A Verifiable Digital Twin-Aided AI (DTAI) Agent-Based Production System Framework With Causally Consistent Symbiotic Simulation

Abstract

Generative AI enables agent-based smart manufacturing to achieve flexible and reactive decision-making with pre-trained models. However, AI agents remain weak in plan verification and are vulnerable to perception errors caused by dynamic environments and inconsistent Industrial Internet of Things data. These limitations critically undermine the effectiveness of AI agent planning in complex production environments. To address the limitations, this study proposes a digital twin-aided AI (DTAI) agent-based production system, in which digital twins provide a causal data provisioning and simulation layer to support situation awareness and decision validation. A causally consistent symbiotic simulation approach is further introduced to align simulation results with actual production processes while reducing monitoring overhead. Controlled experiments show that the proposed approach achieves a higher rate of optimal scheduling while reducing monitoring overhead with production increase, and lowers computational cost compared with alternative digital twin synchronization designs. A large-scale aerospace manufacturing case study demonstrates that DTAI-based scheduling with consistent symbiotic simulation can largely reduce rescheduling rate by identifying and eliminating most resource conflicts, which further improves multi-objective production efficiency, including equipment usage, makespan, etc. These results indicate that the proposed framework provides a practical and verifiable solution for deploying AI agent-based intelligent manufacturing systems.