CN115100176A - Neural network-based reinforced concrete column damage assessment method - Google Patents

Abstract

The invention discloses a damage assessment method for a reinforced concrete column based on a neural network, comprising the following steps: S1, establishing a damage data set of a reinforced concrete column model under an explosion load through a finite element method; S2, sending the damage data set to the construction site Good convolutional neural network model for training, and save the best training weight and the corresponding convolutional neural network model; S3, input the acquired photo of the target object into the saved convolutional neural network model for damage assessment, and output the damage assessment result , which overcomes the shortcomings of time-consuming, labor-intensive, and insufficient accuracy and safety of manual evaluation methods based on expert experience.

Description

A neural network-based damage assessment method for reinforced concrete columns

technical field

The invention relates to the technical field of neural networks, in particular to a damage assessment method for reinforced concrete columns based on neural networks.

Background technique

Reinforced concrete columns are the most basic vertical load-bearing components in various engineering structures such as buildings, bridges, and hydraulics, and their damage degree affects the local and overall safety of the structure. The traditional damage detection of reinforced concrete columns mainly relies on on-site investigation, which is highly dependent on engineering experience, is time-consuming and labor-intensive, and lacks accuracy and reliability. Especially for post-earthquake site damage assessment, the safety problem is even more severe due to the existence of possible subsequent aftershocks.

SUMMARY OF THE INVENTION

In view of the defects in the prior art, the present invention provides a neural network-based damage assessment method for reinforced concrete columns, which can directly assess the damage results of reinforced concrete columns, and overcomes the time-consuming, labor-intensive, accurate, and cost-intensive manual evaluation methods based on expert experience. The disadvantage of insufficient security.

The invention provides a neural network-based damage assessment method for reinforced concrete columns, comprising the following steps:

S1, establish the damage data set of the reinforced concrete column model under the explosion load by the finite element method;

S2, sending the damage data set into the constructed convolutional neural network model for training, and saving the optimal training weight and the corresponding convolutional neural network model;

S3, input the acquired photo of the target object into the saved convolutional neural network model for damage assessment, and output the damage assessment result.

Preferably, the convolutional neural network model includes:

The backbone network is used to extract image features and calculate the volume loss rate;

The damage judgment network is used to simulate the functional relationship between the volume loss rate and the damage, and output the damage level.

Preferably, the backbone network adopts MobilenetV3-Small.

Preferably, the damage judgment network includes two fully connected linear layers.

Preferably, the step S1 specifically includes:

Build reinforced concrete columns and finite element models of explosives;

Adjust the explosive equivalent and the detonation point, and calculate the damaged reinforced concrete column model under the explosion load based on the reinforced concrete column and the finite element model of the explosive;

Apply an axial compressive load to the damaged reinforced concrete column model, and calculate the ultimate bearing capacity of the damaged reinforced concrete column model;

Calculate the ratio of the ultimate bearing capacity of the damaged reinforced concrete column model to the vertical ultimate bearing capacity of the complete reinforced concrete column model to obtain a damage value;

Select a typical angle to obtain image data of a damaged reinforced concrete column model;

The acquired image data of the damaged reinforced concrete column model is classified according to the damage value to obtain a damage data set.

Preferably, the step S1 further includes:

The image data of the classified reinforced concrete column model with damage is processed using a preset edge detection algorithm.

Preferably, the step S1 further includes:

The injury data set is divided proportionally to obtain a training set, a validation set and a test set.

Preferably, the step S2 specifically includes:

The training set and the verification set are sent to the constructed convolutional neural network model for preliminary training;

The training set and the verification set are fused and sent to the convolutional neural network model after preliminary training for further training, and the predicted damage result is obtained;

The test set is sent to the convolutional neural network model after further training, and the best model parameters are selected according to the accuracy and loss of the convolutional neural network model after further training on the test set, and the best model parameters are saved. Train weights and corresponding convolutional neural network models.

