Real-Time Implementation of Nuclear Reactor Models on Embedded GPU Computing Platform: A Pathway to Hardware-in-Loop Control System

This work presents a novel approach for parallel implementation of the neutron diffusion model (NDM) and the neutron telegraph model (NTM) (partial differential equation models) of a nuclear reactor, on an embedded Graphics Processing Unit (GPU) platform, NVIDIA Jetson Orin NX, using Compute Unified Device Architecture (CUDA). These control-oriented models are fundamental for capturing the dynamics of neutron transport in a nuclear reactor and are extensively used to design various control strategies. The computational requirements of these models are significant for real-time simulation or embedded applications. The proposed approach is based on designing algorithms for parallelizing the Separation of Variables (SoV) method for solving these models in real-time and leveraging the massively parallel architecture of the Jetson Orin’s embedded GPU to accelerate their numerical solution. The embedded GPU implementation capability can significantly enhance reactor control responsiveness, support digital twin deployment, improve fault detection, and enable autonomous operation. Unlike classical GPU PDE solvers that parallelize values using neighbouring grid points (stencils), the proposed method prallelizes the full SoV directly across both spatial and temporal dimensions, incorporating warp-level modal reduction and hardware-specific mapping. The focus is on enabling efficient utilisation of GPU cores for concurrent computations, providing a high performance computational framework. Performance evaluation is carried out by comparing the CUDA-based parallel implementation with a baseline serial implementation in C programming. Results demonstrate a substantial reduction in execution time and improved computational efficiency, while maintaining notable numerical accuracy on the embedded GPU platform. This makes the proposed approach suitable for real-time Hardware-in-Loop (HiL) implementation of the reactor models and control applications, where compact, energy-efficient, and high-performance computing is essential.