Predicting rock strength in in-situ stress and thermal environments: Bridging experimental data with failure criteria and machine learning models

Understanding the influence of temperature on rock strength is essential for many rock mechanics applications including geothermal energy, deep mining, and nuclear waste storage, where rocks are exposed to high temperatures and confining pressures. Laboratory studies show rocks generally weaken at elevated temperatures due to thermal damage, but some suggest strength increases in the mild range (room temperature to 300°C), though the mechanisms remain poorly understood. A comprehensive literature review and analysis was conducted, specifically focusing on temperature effects within the mild range. The experimental data were compared against temperature-modified Hoek-Brown and non-linear Mohr-Coulomb failure criteria to evaluate their ability to predict temperature-dependent changes in rock strength. Statistical analysis refined parameter ranges, with machine learning models, including support vector machine, random forest and neural networks, used to predict temperature-dependent triaxial strength. The results indicate that rock strengthening is largely an intrinsic property influenced by various factors, with a transition period at elevated temperatures where rock behavior shifts from strengthening to weakening. Both Hoek-Brown and modified non-linear Mohr-Coulomb criteria can approximate temperature effects. Machine learning outperformed traditional models, showing promise for improving predictions of complex, temperature-dependent rock behavior.