GNGTS 2018 - 37° Convegno Nazionale

774 GNGTS 2018 S essione 3.3 AN ALTERNATIVE PULSE CLASSIFICATION ALGORITHM BASED ON MULTIPLE WAVELET ANALYSIS D. Ertuncay, G. Costa Dipartimento di Matematica e Geoscienze, Università degli Studi di Trieste, Trieste, Italy Vast number of seismic stations near seismically active regions are installed in last decades. These stations allowed us to monitor near-fault behaviors of earthquakes. Occasionally, stations recorded earthquake signals with unexpected patterns, which will be called pulse shape signals henceforward. Such signals are located in the beginning of the earthquake signal. These signals are researched in both classical and engineering seismology. Pulse-shape signals are important to analyze since they can create high demands on structures around the period of the pulse (Kalkan E. and Kunnath, S. K., 2006). However, because of their scarcity, velocity pulses are not taken into account in most of the ground motion prediction equations (GMPE) (Abrahamson, N. et al., 2016). The goal of this study is to create an alternative pulse identification algorithm. Main considerations are determining the time location of the pulse and mimicking impulsive part of the signal with known wavelets. Ricker and Morlet wavelets are used both for spectrum analysis to determine the pulse period and the region with maximum spectral energy and mimicking impulsive part of the signal. Waveform that are identified as pulse shaped are compared with the wavelets which are created with calculated pulse period by checking spectral responses. If the wavelets correspond the features of the long period part of the earthquake signal, the algorithm is considered as successful. The analyzed ground motions are selected from NGA-West2 (Ancheta et al., 2012), GeoNet, ITACA and RAN (Pacor et al., 2011; Luzi et al., 2016) and K-Net databases, which contain data from crustal earthquakes. Earthquake signals that are recorded due to earthquakes with magnitude range between 5.5 and 7.0 with a maximum distance range of 0 km to 150 km from the epicenter, are selected. In total we have 2738 waveform in which 1168, 296 and 1274 are recorded due to strike slip, normal and reverse faulting, respectively. In order to reach these goals, both velocity waveform and spectrum of the waveform are used. Period of maximum wavelet power spectrum where PGV occurred (t pgv ) is considered the pulse period in seconds (T p ). One threshold is applied the waveform which is the minimum PGV amplitude. If the PGV amplitude is less than 30 cm/s, signal is considered as non-impulsive (Shahi, S.K. and Baker, J.W., 2014). Then area between t pgv - T p /2 and t pgv + T p /2 is considered as pulse region on PGV. If the average of energy ratio between the waveform and the pulse region and spectrum energy ratio between the waveform and the pulse region is more than 30%, the signal is considered as pulse shaped signal. Furthermore we investigate the possibility of maximum energy in wavelet spectrum on another region rather than PGV region. We put 25 cm/s minimum amplitude threshold on such region. If it exceeds the threshold and if the region is not inside the PGV region that are determined, then we check the average waveform and spectrum energy ratios. If the energy ratio of the region is bigger than 10% or the average energy of the PGV region, then we consider this part as the impulsive part of the waveform. This method allows us to determine impulsive signals not only in the PGV region but in all waveform. Morlet and Ricker wavelets are used in order to calculate the wavelet spectrum. Morlet wavelet is complex, while Ricker wavelet is real-valued. The complex wavelet function returns both amplitude and phase information, whereas real wavelet function returns only real components. This allows to isolate discontinuities. Since both of the wavelets are giving the same qualitative results on power spectra, both of them can be treated as equal. However, Ricker wavelet can distinguish the discontinuities, since it is a real-valued function, whereas Morlet wavelet can give more smooth results, which is important when there is a high-frequency content in the pulse region (Fig. 1). The resolution of the wavelet function depends on the width