Abstract: Many engineering structures, such as cranes, traffic-excited
bridges, flexible mechanisms and robotic devices exhibit characteristics that
vary with time and are referred to as time-varying or nonstationary. In
particular, linear time-varying (LTV) systems have been often dealt with on a
case-by-case basis. Many concepts and analytic methods of linear time-invariant
(LTI) systems cannot be applied to LTV systems, as for example the conventional
definition of modal parameters. In fact, LTV systems violate one of the
assumptions of the conventional modal analysis, which is stationarity. Subspace-based identification methods, proposed in the 1970s, have been
attracting much attention due to their affinity to the modern control theory,
which is based on the state space model. These methods are now successfully
applied to many industrial cases and may be considered reference methods for
identifying LTI systems. In this paper the use of a subspace-based method for
identifying LTV systems is discussed and applied to both numerical and
experimental systems. More precisely a modified version of the SSI method,
referred to here as ST-SSI (Short Time Stochastic Subspace Identification) is
introduced as well as a method for predicting time-varying stochastic systems
using the angle variation between the subspaces; the latter is able to predict
the system parameter in the "near" future.
