Folien Für SARJ-98 Tokio

Monte Carlo Dynamic Event Tree + MELCOR

The MCDET stochastic module is developed at Gesellschaft für Anlagen- und Reaktorsicherheit (GRS) mbH

An Example of a Stochastic Module*) coupled with an Integral Code#) for PSA Level 2

Martin Sonnenkalb#) GRS Cologne, Germany

Joerg Peschke, Martina Kloos, Bernard Krzykacz*) GRS Munich, Germany

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 1 OECD-WS PSA 2, March 29-31, 2004 Content of Presentation

 Examples of MELCOR Application for German LWR

 Probabilistic Dynamics  What does it mean - Probabilistic Dynamics?  Kinds of Uncertainty within the framework of PSA  Limitations of Event Trees in current PSA Level 2  Other Methods of Probabilistic Dynamics  Example of a Dynamic Event Tree  Basics of New Method MCDET  Dynamic Event Tree with MCDET (and with MELCOR)  Realised Steps in a Prototype Application

 Assumptions/Events of SBO-Sequence

 First Results and Summary

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 2 OECD-WS PSA 2, March 29-31, 2004 Examples of MELCOR Application for German LWR

PWR Project - MCDET and MELCOR since 2000 Test of new Features Severe Acci- 11/2000: MELCOR 1.8.5 dent Analysis Severe Accident Analysis for Plant Spe- for PSA Level 2 Analysis cific PSA Level 2 BWR type 69 Severe Acci- Severe Accident Training dent Training Course by ATLAS Course by AT- Simulator LAS 9/1997: MELCOR 1.8.4 Simulator

Severe Accident Analysis Investigation of AM meas- Severe Accident Analysis ures (e.g. hydrogen issue) for Generic PSA Level 2 BWR Analysis 10/1994: MELCOR 1.8.3 Input Deck develop- ment and code-to- code comparison Severe Accident Analysis 1992: MELCOR 1.8.2 replacing STCP Preparation of PSA Input Deck develop- ment and validation

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 3 OECD-WS PSA 2, March 29-31, 2004 What does it mean - Probabilistic Dynamics?

Probabilistic dynamics is concerned with dynamics (evolution of the physical variables e.g. during a severe accident) and their interaction with the random evolution of parameters (e.g. component behaviour or NPP operator states). Smidts (1994)

Dynamic reliability methods provide a framework for explicitly capturing the influence of time and process dynamics on scenarios. Labeau, Smidts, Swaminathan (2000)

The evolution of an accident is time dependent and determined by the interaction of dynamics and stochastic in a system consisting of man, machine, process and environment. Accidents to be considered in PSA level 2 analysis comprise time dependent events (sequence of -, duration of - and time difference between actions) as well as interactions between process-dynamic and stochastic elements. Hofer (2000)

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 4 OECD-WS PSA 2, March 29-31, 2004 Kinds of Uncertainty within the framework of PSA

 Aleatoric Uncertainty NRC (1998), Parry (1996) The aleatoric uncertainty is that addressed when the events or phenomena being modelled are characterised as occurring in a “random” or “stochastic” manner, and probabilistic models are adopted to describe their occurrence. It is this as- pect of uncertainty that gives PRA the probabilistic part of its name.

 Epistemic Uncertainty NRC (1998), Parry (1996) The epistemic uncertainty is that associated with the analyst’s confidence in the predictions of the PRA model itself, and it reflects the analyst’s assessment of how well the PRA model represents the actual system being modelled. This has been referred to as state-of-knowledge uncertainty.

The customary uncertainty analysis determines the influence of knowledge uncertainties (epistemic uncertainties) in parameters, model formulations, phenomena as well as in numerical solution algorithms on the results of computer model applications (e.g. ATHLET, MELCOR). Analyses of this kind have been performed with the GRS code SUSA since many years. Glaeser(1995), Hofer (1996)

The aleatoric uncertainties are taken into account in the new method MCDET.

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 5 OECD-WS PSA 2, March 29-31, 2004 Limitations of Event Trees in current PSA Level 2

 Characterisation of numerous possible accident scenarios only by:

 coarse assessment scheme related to:  “time”, e.g. of an event - early, late, before or after something happens  “position”, e.g. of a break - high, low  “intensity”, e.g. of a combustion - strong, negligible  ... etc.  simplification of interactions between phenomena, processes, human action  Risk to:

 not consider some contexts of events or scenarios,  not detect events or scenarios which would result from detailed simulation of phenomena, processes, human action, etc.  generate unrealistic events or scenarios due to bad analyst defined conditions

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 6 OECD-WS PSA 2, March 29-31, 2004 Basics of the new method Monte Carlo Dynamic Event Tree (1)

 MCDET:

 use of a stochastic module connected with a dynamic code – MELCOR used in the example  Features of the Stochastic Module:

 modelling of stochastic influences – shake hands with the dynamic code  determination of probability distributions of different dynamic code results like:  occurrence of a failure dependent e.g. on the kind of the failure and its timing  value or time dependent value of a dynamic process variable  a vector of dynamic process variables (release rate of different fission products)  determination of importance of measures  identification of each calculated accident sequence  simplified customary uncertainty and sensitivity analysis

