Influence of non-uniform noise levels on modal identification procedures

Abstract

Considerable deficiencies and errors in measurement systems are frequently encountered in vibration based structural health monitoring (SHM) applications. Some SHM algorithms are capable of considering such kinds of problems as prediction errors and/or sensor channel noise. As a general intention, however, either a uniform channel noise distribution is assumed or the corresponding measurement channel which produces significant noise in the measured data is generally omitted by researchers. From this perspective, this paper presents a comparative study to investigate the performance of two different SHM algorithms in case of non-uniform sensor channel noise spectral densities. In this context, first the considered problem is illustrated based on the disruptions in the spectral coherence between the noisy and noise free data. Then, a numerical example is presented in which the modal identification of a three degree-of-freedom (DoF) system is performed by using Bayesian Fast Fourier Transform Approach (BFFTA) and Covariance-based Stochastic Subspace Identification (SSI–COV). Results show that both techniques can be adversely affected by the non-uniform levels of channel noise. However, SSI–COV performs better in this case.