NuTwin: A Zero-Trust, Confidential Cloud Architecture for Real-Time Nuclear Reactor Digital Twins Using Physics-Informed Neural Networks

Abstract Real-time telemetry diagnostics and predictive anomaly tracking in nuclear reactor cores present a fundamental computational dilemma: high-fidelity numerical neutronics and thermal-hydraulics solvers are too computationally prohibitive for sub-second edge evaluation, whereas conventional black-box deep learning architectures violate physical conservation bounds and lack operational determinism. Simultaneously, transmitting safetycritical nuclear operational data to public cloud infrastructure introduces acute cyber-physical security vectors that breach strict operational air-gaps. This paper introduces NuTwin, a zero-trust, confidential cloud architecture designed for real-time nuclear reactor core digital twins. NuTwin couples a Physics-Informed Neural Network (PINN) surrogate model—constrained directly by point reactor kinetics equations (PRKE) with six delayed neutron precursor groups and coupled thermal-hydraulic reactivity feedback—with hardware-enforced cryptographic isolation. Live operational telemetry, simulated via Python Point Reactor Kinetics (PyRK), is streamed over unidirectional mutual TLS (X.509) connections into AWS IoT Core and processed exclusively inside an ephemeral, attestation-verified AWS Nitro Enclave via an internal virtual socket (vsock) channel devoid of persistent storage or external network interfaces. Experimental validation demonstrates that NuTwin bounds the physical residual loss below 1.4 × 10−4 , achieves sub-second inference latency (42.8 ms per step), and guarantees zero data-leakage surface under zero-trust operational constraints. Index Terms—Digital twins, physics-informed neural networks, confidential computing, AWS Nitro Enclaves, point reactor kinetics, nuclear cyber-physical security, real-time anomaly detection.