Experimental study of virtual closed-loop modeling for parameter estimation

A commonly adopted approach within structural model updating is to estimate a set of model parameters by solving an inverse problem, in which some norm of the discrepancy between target features from the physical system and the corresponding predictions from the model is minimized. In this regard, the inverse problem will often be ill-posed, because the target features are insufficient for providing a unique realization in the model space manifold. One way of resolving this issue is to test the physical system in closed loop with different gains that each yield an eigenstructure from which target features can be extracted. The practical overhead associated with this procedure---which thus far has hindered applicability---can, in principle, be avoided through a virtual implementation where the closed-loop eigenstructures are realized through signal processing of the open-loop transfer functions. The virtual implementation also eliminates the closed-loop stability constraints and allows computation of several closed-loop eigenstructures based on a single open-loop realization. While numerical studies have suggested feasibility of the virtual implementation for closed-loop model updating, no experimental verification has yet been conducted. Therefore, the focus in the present paper is to achieve this through laboratory tests with a steel frame structure, which is instrumented with a feedback system to allow for comparable physical closed-loop testing results.