Application of artificial neural networks in non-destructive testing of layered structures using the surface wave method

Abstract

The surface wave method is an in-situ non-destructive testing procedure for estimation of elastic moduli and layers thicknesses of layered structures such as pavements and natural soil deposits. In this research, MatLab has been employed for applying artificial neural networks in solving the inversion problem of the surface wave test dispersion curve and estimating the soil profile. Multi-layer neural networks along with back propagation training procedure are used to carry out the required inversion process. The networks are trained using the Steepest Descent Gradient Algorithm, Conjugate Gradient Algorithm and Levenberg-Marquardt Algorithm. Eight training functions have been employed and assessed in three, four and five layer networks. The most optimised network with the least error rate and iteration number for convergence was selected and tested for certainty. By employing the selected optimum network, a number of real cases have been studied and the results obtained have been compared with the available actual data. The results show very good match, indicating that the selected back propagation neural network is capable of providing a useful tool for carrying out the inversion process of surface wave method.