Quantification of Model Uncertainty in Shear Strength Predictions for Fiber-Reinforced Sand

Abstract

Several models have been suggested to estimate the improvement brought by fibers to the shear strength of fiber-reinforced sands. To date, the effectiveness and reliability of these models have not been the subject of a comprehensive evaluation. The objectives of this paper are to (1) compile the experimental data available in the literature on the behavior of fiber-reinforced sands into a comprehensive state-of-the-art database, (2) quantify the model uncertainty and bias of current strength prediction models for fiber-reinforced sands, and (3) provide insight regarding possible modifications that could be made to the formulation of available models to improve their predictive effectiveness and reduce their model uncertainty. Two models that are considered to present the best available approaches to predicting sand-fiber shear strength were evaluated, namely, the “energy-based” model and the “discrete” model. The energy-based model was found to underestimate the measured friction coefficient on average by about 10%, while the discrete model overestimated the friction coefficient by 6%, with associated coefficients of variation on bias values of 0.20 and 0.17, respectively. With the introduction of minor modifications to these models, the average bias error was eliminated, and the coefficients of variation in the ratio of predicted to measured shear strength (bias) were reduced to 0.17 and 0.12, respectively, for the two models.