Seite - 1165 - in Book of Full Papers - Symposium Hydro Engineering

implementations ranging from simple to complex and using various numerical methods such as Finite Differences, Finite Elements, Ray theory, Integral solution, etc. The finite difference method is a common method used for solving the wave equation in geophysics and oil exploration as can be seen in works by [6, 7] and [8]. Based on the successful application of Full Waveform Inversion (FWI) in geotechnical exploration and non-destructive testing (NDT), especially with subsurface structures [9, 10, 11, 12, 13, 14, 15, 16], our method aims to extends this application to the identification of damaged regions in dams. 2. THEORY Seismic waves are generally classified into surface waves and body waves. The surface waves as the name implies, generally travel on the while the body waves on the other hand travel through the inner parts of the earth or structure. Seismic inversion can be categorized into seismic reflection and seismic tomography. In seismic reflection the inversion of reflected waves usually gives information on geometry while refracted waves (seismic refraction) give information on elastic parameters. Seismic tomography application on the other hand, employs the inversion of transmitted, reflected and refracted signals from seismic observations to determine the distribution of material properties in a structure. Seismic tomography methods include travel time inversion (kinematic imaging) and Full Waveform Inversion (dynamic imagining). Full waveform inversion (FWI), initially proposed by [17, 18] and further developed in the last decades, is regarded as the modern seismic imaging technique which utilizes a comprehensive representation of the interaction between wave physics and subsurface properties. It offers unique advantages in terms of generality, fidelity, complexity and robustness and is hence capable of imaging arbitrarily heterogeneous compression and shear-wave velocity profiles of the subsurface constituents [9]. The quality and efficiency of the inversion is sensitive to the initial model, acquisition setup, and the formulation of the inverse procedure (inversion parameters, domain, method, etc.). To maintain numerical stability and efficiently use computational resources in simulating wave propagation in unbounded media, it is necessary to truncate the numerical domain. Several absorbing boundary conditions (ABC) has been proposed, however most of these ABC become unstable when considering heterogeneous materials in addition to other limitations. The perfectly matching layer (PML) proposed by [19] is used in this paper. 1165