.png){kind=icon}
.png){kind=icon}
Return
Data-driven method for seismic response prediction of rocking rigid bodies in buildings using deep neural networks
.png){kind=icon}
PRE
AI
Q
Z
L
Y
DOI：10.1007/s11803-026-2407-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Furniture and equipment modeled as rocking rigid bodies within buildings are vulnerable to overturning during earthquakes, and their responses are strongly influenced by floor-level seismic amplification. However, predicting their seismic behavior typically involves analyses of building structural responses and rocking-body dynamics, making conventional approaches computationally expensive. Existing methods also struggle to adequately capture the nonlinear interactions among seismic motion characteristics, structural amplification effects, and the geometric properties of rocking bodies. To address this issue, this study proposes a data-driven method for predicting the seismic response of in-building rocking rigid bodies using a deep neural network (DNN). Floor seismic responses were obtained from city-scale nonlinear time-history analyses and used to calculate corresponding rocking-body responses. A multidimensional database was then established, covering various ground-motion intensity measures, building heights, and rigid-body geometries. Based on this database, a DNN model was developed for rapid overturning prediction. Results show that the proposed model achieves high computational efficiency and an accuracy of 94.37% on the test set, outperforming conventional machine learning methods. Dimensionality reduction further decreases input features and training time while preserving strong predictive performance. The proposed approach provides an efficient and intelligent framework for seismic assessment and risk analysis of rocking components in buildings.
Keywords:
rocking rigid body
seismic response prediction
deep learning
building response
Journal
IF:
3.3
Papers:
115
Citations:
3.0K
Organization
Cited Papers