Identifiability-Enhanced Bayesian Frequency-Domain Substructure Identification

Identification of structural parameters for large structures is challenging due to the large number of unknowns and its possibly ill-conditioned nature. Substructure identification provides a feasible solution to partition a structure into substructures to tackle these difficulties and it has received considerable attention in the recent decades. Although there are variants of substructure identification methods, they share the key to back-calculate the boundary force of the substructure using response measurements. It will then be treated as part of the input to the substructure. However, one major drawback is the deterioration of the identifiability of the inverse problem due to the reduction of effective constraints and the complicated interdependence relationship between the model parameters and the boundary force. In this article, an improved Bayesian substructure identification approach is proposed without the requirement of input or boundary force measurements. The crux is to model the boundary force as filtered white noise because it is an internal response of the system. This provides extra constraints to enhance the identifiability of the inverse problem. Furthermore, the proposed approach provides not only the most probable values of the identified parameters but also their associated uncertainties. This is an important indicator of identifiability, especially for this case. Examples of a 100-story building are presented to demonstrate the proposed method. Comparison is also presented to confirm that the proposed method resolves the ill-conditioned problem encountered in absolute acceleration measurements using existing methods.