| description abstract | In this paper, it is aimed to perform clustering analysis on a large set of near-fault ground motions based on the intensity of velocity pulses. For this purpose, the velocity pulses of these ground motions are extracted based on the asymmetric Gaussian chirplet model (AGCM), the adapted dictionary-free orthogonal matching pursuit (ADOMP) algorithm, and the Newton method. The location of first AGCM atom is considered as the location of real velocity pulse, and its combination with adjacent atoms reconstructs the real velocity pulse of near-fault ground motions. In addition, hierarchical clustering and k-means clustering are used to cluster near-fault ground motions into two, three, and five groups based on the features of their velocity pulses. The results showed that the clustering results for two and three groups are more suitable than the results of five groups according to the silhouette and Davies-Bouldin measures. Also, the clustering results of five groups are shown to be appropriate for the design of structures with different importance factors. According to the results, agglomerative clustering with complete linkage provides more appropriate clustering results for moderate and strong pulselike ground motions. The k-means and agglomerative clustering with complete linkage provide more uniform clustering results for three and five groups, respectively. A sensitivity analysis showed that the k-means provides more stable clustering results when the records for the Chi-Chi earthquake are removed from the data. Also, the clustering results were used to extract the possible relation of the intensity of velocity pulses with the seismological parameters, which were in accordance with the results of previous research. | |
