On nonstationarity corrections and durations in ground motion applications of random vibration theory

Chris Van Houtte, Tam Larkin, & Caroline Holden

Submitted August 15, 2016, SCEC Contribution #6803, 2016 SCEC Annual Meeting Poster #275

Random vibration theory (RVT) is a method for approximating the peak time domain response of a signal from its Fourier amplitude spectrum, based on assumptions of stationarity over a portion of the signal duration, and random phase angles. In ground motion prediction, RVT has primarily been used in the stochastic method of ground motion simulation, to efficiently generate 5%-damped response spectra from a model of the earthquake Fourier amplitude spectrum. RVT methods are also applied in geotechnical earthquake engineering, for 1D site response analyses. This study examines the assumptions in current applications of RVT, and investigates alternative ways of improving the theoretical basis. Using recorded earthquake data in the New Zealand Strong Motion Database, it is found that correcting for an evolutionary power spectral density function (PSDF) improves the method, however applying an evolutionary bandwidth over-corrects the peak response prediction. The definition of the signal duration is also reexamined. Rather than using a 'Drms/Dgm' correction, as is common in the engineering seismology literature, this study derives a model of the oscillator duration directly. D5-75 is found to be the optimal duration metric for prediction of peak response. It is found that, even with the improved theoretical basis of the proposed method, there are still long-period overpredictions of recorded response spectra using RVT. The likely reasons behind these observations are also presented in this study.

Citation
Van Houtte, C., Larkin, T., & Holden, C. (2016, 08). On nonstationarity corrections and durations in ground motion applications of random vibration theory. Poster Presentation at 2016 SCEC Annual Meeting.


Related Projects & Working Groups
Ground Motion Prediction (GMP)