Adaptive job scheduling and resource allocation for industrialized construction processes using digital twin and reinforcement learning

Abstract

Construction processes can be disrupted by unforeseen operational disturbances, such as waiting for shared machines, insufficient workers, material unavailability, and workstation idleness. Such disturbances are difficult to mitigate through predefined schedules alone, as effective responses require real-time diagnosis of production states and coordinated adjustments to jobs, workers, and machines across multiple workstations. To address this challenge, this study proposes a disturbance-triggered job scheduling and resource allocation (DT-JSRA) framework that integrates a digital twin with deep reinforcement learning (DRL). The digital twin continuously synchronizes computer-vision-derived information on workstation activities, worker allocation, machine status, material availability, job progress, and disturbance flags, providing a structured and up-to-date representation of the production system. Based on these synchronized states, the scheduling problem is formulated as a Markov decision process, and a DRL agent learns to select adaptive job dispatching, worker allocation, and shared-machine allocation actions. This study contributes a knowledge-constrained decision-support framework that reconceptualizes industrialized construction scheduling as a disturbance-triggered, multi-resource, and multi-workstation coordination problem. Construction-specific knowledge, including task-resource consistency, human–machine safety, resource availability, allocation priority, transfer-time effects, and production-oriented disturbance logic, is embedded to ensure operationally feasible decisions. The framework was validated using a real-world wooden floor panel production case. Compared with practical construction records and several benchmark methods, the trained policy reduced cumulative non-productive time and completion time under disturbance-specific scenarios, demonstrating its effectiveness in dynamic scheduling and disturbance response through anticipatory resource coordination.