A simulation-based digital twin model for data-driven decision optimization

Digital twins (DTs) are digital representations of physical systems that run in real-time using enterprise data. This work introduces a simulation-based DT framework that supports dynamic manufacturing decision making during a supply chain disruption by combining real-time data with predictive machine learning algorithms. A discrete event simulation (DES) models a hypothetical manufacturing system, which includes condition-based maintenance, worker shifts, inventory policies, and variable job sequences. A supply disruption occurs in the DES, where vital materials are postponed for several weeks. Different mitigation strategies are simulated. Several machine learning models are trained using the simulation results to predict the financial impact of every mitigation strategy. This article presents a novel examination of a DT for manufacturing by simulating real-time updating of a DT by inputting the current state when the required material finally arrives. The results indicate that the optimal set of mitigation strategies, as evaluated via the machine learning model, varies based on different current states. Without a DT, the manufacturer should add an additional worker and implement overtime hours to recover from the supply disruption. With a DT, the manufacturer should add a forklift, increase its reorder quantity, and implement a little overtime if the supply delay lasts 32 days. If the supply delay lasts 48 days, the manufacturer should add a forklift, add an additional worker, and implement overtime hours. The simulation and machine-learning integrated method enables companies to dynamically change their operations depending on current facility conditions, increasing their resilience and profitability.