Integrated Approach for Probability of Failure Analysis in Salt Caverns: API and Predictive Model Framework

Rock salt is a viscoplastic material that cannot withstand deviatoric stresses, leading to creep deformation. This displaces the walls of salt caverns, causing cavern convergence and subsidence. Inelastic volumetric expansion can increase microcracks, resulting in displacement phenomena, posing risks to the population and environment. A predictive model using the POD-RBF metamodel was developed to address these challenges. This model combines Proper Orthogonal Decomposition (POD) and Radial Basis Functions (RBF) to create Reduced Order Models (ROM) from high-dimensional data. An axisymmetric model was built to represent the cavern geometry, considering five evaluation points on the cavern wall. The geomechanical model includes shale, halite, and overburden. Abaqus software was used for simulations, modeling halite's creep behavior over 50 years using the Munson-Dawson model. The metamodel was trained using 200 parameter sets and validated with 20 additional simulations, achieving an R² value of 0.9972 and an NRMSD of 0.0547. An API was implemented to automate parameter updates, reducing computational costs significantly. Reliability-based approaches analyzed uncertainties, with probability of failure assessed using the Monte Carlo method.

Keywords: Geotechnical engineering, Rock mechanics, Salt rock, Reliability analysis, Finite element method, Creep strain, Dilatancy, Internal pressure, Munson-Dawson Model.