City Research Online

Abstract

In this paper the performance of a compressive sensing (CS)-based vis-a-vis a power spectrum blind sampling (PSBS)-based spectral estimation approach is numerically assessed in undertaking operational modal analysis (OMA) using the frequency domain decomposition algorithm. The examined approaches consider response acceleration measurements sampled non-uniformly in time at sub-Nyquist average rates at random time instants (the CS-based), and at deterministically defined time instants through a multi-coset sampling strategy (the PSBS-based), aiming to reduce power consumption in arrays of wireless sensors used in OMA. The modal assurance criterion is adopted to gauge the effectiveness of the two approaches using acceleration time-histories with and without additive Gaussian white noise taken from 15 equidistant recording locations on a white-noise excited linear finite element model of a simply supported beam. It is shown that for a given sub-Nyquist sampling rate the capability of the CS-based approach to extract quality estimates of mode shape depends heavily on the sparsity of the acceleration signals in the frequency domain, which is low for the noisy signals, in relation to the target sparsity level that needs to be assumed in the CS signal reconstruction step. However, the PSBS-based approach, pioneered by the authors, performs equally well and consistently better than the CS-based approach in extracting mode shapes even for noisy signals (at SNR=10db) and for a sampling rate as low as 11% the Nyquist rate. This is because the latter approach is signal agnostic and does not necessitate any target sparsity assumption. Overall, the herein reported numerical results demonstrate that the PSBS-based approach is rather advantageous in practical applications where achieving high signal compression levels is desirable irrespective of the additive noise level.

Publication Type:	Conference or Workshop Item (Paper)
Departments:	School of Science & Technology > Engineering

Export

Downloads