Identification of strong-velocity pulses in seismic ground motion signals

Seismic ground motions in the near-fault region produce strong pulses in the velocity-time history, resulting in severe damage to structures. To accurately and effectively monitor these ground motion signals with strong pulses, Shock-Waveform (SW) method is introduced to quantitatively extract the largest velocity pulse from a given ground motion. SW method is an energy-based and adaptive signal analysis method, which has proven capability of analyzing different physical and engineering signals initiated by sudden actions. It is suitable to identify pulse components in the signal with low error and high efficiency. Three variables are proposed to classify ground motions, which is combined with the Principal Component Analysis (PCA) for data dimensionality reduction and subsequent analysis. In addition, an optimum classification standard on pulse-like and non-pulse-like ground motion is established. To avoid the subjective judgement induced by manual selection, unsupervised machine learning classification method and Support Vector Machine (SVM) are used successively to find the decision boundary. In this study, about 100 pulse-like ground motions with large-velocity pulses are identified from approximately 1000 near-fault ground motion recorded in PEER Next Generation Attenuation-West2 database. It shows that most of the pulse-like ground motions are caused by the directivity effect. Based on the proposed classification approach, new models are developed to forecast the possibility of a single pulse, multi-pulses, and pulse period for a given earthquake event.