Utilization of Cross-Linked Polyethylene (XLPE) Waste in the Production of Sustainable Cementious Construction Materials

Abstract

The increasing generation of cross-linked polyethylene (XLPE) waste, primarily from decommissioned power cables and industrial applications, poses a significant environmental challenge due to its non-recyclable thermoset nature. This study investigates a sustainable pathway for managing XLPE waste by incorporating it into three major construction materials including concrete, fluidized thermal backfill material (FTBM) and controlled low-strength material (CLSM). The study evaluates the feasibility of using XLPE waste as a partial replacement for fine and coarse aggregates, aiming to reduce environmental burdens while maintaining or enhancing material performance.To this end, a comprehensive experimental program was designed to examine the effects of different XLPE replacement levels (0, 5, 10, and 15% by volume) and varying water to cement (W/C) ratios (0.45, 0.50, and 0.55) on the fresh, hardened, and durability properties of concrete, including slump, density, ultrasonic pulse velocity, compressive, tensile, and flexural strengths, as well as permeability, water absorption, and freeze–thaw resistance. FTBM and CLSM mixtures were also developed and tested to assess flowability, unit weight, setting time, thermal resistivity, and suitability for field applications. The leaching potential of XLPE waste was also evaluated to ensure environmental safety. Results demonstrate that incorporating XLPE waste can effectively reduce material density and thermal conductivity, contributing to lightweight and thermally efficient mixtures suitable for backfilling and non-structural applications. Optimal XLPE replacement levels were identified that maintain acceptable strength and durability while promoting waste valorization. To complement the experimental program, advanced machine learning (ML) techniques were employed to develop predictive models for estimating the unconfined compressive strength (UCS) of XLPE-modified construction materials. A comprehensive database was constructed by integrating experimental data from this study with published datasets, encompassing a wide range of mix design parameters. Multiple supervised learning algorithms, including random forest (RF), support vector regression (SVR), gradient boosting (GB), and artificial neural networks (ANN), were trained and optimized using cross-validation techniques. The ML models achieved high predictive accuracy, with the tree-based and ANN-based models showing the strongest generalization performance. Sensitivity analyses and feature importance evaluations revealed that the W/C ratio, XLPE content, and cement dosage were the most influential predictors of UCS. The integration of experimental and data-driven approaches in this research provides a robust framework for both understanding and optimizing the performance of XLPE-incorporated materials. The developed ML models enable efficient prediction of UCS, reducing the need for extensive laboratory testing and supporting rapid mix design optimization. To sum up, this study demonstrates the technical feasibility and environmental benefits of reusing XLPE waste in construction materials while advancing intelligent modeling tools for sustainable material engineering.