A generalized autoregressive model for time-variant dynamical systems

Sat, 01 Aug 2026 00:00:00 +0000

Abstract

Surrogate models are a staple tool for the reliability analysis of complex engineering systems, because they enable the use of otherwise computationally prohibitive algorithms, such as Monte-Carlo simulation. Their prevalence in the context of nonlinear dynamic engineering systems, however, is significantly lower, because approximating their time-dependent response is generally considered more challenging than for their static counterparts. A powerful class of surrogate models is provided by autoregressive with exogenous input models, but their use is mostly limited to system identification tasks, as they tend to underperform in pure prediction tasks. Recent advances in data-driven autoregressive modeling, have vastly improved their prediction capabilities for engineering applications, but they still rely on the assumption of time-invariant dynamics at the core of NARX. In this contribution we demonstrate how the mNARX+ framework can efficiently handle time-variant dynamical systems, such as, e.g., civil or mechanical structures subject to plastic damage or fatigue accumulation, making it an ideal tool for their reliability assessment.

Tip

Click the Cite button above to copy .bib publication metadata into your notes. or directly download here either the bibtex version: Download BibTeX or the plain text one: Download Plain.txt

Time-Variant Dynamical Systems |

A generalized autoregressive model for time-variant dynamical systems