Using the Random Decrement Technique on Short Records with Varying Signal-to-Noise Ratios

Abstract

An earlier paper by the authors discusses potential issues in the practical implementation of the Random Decrement Technique (RDT), particularly when the data has a low signal-to-noise ratio (SNR), and when filtering, which distorts the data and affects the results, is necessary. This is the case for practically every measurement of full-scale buildings, typically done at very low ambient response amplitudes. The RDT procedure used also enables a form of error quantification which allows the engineer or data analyst to judge how reliable the result is. In one of the examples in the earlier paper, it was shown that for a certain SNR, an improvement in the damping estimate is evident when the recording length is significantly longer. For the studied structure, this meant that for better results the recording had to be 72 h (or about 46,000*T where T is the building period of interest) instead of the 4 h initially considered. Even then it required a much longer record duration, estimated to be equivalent to a month of recording, to arrive at an acceptably accurate result. However, such a very long record length is impractical. The current study thus firstly aims to determine what is the minimum SNR (maximum noise level) given say just a 1-h recording, to achieve a certain level of accuracy. This is done by varying the noise level and therefore the SNRs, and applying the mentioned RDT procedure and comparing SNR against the calculated error. The SNRs of ambient response measurements of a full-scale building with and without a supplementary damping system are then viewed considering the finding on minimum SNR.