The beneficial effects of the present invention are:

The damage data set is established by the finite element method, the convolutional neural network model is built, and the model is trained with the damage data set, and the image of the target object is input into the training model for damage assessment, and the damage degree of the reinforced concrete column is directly predicted. The shortcomings of empirical manual evaluation methods are time-consuming and labor-intensive, and their accuracy and safety are insufficient. And the convolutional neural network model includes a backbone network and a damage judgment network, which can calculate the damage of reinforced concrete columns according to the volume loss rate of the input image, which improves the interpretability of the damage assessment results.

Description of drawings

In order to illustrate the specific embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that are required to be used in the description of the specific embodiments or the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. In the drawings, each element or section is not necessarily drawn to actual scale.

1 is a schematic flowchart of Embodiment 1 of the present invention;

2 is a schematic flowchart of Embodiment 2 of the present invention;

Figure 3a is the picture before Canny edge detection algorithm processing;

Figure 3b is a picture processed by the Canny edge detection algorithm;

4 is a schematic structural diagram of a convolutional neural network model according to Embodiment 2 of the present invention;

Figure 5a is a picture of mild injury;

Figure 5b is a picture of moderate damage;

Figure 5c is a picture of a severe injury.

Detailed ways

Embodiments of the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings. The following examples are only used to more clearly illustrate the technical solutions of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

It should be noted that, unless otherwise specified, the technical or scientific terms used in this application should have the usual meanings understood by those skilled in the art to which the present invention belongs.

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

It is to be understood that, when used in this specification and the appended claims, the terms "comprising" and "comprising" indicate the presence of the described features, integers, steps, operations, elements and/or components, but do not exclude one or The presence or addition of a number of other features, integers, steps, operations, elements, components, and/or sets thereof.

It is also to be understood that the terminology used in this specification of the present invention is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural unless the context clearly dictates otherwise.

It should further be understood that, as used in this specification and the appended claims, the term "and/or" refers to and including any and all possible combinations of one or more of the associated listed items .

Example 1:

As shown in FIG. 1 , an embodiment of the present invention provides a neural network-based damage assessment method for a reinforced concrete column, including the following steps:

S1, establish the damage data set of the reinforced concrete column model under the explosion load by the finite element method;

S2, send the damage data set into the constructed convolutional neural network model for training, and save the optimal training weight and the corresponding convolutional neural network model;

S3, input the acquired photo of the target object into the saved convolutional neural network model for damage assessment, and output the damage assessment result.

Embodiment 2:

As shown in FIG. 2 , an embodiment of the present invention provides a neural network-based damage assessment method for reinforced concrete columns, including:

Step 1: Model concrete, steel bars, air domains and explosives through HyperMesh software to obtain finite element models of reinforced concrete columns and explosives;

Step 2: Adjust the explosive equivalent and detonation point, import the finite element model into the LS-DYNA software to calculate the damaged reinforced concrete column model;

Step 3: Import the obtained reinforced concrete column model with damage back into HyperMesh software to add a rigid pressure plate model for applying axial compressive loads, and import it into LS-DYNA software again to calculate the ultimate bearing capacity of the damaged reinforced concrete column model and complete reinforced concrete Calculate the ratio of the vertical ultimate bearing capacity of the column model to obtain the damage value. Use Python program to realize automatic adjustment of explosive equivalent and detonation point and interface connection between HyperMesh and LS-DYNA;

Step 4: Import the damaged reinforced concrete column model into the HyperMesh software, select a typical angle, obtain the image data of the damaged reinforced concrete column model, and write a Python program to call the HyperMesh interface for batch operations;

Step 5: classify the acquired image data of the damaged reinforced concrete column model according to the calculated damage value;

Step 6: Use the Canny algorithm to process the image data of the damaged reinforced concrete column model to obtain a damage data set, and divide the damage data set in proportion to obtain a training set, a verification set, and a test set. Because the two-dimensional image obtained in the finite element model may not be able to restore the imaging effects of the structure in the real environment, such as reflection, shadow, etc., the embodiment of the present invention uses the Canny algorithm to extract the edge information of the image data, obtains the edge lines of the image, and then Follow-up operations can make the trained convolutional neural network model have certain applicability in practical situations, as shown in Figure 3a and Figure 3b.

