Uncertainty quantification of modal characteristics identified from frequency-domain stochastic subspace identification

Stochastic subspace identification has become an industrial standard for operational modal analysis because of its computational efficiency and statistical optimality. For the time-domain version of the algorithm, a computationally efficient method exists for the estimation of (co)variances of the identified system matrices and the related modal characteristics. In the present paper, a computationally efficient uncertainty quantification method is developed for a frequency-domain subspace algorithm that starts from nonparametric positive power spectral density estimates. A connection with the time-domain method is made, and the performance is verified against Monte Carlo simulations in a numerical experiment.