Page - 1170 - in Book of Full Papers - Symposium Hydro Engineering

(a) Time domain (b) Frequency domain Fig. 3 Source wavelet signal in time domain and frequency domain (a) seismogram for shot 14 (b) seismogram for selected traces Fig. 4 Sensor response due to propagated waves in medium. (a) Seismogram of all 165 sensor response (b)(i) Response of selected traces. Dam response recorded by the receivers/sensors due to waves propagating in the medium is obtained in Fig. 4. These synthetically generated seismograms would serve as field data for the inverse analysis (i.e. full waveform inversion). Originally proposed by [30], the FWI is carried out in multiple frequency stages. The inversion is done starting from lower frequencies to higher frequencies (i.e. 0:5-1kHz) to alleviate the non-linearity of the problem. These lower frequencies are less prone to grid dispersion and can identify high velocity regions. The output of low frequency inversion is used as a starting model for higher frequency inversion. Therefore, with each increasing stage of inversion (i.e. increase in frequency), the location/boundaries between different velocity regions becomes more clearly delineated and the quality of the inversion results improves. Inversion results obtained in Fig. 5 shows a good resolution of damages in the dam structure. Anomalies in the dam foundation were not properly resolved, however, from a comparison of the inverted models (i.e. Vp, Vs, ρ) the Vs model can resolve most of the anomalies/damage zones in the dam structure and parts of the foundation. Although less sensors are used in setup 2, better resolution of the damaged regions in addition to more numerical artefacts can be observed. 1170