Data-driven techniques for modeling, control, and
sensor placement
Eurika Kaiser, University of Washington
ABSTRACT:
Nonlinear actuation mechanisms play a critical role
for developing effective closed-loop control strategies. The
probabilistic perspective of dynamical systems is particularly
appealing as it yields a linear model taking into account
nonlinear actuation mechanisms, and thus allows one to make use of
the rich linear control theory to guarantee robust performance
despite model uncertainty and external disturbances.The talk will
describe a data-driven reduced-order modeling approach in which
the high-dimensional flow state is coarse-grained using cluster
analysis, its extension for control, and sensor placement to
enable future real-time applications. Results are shown for
various fluid dynamical systems.