Prediction of freeze–thaw and thermal shock weathering on natural stones through deep learning-based algorithms

Natural stones used as construction materials in outdoor applications and the rock environment in rock engineering applications are subject to weakening such as freeze-thaw (F-T) and thermal shock (TS) due to weather conditions. Predicting the mechanical effects of F-T and TS weathering is important for the design on rock engineering. While the change in mechanical properties can be determined by F-T and TS simulating with experimental studies, it can also be predicted with simple models in the literature and determining initial conditions. While the properties of weakened rock are determined from the models proposed in the literature, a rock-specific experimental study is needed and precise results cannot be obtained. Instead, the predicting of F-T and TS weathering on rocks by using deep learning-based algorithms enables better data for design. In this study, the effects of deterioration on physical and mechanical properties of natural stones after F-T and TS weathering was investigated with experimental simulation. The samples of 15 different rock type were subjected to F-T and TS process for 15, 30 and 45 cycles following standard methods in experimental study. The changes of apparent porosity, water absorption by weight, P-wave velocity, uniaxial compressive strength and elastic module of rocks after each process were investigated and analysed with different deep learning algorithms to predict these properties. It has been determined that AdaBoost is the best algorithm for predicting the properties of natural stone after F-T and TS weathering. Additionally, the stress distribution was modelled numerically to investigate the effect of F-T and TS weathering on rock samples. The study shows that deep learning-based algorithms can be used as an auxiliary tool in prediction in order to perform more precise studies.