Experimental Investigation and Prediction of the Permanent Deformation of Crushed Waste Rock Using an Artificial Neural Network Model

Abstract

The gradual accumulation of permanent deformation in unbound granular material layers is one of the main reasons for flexible pavement rutting. The accurate determination of permanent deformation behavior in pavement materials is critical for the successful design of pavement systems. However, predicting the permanent deformation is complex, and the available empirical regression models have limited accuracy and applicability. In this study, multistage repeated load triaxial tests were carried out under different stress levels in order to evaluate the permanent strain and shakedown ranges of crushed waste rock. The plastic shakedown limit and plastic creep limit were determined so as to estimate the shakedown range of crushed waste rock under certain stress conditions. The Rahman and Erlingsson model (extended using a time-hardening approach) performed better than other models at fitting the accumulated permanent strains (R2 > 0.92), although the prediction accuracy of the shakedown range was relatively low (<85%). An artificial neural network (ANN) model was therefore developed, based on the experimental results, to predict the permanent strain of crushed waste rock. The ANN model consisted of three hidden layers (50 neurons per layer) with Tanh activation function and could predict the permanent strain (R2 > 0.97) and shakedown ranges (accuracy >93%) satisfactorily.