Abstract
The application of sand-clay mixtures is diverse in contemporary engineering practices, with particular emphasis on their shear strength characteristics. This study focused on the estimation of the shear strength of sand-clay mixtures using the artificial neural network (ANN) and low-field nuclear magnetic resonance (NMR) spectroscopy. In this study, NMR tests and triaxial compression tests were carried out on 160 artificial sand-clay mixtures with different mineralogical compositions, water contents, and dry densities in the laboratory to obtain the T(2) spectra and shear strength indices, respectively. Twelve characteristic variables that could reflect the pore structure and water classification in the mixtures were calculated for each T(2) spectrum. A novel predictive model for the shear strength of the mixtures was established using the ANN based on 12 characteristic variables, the Atterberg limits, and the tested shear strengths of mixtures. The Atterberg limits of the mixtures, 12 characteristic variables and shear strengths of the mixtures were defined as the input factors, input covariates and response variables, respectively. The model uses mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R(2)), and Pearson correlation coefficient (R) to prove its accuracy. And the MAE, the RMSE, R(2), and R of the training set were 3.832 kPa, 4.920 kPa, 0.974, and 0.987, respectively. The MAE, the RMSE, R(2), and R of the testing set were 4.920 kPa, 6.164 kPa, 0.962, and 0.981, respectively. This indicated that the accuracy of this model was sufficient enough to predict the shear strength of the sand-clay mixture.