Deep-learning-augmented physics models to predict nonlinear dynamic responses of multi-degree-of-freedom structures

Abstract

Predicting a structure's dynamic response is essential in system identification, structural health monitoring, and structural reliability assessment. In recent years, many deep-learning-based methods have been developed to predict the dynamic responses of structures based on measurement data while guided by physics-based knowledge. However, such an approach has not yet been applied to predict the nonlinear dynamic responses of a large degree-of-freedom (DOF) structure. In this paper, the neural-network-augmented physics (NNAP) model is further developed to incorporate information on multi-degree-of-freedom structural systems by transforming the original DOF into a low-dimensional system using modal truncation. The prediction performance of the proposed method is verified through the numerical example of the Lysefjord bridge structure subjected to response-dependent wind loads. The proposed method is expected to promote further developments of physics-based deep learning approaches for complex structures with large DOFs.