Convergence of DDL forced-response predictions outside training range in sloshing experiment

Determine whether data-driven linearization (DDL) can produce a converged forced-response prediction for the liquid-sloshing tank when the forcing amplitude drives the response significantly outside the range of the unforced training data used to fit DDL.

Background

The authors train DDL on unforced sloshing decay data (single observable from video) to obtain a reduced linearized model on a two-dimensional spectral submanifold. They then use this unforced-data-trained DDL model to predict forced-response curves for several forcing amplitudes.

For small and moderate forcing amplitudes, DDL accurately captures the softening trend and the response branches. However, for the largest forcing amplitude, the response lies outside the domain spanned by the training data, and the authors report that a converged forced-response solution from DDL could not be obtained.