Accumulative plastic strain of freezing–thawing subgrade clay under cyclic loading and its particle swarm optimisation–back-propagation-based prediction model

Abstract

Accurate evaluation of cumulative plastic strains in subgrade soil is crucial for road designs and maintenance. Roads in seasonally frozen regions undergo several freeze–thaw (F–T) cycles annually, resulting in considerable deterioration of soil properties. Under the repeated cyclic loading of moving vehicles, distinct residual deformation occurs in the thawed subgrade. To study the influence of F–T cycles on the cumulative strain characteristics, cyclic triaxial tests were conducted on subgrade clay under the most unfavourable conditions and in situ measured dynamic stress amplitudes. The test results revealed that the cumulative strain (εap) increased with the dynamic stress amplitude. When the dynamic stress amplitude increased from 46 kPa to 70 kPa, εap of unfrozen and F-T specimens increased by more than 14 and five times, respectively. The axial deformation of clay is sensitive to the number of F–T cycles. The freeze-thaw effect was greatest in the first cycle, with εap increasing from 1.42% to 5.63% at the end of the test (dynamic stress amplitude of 46 kPa). In subsequent cycles, relatively small deformations accumulated. A particle swarm optimisation–back-propagation-based (PSO-BP) model for predicting the accumulative plastic strain of clay was developed based on experimental results. In this model, the influences of the F–T process and loading cycles were considered. The prediction performance of the proposed model was validated and compared with that of traditional regression models. The predictions were in good agreement with the experimental data and satisfied the deterioration law for clays subjected to F–T cycles. The maximum error of the PSO–BP model was 2.75%.