Step 7: Adjust the parameters of the damaged reinforced concrete column model, send the damage data set to the convolutional neural network for training, and obtain the predicted damage result; specifically, send the damage data set to the convolutional neural network for training, and obtain the prediction Injury outcomes include:

The training batch size is set to 128, the training set and validation set are sent to the constructed convolutional neural network model for preliminary training, and the number of iterations is set to 30 for preliminary learning;

The training set and the validation set are fused and sent to the convolutional neural network model after initial training for further training, and the predicted damage result is obtained, and the number of iterations is set to 20 for further learning. Adam is used as the optimizer in the initial training phase, the initial learning rate is 1×10 ^-3 , the learning rate is reduced by half when the loss of the neural network model on the validation set exceeds 3 times without decreasing, β ₁ and β ₂ are set to 0.9 and 0.999 , ε is set to 1×10 ^-8 . In the further training phase, SGDM is used as the optimizer, the initial learning rate is 1×10 ^-4 , and the learning rate is reduced by half when the loss of the neural network model on the validation set exceeds 3 times without decreasing, and β is set to 0.9.

Step 8: Send the test set to the convolutional neural network model after further training, select the best model parameters according to the accuracy and loss of the convolutional neural network model after further training on the test set, and save the best training weight and The corresponding convolutional neural network model;

Step 9: Retrieve the saved convolutional neural network model, take a photo of the target object, remove the background of the target object photo through image processing tools such as PS, and send it to the convolutional neural network model after further training for damage judgment and evaluation, and obtain the damage evaluation result.

In the embodiment of the present invention, the convolutional neural network model includes a backbone network and a damage judgment network. As shown in FIG. 4 , the backbone network is used to extract image features to calculate the volume loss rate, and the damage judgment network is used to simulate the relationship between the volume loss rate and the damage. , and output the damage level. The backbone network selects MobilenetV3-Small. The damage judgment network consists of two fully connected linear layers. The input dimension of the first layer is 1, the output dimension is 64, and the input dimension of the second layer is 64 and the output dimension is 3. During the error calculation in the training stage, double errors are used to consider both the volume loss rate error and the damage level error. The calculation method of the volume loss rate error adopts the Mean Absolute Error (MAE), the calculation result is multiplied by the coefficient α, and the damage level The error calculation method adopts cross entropy (Cross Entropy), and the two errors are added to obtain the final error. In this application, the volume loss error and the damage classification error are combined when optimizing the network to balance the calculation accuracy of the image volume loss rate and the damage classification accuracy, so as to balance the overall network. At the same time, the network model is more in line with the process of pushing from volume loss rate to damage level, and the classification effect is improved.

The embodiment of the present invention is used for simulation experiments. The convolutional neural network model runs on a preset platform. The specific parameters of the platform are CPU: Intel(R) Xeon(R) Gold 6248@2.5Ghz×80; RAM: 128GB; GPU: GeForce RTX3090; OS: Ubuntu 18.04.5LTS. The alpha parameter of the model is set to 2 in the first stage and 10 in the second stage. The convolutional neural network model saved in the embodiment of the present invention can achieve an accuracy rate of 99.71% on the test set, and based on the damage assessment model of the convolutional neural network model, the time for structural damage assessment is much shorter than the traditional finite element method, and the It is suitable for quick judgment of structural damage in emergency situations.

