Linking Brinell Creep Indentation to Macroscopic Rock Creep Compression Using Viscoelastic Modelling and Simulation-Informed Machine Learning

Rock creep behaviour affects the long-term behaviour of subsurface systems, yet its reliable characterization remains challenging due to the high cost and long duration of conventional creep testing. Instrumented indentation offers a promising alternative, but localized, non-uniform stress conditions and load-dependent responses complicate the transfer of indentation-derived parameters to macroscopic behaviour. In this study, an integrated numerical and simulation-based machine learning framework is developed to link Brinell creep indentation responses to conventional compression creep properties of rocks. A viscoelastic Burgers creep formulation is incorporated into an elastoplastic numerical framework to simulate time-dependent behaviour under both uniaxial compression and Brinell indentation loading conditions. Parametric sensitivity analyses and the analysis of variance are performed to quantify the roles of Burgers model parameters across scales, followed by the establishment of scale-specific calibration procedures validated against experimental creep data. Using 459 simulation datasets, a neural network model is developed to predict macroscopic Burgers parameters from indentation-derived inputs. Experimental comparisons, including independent data from the literature, provide encouraging evidence of the predictive potential of the proposed approach. The proposed framework provides a potentially practical pathway for estimating macroscopic creep behaviour from indentation tests when conventional long-term creep compression experiments are limited or unavailable.