Comparative Consistency Assessment of TensorFlow, PLAXIS, and YADE for Foundation Suitability in Heterogeneous Ground

https://doi.org/10.46610/JoGS.2026.v011i03.001

Abstract

Foundation decisions in geologically variable ground are often made from a limited number of boreholes, which means that the interpretation between investigated locations can be uncertain. This study re-examines the existing Khare Khola geotechnical dataset from Okhaldhunga, Nepal, through a comparative consistency framework. Four investigated boreholes (BH-1 to BH-4) are used as the observed site basis, while two intermediate locations (BH-5 and BH-6) are evaluated from TensorFlow predictions. PLAXIS finite element results and YADE discrete element results are then interpreted independently and compared in terms of bearing capacity, settlement behavior, stress-deformation conditions, and resulting foundation recommendations. The observed ground changes from dense granular soil at BH-1 toward soft to medium-stiff clay at BH-4, with interpreted bearing capacity decreasing from about 250 to 100 kN/m². TensorFlow estimates approximately 225 kN/m² at BH-5 and 160 kN/m² at BH-6. PLAXIS reports corresponding ultimate capacities of about 225 and 160 kN/m², with settlement ranges of 10–20 mm and 30–50 mm, respectively. YADE gives approximately 225–240 kN/m² for BH-5 and 140–160 kN/m² for BH-6. Despite differences in modeling assumptions, all three approaches identify BH-5 as comparatively favorable for a shallow foundation and BH-6 as a more settlement-sensitive location requiring load-dependent foundation selection. The study therefore argues that the practical value of using several computational approaches lies in the consistency of the engineering decision rather than in exact numerical agreement. Where the methods converge, confidence in preliminary foundation selection increases; where the decision remains borderline, conservative design or additional investigation is warranted.