Backcalculation Analysis of SASW Test Using Artificial Neural Networks

Spectral analysis of surface wave method is an in- situ nondestructive testing technique for site exploration and determination of soil layers stiffness profile and similar systems. The results of SASW measuremants are presented in the form of experimental dispersion curve. This curve is used for determining the soil profile by means of one of the available backcalculation procedures.    In this paper, artificial neural networks have been used for estimation and determination of soil profile. Multi layer neural networks along with back propagation training procedure are suitable choices to carry out the required inversion process. A number of theoretical dispersion curves, which have been obtained using the exact dynamic stiffness matrix method, are used for training the networks. In other words, these curves along with the relevant soil profiles are used as input data for training the networks by means of three training algorithms. These algorithms are the Steepest Descent Gradient Algorithm, Conjugate Gradient Algorithm and Levenbereg Marquardt Algorthim. The comparisons made between the results obtained from this study and those resulted from other backcalculation procedures show very good match indicating that the method employed in this study can be effectively used for determination of soil site profiles.