CN110276743A - A Convolutional Neural Network Based Structural Damage Degree Recognition Method - Google Patents

Abstract

The invention discloses a method for identifying the degree of structural damage based on a convolutional neural network, comprising the following steps: S1: constructing a simply supported beam structure and collecting training sample data according to the simply supported beam structure, the simply supported beam structure including N units , each unit has a different modal strain energy, and the training sample data includes the modal strain energy of each unit; S2: construct the input matrix according to the modal strain energy of each unit; S3: construct the volume according to the input matrix The model of the convolutional neural network; S4: use the training sample data including the modal strain energy of each unit to train the convolutional neural network and save the trained convolutional neural network; S5: use the trained convolutional neural network for the unknown Damage prediction for simply supported beam structures. The invention utilizes the weight sharing method of the convolutional neural network, reduces calculation parameters, and has high calculation speed, which has positive significance for the application of large structures.

Description

A Convolutional Neural Network Based Structural Damage Degree Recognition Method

technical field

The invention relates to the field of neural networks, and more specifically, to a method for identifying the degree of structural damage based on a convolutional neural network.

Background technique

Structural damage recognition is of great significance in the field of damage detection. Structural damage not only threatens people's lives, but also causes huge national property losses. Structure is an important support in people's life. Only by ensuring the safety of structure can people's daily life go smoothly. Therefore, the damage identification of structure is particularly important. How to accurately identify the degree of damage is an important problem currently facing.

The damage characteristics of the structure are usually manifested in the change of the inherent properties of the structure, but due to the influence of the complexity of the structure, the environment, the data processing method, etc., there will be a lot of interference information, resulting in the poor effect of damage identification. At present, structural damage identification is at a low-level stage. It can only judge whether there is damage to the structure and the approximate location of the damage, but cannot make an accurate judgment on the degree of damage. With the continuous improvement of computer algorithms, convolutional neural networks are gradually applied to various fields. The convolutional neural network has excellent nonlinear fitting ability, and can learn the nonlinear change law when the structure is damaged. Therefore, the convolutional neural network can be applied to the field of identification of damage degree.

Contents of the invention

In order to overcome the inability to accurately judge the degree of damage described in the prior art, the present invention provides a method for identifying the degree of structural damage based on volume and neural networks.

In order to solve the problems of the technologies described above, the technical solution of the present invention is as follows:

A method for identifying the degree of structural damage based on a convolutional neural network, characterized in that it comprises the following steps:

S1: Build a simply supported beam structure and collect training sample data according to the simply supported beam structure. The simply supported beam structure includes N units, each unit has a different modal strain energy, and the training sample data includes each unit. Modal strain energy;

S2: Construct the input matrix according to the modal strain energy of each element;

S3: Construct a convolutional neural network model according to the input matrix, the input matrix is the input of the convolutional neural network, and the output of the convolutional neural network is the predicted damage degree of the simply supported beam structure;

S4: Utilize the training sample data including the modal strain energy of each unit to train the convolutional neural network and save the trained convolutional neural network;

S5: Use the trained convolutional neural network to predict the damage of unknown simply supported beam structures.

Preferably, the simply supported beam structure includes 36 units.

Preferably, the training sample data is to collect damage data of 20%, 30%, 40%, 50%, and 60% of damage to each unit, a total of 36×5=180 cases, and 36 cases can be obtained in each case The modal strain energy of each unit, the damage data of 20%, 30%, 40%, 50% of each unit damage and the corresponding modal strain energy are used as the training set, and the damage data of 60% of each unit damage and the corresponding modal strain can be used as a test set.

Preferably, the input matrix is a 6×6 matrix, and the elements of the input matrix are the modal strain energy of each unit.

Preferably, the model of the convolutional neural network in step S3 includes an input layer, a first convolutional layer, a pooling layer, a second convolutional layer, a fully connected layer and a regression layer, and the first convolutional layer includes 50 convolution kernels , the second convolution layer includes 100 convolution kernels.

