"North Carolina State University (NCSU) Nuclear Engineering (NE) faculty will collaboratively support the Enabling Capabilities Consortium���s technical goals by developing new methods in two key areas: 1. Monitoring uranium depletion and plutonium production in advanced reactors to detect diversion of fissile materials from the civilian nuclear fuel cycle for military applications; and 2. Applying inverse modeling and uncertainty quantification (UQ) to enhance the analysis of diagnostic measurements used in nuclear incident response."

The goal of this project is to develop a consistent and rigorous verification, validation and uncertainty quantification (VVUQ) framework for artificial intelligence and machine learning (AI/ML) application in NE to evaluate, establish and enhance ML credibility. Comprehensive, systematic, and quantifiable VVUQ guidelines and metrics will be developed to establish ML predictive credibility by building trustworthy ML models. We will develop innovative uncertainty quantification (UQ) methodologies to quantify the approximation uncertainties in ML models, especially in generalized domains where ML models are used for extrapolation. We will leverage deep generative learning to develop applicable techniques for data augmentation to alleviate the data scarcity issue.

The Department of Energy National Nuclear Security Administration (DOE NNSA) Office of Defense Nuclear Nonproliferation R&D (DNN R&D) will select one consortium of institutions of higher education (IHEs) and DOE national laboratories to receive $25M over 5 years to conduct research supporting DNN's nuclear forensics mission. The consortium will focus on interactions between faculty, students, and national laboratory staff to build capability in critical areas including radiochemistry; geochemistry; nuclear physics, science and engineering; nuclear material science; shock physics; quantum-enabled sensing; and analytical chemistry.

Over the past decade, the use of artificial intelligence techniques in the field of health-monitoring has gained significant interest, especially for structures such as building and bridges. This project proposes development of an Artificial Intelligence (AI) framework for the data-driven condition monitoring of nuclear structural systems and equipment, where the vibration response is governed by multiple localized modes unlike that in buildings and bridges. Hence, techniques such as signal processing and pattern recognition will be employed to extract degradation-sensitive features. Degraded locations can potentially exhibit damage such as localized yielding, cyclic fatigue, or initiation of cracking. Moreover, such locations can at times go undetected by current inspection techniques. Therefore, this research proposes a framework, which utilizes sensor response to generate an AI database for predicting degraded locations and severity in nuclear structural systems and equipment. Degradation severity will be classified as minor, moderate, and severe, along with incorporation of uncertainty.

In this project, we will optimize the fuel cycle of a Westinghouse 4-loop plant with the goal of being able to complete a 24-month fuel cycle with the lowest fuel cycle costs. Some of the specific issues that will be considered are the use of the combination of gadolinia and IFBA poisons, and whether the cycle length can be obtained with current enrichment and burnup limits. In addition to the fuel cycle optimization, we will develop a methodology to optimize single fuel assemblies that have gadolinia and IFBA burnable absorbers to minimize pin peaking factors.

Probabilistic seismic hazard analysis (PSHA) is a key component in seismic design and performance assessment of engineering components. Accurate representations of rupture characteristics, wave propagation, and subsurface soil behavior are necessary to perform an accurate PSHA. However, in traditional PSHA, simplified empirical Ground Motion Models (GMMs) are used to estimate the ground motion levels. These GMMs neglect the inherent physical complexities in earthquake rupture and ground motion properties, such as slip heterogeneity, rupture directivity, and basin depth, among others. In addition, the paucity of ground motions recorded from large magnitude ruptures in the near-fault region, and from stable continentals regions (e.g. French Context) makes GMMs unsuitable for several contexts and applications. The objective of this research is to: 1) apply seismic ground motion methods to generate synthetic ground motions data set based on a real ground motion data set, 2) make use of synthetic ground motions to update GMMs using the Bayesian method already developed in a precedent collaboration (NCSU-EDF) and/or to develop a specific synthetic GMM (using only simulated motions) and/or to develop a hybrid GMM (using synthetic and real ground motions). The challenge of this research is to propose new alternatives to the PSHA based on simulated ground motions to complete the logic tree branches. The weight of logic tree branches of different simulated approaches and GMMs will be assigned to the validity of simulated and empirical models, with respect to the observed ground motions in regions under evaluation.

The objective of this proposal is to develop a novel Artificial Intelligence (AI)-based Process Control and Optimization (PC&O) methodology for Advanced Manufacturing (AM). Process-informed design, which has the potential of enabling transformative manufacturing operations, requires online PC&O. This project will fill the gap of lacking an online interaction mechanism in the process-informed design by providing the capability to intelligently control and optimize AM processes instead of the existing trial and error approach. To achieve this goal, AI-based control algorithms will be developed by employing Deep Reinforcement Learning (DRL). To reduce the computational expense with AM models, physics- informed Reduced Order Models (ROMs) will be developed. The AI-based control algorithms will employ the ROMs������������������ predictions to adaptively inform processing decisions in a simulation environment. In support of this research, an online training integrated capability will be developed based on the Multiphysics Object Oriented Simulation Environment Stochastic Tools Module (MOOSE-STM). Preliminary demonstration and validation of the developed PC&O methodology will be carried out with Idaho National Laboratory (INL)������������������s laser additive manufacturing system (i.e., LENS).

Transient fuel rod analysis addresses fuel rod response during non-Loss of Coolant Accidents (non-LOCA) accidents. The analysis provides time-dependent fuel temperature, rod internal pressure, and rod surface heat flux as input to Departure from Nucleate Boiling (DNB) Ratio (DNBR), fuel melt and cladding stress evaluations under reactor design conditions ranging from nucleate boiling to post-DNB heat transfer as well as high fuel burnup. This project is expected to develop and validate full core CTF models and fluid solutions including the more robust two-phase flow and high-resolution modeling capabilities than the existing Thermal-Hydraulic (T/H) design code.

The primary objective of this proposal is to leverage the recent development in Artificial Intelligence (AI) and Machine Learning (ML) for scientific computing to provide a more comprehensive mathematical representation of model uncertainty during Bayesian inverse Uncertainty Quantification (UQ). Inverse UQ is the process of quantifying the input uncertainties based on model fits to measured data, which is a crucial step to improve computational model accuracy. Model uncertainty has often been ignored due to the lack of satisfactory methods to represent it, which can cause over-fitting in inverse UQ. We plan to adopt Physics-Informed Machine Learning (PIML) and deep neural networks to build metamodels to characterize the model uncertainty. We will organize a workshop on ���AI/ML Applications in Inverse UQ for Nuclear System Modeling and Simulations��� by inviting individuals that participated in OECD/NEA PREMIUM and SAPIUM benchmarks, which are international projects that focused on inverse UQ. There is a keen interest from both the domestic and international nuclear communities to see progress in scientific ML-assisted inverse UQ. The proposed work can benefit the development of advanced nuclear reactors by enhancing the predictive capability of simulation models.

The major objective is to develop Physics-Informed Neural Networks (PINN) techniques to assist the optimization of Additive Manufacturing (AM) process parameters such as applied laser power, traveling speed and scan style, to achieve desired nuclear fuel thermal-mechanical properties and fracture toughness at engineering scale. The project will focus on two tasks: (1) developing and demonstrating efficient Physics-Informed Deep Learning (PIDL) algorithm to solve the underlying Partial Differential Equations (PDEs) in the AM process; (2) determining an optimal set of AM process parameters through optimization using the PINN model.