An Efficient Algorithm to Identify Strong-Velocity Pulses in Multicomponent Ground Motions

Abstract

Abstract Ground motions with strong velocity pulses are of special concern for structural engineers. We describe an efficient and quantitative method to identify such ground motions. Previous algorithms to classify these pulse-like ground motions con-sidered the ground motion in a single orientation, which made classifying pulses in arbitrary orientations difficult. We propose an algorithm that can identify pulses at arbitrary orientations in multicomponent ground motions, with little extra computa-tional cost relative to a single-orientation calculation. We use continuous wavelet transforms of two orthogonal components of the ground motion to identify the ori-entations most likely to contain a pulse. The wavelet transform results are then used to extract pulses from the selected orientations, and a new classification criterion based on support vector machines is proposed. Because we are mostly interested in forward directivity pulses, which are found early in the time history, a criterion to reject pulses arriving late in the time history is also proposed. The procedure was used to classify ground motions in the Next Generation Attenuation-West2 database (Ancheta et al., 2013). The list of pulse-like ground motions was then manually filtered using source-to-site geometry and site conditions to find the pulses most likely caused by directivity effects. Lists of both pulse-like ground motions and directivity ground motions are provided, along with the periods of the pulses and the orientations in which the pulses were strongest. Using the classification results, new models to predict the probability of a pulse and pulse period for a given future earthquake scenario are developed. Online Material: Table of all the pulse-like ground motions identified in the Next Generation Attenuation-West2 database, and HTML files of source-to-site geometry, the recorded time history, extracted pulse, and other directivity parameters for each identified pulse.