The use of neural networks to develop CPT correlations for soils in northern Croatia

The evaluation of soil parameters for design is best undertaken through comprehensive laboratory test programmes. However, due to sampling difficulty, time and cost constraints correlations between in-situ tests and physical-mechanical properties of soils are routinely applied in practice. This paper presents data collected from five sites in Northern Croatia at which Cone Penetration Tests (CPT) and comprehensive laboratory test data was available. One of the advantages of using CPT data in preference to other types of in-situ tests for establishing correlations, is the large volume of high-quality data available at each probe location allows for the application of advanced statistical approaches. In this paper, the use of neural networks in developing such correlations is demonstrated. Using a database of 216 data pairs, obtained from the five sites, a correlation between CPT q _c and soil unit weight is established. A validation exercise was performed in which the correlation was tested against data from the recent Veliki vrh landslide that occurred in the same geographical region as the database sites. In addition, by using the soil behaviour type index, Ic, normalised cone tip resistance, Qtn, and normalised sleeve friction, Fr, the results can be compared to correlations developed for soils from geotechnical diverse regions to check for consistency in the derived correlations.