Method and system for rapidly evaluating post-earthquake building structure damage
By extending the residual convolutional neural network model, and using acceleration data from the bottom and top layers of the building structure for rapid assessment, the problems of high sensor dependence and insufficient damage prediction accuracy are solved, and low-cost, high-efficiency post-earthquake building damage identification is achieved.
Patent Information
- Application Number
- CN202511822058.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-05
- Publication Date
- 2026-02-17
AI Technical Summary
Existing technologies for assessing post-earthquake building structural damage rely heavily on the number of sensors, have limited feature extraction capabilities, and lack sufficient damage prediction accuracy, especially in cases of minor and moderate damage.
An extended residual convolutional neural network model is adopted. Acceleration data of the bottom and top layers of the building structure are obtained, preprocessed to construct a two-dimensional input, and combined with data augmentation and performance optimization to train the extended residual convolutional neural network model and output the damage level.
It enables low-cost and high-efficiency rapid assessment of post-earthquake building structural damage, improves the accuracy of identifying minor and moderate damage states, reduces the dependence on the number of sensors, and enhances feature extraction capabilities.
Smart Images
Figure CN121542704A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of earthquake engineering technology, and more specifically to a method and system for rapid assessment of post-earthquake building structural damage. Background Technology
[0002] Currently, rapid and accurate assessment of building structural damage after an earthquake is crucial for emergency response and post-disaster recovery.
[0003] However, existing structural damage assessment methods based on monitoring data are mainly divided into two categories: one is the parameter inversion method, which judges damage by identifying structural stiffness, modal parameters, etc., but has low identification accuracy and poor noise resistance under sparse sensor arrangement conditions; the other is the feature extraction method, such as autoregressive models and traditional machine learning models, which have certain effects, but rely on manual feature extraction and are difficult to process raw acceleration time history data.
[0004] In recent years, deep learning technology has been introduced into the field of structural damage identification. However, existing deep learning methods still have the following shortcomings: 1) They are highly dependent on the number of sensors, which usually need to be arranged layer by layer, resulting in high costs and difficulty in promotion in practical engineering; 2) Traditional convolutional kernels have limited receptive fields, making it difficult to capture long-range temporal dependencies in seismic responses; 3) Most methods only use seismic motion input and lack structural response information, resulting in insufficient prediction accuracy for minor, moderate and severe damage states.
[0005] Therefore, how to provide a rapid assessment method for post-earthquake building structural damage that can solve the above problems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] In view of this, the present invention provides a method and system for rapid assessment of post-earthquake building structural damage, in order to solve the problems of high dependence on the number of sensors, limited feature extraction capability, and insufficient damage prediction accuracy in the prior art.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: A rapid assessment method for post-earthquake building structural damage includes the following steps: S1: Obtain the first acceleration data of the bottom layer of the building structure when the ground motion acceleration is input, and the second acceleration data of the top layer of the building structure when the ground motion acceleration is input, and preprocess the first acceleration data and the second acceleration data to form two-dimensional input data; S2: Construct and train an extended residual convolutional neural network model; S3: Input the two-dimensional input data into the dilated residual convolutional neural network model for processing, and output the damage level of the building structure.
[0008] Preferably, S2 includes: S21: Construct a dilated residual convolutional neural network model, wherein the dilated residual convolutional neural network model includes an input layer, multiple dilated residual modules, a global pooling layer, and a fully connected layer; S22: Acquire seismic response data from various building structure types, and sequentially perform data augmentation and nonlinear time history analysis on the seismic response data to generate a training set; S23: Train the dilated residual convolutional neural network model using the training set; S24: Optimize the performance of the dilated residual convolutional neural network model during training.
[0009] Preferably, S21 specifically includes: The extended residual module includes a first convolutional layer, an extended convolutional layer, a second convolutional layer, and a residual connection layer connected in sequence.
[0010] Preferably, S24 specifically includes: Construct a loss function and use the loss function to optimize the training process of the dilated residual convolutional neural network model; The Adam optimizer is used to iteratively optimize the parameters of the dilated residual convolutional neural network model.
