Experimental spectral submanifold reduced order models from machine learning

Abstract

Nonlinear system identification is a challenging problem in experimental modal analysis. It is currently tackled using a toolbox approach, where different techniques are employed depending on the structural system under investigation, the identification goals and the type of excitation used. In this contribution, we exploit analytic reduction to spectral submanifolds combined with machine learning techniques in order to obtain the nonlinear coefficients up to cubic order of a single-degree-of-freedom reduced order model. The system measurements aimed at model fitting can be performed using any type of excitation techniques, ranging from free-decay to sine-sweeps or random shaker testing. We illustrate the accuracy of our method using both simulated and real experimental data.