Adaptive importance sampling based neural network framework for reliability and sensitivity prediction for variable stiffness composite laminates with hybrid uncertainties
Abstract:In this work, we propose to leverage the advantages of both the Artificial Neural Network (ANN) based Second Order Reliability Method (SORM) and Importance sampling to yield an Adaptive Importance Sampling based ANN, with specific application towards failure probability and sensitivity estimates of Variable Stiffness Composite Laminate (VSCL) plates, in the presence of multiple independent geometric and material uncertainties. The performance function for the case studies is defined based on the fundamental fr… Show more
“…Xiao, Oladyshkin, and Nowak (2020) investigated an adaptive Kriging approach along with a stratified importance sampling. Mathew, Prajith, Ruiz, Atroshchenko, and Natarajan (2020) developed a framework coupling neural network with adaptive importance sampling, and so forth. Inspired by this research line, this study investigates a highly efficient approach, dubbed SS-DL, that fuses the subset simulation with a DL-based surrogate model for tackling the seismic reliability problem which is still scarce in the literature to the best of the authors’ knowledge.…”
The reliability of building structures subjected to ground motion is a time-consuming, complex, and iterative computation problem involving multidiscipline theories such as probability theory, reliability theory, and dynamic structural analysis. In order to address this challenge, this study proposes a highly efficient approach based on the subset simulation and a deep learning-based surrogate model. The subset simulation is an efficient sampling strategy that significantly diminishes the number of samples to compute when determining the failure probability, especially for very small ones. On the other hand, the surrogate model based on a deep learning algorithm can deliver equivalently accurate structures’ responses compared to the well-known finite element method with markedly smaller time complexity, given appropriate available training data. The efficiency and effectiveness of the proposed approach are demonstrated in detail through three case studies with increasing complexity: a 1-DoF problem, a 2D frame, and a 3D structure based on experimental data showing a reduction up to two orders of magnitude in time complexity compared to the Monte Carlo simulation using finite element method.
“…Xiao, Oladyshkin, and Nowak (2020) investigated an adaptive Kriging approach along with a stratified importance sampling. Mathew, Prajith, Ruiz, Atroshchenko, and Natarajan (2020) developed a framework coupling neural network with adaptive importance sampling, and so forth. Inspired by this research line, this study investigates a highly efficient approach, dubbed SS-DL, that fuses the subset simulation with a DL-based surrogate model for tackling the seismic reliability problem which is still scarce in the literature to the best of the authors’ knowledge.…”
The reliability of building structures subjected to ground motion is a time-consuming, complex, and iterative computation problem involving multidiscipline theories such as probability theory, reliability theory, and dynamic structural analysis. In order to address this challenge, this study proposes a highly efficient approach based on the subset simulation and a deep learning-based surrogate model. The subset simulation is an efficient sampling strategy that significantly diminishes the number of samples to compute when determining the failure probability, especially for very small ones. On the other hand, the surrogate model based on a deep learning algorithm can deliver equivalently accurate structures’ responses compared to the well-known finite element method with markedly smaller time complexity, given appropriate available training data. The efficiency and effectiveness of the proposed approach are demonstrated in detail through three case studies with increasing complexity: a 1-DoF problem, a 2D frame, and a 3D structure based on experimental data showing a reduction up to two orders of magnitude in time complexity compared to the Monte Carlo simulation using finite element method.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.