Large Language Model Based Prediction of SPT N value from Electrical Resistivity Correlations

Subsurface investigation and characterization of subsoil are the fundamentals for designing appropriate foundation systems for any Civil Engineering construction. Commonly employed geotechnical investigations are costly, laborious and time consuming whereas non-invasive methods like Electrical resistivity geophysical methods are widely used as it is a low cost and fast technique used to identify subsoil strata. This method evaluates the apparent resistivity of soil from the measured resistance and used for characterization of subsoil strata. The aim of the present work is to identify the subsoil profile by measuring the resistivity of soil, establishing correlations between geoelectrical parameter and SPT N value thereby estimating the Safe Bearing Capacity (SBC) of soil. Multiple linear regression with interaction terms was employed using Large Language Model (LLM) to develop possible correlations between apparent resistivity and SPT N value. The study employed Electrical Resistivity geophysical method at selected locations complemented by borehole investigations to depths ranging from 0.0m to 12.0m. Borehole investigations were carried out at 19 locations using wash boring and conventional sounding tests like SPT to obtain the stiffness of soil at every 1.5m depth measured in terms of SPT N value. The resistivity values are integrated with borehole data to develop correlations, and the developed correlations predict SPT N using parameters such as depth, apparent resistivity, and interaction coefficients. Two electrode arrays namely Wenner and Schlumberger were used in this study. Among the two methods, the Schlumberger array showed higher resistance at 1.5m at all locations, while Wenner array gave higher resistance beyond 1.5m depth and the resistivity varied from 1.649Ωm to 246.176Ωm depending upon the soil types. Silty sand showed resistivity of 30.257Ωm whereas the presence of clay reduced the resistivity showing a minimum value of 19.97Ωm. Positive slopes showed increase in water content with respect to depth. It varied from 8.7% to 18.18% following increasing trend at location 15 but negative slope did not follow a definite pattern. There is no general relationship for correlating electrical resistivity and SPT N value for homogeneous soil layers and use of AI technique in establishing relationships has not been explored. The mathematical approach employed least squares optimization with depth resistivity interaction terms and zero-shot prompting technique to find relationships between geoelectrical and SPT N. The fitness of the model measured in terms of coefficient of correlation R2 is observed as 0.9124. Keywords: Electrical Resistivity, Standard Penetration Test, Wash Boring, Wenner, Schlumberger, Safe Bearing Capacity, Large Language Model