Advancing nuclear safety

Safety is a mandatory requirement for nuclear engineering applications in any industrial and medical fields, i.e., for the operation of nuclear power plants, nuclear medical and industrial devices, radioactive waste repositories, nuclear propellers, etc. Then, it is strategic for the development of nuclear engineering.

With a specific focus on nuclear facilities related to energy production, the present paper aims at providing perspectives on a number of issues and topics of safety, which are receiving attention from the researchers and practitioners.

Nuclear safety comprises all measures needed for nuclear facilities to be operated in normal conditions. Such measures aim at preventing accidents of technical, external and human (including malicious) origin, and mitigating the effects that any released radiation may have on the workers, the general public and the environment. For this, technical and organizational measures are put in place for all phases of the lifecycle of a nuclear facility, from design to manufacturing and construction, to operation and decommissioning.

In the current era of strong technological evolution and transition (e.g., energetic and digital), partly motivated by the concerns of climate change, new designs of fission-based nuclear power plants (Small Modular Reactors, SMRs, and Micro Reactors, MRs), and prototype fusion-based facilities for energy production are being put forward. Their licensing for deployment requires verification of the design objectives for operation and safety (Pioro, 2023). With regards to the latter, any new design for safety is expected to conform to the principles of defense in depth, which means it should contain: inherent safety features, passive safety features, specified procedural actions, actions of control systems, actions of safety systems, actions of complementary design features. In particular, passive safety systems are extensively used in most SMRs and MRs because of their simplicity and reliability. However, passive safety systems present some challenges with regards to the assessment of their reliability itself (Di Maio et al., 2021), like lack of data related to relevant functional phenomena, need to know performance under a wide range of conditions and difficult testability to verify the intended function that they must provide. This leads to a significant use of expert judgement and subjective assumptions in the reliability assessment of passive systems and the consequent need of identifying, modeling and propagating the associated uncertainties. This can be done by simulating numerous realizations of the system response under different operating conditions, within a functional failure-based approach based on the integration of Monte Carlo sampling and the mechanistic code of the system physical process, which however requires significant computational efforts. Thus, advanced sampling, empirical metamodelling (e.g., by Artificial Intelligence, AI), advanced sensitivity and uncertainty analyses need to be developed and validated, including efficient and sound Inverse Uncertainty Quantification (IUQ) analyses to find a characterization of the input parameters uncertainty that is consistent with the experimental data, while limiting the associated computational burden (Roma et al., 2022). Clearly, resorting to empirical, fast-computing metamodels for estimating passive systems failure probabilities and for carrying out uncertainty and sensitivity analyses can be very efficient computationally but requires that the regression error be carefully quantified (and possibly controlled), in order to reduce its impact on the quality of the final reliability assessment. Furthermore, the higher the input dimensionality (e.g., when dealing with time series data), the higher the size of the training dataset needed for accuracy, so that approaches to reduce the input dimensionality (e.g., Principal Component Analysis, PCA, or Stacked Sparse Autoencoders) should be applied.

For what concerns active and passive safety systems, it is important that their design and testing be advanced in a way to address hazards and vulnerabilities characteristic of the new reactors technological solutions, in terms of materials, components and systems therein adopted. For the passive systems (IAEA, 2009), in particular, also the effects of climate change on their functionality need to be considered in the expected time frame of operation (Sahlin et al., 2015; Vagnoli et al., 2018; https://utilitiesone.com/the-impact-of-climate-change-on-nuclear-power-plant-safety).

This requires also the advancement of deterministic and probabilistic safety assessment capabilities, to allow relaxing excessive conservatism for more realistic analysis (e.g., by Best Estimate Plus Uncertainty, BEPU methods; D’Auria, 2019; Sun et al., 2020), on one side, and verifying and validating (V&V) the quality of the methods used for the assessment, on the other side (Paul et al., 2016). These advancements should regard also the confirmation of the added value of integrated deterministic and probabilistic safety assessment (IDPSA, Zio, 2014; Di Maio et al., 2015; Heo et al., 2021) and dynamic probabilistic safety assessment (DPSA, Wiltbank and Palmer, 2021; Parhizkar et al., 2022), in certain cases and for certain scenarios to be definitely identified, and their final deployment based on the power of computational risk assessment (CRA, Sezen et al., 2019). Indeed, BEPU, IDPSA, DPSA and CRA are expected to contribute significantly to robust risk-informed decision making for nuclear safety in practice, by enabling the integrated consideration and treatment of the time-dependent interactions of the stochastic processes of hardware component failures, the deterministic responses of the system process, the effects of the control and operator actions, software and firmware. Yet, to bring these computational frameworks to industrial use requires further efforts to increase their efficiency for the generation of multiple scenarios of system behavior by combining advanced Monte Carlo simulation with AI metamodelling for guided design of experiment (DoE) and system response simulation. Also, transparent post-processing by clustering (e.g., by AI) of the multiple scenarios output is necessary for classification (e.g., by Machine Learning, ML) so as to render the results treatable and useable for risk-informed decision making. Finally, in very practical terms it is needed that the computational codes be user-friendly and easy to link with existing PSA/DSA codes.

