A feasibility study of the effect of particle shape onthe shear modulus of sand using dynamic simpleshear tests and artificial intelligence

By using dynamic simple shear tests and an artificial method, this study examined the effect of particle shape on secant shear modulus of dry sands. The monotonic behavior of sand was first examined, followed by cyclic simple shear tests of samples. Under constant-stress and controlled-stress modes, the tests were conducted at different vertical stresses and cyclic stress ratios (CSRs). A total of ۱۰ sand particles were then randomly selected in two stages: (۱) before the first test and before the second test. To quantify the particle shapes, three shape descriptors were used, including roundness (R), sphericity (S), and regularity (ρ). Each cyclic test was followed by the drawing of hysteresis loops and the determination of secant shear modulus. Results show that the sand had a dilative behavior under cyclic load, with the particles becoming slightly rounded after each cyclic test. The three particle shape parameters were increased by approximately ۵ to ۱۰%, which resulted in a significant reduction in secant shear modulus. Furthermore, the results of classification and regression random forests (CRRF) as an artificial intelligence method show that the CRRF model could predict shear modulus of sand with the coefficient of determination (R۲) of ۰.۹۱ and the mean absolute error (MAE) of ۵۴۹.۱۲. These results showed the great performance of AI methods to predict the dynamic behavior of sands.