Seismic response prediction method for building structures using convolutional neural network

Abstract

Summary In this study, a method of predicting the seismic responses of building structures based on a convolutional neural network (CNN) is proposed. In the method, the time histories of acceleration responses previously measured in a building during earthquakes are used in the CNN input layer, with the corresponding time histories of the displacement responses being used in the CNN output layer. The correlations between the features automatically extracted from the acceleration responses by the convolution and pooling operations in the various CNN layers and the displacement responses in the CNN output layer are established through CNN training. The trained CNN predicts the displacement responses for a future earthquake event using only the measured acceleration responses. To verify the validity of the proposed method, a numerical study on the ASCE benchmark model and an experimental study on a reinforced concrete frame structure are conducted. In the numerical study, the structural responses of the ASCE model subject to artificial earthquakes are utilized for CNN training, and the performance of the trained CNN is verified through seismic response prediction. In the experimental study, shaking table tests are performed for various earthquakes to measure the seismic responses of the reinforced concrete test structure. The seismic response prediction performance of the proposed method is examined using CNN trained with the measured seismic responses. Furthermore, the performances of CNNs trained with different numbers of datasets generated by the proposed data generation method based on data overlapping with the same data pool are discussed with related to the number of CNN training iteration number.

Multi-Objective Optimization Method to Search for the Optimal Convolutional Neural Network Architecture for Long-Term Structural Health Monitoring

“…Therefore, in the fitness evaluation of one individual in the solution group within one generation, that is, the CNN training, the training was set to stop early. In the author's previous studies [13], [14], it was confirmed that in the case of a CNN for prediction using measured structural responses, as in this study, training proceeds with a sharp decrease in the loss function value in the early stages of the training process. Therefore, based on the premature convergence characteristics of the CNN for the structural response estimation, this research method allowed the CNN training of one individual in the solution group to stop in its early stages, which can reduce the huge computational costs that can be incurred in the GA optimization process that operates multiple solutions.…”

“…Convolutional neural networks (CNN), which are deep learning algorithms, are actively being used in various research fields due to their excellent performance. For example, they have been used to build a structural response estimation model using measured structural responses in advance, and predict the structural responses when sensing is impossible or when data loss occurs [13], [14]. CNN was also used to predict the lateral displacement of a structure, which indicates the seismic performance of the structure, by identifying the relationship between the displacement and the acceleration structural response, which CNN can measure relatively easily [13].…”

“…For example, they have been used to build a structural response estimation model using measured structural responses in advance, and predict the structural responses when sensing is impossible or when data loss occurs [13], [14]. CNN was also used to predict the lateral displacement of a structure, which indicates the seismic performance of the structure, by identifying the relationship between the displacement and the acceleration structural response, which CNN can measure relatively easily [13]. In addition, to evaluate the safety of bridges, CNN was used to develop a technique for establishing the relationship between the strain responses measured with multiple fiber optic sensors installed in the bridge [14].…”

“…While optimization techniques have recently been introduced to automatically search for CNNs with optimal performance, studies have focused mainly on image classification, and the optimal CNN model for solving structural engineering problems, including SHM, has not yet been searched for. The studies on data recovery and structural response estimation for long-term SHM [12]- [13][14] yielded remarkable results using CNN, but no specific and rational rules or techniques for building the suggested CNN architecture were provided, and the CNN architecture was presented mainly by trial and error. Therefore, the proposed architecture cannot be said to have exhibited the best performance.…”

Absolute Displacement-Based Formulation for Peak Inter-Story Drift Identification of Shear Structures Using Only One Accelerometer

“…In addition, some researchers combined other responses data (e.g., displacement or strain) with acceleration responses to evaluate the structures [ 26 , 27 , 28 , 29 ]. Moreover, Oh et al [ 30 ] used a convolutional neural network to predict the time histories of relative displacements from the recorded absolute accelerations, but relative error of the maxima (up to 16%) in the numerical simulation was unsatisfactory. Sun et al [ 31 ] used kernel-based machine-learning methods to estimate the seismic response demands, such as peak inter-story drift.…”