[0011] Preferably, in S3, the output structural damage level is determined based on the maximum inter-story drift angle of the structure, including intact, slightly damaged, moderately damaged, and severely damaged.
[0012] This invention also provides a rapid assessment system for post-earthquake building structural damage, comprising: The acquisition module is used to acquire the first acceleration data of the bottom layer of the building structure when the input seismic acceleration is applied, and the second acceleration data of the top layer of the building structure when the input seismic acceleration is applied, and to preprocess the first acceleration data and the second acceleration data to form two-dimensional input data; The model building module is used to build and train dilated residual convolutional neural network models; The evaluation module is used to input the two-dimensional input data into the dilated residual convolutional neural network model for processing and output the damage level of the building structure.
[0013] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for rapid assessment of post-earthquake building structural damage, which effectively solves the problems of high dependence on the number of sensors, limited feature extraction capability, and insufficient accuracy in identifying non-intact damage states in the prior art. It provides a practical solution for achieving low-cost, high-efficiency, and automated post-earthquake building safety assessment in engineering practice. Attached Figure Description
[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0015] Figure 1 This invention provides an overall flowchart of a method for rapid assessment of post-earthquake building structural damage. Figure 2 This is a schematic diagram of the extended residual module structure provided in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the receptive fields of dilated convolution and ordinary convolution provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the confusion matrix results of the model provided in this embodiment of the invention on a two-layer RC frame structure test set; Figure 5 A schematic diagram of the confusion matrix results of the model provided in this embodiment of the invention on a five-story steel frame structure test set; Figure 6 A schematic diagram of the confusion matrix results of the model provided in this embodiment of the invention on a test set of a twelve-story frame-shear wall structure; Figure 7 A comparison chart of prediction results using only seismic input and the method (seismic motion + top-level response) provided in this embodiment of the invention. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0017] See Figure 1 The embodiment of the present invention disclosed a method for rapid assessment of post-earthquake building structural damage, including the following steps: S1: Obtain the first acceleration data of the bottom layer of the building structure when the ground motion acceleration is input, and the second acceleration data of the top layer of the building structure when the ground motion acceleration is input, and preprocess the first acceleration data and the second acceleration data to form two-dimensional input data; S2: Construct and train an extended residual convolutional neural network model; S3: Input the two-dimensional input data into the dilated residual convolutional neural network model for processing, and output the damage level of the building structure.
[0018] In a specific embodiment, the specific implementation process of S1 includes: Accelerometers are placed on the bottom (or foundation) layer of the building structure to be evaluated to record the input ground motion Ag(t) as the first acceleration data, while acceleration sensors are placed on the top layer of the building structure to record the acceleration response At(t) at the top of the structure as the second acceleration data. This arrangement ensures the capture of key structural dynamic information while minimizing the number of sensors and reducing costs. Preprocessing of the first acceleration data and the second acceleration data includes the following steps: (1) Duration uniformity: The duration of all acceleration records is uniformly truncated or padded with zeros to the same length, such as 30 seconds, to ensure the consistency of the input data dimensions; (2) Sampling frequency standardization: unify the sampling frequency of all data to the same standard, such as 100Hz, in order to avoid frequency differences from interfering with the model; (3) Data combination: The one-dimensional bottom ground motion time history Ag(t) and the one-dimensional top acceleration response time history At(t) after the above processing are stacked in the channel dimension to form a two-dimensional array Xinput (size is [data length, 2]) as the input of the model. This process does not require any complex manual feature extraction (such as Fourier transform) and realizes true end-to-end processing.
