Predictive Models for Estimating Swelling Potential of Expansive Soils Based on Geotechnical Properties
Résumé
Expansive soils pose significant challenges in geotechnical engineering due to their tendency to undergo notable volume changes with moisture variation, often resulting in severe damage to infrastructure. Traditional laboratory methods, while reliable, are time-consuming, costly and impractical for routine investigations. This study first reassesses four widely used empirical correlations (Seed et al., Chen, Nayak & Christensen, Schneider & Poor) on a combined set of natural and experimental expansive clays and shows that they provide limited quantitative accuracy, with R2 around 0.6, normalized errors (NRMSE) typically between about 45% and 100%, and individual swelling misestimations up to 14.5%. New regression equations are then developed using only routine index properties (liquid limit, plastic limit, plasticity index and clay content). A first model calibrated on five controlled soils reproduces oedometer swelling with errors below 0.2%. An extended global model, calibrated on 12 soils from diverse geological settings and incorporating nonlinear and interaction terms (CL2, LL × PI, PL × CL and LL × CL), achieves R2 ≈ 0.92 with NRMSE ≈ 8% and maximum errors below about 2% on this calibration set. When applied to a separate set of six independent validation soils not used for calibration, the same model maintains NRMSE ≈ 10% with maximum errors of about 1.5–2%, representing a four- to eightfold reduction in normalized error compared with the classical equations. A variant including VBS yields comparable accuracy while explicitly representing clay activity through a simple additional test. The proposed equations therefore offer practical, index-based tools for estimating laboratory swelling potential, improving the reliability of expansive-soil screening and reducing the need for extensive oedometer testing in routine design.