Key points are not available for this paper at this time.
• Autoencoder-predictor model that forecasts end-of-life pathways from material compositions. • Additional input features include circular origin and dismantlability potential. • Findings show strong model robustness and semantic material learning. • Identifies key material enablers of circularity in construction. • Enables material traceability and provides probabilistic guidance aiding decision-making. Construction and demolition activities are a major source of industrial waste, yet material end-of-life circularity and traceability remain poorly understood. This study addresses the challenge of forecasting end-of-life pathways from building material compositions by introducing a hybrid autoencoder–predictor model. The approach encodes material profiles into continuous embeddings and considers additional design parameters to predict probable end-of-life scenarios. Trained on 8,680 environmental product declaration-derived samples, the model achieved a mean error of 0.01%, MAE of 3.3%, RMSE of 6.2%, and R² = 0.82. Results identify key materials that enable recycling and highlight the importance of design-for-disassembly and recycled content in guiding end-of-life decisions. Besides, findings also reveal that end-of-life reporting practices are somewhat inconsistent, especially for reuse, filling, reconditioning, and composting, highlighting opportunities for policy and reporting standard enhancements. By enabling probabilistic forecasting of end-of-life outcomes, this tool supports transparent material traceability and informs procurement, policy development, and sustainable design.
Vergés et al. (Wed,) studied this question.