Explainable AI-integrated multi-objective optimisation for circular material routing in modular construction

This is the authors' abstract. We don't add key points for this paper.

The rising digitisation of the construction sector has increased the demand for automated, data-driven decision-making in BIM-enabled operations. However, material routing decisions in modular construction are still primarily manual or based on opaque optimisation models, which limits their integration with construction automation systems. To solve this, the Explainable AI-Integrated Multi-Objective Optimisation (XAIOM) framework is presented, which allows for automated material routing straight from BIM-derived data. The system combines machine learning (ML), SHAP and LIME explainability, non-dominated sorting genetic algorithm II (NSGA-II) optimisation, and Monte Carlo uncertainty analysis into a single pipeline. Explainability is built into the optimisation process to aid the development of Pareto-optimal solutions based on feature contributions. Validation of a modular residential construction in Finland shows clear trade-offs between CO₂ reduction and economic value under uncertainty. The findings confirm that the framework enables transparent, reliable, and scalable automated decision-making, advancing AI-driven construction automation by linking BIM-based digital models with intelligent optimisation workflows.