Evaluation of damping estimates in the presence of closely spaced modes using operational modal analysis techniques

The Operational Modal Analysis (OMA) techniques provide in most cases reasonably accurate estimates of structural frequencies and mode shapes. They are however known to produce erroneous structural damping estimates, which are presumably thought to be due to inherent random- or bias errors that have
varying significance for different techniques. This paper evaluates the sensitivity of damping estimates of closely spaced modes for two existing OMA techniques derived in the time and frequency domain; namely Eigensystem Realization Algorithm (ERA) and Frequency Domain Decomposition (FDD). The evaluation is based on identification using random response from white noise loading of a three degree-of-freedom (3DOF) system numerically established from specified modal parameters for a range of natural frequencies. The numerical model provides comparisons of the effectiveness of damping estimation for a variety of damping levels, signal noise and the sensitivity to closely spaced modes. It is shown that FDD has a tendency to overestimate damping due to leakage in the estimated spectral density function and it is a more sensitive technique to system changes than the ERA. The accuracy of damping estimates converges with increased frequency of the system, which is mainly a result of the problematic regions in the correlation function estimation. These regions cause amplification of the damping estimation errors at higher levels of damping. This emphasizes the importance of correctly estimating the correlation function and spectral density as bias will potentially result in large errors in the estimation of highly damped systems. It is concluded that damping estimated are sensitive to closely spaced modes. In addition, it is found that two closely spaced modes will also disturb the estimation of damping of the remaining modes in the system.