Probabilistic incremental dynamic analysis of structures using temporal surrogate model

• Published in: Applied intelligence (Dordrecht, Netherlands) Netherlands), 2023-06, Vol.53 (12), p.15011-15026
• Main Authors: Nguyen, Truong-Thang, Dang, Viet-Hung
• Format: Article
• Language: English
• Subjects: Artificial Intelligence; Computer architecture; Computer Science; Computing costs; Data integration; Earthquakes; Finite element method; Fragility; Machines; Manufacturing; Mathematical models; Mechanical Engineering; Neural networks; Probabilistic analysis; Processes; Seismic response
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-022-04264-y

This study develops a highly efficient framework, termed iDANS, for Incremental Dynamic Analysis (IDA) of civil structures subjected to earthquakes using a physical-induced data-driven surrogate model. IDA is a reliable tool for assessing the structure’s seismic performance; however, it requires extensive calculations to model the structures’ behavior from linear to non-linear ranges, leading to a high computational cost. Therefore, establishing a surrogate model that is able to provide the structure’s responses rapidly, is helpful. To this end, one leverages multiple advanced techniques from data analytic, artificial intelligence, and structural dynamic such as rolling window strategy, data fusion, a temporal neural network architecture, and a length- and magnitude-agnostic loss function. The surrogate model is trained on a dataset generated by finite element models carefully calibrated with experimentally measured data in advance. The proposed approach’s accuracy and efficiency are quantitatively demonstrated through a case study of a six-story steel building. The computed results show that iDANS can reduce the computational complexity by three orders compared to conventional IDA methods. Furthermore, iDANS is employed to perform probabilistic analysis for assessing the impact of input uncertainty on the structure’s fragility curves.