An uncertainty-aware 4D/5D digital-twin framework for cost estimation and probabilistic schedule control: field and benchmark validation from a Texas mid-rise project

Persistent cost and schedule overruns in building projects highlight the limitations of document-based estimating and deterministic Critical Path Method (CPM) scheduling under uncertainty and dynamic site conditions. This study proposes an uncertainty-aware 4D/5D digital-twin framework that integrates BIM with probabilistic schedule control and data-driven progress sensing. The main contribution is a Bayesian–Monte Carlo CPM forecasting approach supported by automated 5D cost integration and field-updated progress signals. The framework combines transformer-based NLP for cost-item classification (weighted F1 = 0.883), scan-to-BIM reconciliation using photogrammetry and LiDAR, and computer-vision-based activity recognition (micro-accuracy = 0.891). These inputs continuously update P₅₀ and P₈₀ schedule forecasts through Bayesian inference. A deep reinforcement learning (DRL) module provides human-in-the-loop decision support for crew allocation and overtime mitigation. The approach was evaluated on a 5–8 story mixed-use mid-rise project in Texas. Results show a 43% reduction in estimating labor effort and project completion aligned with the probabilistic P₅₀ forecast of 128 days. DRL-assisted scheduling reduced overtime by approximately 6%. Although validated on a single project, the results demonstrate the feasibility of uncertainty-aware 4D/5D digital twins for predictive building project control.