Proposing New Artificial Intelligence Models to Estimate Shear Wave Velocity of Fine-grained Soils: A Case Study

Dynamic parameters are the most important geotechnical data used to understand the behavior of soil media under dynamic loads and to recognize the seismic response of the soil. Several in-situ and laboratory geophysical tests, such as the down-hole test, are used to determine these parameters. Since this experiment is costly and time-consuming and the preparation of appropriate boreholes is not easy, it is preferable to estimate the results of this test with the help of empirical correlations or experimental models. The main output of the down-hole test is the shear wave velocity (VS) of soils, which can be used to obtain the dynamic shear modulus (Gs) indirectly. The relationship between physical properties and mechanical specifications of soils is a well-known principle of geotechnical engineering. Utilizing the results of ۱۹ down-hole experiments and available geotechnical data in the southern regions of Tehran, as well as the inputs of an adaptive neuro-fuzzy inference system (ANFIS). This study attempts to provide practical models to predict shear wave velocity of fine-grained soils in Tehran. Two new models have been proposed as a result of preprocessing and smart modeling. The independent variables of the first suggested model included the moisture content, plasticity index (PI), liquid limit (LL), depth of test, and grain size distribution of soils. In the second model, the number of standard penetration test (NSPT) is also used in addition to the mentioned independent variables. The proposed models had coefficients of determination (R۲) of ۰.۷۴ and ۰.۸ for the total training and validation data, respectively.

Keywords:

Fine-grained soil , Shear wave velocity , Down-hole test , Geophysics , adaptive neuro-fuzzy inference system