Buy article PDF
The purchased file will be sent to you
via email after the payment is completed.
US$ 35
|
Earthquakes and Structures Volume 31, Number 2, August 2026 , pages 253-282 DOI: https://doi.org/10.12989/eas.2026.31.2.06 |
|
|
 open accessPrediction of inter-story drift of steel-bundle tube structures under seismic actions based on machine learning |
||
Xiang Li, Zhenhua Ma, Xiaohui Qin, Weihao Liu, Yong Hao, Jinhao Zhou
|
||
| Abstract | ||
| As a key parameter reflecting both the structural deformation characteristics under seismic loading and the serviceability for daily occupancy comfort, inter-story drift serves as a crucial indicator for evaluating the seismic performance and serviceability of structures. To enable fast and high-accuracy prediction of inter-story drift in steel bundle-tube structures, this study proposes a systematic machine-learning-based framework integrating data-driven prediction, physical interpretability analysis, and performance-oriented inverse design. First, a high-fidelity finite element model is established in Abaqus and nonlinear time-history analyses are conducted under three ground motions and nine PGA levels, yielding 2700 samples from 27 analysis cases. On this basis, eight predictive models, including four conventional and four deep learning algorithms, are developed and evaluated using R2, MAE, MSE, and RMSE. The results indicate that SVR exhibits the best overall performance, with an R2 of 0.966, while deep learning models generally outperform conventional models in prediction accuracy and generalization, with CNN achieving the highest deep learning accuracy, with an R2 of 0.956. Compared with approximately 3h required by Abaqus per analysis, SVR and CNN inference required 60.25 ms and 95.35 ms, respectively. SHAP and Sobol analyses further verify the physical consistency and interpretability of SVR predictions with respect to structural dynamic mechanisms. Under a code-specified inter-story drift limit of 72 mm, a surrogate inverse model for PGA is constructed, yielding critical PGAs of 2077 gal at the 8th story under HWA043 and 1730 gal at the 65th story under TTN024; verification at a PGA of 1730 gal showed that the maximum inter-story drift at the 65th story was close to the limit. The proposed framework provides an efficient and reliable pathway for seismic performance assessment and design optimization of steel bundle-tube structures, and offers an extendable paradigm for performance prediction and performance-based design of other high-rise structural systems. | ||
| Key Words | ||
| deep learning; inter-story drift; machine learning; prediction model; steel-bundle tube | ||
| Address | ||
| Xiang Li, Weihao Liu: School of Information Engineering, Hebei University of Architecture, Zhangjiakou 075000, China Zhenhua Ma, Xiaohui Qin: 1) School of Information Engineering, Hebei University of Architecture, Zhangjiakou 075000, China; 2) Key Laboratory of Smart City Perception and Intelligent Computing of Hebei Province, Zhangjiakou 075000, China Yong Hao, Jinhao Zhou: School of Civil and Engineering, Hebei University of Architecture, Zhangjiakou 075000, China | ||