Develop surrogate models and active learning for nonlinear dynamical reliability

Develop machine-learning surrogate models and suitable active-learning functions for large computational models of nonlinear dynamical systems used in time-variant structural reliability estimation.

Background

The review observes that surrogate modelling and active learning have been developed primarily for static systems and time-invariant reliability. Nonlinear dynamical systems introduce additional difficulties because each evaluation may require long time integrations, responses may be nonstationary or discontinuous, and random excitations and system parameters may jointly affect the trajectory.

The authors identify the extension of surrogate-based reliability modelling to these systems as a future research direction, while specifically noting that appropriate learning functions for active learning are not yet evident.