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 7 OECD-WS PSA 2, March 29-31, 2004 Basics of the new method Monte Carlo Dynamic Event Tree (2)

 Application of MCDET to MELCOR  preparation of an interface program between MCDET and MELCOR:  to generate MELCOR input  to read the MELCOR message file with information needed in MCDET  to automatically start and stop all sequences  manipulation of system availabilities by Control Functions in MELCOR  calculation of base sequence first, other sequences continuing from Restart messagef '/temp/pej/message/m#*-99' cf43101 0.0 edf00101 '/temp/pej/out/r#*-98' cf44100 'srv2 logic' l-a-ifte 3 #*007 #*008 exacttime1 #*002 cf44101 0.0 restartcf 438 * Restart at SG-Signal o. Valve ... * HP-Pump 1 HL restart time #*001 cf04100 '1x HP-SiP' tab-fun 1 #*009 #*010 tend #*002 cf04101 0.0 * Pres. Valve PORV, SRV1 + SRV2 * HP-Pump 2 HL cf42100 'porv logic' l-a-ifte 3 #*003 #*004 cf04200 '1x HP-SiP 3A' tab-fun 1 #*011 #*012 cf42101 0.0 cf04201 0.0 cf43100 'srv1 logic' l-a-ifte 3 #*005 #*006

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 8 OECD-WS PSA 2, March 29-31, 2004 Other Methods of Probabilistic Dynamics

 explicit methods

 requiring, that the analyst explicitly defines all system states and its transitions   well known as “ Markov-Models”  implicit methods

 implicit definition of the system states and its transitions dependent on the rules provided by the analyst   known as “Dynamic Event Trees”   methods or codes described in the literature are e.g.: DYLAM [Cojazzi (1996)], DETAM [Siu (1994)], ADS [Chang (1998)], ISA [Asensio [1997)]

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 9 OECD-WS PSA 2, March 29-31, 2004 Example of a Dynamic Event Tree

# Dynamic Event Trees consider the influence of time in detail not only be characteristic states Dynamic Event Tree with MCDET (and with MELCOR)

fts – failure to start (a valve fails to open at request); ftr – failure to run (a valve fails after x cycles); threshold – branch not considered fur- ther due to low probability

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 10 OECD-WS PSA 2, March 29-31, 2004 fts – failure to start; ftr – failure to run; threshold – branch not considered due to low probability det – deterministic prob – probabilistic MC – Monte Carlo PH - Phenomena

# Dynamic Event Trees consider the influence of time and the interaction between dynamic and stochastic events # 1 of N Dynamic Event Trees with different types of branch points according to MCDET

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 11 OECD-WS PSA 2, March 29-31, 2004 Realised Steps in a Prototype Application (1)

1. Specification of Initial Conditions, e.g.:  PWR-1300 at nominal power  initial event: total station black out

2. Identification of important process variables and system states, e.g.:  core exit temperature  core or RPV water level  degree of core degradation (amount of melt, hydrogen, ...)  probability distributions of process parameters dependent on time, location, etc.

3. Determination of dynamic code, e.g.:  MELCOR with PWR-1300 input deck

4. Identification of stochastic elements, e.g.:  Active Components (AC)  Phenomena (PH) not modelled  pressuriser relief and safety e.g. in MELCOR valves  a molten pool inside reactor core,  pumps and valves of HP and LP (coolability, recriticality) safety injection systems  hydrogen detonation  valves in accumulator system  steam explosion  Passive Components (PC)  External (EX) stochastic events  reactor pressure vessel  recovery of external power sup-  hot leg, pressuriser surge line ply  containment  switch back of each of the 4 different ECC trains  Human Action (HA)  status of systems at time of  determination of plant status power recovery  activation of systems  AM measure: primary bleed

5. Identification of stochastic variables, e.g.:  AC – function of pressuriser relief valve:  normal operation  manual depressurisation OK  failure to open  failure to open during manual  failure to close (stuck open) operation  PC – failure criteria for RPV and reactor circuit components  HA – time needed to perform AM measure “primary bleed”  EX – time difference between initial event and power recovery

6. Identification of time and system states where a stochastic impact will occur

7. Identification of time-state-windows (combinations of time and system states) for different distributions of stochastic variables – failure function of valves  How many times the pressurizer relief valve has opened/closed already during pressure limitation process at ~16 MPa?

