Noise-bias compensation in physical-parameter system identification under microtremor input

Abstract

A direct method of physical-parameter system identification (SI) is developed in the case of containing noises at both floors above and below a specified story. To investigate the effect of the level of noise on the accuracy of identification, numerical simulations are performed in the frequency domain by generating two stationary random processes with the specified levels of power spectra. When the previous method of physical-parameter SI is applied to the case contaminated by noise at both floors, just above and below a specified story, it is difficult to evaluate the true stiffness and damping coefficients depending on the intensity. To overcome this difficulty, a new noise-bias compensation method is proposed which enables one to evaluate the intensity of noise in addition to the identification of story stiffness and damping. This method can be used in the case of small SN (Signal to Noise) ratios such as actual microtremor records. The effectiveness of the proposed method is demonstrated through a numerical example and an experiment using actual microtremor records for a reduced two-story model.