The effect of selecting different stress states in laboratory tests on the predicted permanent deformation of soils

Abstract

Modeling permanent deformation (PD) in pavement sublayer materials is essential for predicting pavement performance. The present study evaluated the influence of stress state, by means of the stress pair (confining stresses, σ3, and deviator stresses, σd) considered on the calibration of a permanent deformation model widely used in Brazil. A total of 54 repeated load triaxial (RLT) tests were performed on 6 different soils, each subjected to 9 stress pairs, resulting in 130 combinations per material. Model coefficients were obtained by multiple linear regression. The results indicated that reducing the number of stress pairs improves the coefficients of determination (R2), but may compromise the model’s global predictive capacity, especially for pairs not included in the fitting. In 61% of cases analyzed, predictions fell outside the fitted range, highlighting the model’s sensitivity to pair selection. Some coefficients showed high variability, with differences exceeding 100% even for the same material, compromising their physical interpretation. Application of this methodology to literature data confirmed these trends and showed that different stress pair selection significantly affects the regression coefficients, even when R2 values are high. Numerical modeling with MeDiNa software reinforced the practical relevance of these findings, revealing that changes in the coefficients of just one layer resulted in differences of up to 22% in total pavement deformation and 92% in the deformation of the base layer. It is concluded that defining technical criteria for stress pair selection is essential for consistent predictions. Further studies are recommended to deepen the understanding of pavement stresses and their impact on permanent deformation.