[0019] See Figure 1-2 As shown, in a specific embodiment, S2 includes: S21: Construct the dilated residual convolutional neural network model ResNet-Di, which includes an input layer, multiple dilated residual modules, a global pooling layer, and a fully connected layer. The input layer is used to receive two-dimensional acceleration data of size [L,2] (i.e., two-dimensional input data), where L is the number of standardized data points and 2 represents two input channels (bottom ground motion and top response). After feature extraction by multiple dilated residual modules, the feature map is compressed into a one-dimensional feature vector using a global average pooling layer and connected to a fully connected layer. S22: Acquire seismic response data from various building structure types, and sequentially perform data augmentation and nonlinear time history analysis on the seismic response data to generate a training set; S23: Train the dilated residual convolutional neural network model using the training set; S24: Optimize the performance of the dilated residual convolutional neural network model during training.
[0020] See Figure 3 As shown, in a specific embodiment, S21 specifically includes: The dilated residual module comprises a first convolutional layer, a dilated convolutional layer, a second convolutional layer, and a residual connection layer connected in sequence. Its specific working principle includes: (1) First, the input features are reduced in dimensionality by using a 1×1 convolutional layer (i.e., the first convolutional layer) to reduce the amount of computation; (2) See Figure 3 As shown, a 3×3 dilated convolutional layer is then used for feature extraction. In this embodiment of the invention, the optimal dilation rate is set to 3. When the dilation rate is 3, the receptive field of a 3×3 convolutional kernel can be expanded to 7, which can effectively capture the temporal dependencies with large spans in seismic response data, which is something that traditional convolutional kernels cannot achieve. (3) Use a 1×1 convolutional layer (i.e., the second convolutional layer) again to restore the dimension of the feature map to match the original input; (4) Finally, the input of the module is added element by element to the output obtained by the above convolution operation to realize residual connection. This structure effectively alleviates the gradient vanishing problem in deep networks and ensures that shallow features can be directly transmitted to deep layers, thereby enhancing the training stability and feature fusion capability of the model.
[0021] Each convolutional layer is followed by a batch normalization layer and a ReLU activation function.
[0022] In one specific embodiment, S22 specifically includes: To improve the generalization ability of the model, the training data should cover a variety of structural types. Therefore, the embodiments of the present invention target two-story RC frame, five-story steel frame and twelve-story frame-shear wall structure respectively. Using nonlinear time history analysis, thousands of ground motion records selected from strong earthquake databases such as KiK-net, K-NET and PEER are input to generate structural response data.
[0023] To address the data imbalance problem caused by the scarcity of severe damage samples in actual earthquakes, this invention employs amplitude scaling as a data augmentation technique. This involves amplifying the peak ground acceleration of the original seismic records by 1x (original), 2x, 3x, and 4x, respectively, before performing time-history analysis. This method effectively increases the proportion of minor, moderate, and severe damage samples in the dataset, resulting in a more balanced distribution of data across categories and thus training a more robust model.
[0024] In one specific embodiment, S24 specifically includes: A loss function is constructed and used to optimize the training process of the dilated residual convolutional neural network model. The loss function employs a multi-class cross-entropy loss function as the metric for the training objective, and its specific expression is as follows:
[0025] In the formula, p i Let y represent the probability that the model predicts a sample belongs to the i-th class. i These are the true labels of the samples; The Adam optimizer was used to iteratively optimize the parameters of the dilated residual convolutional neural network model. When the learning rate was set to 0.00001, the model training process was the most stable and the convergence effect was the best. Other parameters in the Adam optimizer were set as follows: the exponential decay rate β1 for the first moment estimation was 0.9 and the exponential decay rate β2 for the second moment estimation was 0.999.
[0026] In a specific embodiment, in S3, the output structural damage level is determined based on the maximum inter-story drift angle of the structure, including intact, minor damage, moderate damage, and severe damage.
[0027] The present invention also provides a system for rapid assessment of post-earthquake building structural damage using any of the above embodiments, comprising: The acquisition module is used to acquire the first acceleration data of the bottom layer of the building structure when the input seismic acceleration is applied, and the second acceleration data of the top layer of the building structure when the input seismic acceleration is applied, and to preprocess the first acceleration data and the second acceleration data to form two-dimensional input data; The model building module is used to build and train dilated residual convolutional neural network models; The evaluation module is used to input the two-dimensional input data into the dilated residual convolutional neural network model for processing and output the damage level of the building structure.