Preferably, the specific steps of step S4 are as follows:

S4.1: The training set enters the first convolutional layer from the input layer, and obtains 50 3×3 feature matrices through 50 convolution kernels of the first convolutional layer;

S4.2: Input the 50 3×3 feature matrices obtained in step S4.1 to the pooling layer to obtain 50 2×2 feature matrices after pooling;

S4.3: Input 50 pooled feature matrices of 2×2 to the second convolutional layer, and obtain 100 feature values from 100 convolution kernels in the second convolutional layer;

S4.4: Take 100 eigenvalues as the input of the fully connected layer and output a vector representing the degree of damage from the regression layer;

S4.5: Compare the vector output by the regression layer with the corresponding actual damage degree, and obtain the mean square error of the two;

S4.6: Adjust the volume and weight parameters of the convolutional neural network according to the mean square error, repeat S4.1 to S4.5, complete the training until the mean square error no longer changes, and save the trained convolutional neural network.

Preferably, in step S4.6, stochastic gradient descent method is used to adjust the volume and weight parameters of the neural network.

Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

The present invention utilizes the nonlinear learning ability of the convolutional neural network to obtain the nonlinear change law under different damage degrees, and saves the obtained weight parameters. At this time, the convolutional neural network can be regarded as a nonlinear function. When a new damage is input When the modal strain energy data corresponding to the degree is used, the prediction results close to the real one can be obtained; at the same time, the convolutional neural network uses the weight sharing method to reduce the calculation parameters, and the calculation speed is very fast, which has positive significance for the application of large structures.

Description of drawings

Figure 1 is a flowchart of a method for identifying the degree of structural damage based on a convolutional neural network.

Figure 2 is a schematic diagram of a simply supported beam structure.

Fig. 3 is the construction schematic diagram of input matrix;

In the figure, E1-E36 are the modal strain energies of 36 units; the A matrix is used as the input of the convolutional neural network.

Figure 4 is a schematic diagram of a convolutional neural network model.

Fig. 5 is a schematic diagram of the principle of convolution operation.

Fig. 6 is a schematic diagram of the pooling operation principle.

Detailed ways

The accompanying drawings are for illustrative purposes only and cannot be construed as limiting the patent;

In order to better illustrate this embodiment, some parts in the drawings will be omitted, enlarged or reduced, and do not represent the size of the actual product;

For those skilled in the art, it is understandable that some well-known structures and descriptions thereof may be omitted in the drawings.

The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Example 1

A method for identifying the degree of structural damage based on a convolutional neural network, as shown in Figure 1, includes the following steps:

S1: Build a simply supported beam structure and collect training sample data according to the simply supported beam structure, the simply supported beam structure includes 36 units, as shown in Figure 2, each unit has a different modal strain energy, the training sample data includes The modal strain energy of each unit, the training sample data is to collect the damage data of 20%, 30%, 40%, 50% and 60% damage of each unit respectively, a total of 36*5=180 kinds of situations, each In this case, the modal strain energy of 36 units can be obtained, and the damage data of 20%, 30%, 40%, 50% of each unit damage and the corresponding modal strain energy are used as the training set, and the damage data of 60% of each unit damage and the corresponding modal strain energy as the test set;

S2: Construct an input matrix according to the modal strain energy of each unit, the input matrix is a 6×6 matrix, as shown in Figure 3, the elements of the input matrix are the modal strain energy of each unit;

S3: Construct a convolutional neural network model according to the input matrix, the input matrix is the input of the convolutional neural network, the output of the convolutional neural network is the predicted damage degree of the simply supported beam structure, and the convolutional neural network model includes Input layer, first convolution layer, pooling layer, second convolution layer, fully connected layer and regression layer, the first convolution layer includes 50 convolution kernels, the second convolution layer includes 100 convolution kernels, As shown in Figure 4, different damage conditions are set to different vectors, for example, 20% of unit 1 damage is set as vector [0.2, 0, 0..., 0], and 20% of unit 2 damage is set as vector [0, 0.2,0...,0],..., and so on;

S4: Use the training sample data including the modal strain energy of each unit to train the convolutional neural network and save the trained convolutional neural network; where the convolution and pooling operation principles are shown in Figures 5 and 6, the specific steps are as follows:

S4.1: The training set enters the first convolutional layer from the input layer, and obtains 50 3×3 feature matrices through 50 convolution kernels of the first convolutional layer;

S4.2: Input the 50 3×3 feature matrices obtained in step S4.1 to the pooling layer to obtain 50 2×2 feature matrices after pooling;