Within the safety assessment of existing multi-units and future modular nuclear installations, two issues that are still open and, thus, deserving attention are multi-hazard and multi-unit risk assessment, for achieving a scientifically solid evaluation of the aggregated site risk (Yoo et al., 2021; Ming et al., 2022; Hochrainer-Stigler et al., 2023). This puts under the spotlight also the assessment of the risk contribution from external events, e.g., earthquakes and extreme meteorological events (also in relation to the climate change that is being experienced and which increases the relevance of the latter), and the resilience of nuclear power plants (Di Maio et al., 2022; 2023).

From another perspective, there are continuous advancements in sensor electronics for component and process state monitoring, AI analytics and machine learning ML algorithms for elaborating the monitored data, and computational power availability to fuel these elaborations (Zio, 2018). This exciting combination offers a stream of opportunities for advanced solutions of operation and maintenance of nuclear facilities, e.g., autonomous operation supported by digital twins (DTs) and data-driven condition-based or predictive maintenance (Coble et al., 2012; Hu et al., 2022; Zio, 2022; Kim et al., 2023; Kropaczek et al., 2023). Yet, the issues of the qualification and V&V of AI models and ML algorithms, which are typically black boxes, for their safe use is still very much open even if they were used “only” to assist the operators and decision makers, and not for autonomous operation (Huang et al., 2023). On one side, this calls for a proper treatment of the uncertainty in the predictions used in the decision making, e.g., by Bayesian neural networks and deep gaussian processes. On another side, the safety criticality of the application of AI models and ML algorithms in the nuclear field motivates research for improving the transparency and interpretability of the predictions so as to build trust on their use. Methods for injecting physical information in learning models and algorithms, post hoc sensitivity approaches and visualization techniques need to be further developed and used to provide interpretability from different perspectives, including explaining the learned input-output relation representations, explaining the individual outputs, explaining the way the output is obtained. Regarding digital twins, one challenge is the seamless integration of DTs with existing systems and processes. Ensuring that DTs can communicate and share data with existing systems and methods can be a complex task, as it may require the development of new interfaces and protocols. On the other hand, interoperability is one of the most promising features that DTs can offer. It is crucial to ensure that data is being shared effectively so that it can be used to support decision-making and control tasks to fully exploit this technology. Also, the role of operators and users in using and implementing DTs must be carefully considered.

Also, clearly the ongoing digitalization of the industry increases the attention to cybersecurity of nuclear installations (Kim et al., 2020). Indeed, cyber risk analysis of digitalized assets is still an immature field with unproven techniques due, in part, to the continuously changing threat environment. On the other hand, nuclear installations continue to digitalize and, thus, it becomes increasingly important to have techniques of analysis whose outcomes can support risk management decisions for implementation of prioritized security controls to prevent and mitigate cyber risk within the unique constraints of the nuclear industry (Ayodeji et al., 2023).

Given the critical role that safety plays for the development of nuclear power to the benefit of sustainable progress, the expectation is to develop significant advances, both in theory and in practice, in the areas discussed above.

Statements

Summary

Keywords

nuclear safety, risk assessment, active and passive safety systems, integrated deterministic and probabilistic safety assessment (IDPSA), dynamic probabilistic safety assessment (DPSA), digital twin (DT), uncertainty and sensitivity analysis (UA and SA), climate change

Citation

Zio E (2024) Advancing nuclear safety. Front. Nucl. Eng. 2:1346555. doi: 10.3389/fnuen.2023.1346555

Received

29 November 2023

Accepted

12 December 2023

Published

04 January 2024

Volume

2 - 2023

Edited and reviewed by

Sergei Dudarev, United Kingdom Atomic Energy Authority, United Kingdom

Updates

Check for updates

Copyright

© 2024 Zio.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Enrico Zio, enrico.zio@polimi.it

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.