On the value of data fusion and model integration for generating real-time risk insights for nuclear power reactors

Abstract The integration of data science, analytics, and model-based reasoning provides a mechanism for enhanced understanding of systems and improved decision-making, but its potential has not been thoroughly explored for improving the safety and operational efficiency of nuclear power reactors. Nuclear power owners, operators, regulators, and researchers have made significant investments in probabilistic risk assessments, numerical models, computational simulations, and development of databases that capture industry-wide component performance and operating experience. The nuclear industry is relatively unique in the size, variety, scope, technical sophistication, and quality of available data and models that capture system performance under normal operations and a wide-range of adverse event conditions. However, to date, these resources have been used in a largely static and siloed manner. Data science, analytics, and model-based reasoning provide a mechanism for fusing diverse data sources and models to develop new insights on a variety of topics. Of particular interest to the nuclear industry is the ability to leverage these resources to enhance the safety and operational efficiency of nuclear power reactors. In this paper, we present a challenge to the nuclear energy community to better leverage the existing investments in data and models to enhance decision-making. In particular, we propose that integration of recent advances in data science, analytics, and model-based reasoning provides a valuable opportunity for the nuclear industry to build upon their existing investments by accessing the power of modern data integration and risk assessment tools. We begin by describing common data and model resources available in nuclear power operations and safety analysis and offer commentary on the potential power of using Bayesian networks as a structured framework for data fusion and model integration. Then we present two example problem structures for modeling risk-informed operational decisions using heterogeneous data and models to provide simple illustrations of the means by which information streams can be leveraged in real-time to provide online assessment of risk and to increase diagnostic capabilities. Illustrative model formulations are presented for decisions under adverse events and normal operational contexts. We conclude by identifying research activities that will enable the transformation of decision-making by applying new computational and modeling tools to existing data and models.