Enhancing U.S. Manufacturing Resilience: A Reinforcement Learning Approach to Real-Time Supply Chain Rerouting and Predictive Asset Lifecycle Management

• Billy Elly

  Lautech

  Author

The resilience of U.S. manufacturing supply chains has emerged as a critical national priority, yet traditional risk management approaches—including supplier diversification and inventory stockpiling—have proven inadequate against the compounding effects of geopolitical tensions, climate disruptions, and demand volatility. This research addresses the gap between static risk mitigation strategies and the dynamic, real-time decision-making requirements of modern manufacturing networks by developing a reinforcement learning-based framework for integrated supply chain rerouting and predictive asset lifecycle management. The study employs a bilevel deep reinforcement learning architecture combining Soft Actor-Critic optimization with attention-enhanced Long Short-Term Memory networks for remaining useful life prediction, validated through a simulation environment modeling a 75-asset manufacturing network with 12 supplier nodes and 8 distribution centers. The proposed framework achieved 89.4% accuracy in disruption prediction and demonstrated a 37.2% reduction in unplanned downtime, with computational inference times averaging 1.2 seconds for rerouting decisions. These findings establish a replicable, data-driven approach to supply chain resilience that bridges predictive analytics with real-time adaptive control, offering actionable insights for manufacturing practitioners and policymakers seeking to operationalize resilience under uncertainty.

• PDF

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.