Geotechnical engineers recognize the variability of the geological materials they work with, including uncertainties associated with subsurface characterization tasks. These uncertainties include data scattering, such as real spatial variation in soil properties, or random testing errors. Systematic errors, as can occur in bias measurement procedures, are also common. In almost all construction projects, penetration tests play a major role in subsoil characterization. Interpretation of test results is mostly empirical, and it is therefore prudent to find a suitable computational method to minimize the error in predicting values at points away from actual test locations. In this research, the capabilities of artificial neural networks (ANNs) are assessed as a computational method for predicting standard penetration test (SPT) results at any point ( x, y, z) in a field where a set of SPTs is performed. SPT and moisture content data for five bore holes are used to train and test the developed three-dimensional network models. To graphically visualize the underlying soil strata, select contour maps of blows and moisture content values at various locations are presented. The results obtained indicate the viability and flexibility of ANN methodology as an efficient tool for site characterization tasks.
Prediction of Dried Durian Moisture Content Using Artificial Neural Networks