8. Specification of probability distributions for each stochastic variable of the time- state-window - defined

9. Do we need additional physical models?  PH – to simulate coolability of a molten pool inside the core region? – yes, but not done in example  PH – to simulate re-criticality? – no

10. Implementation of needed physical models – no model in this application

11. Determination of additional stochastic variables in added models – no variables

12. Determination of conditions indicating where the dynamic calculation should end

 successful core cooling or RPV failure or process time greater 12000 sec

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 15 OECD-WS PSA 2, March 29-31, 2004 Assumptions/Events of SBO-Sequence (1)

T r i p l e L o o p R e a c t o r S i n g l e L o o p P r e s s u r i z e r + T a n k RV S t e a m RV S t e a m EBS -Spray  MELCOR Input Deck Requirements Burst Disks VCS -Spray SRV S t e a m C o l l e c t o r SRV  practically no limitations F W F W  all states of systems (AC, PC) mod- 3x Accu's Accu

elled by Control Function PORV, SRV1+2 3x SI

U m g e b u n g 4 0 Heater 16 M C P K u p p e l t o p M C P

VCS -Spray EBS -Spray Spray 2 7 2 9 EBS - Extra Borating System 4 3 4 5 10 14 15 11 EBS VCS 75 Node Reactor Core VCS - Volume Control System K u p p e l A K u p p e l B SIS - HP+LP Safety Injection System

6 0 R P V F a i l u r e 2 4 4 2 4 1 2 5 4 4 2 3   Control Function values are 4 1 9 5 3 1 modified during Restart 3 4 13 3 5 3 7 3 9 8 9  modifications are generated auto- R e a k t o r R a u m matically by one of the MCDET D E A D E B D H P e r i f . B P e r i f . A H K P A H K P B modules 1 0 1 2 12 1 5 1 3 1 6 2 0 1 2  main interesting parameter writ- 6 R e a k t o r 7 3 S u m p f 1 1 4 GC ar u v b i t ey ten by External Data File (pre- R i n g r a u m 2 2 defined !!)

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 16 OECD-WS PSA 2, March 29-31, 2004 Assumptions/Events of SBO-Sequence (2)

 Timing of Events:  SBO, MCP off, RESA: 0 s - 3 s  begin of FP gap release w/o  SG empty: ~3600 s reflooding : ~8400 s  1st opening of press. RV ~4000 s  begin of core melting w/o reflooding : ~9000 s  1st release into containm. ~5300 s  threshold for path-probability:  AM measure - Bleed: ~6220 s < 1. E-04  Stochastic Events – First Event-Tree out of N with randomly defined values  different failure modes of press.  different states of accumulators valves after M cycles of operation  different time until recovery of 48 - RV, 43 - SV1, 4 - SV2 power in each of the 4 different  only “available” valves can be ECC trains: 7038 s, 7488 s, opened by AM measure “Bleed” 7938 s, 8388 s  time needed to initiate AM meas-  different states of valves and ure after process signal: 325 s safety injection pumps in each  different states of “available” ECC train after power recovery valves after AM measure “Bleed”  failure criteria for MCL and surge line (not relevant in this set)

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 17 OECD-WS PSA 2, March 29-31, 2004 Containment Pressure – Results of all sequences calculated with 1 of N sets of stochastic data

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 18 OECD-WS PSA 2, March 29-31, 2004 RPV Pressure – Results of all sequences calculated with the 1st set of stochastic data out of N

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 19 OECD-WS PSA 2, March 29-31, 2004 M C D E T / M E L C O R a l l e P f a d e a u s L a u f 7 6 0 0

0 4 0 0 0 . 0 6 0 0 0 . 0 8 0 0 0 . 0 1 0 0 0 0 . 0 1 2 0 0 0 . 0 Z e i t [ s ]

st Total H2 mass – Results of all sequences calculated with the 7 set of stochastic data out of N

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 20 OECD-WS PSA 2, March 29-31, 2004 RPV Water Level – Results of all sequen. calculated with the 1st set of stochastic data out of N

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 21 OECD-WS PSA 2, March 29-31, 2004 cumulative probability

Probability Distribution - Minimum of RPV Water Level Results of all sequences calculated with the 1st set of stochastic data out of N

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 22 OECD-WS PSA 2, March 29-31, 2004 cumulative probability

Probability Distribution - Maximum of Core Exit Temperature Results of all sequences calculated with the 1st set of stochastic data out of N

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 23 OECD-WS PSA 2, March 29-31, 2004 Summary

 MCDET allows:  to be added as module to different dynamic codes (MELCOR, ATHLET, ...)  a consideration of a wide spectrum of stochastic influences (AC, PC, HA, PH, EX)  to use multiple branching points  consideration of continuous stochastic transitions by Monte Carlo method  different dependencies of stochastic elements (time, status, history of accident)  status change of multiple number of stochastic elements within one time step  uncertainty and sensitivity analysis (simplified)  use of parallel processor units for speed up  CPU time required for calculations with first set of stochast. data  ~150 sequences calculated with 1st set of stochastic data  only a very few abnormal ends of MELCOR runs  ~ 3 - 5 d needed on a DEC-ALPHA Workstation here without parallel jobs  more information in next paper of Löffler/Peschke GRS

M. Sonnenkalb ([email protected]), J. Peschke ([email protected]) 24 OECD-WS PSA 2, March 29-31, 2004