S4.3: Input 50 pooled feature matrices of 2×2 to the second convolutional layer, and obtain 100 feature values from 100 convolution kernels in the second convolutional layer;

S4.4: Take 100 eigenvalues as the input of the fully connected layer and output a vector representing the degree of damage from the regression layer;

S4.5: Compare the vector output by the regression layer with the corresponding actual damage degree, and obtain the mean square error of the two;

S4.6: Adjust the weight parameters of the convolutional neural network using the stochastic gradient descent method according to the mean square error, repeat S4.1 to S4.5, complete the training until the mean square error no longer changes, and save the trained convolutional neural network .

S5: Use the trained convolutional neural network to predict the damage of unknown simply supported beam structures.

The same or similar reference numerals correspond to the same or similar components;

The terms describing the positional relationship in the drawings are only for illustrative purposes and cannot be interpreted as limitations on this patent;

Apparently, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limiting the implementation of the present invention. For those of ordinary skill in the art, on the basis of the above description, other changes or changes in different forms can also be made. It is not necessary and impossible to exhaustively list all the implementation manners here. All modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (7)

1. A method for identifying the degree of structural damage based on a convolutional neural network, comprising the following steps: S1: Build a simply supported beam structure and collect training sample data according to the simply supported beam structure. The simply supported beam structure includes N units, each unit has a different modal strain energy, and the training sample data includes each unit. Modal strain energy; S2: Construct the input matrix according to the modal strain energy of each element; S3: Construct a convolutional neural network model according to the input matrix, the input matrix is the input of the convolutional neural network, and the output of the convolutional neural network is the predicted damage degree of the simply supported beam structure; S4: Utilize the training sample data including the modal strain energy of each unit to train the convolutional neural network and save the trained convolutional neural network; S5: Use the trained convolutional neural network to predict the damage of unknown simply supported beam structures. 2. The method for identifying the degree of structural damage based on a convolutional neural network according to claim 1, wherein the simply supported beam structure comprises 36 units. 3. The method for identifying the degree of structural damage based on convolutional neural networks according to claim 2, wherein the training sample data is to collect damages of 20%, 30%, 40%, 50%, 60% of each unit respectively. % damage data, a total of 36 × 5 = 180 cases, in each case the modal strain energy of 36 units can be obtained, and the damage data of 20%, 30%, 40%, 50% of each unit damage and the corresponding The modal strain energy is used as the training set, and the 60% damage data of each unit damage and the corresponding modal strain energy are used as the test set. 4. The method for identifying the degree of structural damage based on convolutional neural networks according to claim 3, wherein the input matrix is a 6 × 6 matrix, and the elements of the input matrix are the modal strain energy of each unit . 5. the structural damage degree recognition method based on convolutional neural network according to claim 4, is characterized in that, the model of convolutional neural network in step S3 comprises input layer, the first convolutional layer, pooling layer, the second Convolution layer, fully connected layer and regression layer, the first convolution layer includes 50 convolution kernels, and the second convolution layer includes 100 convolution kernels. 6. the structural damage degree recognition method based on convolutional neural network according to claim 5, is characterized in that, the concrete steps of step S4 are as follows: S4.1: The training set enters the first convolutional layer from the input layer, and obtains 50 3×3 feature matrices through 50 convolution kernels of the first convolutional layer; S4.2: Input the 50 3×3 feature matrices obtained in step S4.1 to the pooling layer to obtain 50 2×2 feature matrices after pooling; S4.3: Input 50 pooled feature matrices of 2×2 to the second convolutional layer, and obtain 100 feature values from 100 convolution kernels in the second convolutional layer; S4.4: Take 100 eigenvalues as the input of the fully connected layer and output a vector representing the degree of damage from the regression layer; S4.5: Compare the vector output by the regression layer with the corresponding actual damage degree, and obtain the mean square error of the two; S4.6: Adjust the weight parameters of the convolutional neural network according to the mean square error, repeat S4.1 to S4.5, complete the training until the mean square error no longer changes, and save the trained convolutional neural network. 7. The method for identifying the degree of structural damage based on convolutional neural network according to claim 6, characterized in that in step S4.6, stochastic gradient descent method is used to adjust the volume and weight parameters of the neural network.