Variance and Bias Computation for Improved Modal Identification using ERA/DC

A series of variance and bias confidence criteria were recently developed for the Eigensystem Realization Algorithm (ERA) identification technique. These criteria are extended here for the modified version of ERA based on data correlation, ERA/DC, and also for the Q-Markov Cover algorithm. The importance and usefulness of the variance and bias information is demonstrated in numerical studies. The criteria are shown to be very effective not only by indicating the accuracy of the identification results, especially in terms of confidence Intervals, but also by helping the ERA user to obtain better results by seeing the effect of changing the sample time, adjusting the Hankel matrix dimension, choosing how many singular values to retain, deciding the model order, etc.