ABSTRACT
Accurate estimates of soil unit weight are fundamental for correctly post process CPTu data and making use of Soil Behavior Type-based classification systems. Soil-specific and global regressions have been proposed for this purpose. However, soil-specific correlation might pose a problem of pertinence when applied at new sites. On the other hand, global correlations are easy to apply, but generally carry large systematic uncertainties. In this context, this work proposes a data clustering technique applied to geotechnical database aiming to identify hidden linear trends among dimensionless soil unit weight and normalized CPTu parameter according to some unobservable soil classes. Global correlations are then revisited according to such data subdivision aiming to improve accuracy of soil unit weight prediction while reducing transformation uncertainty. A new probabilistic criterion for soil unit weight prediction is also obtained. The potential benefits of the proposed procedure are illustrated with data from a Llobregat delta site (Spain).
