Backcalculation of Pavement Profiles from Spectral-Analysis-of-Surface-Waves Test by Neural Networks Using Individual Receiver Spacing Approach

Abstract

The Spectral-Analysis-of-Surface-Waves (SASW) method is a seismic technique for in situ evaluation of elastic moduli and layer thicknesses for layered systems, such as pavements and soils. The objective of the SASW test is to obtain the experimental dispersion curve and, through an inversion procedure, obtain the profile of an elastic moduli of the layered system. The inversion process in practice uses an average of dispersion curves for different receiver spacings. Results of theoretical studies indicate that differences in dispersion curves for various spacings are a result of interference of a number of body and surface waves. The development and application of neural networks to perform the inversion procedure for SASW testing of asphalt concrete (AC) pavements is presented. The most important feature of the developed network is that training of the network was done by the dispersion curves for individual receiver spacings. The training set consists of dispersion curves for seven receiver spacings and 78 dimensionless frequencies, while output is presented by elastic moduli and layer thicknesses of a four-course AC pavement. The dispersion curves used to train the neural networks are synthetic dispersion curves developed from numerical simulations of the SASW test. The obtained neural network model is compared to the previously developed model for backcalculation of moduli from the SASW test based on the averaged dispersion curve. Although both approaches can accurately define profiles, each has some advantages in evaluation of the thickness of the subbase.

Iterative prior resampling and rejection sampling to improve 1-D geophysical imaging based on Bayesian evidential learning (BEL1D)

“…Machine learning algorithms have been applied early in the 1990 s to the interpretation of surface waves (Meier & Rix 1993;Williams & Gucunski 1995;Gucunski & Krstic 1996;Michaels & Smith 1997). It was also among the first to see an application of a probabilistic neural network (Devilee et al 1999).…”