Modeling of the Uniaxial Compressive Strength of Carbonate Rocks Using Statistical Analysis and Artificial Neural Network

Carbonate formations are widespread in Iran's structural zones. Given their significance in infrastructure development and hydrocarbon reservoirs, understanding the geological and engineering geological characteristics of these formations is crucial. This study focuses on the Ilam Formation, a prominent carbonate formation in the Zagros Fold-Thrust Belt, due to its extensive distribution and association with Iran's oil fields. A comprehensive investigation, involving field surveys and laboratory analyses, was conducted on Ilam Formation outcrops in the central Zagros region, encompassing Ilam, Kermanshah, and Lorestan provinces. Microscopic studies identified seven microfacies types, categorized into four groups: mudstone, wackestone, wackestonepackstone, and packstone. Physical tests, including porosity, water absorption, density, P-wave velocity, and S-wave velocity, were performed on these samples. Mudstone samples exhibited higher porosity and water absorption, while packstone samples had higher P-wave velocity and density. The superior physical characteristics of packstone samples can be attributed to their formation in a higher-energy sedimentary environment, which resulted in a higher percentage of grains and increased interlocking between them. Mechanical tests, such as uniaxial compressive strength, modulus of elasticity, point load index, and Brazilian tensile strength, were also conducted. Packstone samples demonstrated the highest average strength, whereas mudstone and wackestone samples exhibited the lowest strength. According to Deere and Miller's classification, the samples were categorized as B and C, falling within the modulus ratio range of 200-500. The physical and mechanical properties-based multiple linear regression analysis revealed that P-wave velocity and point load index were the most influential parameters in predicting uniaxial compressive strength, respectively. Additionally, an artificial neural network model was developed to predict uniaxial compressive strength using porosity, P-wave velocity, modulus of elasticity, point load index, and Brazilian tensile strength as input parameters. According to the result, point load index was the most influential parameter in the model. Both multiple regression and artificial neural network models demonstrated satisfactory performance in predicting uniaxial compressive strength. However, the simplicity and ease of use of multiple regression make it a more practical choice for engineering applications, despite the comparable performance of the neural network model.