Comparison study on identification methods applied to linear time-varying structures

A detailed comparison study on the global method and the recursive identification methods for identifying linear time-varying dynamic models of mechanical structures with modal experimental data is presented. The global method is performed using arbitrary input and output data, while the recursive methods are implemented with the ordinary MIMO Output-Error State space model identification (MOESP) algorithm or the subspace-based state space system identification (4SID) algorithm. The comparison is performed by simulations using a robotic manipulator with varying inertia links. The general conclusion is that the global method can estimate the pseudo-modal parameters more accurately than the recursive methods, but it is sensitive to noise and inconvenient to be used on line. The recursive methods perform well if parameters are appropriately selected (such as forgetting factor) and it can be used on line conveniently.