Including noise measurements for improved experimental frequency response function estimation

When estimating the frequency response function using experimental modal analysis, the classical approach is to use either H₁-, H₂-, or other estimators. These estimators have different assumptions regarding noise level and distribution in input and output. To choose the appropriate frequency response function estimator, some information about the noise in the experimental setup must be known prior to the experiments. In this work, a different approach is presented to generalize the frequency response function estimation to obtain the ‘best’ estimation, independent of the experimental setup in use. By default, the assumption is that both input and output measurements contain normally distributed noise. In this approach the input and output noise need to be measured prior to the experiments, to obtain better frequency response function estimates in terms of bias and uncertainty. The approach is illustrated using a Monte-Carlo simulation and a simple finite element model and compared with the classical estimators H₁ and H₂