We show in this paper how a recent class of time domain methods known as subspace identification methods, overcomes the limitations of classical time-domain techniques. We shall show that these methods deliver accurate models with a model order close to the number of physically excited modes. This improvement is mainly attributed to the ability of these subspace methods to asymptotically cancel both the process and measurement noise effects prior to the identification step. These advantages will be pointed out by applying two different sub-space algorithms, namely MOESP and N4SID, to the modal identification of a flexible subframe structure. The results given by these two algorithms are compared to those given by ERA/OM.
展开▼