Nouveau Edf

Bridging the gap between traditional and dynamic Probabilistic Safety Assessments - H/F

  • Internship
  • Palaiseau (Essonne)
  • Master, Titre d'ingénieur, Bac +5
  • Études / Statistiques / Data

Description de l'offre

1 Context

EDF is one of the largest producers of electricity from nuclear power in the world. Safety of its nuclear fleet is a top priority of the company.

In France, Probabilistic Safety Assessment (PSA) [1] is an obligatory [2] part of the safety demonstration for nuclear installations. Recent changes in the legislation [3] increase the exposure of PSA results to the general public. Thus, it is very important to assure the high degree of realism of PSA results.

Traditionally Probabilistic Safety Assessment uses transient calculations (ther- mal hydraulic or severe accident) to construct the accidental sequences. These sequences and the missions of the safety systems are later used to construct Event Trees/Fault Trees [4] of PSA models. Event trees represent accidental sequences and fault trees represent safety system missions.

The quality and the realism of the PSA largely depend on the initial choice of the transient calculations (see for example Sections 5.53-5.58 of [5]). Ideally the construction of a PSA model and the choice of the transient calculations have to be done in an iterative way. Unfortunately due to the the time and budget constraints the iterative process is avoided in favour of the conservative choice of transient data. That in turn, leads to over-conservative PSA results.

There has been over 30 years of development of advanced methods for risk assessment. [6] At EDF R&D, we have an experience of development and ap- plication of advanced methods for Probabilistic Safety Assessment [7]. These methods explicitly bring temporal evolution of the system into the considera- tion. Therefore these methods should provide by far larger degree of realism of the PSA. Unfortunately the industrial applications of these methods are yet to appear due to many reasons.

2 Internship Objective

The internship has an ambitious goal to bridge the gap between the dynamic and traditional PSA in order to improve the quality of PSA models. A short or mid term industrial application of this methodology should be guaranteed by the conservation of traditional Boolean PSA quantification tools [8]. To achieve this goal the selected candidate should

perform an application of dynamic methods [10] for PSA level 1 or PSA level 2 test case using existing tools [9].

perform the Analysis/PostProcessing/Clustering of the results/Definition of sequences and success criteria

Translation of the success criteria into the functional diagrams and inte- gration into the Boolean PSA models using e.g. Andromeda tool [12]

3 Environment

Internship will take place at EDF Lab Paris Saclay in the department PERI- CLES of EDF R&D.

4. Duration

Internship duration is 6 month.


[1]  Probabilistc Risk Assessment, Wikipedia,

[2]  Arrete du 7 fevrier 2012 fixant les regles generales relatives aux installations nucl eaires de base, Article 3.3

[3]  Arrete du 11 janvier 2016 portant homologation de la decision No 2015- DC-0532 de l’Autorite de surete nucleaire du 17 novembre 2015 relative au rapport de suˆrete des installations nucleaires de base, Sous-section 4 : Analyses probabilistes de surete

[4]  Event tree analysis, Wikipedia, Fault Tree analysis, Wikipedia

[5]  Development and Application of Level1 Probabilistic Safety Assessment for Nuclear Power Plants, Specific Safety Guide, No. SSG-3, IAEA

[6]  T. ALDEMIR, “A survey of dynamic methodologies for probabilistic safety assessment of nuclear power plants”,Annals of Nuclear Energy,52, 113, (2013).

[7]  PSA Level 2 with dynamic event trees. Lessons learned and perspectives. 

[8]  Risk Spectrum software for PSA

[9]  RAVEN - Risk Analysis Virtual Environment

[10] Coupling of RAVEN and MAAP5 for the dynamic event tree analysis of nuclear power plants 

[11] Andromeda : Advanced PSA modeling

[12] Automated Generation of Event Trees From Event Sequence/Functional Block Diagrams and Optimisation Issues. Friedlhuber T., Hibti M. and Rauzy A., PSAM 2013.

Profil recherché

An ideal candidate should have the following skills:

Solid knowledge of Linux environment and Python (or any other object oriented) programming language
experience in TH calculations using any TH/SA code (CATHARE, RELAP, MAAP,MELCORE etc.) is desired but not required.
knowledge of statistics
critical thinking and the good level of autonomy
A PhD possibility

Based on the obtained results the internship can be extended into a PhD project starting from the end of 2019/beginning 2020.