AI-Driven Digital Twin Framework for Safety Stock Optimization in Multi-Stage Manufacturing Systems

Safety stock optimization in multi-stage manufacturing systems remains a critical challenge due to demand uncertainty, production variability, and interdependencies across production stages. Traditional inventory control approaches often rely on static assumptions and limited adaptability, which restrict their effectiveness in dynamic industrial environments. This paper proposes an AI-driven digital twin framework for adaptive safety stock optimization in multi-stage manufacturing systems. The framework integrates real-time operational data, analytical inventory models, and machine learning-based predictive mechanisms within a synchronized digital twin environment. By continuously updating system states and forecasting uncertainty, the proposed approach enables proactive and dynamic adjustment of safety stock levels across multiple stages. Experimental evaluation using simulated multi-stage manufacturing data demonstrates that the proposed framework significantly improves service levels while reducing inventory holding costs and stockout occurrences compared to conventional safety stock policies. Statistical validation using paired t-tests confirms that the observed improvements are significant and robust under varying demand and lead-time uncertainty conditions.