Quantitative classification of velocity pulse-like ground motions based on an adaptive response spectrum-based decomposition method

Abstract

Pulse-like ground motions generally contain low-frequency velocity pulse with large amplitude, which can lead to severe damage or even collapse of near-fault structures. This study aims to proposed an adaptive decomposition method and corresponding identifying approach for pulse-like records. Firstly, an adaptive response spectrum-based decomposition (ARSD) method is proposed to decompose the records into several intrinsic mode functions (IMF) and their periods are identified by peak-point method. The decomposed results are compared with two representative methods, ensemble empirical-mode decomposition (EEMD) and variation mode decomposition (VMD). The results show that the proposed method can effectively avoid the mode mixing appears in the decomposed results by EEMD and generates similar results with those by the VMD. Then the low-frequency pulse is extracted by composing the key IMFs with the period and energy change larger than 0.5 s and 0.1. Based on the results of visual inspection method, the threshold of energy-based pulse indicator factor in general form and exponential form is calibrated to 0.26 and 0.30 respectively. The results show that the exponential form can improve identification rate of pulse-like records. The pulse extraction and classification results of the proposed method are comprehensively compared with three benchmark methods. The classification results by the proposed method are generally consistent with the three methods. Moreover, all the four methods can accurately detect the non-pulse records. And the proposed method could detect the most pulse-like records, followed by the Zhai's method, while the Feng's and Baker's methods detected the least pulse type records. The corresponding identification rates are 91.04%, 80.60%, 63.43%, and 63.06%, respectively. Finally, the periods of the records with single mode and multiple modes estimated by the proposed method and three benchmark methods are compared. The estimated periods of the records with single mode by the proposed method are highly consistent with those by the three benchmark methods, while this consistency for the estimated periods of records with multiple modes is reduced at some extent. This is due to the interference of multiple modes embedded in the records to the pulse identification.