Seismic fragility analysis using mNARX modelling
OPEN ACCESS
Loading...
Author / Producer
Date
2024-05-29
Publication Type
Other Conference Item
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
Assessing the seismic vulnerability of civil structures is crucial for safeguarding human lives and ensuring the long-term functionality of essential infrastructure. Nevertheless, quantifying the capability of a structure to withstand a potentially large spectrum of seismic events still poses a significant challenge, due to the high level of uncertainty involved. In the current state-of-the art, the uncertainty in the occurrence and magnitude of seismic events is modelled through a statistical ground-motion model (SGMM), which is then propagated through detailed computational models of the structure under investigation. Still, conducting Monte Carlo simulation using an SGMM model is often unfeasible on complex structures due to the high computational costs associated, e.g. due to high-resolution finite-element modelling (FEM). To tackle this problem, surrogate models have emerged as computationally efficient proxies for FEM simulations. These models are trained on a relatively small dataset, typically a few hundred to a thousand FEM simulations, and focus on mapping SGMM parameters directly to scalar building performance metrics, such as maximum interstory drift or other damage measures. Traditional surrogate models may however struggle to capture the stochastic nature of SGMMs, which exhibit significant latent variability. In other words, to each set of SGMM parameters corresponds an infinite number of ground motions. Additionally, these surrogates often provide only selected scalar properties of the time-dependent structural responses, rather than their complete time history. To address these limitations, we propose to take advantage of the recently developed mNARX surrogate modelling strategy [1] to approximate the full history of the system response. mNARX offers two key advantages over traditional surrogates. First, it acts as an emulator for the full FEM, providing full-time history predictions, hence offering a deeper insight into the structural behavior. Second, it allows for incorporating prior knowledge of the physical system, through the construction of an exogenous input manifold, which results in exceptional data efficiency, significantly decreasing the training data needed with respect to traditional surrogates. To illustrate the effectiveness of mNARX in seismic fragility analysis, we present a case study involving a three-story steel frame simulated using the open-source software OpenSees. The structure is exposed to real earthquake data from the PEER ground motion database. Our results show that the mNARX surrogate accurately emulates the quantities like the interstory drift, even when trained on very small datasets. [1] Schär, S. et al. “Emulating the dynamics of complex systems using autoregressive models on manifolds (mNARX)”, Mech. Syst. Signal Proces., 2023
Permanent link
Publication status
published
External links
Editor
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich, Chair of Risk, Safety and Uncertainty Quantification
Event
Engineering Mechanics Institute Conference and Probabilistic Mechanics & Reliability Conference (EMI/PMC 2024)
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Organisational unit
03962 - Sudret, Bruno / Sudret, Bruno
Notes
Funding
101006689 - HIghly advanced Probabilistic design and Enhanced Reliability methods for high-value, cost-efficient offshore WIND (EC)