System Identification Via Auto-Encoders: A Comparison Between Data-Driven and Physics-Informed Solutions

Abstract

Variational Auto-Encoders (VAE) and Long Short-Term Memory (LSTM) are investigated in this article in the framework of Structural Health Monitoring (SHM). The presented approaches aims to combine sensor data with numerical modelling of the dynamical system in a Reduced Order Modelling setting. Two Finite Element case-of-study are proposed with the scope of identify non-linear forces and physical parameter degradation. Starting from a reduced set of sensors data and different levels of knowledge of the physical system, the reconstruction capabilities of time series data are presented and compared for both proposed architectures.