In order to test the application of the trained convolutional neural network model in reality, 3D printing technology was used to print one model with mild, moderate and severe damage. Figures 5a, 5b, and 5c show the photos of the three samples. Image processing removes the background. A total of 24 test photos were obtained by shooting at different angles. The test photos are sent to the best model trained above, and the final accuracy rate is 70.83%. It can be considered that the convolutional neural network model saved in the embodiment of the present invention has certain application value in actual samples.

The invention provides a damage assessment method for reinforced concrete columns based on a neural network. A damage data set is established by a finite element method, a convolutional neural network model is built, and the model is trained by using the damage data set, and the image of the target object is input into the model. The damage assessment can directly predict the damage degree of reinforced concrete columns, which overcomes the shortcomings of time-consuming, labor-intensive, and insufficient accuracy and safety of manual evaluation methods based on expert experience. Moreover, in the embodiment of the present invention, the convolutional neural network model includes a backbone network and a damage judgment network, which can calculate the damage of the reinforced concrete column according to the volume loss rate of the input image, which improves the interpretability of the damage assessment result.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The technical solutions described in the foregoing embodiments can still be modified, or some or all of the technical features thereof can be equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention. The scope of the invention should be included in the scope of the claims and description of the present invention.

Claims (8)

1. a reinforced concrete column damage assessment method based on neural network, is characterized in that, comprises the following steps: S1, establish the damage data set of the reinforced concrete column model under the explosion load by the finite element method; S2, sending the damage data set into the constructed convolutional neural network model for training, and saving the optimal training weight and the corresponding convolutional neural network model; S3, input the acquired photo of the target object into the saved convolutional neural network model for damage assessment, and output the damage assessment result. 2. A kind of reinforced concrete column damage assessment method based on neural network according to claim 1, is characterized in that, described convolutional neural network model comprises: The backbone network is used to extract image features and calculate the volume loss rate; The damage judgment network is used to simulate the functional relationship between the volume loss rate and the damage, and output the damage level. 3. A neural network-based damage assessment method for reinforced concrete columns according to claim 2, wherein the backbone network adopts MobilenetV3-Small. 4. A neural network-based damage assessment method for reinforced concrete columns according to claim 2, wherein the damage judgment network comprises two fully connected linear layers. 5. A neural network-based reinforced concrete column damage assessment method according to claim 1, wherein the step S1 specifically comprises: Build reinforced concrete columns and finite element models of explosives; Adjust the explosive equivalent and the detonation point, and calculate the damaged reinforced concrete column model under the explosion load based on the reinforced concrete column and the finite element model of the explosive; Apply an axial compressive load to the damaged reinforced concrete column model, and calculate the ultimate bearing capacity of the damaged reinforced concrete column model; Calculate the ratio of the ultimate bearing capacity of the damaged reinforced concrete column model to the vertical ultimate bearing capacity of the complete reinforced concrete column model to obtain a damage value; Select a typical angle to obtain image data of a damaged reinforced concrete column model; The acquired image data of the damaged reinforced concrete column model is classified according to the damage value to obtain a damage data set. 6. A neural network-based damage assessment method for reinforced concrete columns according to claim 5, wherein the step S1 further comprises: The image data of the classified reinforced concrete column model with damage is processed using a preset edge detection algorithm. 7. a kind of reinforced concrete column damage assessment method based on neural network according to claim 1, is characterized in that, The step S1 also includes: The injury data set is divided proportionally to obtain a training set, a validation set and a test set. 8. A neural network-based reinforced concrete column damage assessment method according to claim 7, wherein the step S2 specifically comprises: The training set and the verification set are sent to the constructed convolutional neural network model for preliminary training; The training set and the verification set are fused and sent to the convolutional neural network model after preliminary training for further training, and the predicted damage result is obtained; The test set is sent to the convolutional neural network model after further training, and the best model parameters are selected according to the accuracy and loss of the further trained convolutional neural network model on the test set, and the best model parameters are saved. Train weights and corresponding convolutional neural network models.

Images