Argonne’s Nuclear Science and Engineering division is developing a framework and methods that use artificial intelligence (AI) and scientific machine learning (ML) techniques to improve safety modeling and simulation (M&S) of advanced nuclear reactors. The objective is to facilitate the development and deployment of advanced reactors by improving economics (through accurate safety margin predictions) and reducing the licensing burden (through improved uncertainty quantification). These methods have at their core the development of physics-guided reduced order machine learning models that capture the important physical phenomena and are accompanied by an uncertainty quantification.
Thermal fluid phenomena play major roles in advanced reactor safety. In this work, high-fidelity calculations are used to probe the micro-details of T/F phenomena and generate training data to develop reduced-order models (ROM) for more accurate reactor safety analyses and to better quantify the uncertainties associated with the analysis. The developed data-driven models are being integrated into modern transient safety analysis code SAM to facilitate the design, licensing, and safety analysis of advanced nuclear reactors. The figure shows our physics-informed machine learning workflow with application of a neural network model into SAM for multi-dimensional flow simulation.
Recent work is described in Surrogate Models.
Sensor Set Design
The system sensor set is an important factor that needs to be considered when developing improved maintenance procedures for cost reduction. Each procedure must be supported with the appropriate set of sensors to not only permit current equipment health status to be determined but also to predict going forward. The adjacent figure suggests how such a sensor set might be designed to accomplish this. Essentially different sensor configurations are cycled through, each time determining the utility of a configuration for assessing the status of equipment condition that is addressed by the maintenance procedure.
In application to the current fleet this process can identify the required additional sensors to augment the existing set to meet new maintenance procedures aimed at reducing equipment rounds and manual surveillance. In application to the advanced reactor the opportunities for realizing the ideal sensor set are greater since the design of the sensor set can be made an integral part of the plant design process and is not constrained by physical limitations on equipment layout such as may be present in an existing plant.
An application of sensor set assignment for an excising nuclear plant is described in Sensor Assignment. There the problem of upgrading an existing señor set needed to deliver a required diagnostic capability is described.