Experimental investigation and modelling the deformation properties of demolition wastes subjected to freeze–thaw cycles using ANN and SVR

Abstract

Construction and demolition (C&D) waste materials are widely being used as a pavement construction material. The aim of this research is to investigate the effect of freeze–thaw (FT) cycles on the deformation properties of C&D materials. Two artificial intelligence (AI) techniques, namely artificial neural network (ANN) and support vector regression (SVR) were developed for prediction of resilient modulus (Mr), which have been rarely used for modelling the behaviour of C&D materials. Laboratory repeated load triaxial (RLT) tests were performed on two types of C&D materials including recycled concrete aggregate (RCA) and crushed brick (CB) to investigate the permanent deformation and Mr. The influence of up to 20 FT cycles on the behaviour of C&D materials, for freezing to −15 ℃ and thawing to 20 ℃, was investigated. It was observed that deformation properties of RCA improved consistently as the number of FT cycles increased. Increasing the number of FT cycles reduced the permanent deformation and increased the Mr of RCA. For CB, application of up to 10 FT cycles resulted in an increase of resilient modulus (Mr) and a decrease of permanent deformation. However, application of more than 10 FT cycles, i.e., 15 and 20 FT cycles, had detrimental effects on the deformation characteristics of CB. An ANN model, as well as SVR models with different kernels, were developed for predicting the Mr of C&D materials exposed to FT cycles and investigating the effect of test variables. The developed models included number of FT cycles (NFT) and stress states, i.e., confining pressure (σ3), and deviator stress (σd) as input parameters, and the Mr was the model output. Results of numerical modelling indicated that ANN and SVR were highly capable of predicting the Mr of C&D materials subjected to FT cycles. Several supplementary analysis and verification phases were conducted to examine the reliability and precision of the developed models.