[0028] To verify the performance of the method provided in this embodiment of the invention, the ResNet-Di model proposed in this embodiment of the invention was compared with mainstream deep learning models such as VGG16, VGG16_Di (VGG16 with dilated convolution), DenseNet, and ResNet-50 on the same experimental platform and dataset. The specific verification process is as follows: See Figure 4 , Figure 5 , Figure 6 As shown, ResNet-Di significantly and consistently outperforms other comparative models in predicting all damage levels (especially minor, moderate, and severe damage) on test sets with three different structure types.
[0029] Figure 7Comparative experiments further demonstrate that using only seismic motion as input (see...) Figure 7 The model shown in (a) has a much lower predictive performance than the method proposed in this embodiment of the invention that uses both ground motion and top-level response as inputs (see Figure a). Figure 7 (As shown in b). This highlights the importance of incorporating structural response information for accurately assessing damage, especially in imperfect states.
[0030] In summary, the embodiments of the present invention provide a rapid structural seismic damage assessment method based on sparse acceleration and dilated residual convolutional networks. This method effectively solves the problems of high dependence on the number of sensors, limited feature extraction capabilities, and insufficient accuracy in identifying non-intact damage states in existing technologies. It provides a practical solution for achieving low-cost, high-efficiency, and automated post-earthquake building safety assessment in engineering practice.
[0031] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0032] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for rapid assessment of damage to a building structure after an earthquake, characterized by, The method comprises the following steps: S1: obtaining first acceleration data of a bottom layer of a building structure when inputting seismic motion acceleration and second acceleration data of a top layer of the building structure when inputting the seismic motion acceleration, and preprocessing the first acceleration data and the second acceleration data to form two-dimensional input data; S2: constructing and training an expanded residual convolutional neural network model; S3: inputting the two-dimensional input data into the expanded residual convolutional neural network model for processing to output a damage level of the building structure.
2. The method for rapid damage assessment of post-earthquake building structures according to claim 1, wherein, The S2 comprises: S21: constructing the expanded residual convolutional neural network model, wherein the expanded residual convolutional neural network model comprises an input layer, a plurality of expanded residual modules, a global pooling layer, and a full connection layer; S22: obtaining seismic response data from a plurality of building structure types, and sequentially performing data enhancement processing and nonlinear time history analysis on the seismic response data to generate a training set; S23: training the expanded residual convolutional neural network model by using the training set; S24: optimizing the performance of the expanded residual convolutional neural network model during the training process.
3. The method for rapid damage assessment of post-earthquake building structures according to claim 2, wherein, The S21 specifically comprises: The expanded residual module comprises a first convolutional layer, an expanded convolutional layer, a second convolutional layer, and a residual connection layer connected in sequence.
4. The method for rapid damage assessment of post-earthquake building structures according to claim 2, wherein, The S24 specifically comprises: constructing a loss function, and optimizing the training process of the expanded residual convolutional neural network model by using the loss function; performing parameter iterative optimization of the expanded residual convolutional neural network model by using an Adam optimizer.
5. The method for rapid damage assessment of post-earthquake building structures according to claim 1, wherein, In the S3, the output structure damage level is determined according to the maximum inter-story drift angle of the structure, and comprises intact, slight damage, moderate damage, and severe damage.
6. A system for rapid damage assessment of a post-earthquake building structure using the method of any one of claims 1-5, wherein, The method comprises: an acquisition module configured to obtain first acceleration data of a bottom layer of a building structure when inputting seismic motion acceleration and second acceleration data of a top layer of the building structure when inputting the seismic motion acceleration, and to preprocess the first acceleration data and the second acceleration data to form two-dimensional input data; a model construction module configured to construct and train an expanded residual convolutional neural network model; an evaluation module configured to input the two-dimensional input data into the expanded residual convolutional neural network model for processing to output a damage level of the building structure.