The Prediction of the Tensile Strength of Sandstones from their petrographical properties using regression analysis and artificial neural network

Abstract

This study investigates the correlations among the tensile strength, mineral composition, and textural features of twenty-nine<br />sandstones from Kouzestan province. The regression analyses as well as artificial neural network (ANN) are also applied to evaluate<br />the correlations. The results of simple regression analyses show no correlation between mineralogical features and tensile strength.<br />However, the tensile strength of the sandstone was decreased by cement content reduction. Among the textural features, the packing<br />proximity, packing density, and floating contact as well as sutured contact are the most effective indices. Meanwhile, the stepwise<br />regression analyses reveal that the tensile strength of the sandstones strongly depends on packing density, sutured contact, and cement<br />content. However, in artificial neural network, the key petrographical parameters influencing the tensile strength of the sandstones are<br />packing proximity, packing density, sutured contact and floating contact, concave-convex contact, grain contact percentage, and<br />cement content. Also, the R-square obtained ANN is higher than that observed for the stepwise regression analyses. Based on the<br />results, ANN were more precise than the conventional statistical approaches for predicting the tensile strength of these sandstones from<br />their petrographical characteristics.