Identification of Near-Fault Impulsive Signals and Their Initiation and Termination Positions with Convolutional Neural Networks

Ground motions recorded in near-fault regions may contain pulse-like traces the velocity domain. Their long periodicity can identify such signals with large amplitudes. Impulsive be hazardous for buildings, creating demands due to their periods. In this study, a dataset was collected from various data centres. Initially, all impulsive signals, which are reality rare, manually identified. Furthermore, then, synthetic waveforms created increase number of by using model developed Mavroeidis and Papageorgiou, k−2 kinematic modelling. accordance, convolutional neural network (CNN) trained detect these ordinary signals. labelled used initiation termination positions To do so, waveform position amplitude information maximum minimum points used. Once detects positions, period pulse is calculated analysing spectral Although our detection algorithm works relatively worse than three robust algorithms benchmarks, it significantly better determination positions. At moment, models understand features location without any thresholds or formulations that heavily previous studies.