Development of Artificial-Neural-Network-Based Permanent Deformation Prediction Model of Unbound Granular Materials Subjected to Moving Wheel Loading

Abstract

The majority of existing regression models for unbound granular materials (UGMs) consider only the effects of the number of loading cycles and stress levels on the permanent deformation characteristics and are thus unable to account for the complexity of plastic deformation accumulation behavior influenced by other factors, such as dry density, moisture content and gradation. In this study, research efforts were made to develop artificial-neural-network (ANN)-based prediction models for the permanent deformation of UGMs. A series of laboratory repeated load triaxial tests were conducted on UGM specimens with varying gradations to simulate realistic stress paths exerted by moving wheel loads and study permanent deformation characteristics. On the basis of the laboratory testing database, the ANN prediction models were established. Parametric sensitivity analyses were then performed to evaluate and rank the relative importance of each factor on permanent deformation behavior. The results indicated that the developed ANN prediction model is more accurate and reliable as compared to previously published regression models. The two major factors influencing the magnitude of accumulated plastic deformation of UGMs are the shear stress ratio (SSR) and the number of loading cycles, of which the calculated influence coefficients are 38% and 41%, respectively, while the degree of influence of gradation is twice that of the confining pressure.

Keywords

• unbound granular materials
• cyclic triaxial test
• permanent deformation
• prediction model
• artificial neural network (ANN)