Identification of linear models of structural systems using a limited number of input/output time histories

This work is an attempt to develop a robust and practical methodology to identify structural systems with a limited number of sensors and actuators (incomplete set of measurements), particularly for applications to civil engineering structures. A full set of instrumentation is required in typical identification approaches to obtain a physical model of the structure (mass, damping and stiffness matrices). However this condition cannot be often satisfied in real life applications. The method proposed in this thesis provides a solution to this significant obstacle by developing, first, "reduced-order" models of a structural system and then expanding such models to "full-order" ones that are quite useful in damage detection. A sensitivity analysis and an optimization procedure are adopted to identify reliable models. Furthermore, topics including the dependency of the second order system matrices on the unmeasured information, the uniqueness of solution and noise effects are analyzed. Results are supported by means of numerical examples.