Hybridizing Neural Network with Trend-Adjusted Exponential Smoothing for Time-Dependent Resistance Forecast of Stabilized Fine Sands Under Rapid shearing

A comprehensive experimental research was undertaken to investigate the association of undrained shear strength with B-ratio, void ratio, confinement pressure, and principal stress difference at failure of zeolite-lime-treated fine sands. For this purpose, a series of undrained triaxial shearing tests were performed on samples comprising several lime-activated zeolites. With regard to the experimental evidence, a novel trend-adjusted (TA) growth forecast was performed with exponential smoothing to extend curing ages beyond the conditions of the experimental program. Then, hybridization of the artificial neural network (ANN) was done by feeding the network with feature adjusted shear resistance values against void ratio, B-ratio, confining pressure, and binding agents over extension of projected curing period. The proposed TA-ANN model showed a boosted optimization in input selection and high accuracy and confidence rate in predicting undrained shear resistance while including extended curing periods. Finally, results of variable importance and sensitivity analysis indicate a significant impact of underlying degree of saturation to the final shear resistance followed by void ratio, confinement pressure, and zeolite content.