. However, if the damage is caused by corrosion (other components have also appeared weak links), the redistribution of force will lead to or accelerate the damage of more components. The integrity of the structure will be seriously damaged and may cause the gradual collapse of the structure. In the past, the use of neural networks to study damage was mainly aimed at single damage. This is because the regularized frequency in single damage is only related to the damage location, but not to the damage degree, which makes the damage location easier to identify. In general, for a structural system composed of multiple components, the damage situation is often not a single damage. For example, damage caused by corrosion may involve multiple components at the same time, which is a typical multiple damage situation. For multiple damage situations, the regularized frequency is not only related to the damage location, but also to the damage degree. Therefore, it is more difficult to identify the damage location using the usual method than for single damage situations. However, since this paper uses the Monte Carlo method to generate enough samples, it is not important whether the samples are only related to the damage location. We use the deep learning method, and the sample parameters can be arbitrarily selected as long as there are a large number of samples. In the subsequent analysis, we mainly use the regularized frequency change rate, the change ratio of the second-order frequency, and the square ratio of the second-order frequency change. Based on the Monte Carlo method, we can directly study the multi-damage situation, or we can only study the damage identification of a single damage. The basic idea of ​​the approach is to reduce the problem to two problems: classification and regression.

[0078] Among them, regarding the classification problem:

[0079] (1) The Monte Carlo method is used to generate damage data in each substructure. The damage data here refers to the natural frequency obtained when the elastic modulus of important components of the structure changes. The damage data can be used to calculate the regularized frequency change rate, the change ratio of the second-order frequency, and the square ratio of the second-order frequency change. They can be used as input samples of the network, and the labels are encoded using one-hot encoding.

[0080] (2) Divide the damaged samples into training samples and test samples (also called validation samples). Usually 80% of the total samples are used as training samples and 20% as test samples.

[0081] (3) Select the number of neural network layers, the number of units in each layer, and the activation function. The loss function is categorical crossentropy. The network is trained using the samples and labels in (1). The last layer is a softmax layer with n units, and an array of n probability values ​​(summing up to 1) is returned. The largest element in the array indicates that the corresponding substructure (or component) is damaged.

[0082] Among them, about the regression problem:

[0083] The component damage degree is a regression problem. The input can be the same as the classification problem, but the label is no longer a discrete value encoded by one-hot, but a continuous value such as the damage degree. The loss function uses MSE, that is, mean squared error (MSE), which is the square of the difference between the predicted value and the target value. The number of output units of the network is the number of important components of the substructure, and the output of the unit is the component damage degree.

[0084] The present invention provides a multiple-step identification method for structural damage based on sensitivity. First, the entire structure is divided into multiple substructures, and then the components in the substructures are divided into important components and general components according to their own sensitivity. Damage identification will only be performed on important components, and the damage problem is solved separately as a substructure classification identification problem and a component damage degree regression problem.

[0085] As a second embodiment of the present invention, a multiple-step identification system for railway station building structural damage is provided, the system comprising a classification module and a step-by-step identification module;

[0086] The classification module is used to divide the railway station building structure into a plurality of substructures according to the different characteristics of each part of the railway station building structure;

[0087] The step-by-step identification module is used to determine the damaged substructure in the railway station building structure, then determine the damaged components in the substructure known to be damaged, and finally determine the damage degree of the damaged components.

[0088] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-step identification method for railway station building structural damage, characterized in that: The method comprises: Step 1, dividing the railway station building structure into multiple substructures according to the different characteristics of each part of the railway station building structure; Step 2, determine the damaged substructure in the railway station structure, then determine the damaged components in the known damaged substructure, and finally determine the damage degree of the damaged components; Wherein, in step 2, determining the damaged substructure in the railway station building structure includes: determining a first damage indicator affecting the railway station building structure, using the first damage indicator as a training sample to train a first neural network; obtaining the first damage indicator of each substructure as an input parameter of the trained first neural network, and determining the damaged substructure in the railway station building structure based on the output result of the first neural network, wherein the first damage indicator is a damage indicator related to the damage location; Wherein, in step 2, determining the damaged component in the substructure known to be damaged includes: determining a second damage indicator affecting the component in the substructure, and using the second damage indicator as a training sample to train the second neural network; obtaining the second damage indicator of the substructure as an input parameter of the trained second neural network, and determining the damaged component in the substructure based on the output result of the second neural network, wherein the second damage indicator is a damage indicator related to the damage position; Wherein, in step 2, determining the damage degree of the damaged component includes: determining a third damage index that affects the damage degree of the component in the substructure, using the third damage index as a training sample to train the third neural network; obtaining the third damage index of the damaged component in the substructure as an input parameter of the trained third neural network, and determining the damage degree of the component based on the output result of the third neural network; Wherein, the damage index related to the damage position is the regularized frequency change rate or the change ratio of the second-order frequency; the third damage index for determining the degree of damage to the components in the substructure is the square ratio of the second-order frequency change; Among them, for the case of multiple damages: Based on the Monte Carlo method, the basic idea of ​​studying the case of multiple damages is to reduce the problem to two problems: classification and regression; Among them, regarding the classification problem: A1. Monte Carlo method is used to generate damage data in each substructure. The damage data refers to the natural frequency obtained when the elastic modulus of important components of the structure changes. The damage data is used to calculate the regularized frequency change rate, the change ratio of the second-order frequency and the square ratio of the second-order frequency change as the input samples of the network. The labels are encoded using one-hot encoding. B1. Divide the damage samples into training samples and test samples, where 80% of the total samples are used as training samples and 20% as test samples; C1, select the number of neural network layers, the number of units in each layer and the activation function. The loss function uses the classification cross entropy. The network is trained with the samples and labels in A1. The last layer uses a softmax layer with n units. At this time, an array of n probability values ​​is returned. The largest element in the array indicates that the corresponding substructure or component is damaged. Among them, about the regression problem: The damage degree of the component is a regression problem. The input in this case is the same as that of the classification problem, but the label is no longer a discrete value encoded by one-hot, but a continuous value of the damage degree. The loss function uses mse, that is, mean square error, which is the square of the difference between the predicted value and the target value. The number of output units of the network is the number of important components of the substructure, and the output of the unit is the damage degree of the component.

2. The multi-step identification method for railway station building structural damage according to claim 1 is characterized in that: The method further includes: dividing the components in each substructure into important components and general components based on sensitivity analysis, and performing damage identification only on the important components.

3. The multi-step identification method for railway station building structural damage according to claim 1 is characterized in that: Based on sensitivity analysis, they are divided into important components and general components. Specifically, based on the Spearman rank correlation coefficient as the standard for distinguishing important components from general components, if the Spearman rank correlation coefficient is equal to or greater than 0.025, it is an important component, otherwise it is a general component.

Images