Prediction of drained soil shear strength parameters of marine deposit from CPTu data using GMDH-type neural network

Abstract

ABSTRACTDue to a few limitations through laboratory testing for determining soil shear strength parameters, cone penetration test (CPT) has been realized as a multipurpose apparatus with an acceptable performance, especially in soft to medium deposits. The known methods for obtaining shear strength parameters using CPT data, estimate conservative values for φ (internal friction angle) in granular and Su (undrained soil shear strength) in cohesive soils. As the determination of the shear strength parameters (shear strength (τ), Cohesion (C), and friction angle (φ))—especially at depths—is relatively costly and time-consuming, there is a need to develop models that can handle simply determinable properties. In the present study, the suitability of the group method of data handling (GMDH)-type neural network (NN) and genetic algorithms (GAs) to predict the C and φ of marine soils have been investigated. To derive GMDH models, a database containing 50 datasets compiled from geotechnical investigation sites in Iran, has been used. Results show that the polynomial models introduced in the current study are appropriate models to estimate shear strength parameters (C and φ) which represented accurate results. To achieve better correlation, using ln form of cone tip resistance (ln qc) have been suggested as input parameter. The performance of the GMDH models has been compared with other available correlations, and it has been demonstrated an improvement in estimating the marine soil shear strength parameters. Finally, a sensitivity analysis has been conducted on the proposed models, showing that the proposed φ is considerably influenced by changing the Rf value.