AI-driven prediction of shear parameters for geosynthetic-reinforced recycled sand

The rapid pace of urbanisation has intensified the demand for construction materials, leading to the depletion of natural sand (NS) reserves and necessitating sustainable alternatives. Recycled sand (RS), derived from the washing of construction and demolition waste, has emerged as a promising substitute for geotechnical applications. This study examines the interface behaviour of RS and NS reinforced with ten types of geosynthetics using large-scale direct shear (LSDS) testing, generating a comprehensive dataset from 66 tests conducted under varying normal stresses. Considering the labour and resource intensive nature of LSDS testing, a set of artificial intelligence (AI) models such as linear regression, decision tree, support vector machine, random forest, ensemble learning, and deep neural networks (DNN) was developed to predict peak shear strength and interface parameters. Among these, the DNN model demonstrated superior predictive performance, achieving R2 values between 0.89 and 0.99 for NS and between 0.92 and 0.96 for RS. The RS interfaces exhibited shear strength characteristics comparable to NS, validating its potential for sustainable ground improvement applications. The integration of AI-based modelling with experimental testing establishes a reliable and time-efficient framework for predicting soil–geosynthetic interface behaviour. The findings advance performance-based design methodologies for recycled geomaterials and underscore the role of AI in promoting sustainability within geotechnical engineering.