Mission-critical supply chains, such as those supporting defense deployments, emergency medical response, and critical energy infrastructure, face ever-evolving disruptions from both cyber-physical attacks and unpredictable operational shocks. A single node or link failure can cascade through the network to severe delays, resource shortages, or even life-threatening consequences. Traditional monitoring approaches struggle to detect subtle anomalies in real time, while pure optimization engines lack the flexibility to learn from noisy telemetry. Therefore, we present a NeuroSymbolic Self-Healing framework for mission-critical supply chains that unites data-driven anomaly detection with provably safe recovery planning. An LSTM autoencoder is trained on benign operational telemetry to flag deviations via reconstruction error, while a lightweight symbolic rule engine encodes hard domain constraints to score risk and synthesize rerouting plans around compromised nodes. These neural and symbolic insights are fused into a hybrid metric for downstream classifiers that are further robustified through synthetic-data augmentation, GAN-based adversarial examples, and transfer-learning fine-tuning. Across extensive experiments on a Department of Defense (DoD) contracts dataset, our approach achieves state-of-the-art detection performance around 99% and demonstrates superior recovery rates that outperform purely neural or symbolic baselines in both anomaly detection and node recovery. Overall, the proposed framework provides real-time, explainable, and high-assurance self-healing for high-stakes supply chain networks.
