Premium
Artificial neural network and statistical models for predicting the basic geotechnical properties of soils from electrical measurements
Author(s) -
Boadu Fred Kofi,
OwusuNimo Frederick,
Achampong Francis,
Ampadu Samuel I.K.
Publication year - 2013
Publication title -
near surface geophysics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.639
H-Index - 39
eISSN - 1873-0604
pISSN - 1569-4445
DOI - 10.3997/1873-0604.2013011
Subject(s) - geotechnical engineering , artificial neural network , electrical resistivity and conductivity , soil science , soil water , geotechnical investigation , linear regression , water content , geology , amplitude , engineering , machine learning , statistics , mathematics , computer science , electrical engineering , physics , quantum mechanics
The ability to predict the geotechnical properties of subsurface soils using non‐invasive geophysical measurements can be undeniably useful to the geotechnical engineer. Using laboratory data, we assess the potential of artificial neural networks to investigate the relations between geotechnical and electrical parameters characterizing a variety of soils. The geotechnical parameters are: fines content, mean grain size, mean pore size and the specific surface area. The electrical parameters obtained from low‐frequency electrical measurements (4 Hz) include the resistivity amplitude, phase shift and the loss tangent. Relations that can be used to predict the geotechnical parameters of a soil given its electrical parameters are developed. The predictive capabilities of the neural networks are compared with traditional multivariate regression models. The performances of the neural network and regression models in predicting (a) the geotechnical parameter given the same electrical parameters as inputs and (b) the electrical parameters given the same geotechnical parameters as inputs are compared. In both cases, the neural network outperforms the multivariate regression as the neural network is able to capture and model the non‐linear and complex relationships among the variables. The relative importance of the geotechnical parameters on the overall electrical conduction was examined using the neural networks. The results indicate that mean grain size and fines content are the two geotechnical parameters that influence phase‐shift values the most; fines content and mean pore size influence the resistivity amplitude the most, whilst fines content and mean grain size influence the loss tangent the most. It was observed that of the four geotechnical parameters, the mean grain size influences the measured resistivity values the least.