Deep Learning-Assisted Computational Modeling of Granular Soil Behavior for Foundation Design Optimization in Kharke Khola Region

Gaining insight into granular soil behavior is crucial for designing cost-effective and safe foundations, particularly in geotechnically diverse regions like Kharke Khola, Nepal. The research aims to develop a coupled computational method that integrates field exploration, deep learning, and numerical simulation to optimize foundation design. Four critical site field borehole data (BH-1 to BH-4) were analyzed to determine soil classification, bearing capacity, groundwater level, settlement risk, and excavation complexity. Inferred from this data set, supervised deep learning models in TensorFlow were developed to predict subsurface conditions at two new sites (BH-5 and BH-6). These predictions were validated using numerical simulations using YADE (Discrete Element Method) and PLAXIS (Finite Element Method). These tests showed that BH-1 and BH-5 consist of dense granular soils, which are suitable for shallow foundations with a low settlement risk, while for BH-4, soft clay is present, requiring deep foundations. For BH-6, there were uncertain soil conditions with weak strength, and it was made evident that flexible and reinforced foundation schemes are required. The approach enhances the validity of subsurface predictions, reduces reliance on extensive amounts of physical testing, and supports more informed, cost-saving geotechnical decision-making.