Factors Affecting the Dynamic Young's Modulus of Sandstones Using Intelligent and Statistical Methods

Abstract

This study analyzed how the sandstone's mineralogy and physical properties affect dynamic Young's modulus (Ed). To model these effects, back propagation multilayer perceptron (BPMLP), K-nearest neighbor (KNN), classification and regression tree (CART), simple and multivariate linear regression (MVLR) were employed. After microscopic analysis, compressional and shear wave velocities, water absorption, porosity, and density were measured on the samples. Results showed the calc-litharenite sandstones exhibited a lower Ed compared to the feldspathic litharenite types. Quartz showed a stronger influence on Ed compared to the porosity and water absorption. A strong correlation between Es and Ed was proposed. The dynamic modulus was found to exceed the static Young's modulus (Es) with a calculated ratio of 2.56 (Ed /Es=2.56). Based on the RMSE, Nash-Sutcliffe index, A20 index, and determination coefficient, the methods were appraised. Within the mineralogical types, quartz had the highest correlation with Ed. Mica, opaque, and fragments showed a negative effect on the Ed, while chert and cement showed a positive effect on the Ed. Among the tested models, BPMLP achieved the highest accuracy, with an R of 0.94, RMSE of 0.08, CPM of 1.73GPa, and A20 value of 0.98, along with NSE of 97.04%, confirming the superiority of AI-based approaches over statistical techniques in predicting Ed.