In the wide framework of nuclear power plants, the dynamic probabilistic risk assessment could answer the time dependence deficiency of the event tree and fault tree analysis. The basic event tree approach relies on expert's pre-constructed accident sequences, without exploring the real nature of an accident scenario in which the time dependence on the occurrence of a selected number of events could strongly affect the accident sequence. Conversely, effects of events' timing can be studied adopting a Dynamic Event Tree approach.
The development of a Dynamic Event Tree methodology requires the integration between a system code capable of replicating an accident scenario, and a logic-driver code able to generate the event tree sequence, triggering plant safety systems and managing other relevant events throughout the simulation.
The project aims to develop a tool for the integration and communication between a system code (MELCOR) and a logic-driver code (RAVEN) to perform a dynamic probabilistic risk assessment of nuclear and other industrial plants. The tool will be developed using Python as programming language because of its versatility in transferring and handling huge amounts of data.
RAVEN (Risk Analysis Virtual Environment) is an open-source software tool developed at the Idaho National Laboratory (INL) to act as a control logic driver and post-processing tool for different applications. MELCOR is a fully integrated severe accident code that simulates thermal-hydraulic transients and self-consistently accounting for aerosol transport in industrial facilities and reactor cooling systems.
The developed tool will be used to perform dynamic event tree studies during accident transients in fission and fusion reactors. The new coupling between these two codes will provide a wide range of nuclear power plant risk assessment analyses and could establish a best practice.
A major limitation of traditional Event Tree (ET) basic methods is the poor consideration of physical, temporal, and spatial dependencies [1]. Each element of the tree is previously hypothesized considering the previous experience and the designer's ability to model the system. In addition, the operator's response, and the difficulties that the operator may suffer during an accident scenario, are not accurately represented [2].
To overcome these limitations, a new PRA methodology called Dynamic PRA has been developed. DPRA methods combine stochastic analyses with plant simulation codes to determine the risk associated with the operation of complex systems such as nuclear plants [3]. Compared to the classic PRA, the DPRA can evaluate the risk with greater precision through a greater resolution of the physical space, of time as a variable, and of the dynamic sequence of events; without introducing excessively conservative assumptions.
In particular, we will no longer use a static event tree but move on to a dynamic event tree DET in which each branch coincides with the occurrence of a certain event simulated in this case by the code itself, and not previously assumed. Great importance must be given to the exact description of the system and to the reliability analysis of its components. However, it must be considered that an excessively precise description of the plant and an extremely complex DET require high time and computational skills. An optimum between the real description of a process and not high computational time must be achieved.
Dynamic Probabilistic Risk Analysis improves the standard PRA by introducing "Time" as a main parameter [4]. An accident sequence evolves through different states and the intervention time of an emergency system is crucial to the success or failure of these systems. Therefore, a complete and exhaustive risk analysis must account for the chance of emergency systems to intervene at different time frames during an accident sequence.
The DPRA, in general, and the DET methodologies, in particular, are adopted to take the timing of events as the main variable to understand how the system could react in the different circumstances and becomes particularly relevant when uncertainties in complex phenomena are considered.
To better understand the innovation of this project, we can compare the DPRA and CPRA defining an issue space represented by all the possible final states of the process. The issue space is subdivided into two mutually exclusive regions, one in which the NPP has been affected by Core Damage (CD) and the other in which the safety systems have intervened successfully (without CD). A CPRA evaluates the Minimal Cut Sets (MCSs) that can lead to CD, while a DPRA expands the issue space adding new degrees of freedom (e.g. timing of events) and identifies a new set of final states. Theoretically, the CD regions generated by both approaches should coincide although conservative assumptions (e.g. success criteria) and the impact of timing of events and physical nature simulated by a DPRA, alter the overlap of the CD region. Also, the different approach of DPRA that focuses on either accident progression or failure propagation separates further the two-CD regions, given that PRA focuses on both accident progression and failure propagation together.
This new methodology has never been done before using MELCOR as a system code, offering a wide range of possibilities.
Riferimenti:
[1] A. Alfonsi, "Dynamic Event Tree Analysis through RAVEN," in International Topical Meeting on Probabilistic Safety Assessment and Analysis, Columbia, 2015.
[2] A. Mosleh, "PRA: a perspective on strenghts, current limitations, and possible improvements," Center for Risk and Reliability, p. 10, 2014.
[3] D. Mandelli, "Mutual Integration of Classical and Dynamic PRA," Nuclear Technology, p. 14, 2020.
[4] Y. Vorobyev, "Development and Application of a genetic algorithm based dynamic PRA methodology to plant vulnerability search.," in International Topical Meeting of Probabilistic Safety Assessment and Analysis, Wilmington, NC, 2011.