Andromeda : Advanced PSA modeling

Transcription

1 Andromeda : Advanced PSA modeling A PSA management tool EDF R&D - Edgemind October 6, 2017 UK s 8th Risk Forum Next Generaion Risk Analysis 1 / 1

2 2 / 1

3 PSA Models are complex!! Me starting to use PSA models for the first time 3 / 1

4 PSA Models are complex!! Me starting to use PSA models for the first time 3 / 1

5 An event tree 4 / 1

6 A fault tree 5 / 1

7 Increasing complexity and transparency needs Difficulties Size of the models beyond the capacity of the reference tools Increasing Computational complexity and running times no more appropriate for industrial use The models become more and more obscur for neophytes Model maintenance difficulties The development of the models become more tense (work in serie) 6 / 1

8 Increasing complexity and transparency needs Difficulties Size of the models beyond the capacity of the reference tools Increasing Computational complexity and running times no more appropriate for industrial use The models become more and more obscur for neophytes Model maintenance difficulties The development of the models become more tense (work in serie) Scientific and technologic obstacles We are at the frontier of untractable problems (NP-hard) Need to explore new concepts Get out from the black box to a wide open world Make a bridge to complexity science Allow more flexibility for model exchange and benchmarking 6 / 1

9 Open PSA Open PSA initiative Antoine Rauzy Woody Epstein 7 / 1

10 Modular PSA NPSAG workshop Transparencey And User Friendly PSAs & Lamda Mu 8 / 1

11 Modular PSA NPSAG workshop Transparencey And User Friendly PSAs & Lamda Mu Model 8 / 1

12 Modular PSA NPSAG workshop Transparencey And User Friendly PSAs & Lamda Mu Model 8 / 1

13 Andromeda : a set of modeling and management tools Thesis : Thomas Friedlhuber (Rauzy & Hibti) 9 / 1

14 Andromeda : a set of modeling and management tools Thesis : Thomas Friedlhuber (Rauzy & Hibti) Diagnosis Visualisation Evaluation Documentation Modeling 9 / 1

15 Andromeda : a set of modeling and management tools Thesis : Thomas Friedlhuber (Rauzy & Hibti) Comparison Merge Diagnosis Cartography Visualisation Andromeda Core Fault-Event Trees Evaluation Documentation Version Management Diagram Modeling 9 / 1

16 Andromeda : a set of modeling and management tools Diagnosis Cartography Comparison Merge Visualisation Version Management Andromeda Core Fault-Event Trees Evaluation Documentation Diagram Python (jython inside) API Shell APPS Clojure (standalone outside) Modeling 10 / 1

17 A global integrated vision Quantification engine Sequence Diagrams calc Master FT calc MINCS apply approximations calc sensibility calc contributions... link Variant Manager TOP Gate Quantitative Risk Assessment: Quantify link link Basic event BE1 Exponential Distribution Markov Chain can be analysed in here event event end state event end state Internet Intranet link link link link System diagram Simulation 11 / 1

18 12 / 1

19 A global integrated perspective 13 / 1

20 Integrated wiki documentation 14 / 1

21 Complex model comparison : event trees, fault tree and event sequence diagrams 15 / 1

22 Functional diagrams 16 / 1

23 Accessibilité : INTERNE Document issu de la GED DPI Nucléaire le 07/12/2016 RDU : D NOTE D'ETUDE EPS EPR BK - ANALYSE QUALITATIVE DES SÉQUENCES DE PERTE DE REFROIDISSEMENT DE LA PISCINE DE DÉSACTIVATION CNEN_SNE Référence: D ,QGLFH A 3DJH60 35 ::)-*IK-(E=#'#-(K(? &HGRFXPHQWFRQWLHQWGHVLQIRUPDWLRQVVHQVLEOHVUHOHYDQWGXVHFUHWHWMXULGLTXHPHQWSURWpJpHV,OHVWUpVHUYpjO XVDJHH[FOXVLIGHVSHUVRQQHVGpVLJQpHVFRPPHGHVWLQDWDLUHVGXGRFXPHQWHWRXDXWRULVpHVj\DFFpGHU,OHVWLOOpJDOGHSKRWRFRSLHU GLVWULEXHUGLYXOJXHURXGXWLOLVHUGHWRXWHDXWUHPDQLqUHOHVLQIRUPDWLRQVFRQWHQXHVGDQVFHGRFXPHQWVDQVDFFRUGGXVHUYLFHpPHWWHXU &RS\ULJKW(')6$±&HGRFXPHQWHVWODSURSULpWpG (')6$ 2016 Variant management DOCUMENT NON APPROUVE 17 / 1

