데이터 기반 모델축소법을 이용한 효율적인 비선형 구조해석 개발

In this paper, a data-driven model order reduction framework is proposed for efficient nonlinear structural analysis. The data-driven model order reduction framework consists of two stages: data mining/analysis with reduced-order modeling (offline) and parametric simulation (online). Herein, the reduced-order modeling is performed using proper orthogonal decomposition and an autoencoder in the offline stage. Furthermore, a variational autoencoder is considered as an artificial neural network-based model order reduction to improve the efficiency within the offline stage. The proposed approaches are compared to the full-order model by analyzing nonlinear numerical examples to demonstrate their efficiency and accuracy.