There are four main research thrust areas in ARDOR Lab: computational reactor physics, multiphysics modeling and simulation capabilities, advanced reactor design and fuel cycle analysis, and machine learning for reactor operation and maintenance.

Computational Reactor Physics

Our innovation lies in the integration of high-fidelity codes that aim to reproduce the physics at its maximum and fast-running low-fidelity codes that are suitable for routine safety calculations. The simulation capabilities that we produce satisfy the current nuclear industry needs and performance requirements.

Selected Research Projects

Multiphysics Modeling and Simulation Capabilities

Our goal is to develop multiphysics predictive capabilities to capture the interactions between different physics phenomena at various scales in nuclear power plants (NPPs). The research focuses on establishing tight, flexible, and massively parallelized coupling mechanism oriented towards high-fidelity reactor physics and thermal-hydraulics codes.

Selected Research Projects

Advanced Reactor (AR) Design and Fuel Cycle Analysis

The economic and safety benefits offered by ARs cannot be fully realized until its performance is coupled with appropriate waste management and/or spent fuel reprocessing and recycling strategies. The reactor performance largely depends on their specific core and system design, which presents a large, non-linear variable space and thus infeasible to find optimal solution through brute-force or manual design methods. We aim to produce “smart” designs of ARs by leveraging optimization theory, sensitivity and uncertainty (S/U) analysis, and multiphysics M&S capability.

Selected Research Projects

Artificial Intelligence (AI) and Machine Learning (ML) in Nuclear Engineering

We have been actively incorporating AI/ML to solve challenging real-world nuclear engineering problems. In a high-profile ARPA-E project, advanced data analytics are being applied to aggressively reduce the high cost associated with AR operation and maintenance (O&M) by continuously detecting and characterizing anomalies in O&M and planning for corrective actions based on the outcome of the prognostic procedure. At the heart of the predictive maintenance (PdM) framework lies the deep Bayesian Neural Network (BNN), which can learn from real-time equipment monitoring data and make predictions of the remaining useful life (RUL) and its uncertainty. High-fidelity neural network surrogate models were developed and tested to precisely control the distribution of the crud deposition in the LWR through coupling with optimization algorithms.

Selected Research Projects