Hierarchical Bayesian inference of rockfill strength parameters from large-scale triaxial test data

The shear strength of rockfill is often estimated using the empirical Barton-Kjærnsli (B-K) criterion to address testing limitations associated with the large particle size of these materials. In this criterion, the equivalent roughness (R) and strength (S) are the stress-dependent structural component parameters added to the basic friction angle (φb) of the parent rock to determine the shear strength of the rockfill material. However, for materials intended for use in tailings dam construction, the strength parameters must be supported by triaxial testing results. Due to the characteristics of the materials, large-scale tests are often limited in number. This challenge is compounded by the heterogeneity of the materials and differing proportions of rock types used in a project area. In previous papers (Linero-Molina et al., 2022; Contreras et al., 2022), the authors discussed the application of a Bayesian approach for inferring B-K parameters from large-scale laboratory testing results. This paper discusses the advantages of applying a hierarchical Bayesian model for optimal estimation of B-K parameters from large-scale triaxial test results, particularly when multiple rockfill mix proportions and lithological types are involved. The Bayesian framework integrates the B-K model, laboratory data from large-scale triaxial and basic friction angle testing, and allowable parameter ranges (priors) to create a posterior probability function. This posterior is evaluated using the Markov Chain Monte Carlo (MCMC) method to estimate the most likely parameter values that align with observed material behaviour. This approach provides insights into parameter correlations and uncertainty, thereby streamlining parameter estimation in an objective and rational manner. The methodology is demonstrated through a case study in which four rockfill material groups comprising mixtures of five rock types are investigated using the available laboratory testing data.