Identification of Soil Strata Based on General Regression Neural Network Model From CPTU Data

Abstract

In geotechnical engineering, assessment of the depth location of stratigraphic interfaces and the depth and thickness of thin layers can be critical in the design process. For example, stratigraphic interfaces can promote anisotropic soil strength response and potentially provide preferential slip planes that create slope instability. The piezocone penetration test (CPTU or PCPT) is an extension of the cone penetration test (CPT) and is able to measure cone tip resistance, sleeve friction, and generated pore-water pressures simultaneously. In this paper the relationship between CPTU parameters and soil types and strata is analyzed, and the structure of a general regression neural network (GRNN) is designed for soil classification and soil strata identification. This research discusses soil strata automation identification, supplies a new way to deal with the CPTU data, and has actual significance in promoting efficiency and precision of CPTU data processing. The example results are presented and verified using backup laboratory and in situ data for case study to identify soil strata by CPTU and have confirmed that GRNN can be used to carry out automatic soil strata identification. GRNN-based model was found to be correlating well for the 87% of the cases with the USCS classification system results.