Polynomial neural network model to estimate the stress–strain behavior of zeolite-cement injected sand

Abstract

The process of cement production is costly and one of the main factors in carbon dioxide emission. Hence, part of it should be replaced with eco-friendly pozzolanic materials like zeolite. As the determination of shear behavior of injected sands is time-consuming and laborious, in the present research the polynomial neural network (PNN) model was used to predict stress (q)-strain (ε) behavior of zeolite-cement injected sand. For this purpose, a number of consolidated undrained (CU) triaxial tests was performed on sand samples injected with zeolite-cement grout. Due to the difference in the shear behavior of the injected sand before and after the yield point (YP), the stress–strain curves were divided into two parts (up to the YP and beyond the YP), and the curves of each part were predicted with separate relationships. The results revealed that the PNN-based equations can accurately estimate the q-ε curves of sand samples injected with zeolite-cement grout, such that the mean absolute percent error (MAPE) for testing data sets to estimate pre- and post-YP q was 7.79 and 5.38%, respectively. Sensitivity analysis indicated that up to the YP, the confining pressure (CP) was the most important parameter affecting the injected sand strength. The importance of water to cementitious materials ratio (W/CM) and cement replacement with zeolite content (Z) on the pre-YP q predicted by the PNN model was close to each other and less than the CP. Beyond the YP, the effect of W/CM, CP and Z was almost the same.