24 A script language à la python dict = {" " : "_",">":"S","<":"I", " ":"M","\"":" Sec "} # Incomplete list def correct_id ( model ): list_ids = model. get_global_elements (); pattern_reference = re. compile (".* Reference ") for element in list_ids : name = element. get_name () for key, value in dict. iteritems (): if key not in name : continue else : new_char = dict [ key ] new_name = name. replace (key, new_char ) mtype = element. get_type () print " dealing with " + name if len ( new_name ) > 20 \ and mtype is not " Sequence " \ and mtype is not " Analysis " \ and not pattern_reference. match ( mtype ): print " new_name " + new_name + " Has more than 20 characters " print " You have to modify your dictionary accordingly " sys. exit () element. set_name ( new_name ) print name + " modified " + "( character " + key + " replaced by " + new_char + ")" 18 / 1

25 19 / 1

26 Maps and Services Tesnim Abdellatif 20 / 1

27 Maps and Services Tesnim Abdellatif 20 / 1

28 Following Dependencies From parameters to consequences Search and reach dependencies for validation processes Understand dependencies for diagnosis purposes Use cartography to break computational complexity split analysis cases use modularity use network metrics ans spacialisation 21 / 1

29 INEC, CEI and CDG Networks 22 / 1

30 Networks for automatic parallel analysis specifications 23 / 1

31 Analysis cases generation 24 / 1

32 Split huge cases 25 / 1

33 26 / 1

34 PSA project development In theory Initiators Fire Internal Flood Seismic other Level 1 PSA Level 1 ET model for core damage states Core Damage States (CDS) Level 1+ Interface Containment system ET (CSET) Plant Damage States (PDS) Level 2 Containment Phenomenological ET (CPET) 27 / 1

35 PSA project development In theory In the real life Initiators Fire Internal Flood Seismic other Level 1 PSA Level 1 ET model for core damage states Core Damage States (CDS) Level 1+ Interface Containment system ET (CSET) Plant Damage States (PDS) Level 2 Containment Phenomenological ET (CPET) 27 / 1

36 An extension of the project boundaries New boundaries for a version-controlled PSA model Include LOG files (logbook) Include peripheric documentation (assumptions, links... ) Include Diagrams for qualitative sequence analysis Include scripts if massive modifications were applied Include any material that served for some version Characteristics of a reproducible PSA The possibility to keep track of all the history of the model and its genesis Garantee a better quality assurance Enhance awareness of the modeling, assumptions, theoretical arguments, Better modeling principles 28 / 1

37 A decentralized model A decentralized model Need to Work in parallel in a decentralized manner Need to share information between developpers [and users] Need to work on a regularly updated basis 29 / 1

38 A decentralized model A decentralized model Need to Work in parallel in a decentralized manner Need to share information between developpers [and users] Need to work on a regularly updated basis A decentralized model Quality assurance : State of the art of Best Software Quality Assurance Model documentation and progressive logbook Be informed of the relevant updates Logs and history trackers Working on integraded or splited models 29 / 1

39 Version management 30 / 1

40 A model 31 / 1

41 Model merge 32 / 1

42 33 / 1

43 Computational Complexity breaking More advanced computation parallelisation Now the fractionning is up to the analyst using intuitive heuristics A priori computation of complexity metrics for model pieces Clustering and web-based computation Results Analysis and understanding Tools for smart representation of results and insights (CRA : Communicating PSA results to non experts Open PSA workshop 2016) Diagnosis tools for a better understanding of accident scenarios 34 / 1

44 Complex/simplex Networks "The more you refine your model the more you lower your risk" Antoine Rauzy Yesterday 35 / 1

45 Complex/simplex Networks "The more you refine your model the more you lower your risk" Antoine Rauzy Yesterday Figure: LOCA example 35 / 1

46 Excentric components using different spacialisation Excentric components Some components are excentric in the sense that they are linked to many other following the same direction Many main components of the support systems fall in this category 36 / 1