Near-fault pulse extraction based on Adaptive Chirp Mode Pursuit algorithm and Asymmetric Gaussian Chirplet Model

Abstract

Previously, time-domain algorithms were proposed to extract long-period waveforms, which do not consider frequency content of near-fault ground motions effectively. In this study, we developed an algorithm extracting high-energy long-period locations of near-fault ground motions in the time-frequency domain combining the Adaptive Chirp Mode Pursuit (ACMP) and Asymmetric Gaussian Chirplet Model (AGCM). The ACMP is an adaptive time-frequency filter, decomposes ground motions into long-period and short-period components, extracts nonlinear components one by one, and updates the bandwidth of each component to reach to the actual value. The AGCM is used to extract a velocity pulse from the long-period component. Because of the high flexibility of AGCM’s chirplet atoms, the AGCM can simulate asymmetric, irregular, and multi-cycle waveforms with one atom and time-varying frequency content of waveforms. The results of the proposed algorithm showed superior performance in the detection and extraction of pulse-like waveforms, compared to the previous methods. By analyzing 1398 ground motions, we propose a linear equation for the prediction of pulse-period as a function of moment magnitude for soil and rock sites, and for all pulse-like data. The results showed that the proposed prediction equations have greater slop than the previous ones. Moreover, it was shown that for a greater moment magnitude, the energy ratio and pulse-period are distributed in a wider range, the skewness of the distribution becomes less, near-fault records have a longer pulse-period, and the velocity pulse contributes more to the total energy of the record. Investigation of the Fourier spectrum of ground motions showed that the period corresponding to the peak amplitude can be considered as the initial estimate of a pulse-period, especially for intense velocity pulses.