Probabilistic calibration of a damage rock mechanics model

Current practice in the calibration of damage models requires downscaling the effects of experimental observations from macro/meso to micro. This process introduces uncertainty that is seldom quantified to reflect the expert's confidence in the model predictions. A probabilistic calibration methodology can be introduced to overcome this problem. This paper shows a case study based on a damage rock mechanics model and triaxial experimental data on sandstone, where this approach is implemented to illustrate the impact of varying states of evidence (i.e. model complexity, experimental observations and expert's judgement) on the model predictions. The probabilistic calibration method relies on the use of the Bayesian paradigm to assimilate experimental observations into the probabilistic definition of the model parameters. Results of this approach can be encapsulated into a single probability distribution or posterior, which is later used to assess the model performance. The proposed approach shows the potential to improve current practice in risk analysis since it allows tracing of changes of the model performance for varying evidence conditions in damage-sensitive geostructures such as nuclear waste disposals, landfills, geothermal wells and unconventional oil and gas formations, among others.