Output only dynamic identification in time-frequency domain

Classical structural identification procedures, which use transfer functions, require the knowledge or a reliable measure of the forcing actions. This has led to the use of controlled input signals obtained through excitation instrumentation whose production may prove very costly. Some newly developed techniques exploit the natural excitation that any structure is subjected to, and therefore do no require any special instrumentation. Some of them are based on an analysis of the response signal in the time domain and on the determination of a time series approximating its evolution through the use of appropriate statistical models, whilst some others use the concept of a random decrement function to extend the applicability domain of techniques that originally had been designed to analyse free-decay response signals. The techniques previously mentioned typically require the hypothesis of stationarity, though the field of civil engineering excitations is generally non-stationary. This consideration prompted some proposals for new identification methods designed to handle these types of non-stationary excitation, which are based on special amplitude and phase estimators defined in the time-frequency domain. The advantage of these methods is that identification is based solely on the properties of the modal signals, that are dealt with as modulated harmonics. This paper briefly summarises the hypotheses and theories that constitute the basis of the main "output-only" techniques. The problems usually encountered in this kind of identification are stated and critically discussed. The final part of the paper reports a few numerical simulations