Title: SARNET WP5 - Level 2 PSA Comparison between classical and dynamic reliability methods. Specification and results of a benchmark exercise on consequences of hydrogen combustion during in-vessel core degradation E. Raimond, T. Durin IRSN, BP 17
1SARNET WP5 - Level 2 PSA Comparison between
classical and dynamic reliability methods.
Specification and results of a benchmark exercise
on consequences of hydrogen combustion during
in-vessel core degradationE. Raimond, T.
DurinIRSN, BP 17 92265 Fontenay-aux-Roses
2Background
- SARNET WP 5.3 Dynamic methods for level 2 PSA
- Stage 1 Review of existing approaches (cf. P.E
Labeau ERMSAR-05) - Stage 2 Practical application in a benchmark
exercise - Benchmark specification End 2005 / Mid 2006
- Solutions proposal by participants 2006
- Synthesis and additional contributions Mid-2007
3Background
- THE CHALLENGE
- TO PROVIDE A SIMPLE EXAMPLE
- THAT DEMONSTRATE THE LIMITATION
- OF CLASSICAL EVENT TREE METHODS
- AND GAIN OBTAINED BY DYNAMIC METHODS
4- PRESENTATION OF THE BENCHMARK EXERCISE
5Benchmark description
- A basic transient
- A French 900 MWe PWR (3 loops, with Passive
Autocatalytic Recombiners PAR) operating at
nominal power before the initiating event - Loss of coolant accident (LOCA) after a 3 break
size on cold leg of RCS, - Failure of all water injection system and spray
system - An ASTEC calculation provides basic information
on - The kinetic of core degradation process
- The kinetic of hydrogen and vapor releases in
containment - The delay before vessel rupture
- The pressure evolution in containment (and
atmosphere composition)
6INFORMATION FROM ASTEC
Maximum Hydrogen mass (100 Zr oxydation) May
be increased by steel oxydation
Pressure evolution in containment (between 2 or
3 bar)
Hydrogen mass released in containment
Time for core degradation beginning
Vessel Rupture time
7Benchmark description
For the basic transient (no water injection, no
spray) the containment atmosphere is not
flammable no combustion
8The issue
- After reparation, water injection and spray
system are available after beginning of core
degradation - QUESTION what is the dominant risk of
containment failure - Three independent events are considered
- Water injection
- Spray system start
- Ignition of H2-H20-Air mixture by recombiners
- No chronological link between the events is
assumed - Consequences of each event is described by
analytical models
9Water injection
- For the benchmark, consequence of water injection
is only hydrogen production. An analytical model
has been proposed
Maximum Hydrogen mass (100 Zr oxydation) May
be increased by steel oxydation
New hydrogen source term as a function of time
reflooding
Hydrogen mass released in containment without
reflooding
Vessel Rupture time
Time for core degradation beginning
10Spray system effect
- In the benchmark, consequence of spray system
start is atmosphere containment depressurization,
cooling and composition modification - An analytical model has been proposed based on
ASTEC results
Pressure evolution in containment (between 2 or
3 bar)
1 bar
New evolution of pressure in containment
Time for core degradation beginning
Vessel Rupture time
11Ignition (by recombiners or other)
- Classical physical criteria have been used to
precise if combustion is possible and probable - Combustion can be total or local only
- Multiple combustions can also occur
- Delay before combustion is unknown
- Pressure peak in containment due to combustion
are evaluated by PAICC model
12List of used physical information
- A representative ASTEC transient without spray
and reflooding - Beginning of core degradation Vessel Rupture
- Hydrogen mass released in containment
- Containment Pressure as a function of time
- A simple law that allows to predict pressure
evolution as a function of time after spray
system start - A simple law that allows to predict H2 release
after core reflooding - A simple law that allows to predict recombiners
efficiency in function of H2 and H20
concentrations - Criteria for hydrogen combustion Shapiro,
ignition by recombiners - The probability of containment failure as a
function of pressure peak
13Stochastic events
- Water injection
- Probability 0.5 to have water injection before
vessel rupture - uniform probability distribution
- Spray system start
- Probability 0.5 to have water injection before
vessel rupture - uniform probability distribution
- Ignition of hydrogen combustion
- Atmosphere flammability is defined with a
Shapiro diagram and a criteria for ignition by
recombiners - (the atmosphere ignition within a short delay is
very problable if the recombiners ignition
criteria is achieved for average H2
concentration) - Local ignition (partial) have been taken into
account
14- ??
- CAN THIS SIMPLE QUESTION BE SOLVED WITH A
CLASSICAL EVENT TREE METHOD - ??
15Difficulties for the classical event tree
approach
- The chronological links between events have to be
defined - 12 situations are possible from the chronological
point of view - Spray water injection ignition
- Spray water injection no ignition
- Ignition Spray water injection
- Ignition Spray no water injection
- Spray ignition water injection
- Spray ignition no water injection
- Ignition water injection Spray
- Ignition water injection no Spray
- water injection ignition Spray
- water injection ignition no Spray
- water injection Spray ignition
- water injection Spray no ignition
16Difficulties for classical event tree approach
- The treatment of all chronological issues is
difficult. - !!!! The treatment of multiple combustions is
impossible !!! - The only way for a classical approach is a
conservative approach - TO CHECK THAT IN THE WORTH CASE, THE
CONTAINMENT FAILURE RISK IS RESIDUAL
172 steps
- STEP 1 First implementation of the problem with
dynamic or classical method, with simple
analytical model for the physics - STEP 2 Second implementation of the problem
with complements in the analytical models for
epistemic uncertainties - ?? Possible STEP 3 Implementation of ASTEC
modules instead of simple analytic model ??
18- SYNTHESIS OF THE BENCHMARK RESULTS
1910 PARTICIPANTS
- IRSN
- GRS
- CEA
- AREVA
- VEIKI
- ULB
- (CSN)
- LEI
- UJV
- INR
20Five categories of solutions
- Direct calculation
- Classical event tree methods
- Macro-event method with classic tools
- Monte Carlo Dynamic Event Tree (MCDET)
- Stimulus-Driven Theory of Probabilistic Dynamics
(SDTPD)
21Methods1- Direct calculation (CEA, INR)
- Monte Carlo simulations (C or Fortran program)
- time of water injection and spray system
activation are randomly sampled - For each sequence
- The evolution of the system is calculated for
each time step (1 second) - If the containment atmosphere is flammable, a
delay before combustion is determined - Overpressure is calculated
- Containment integrity is determined
- simple method, the validation of the model is
easy - - few information for analysis, implementation
in global event tree ?
22Methods2- Classical event tree method (AREVA,
VEIKI INR)
- Specific calculations of containment composition
evolution are performed for a few sequences (with
EXCEL) - Results are used for quantification of
probabilities of branching nodes in a containment
event tree - Stochastic events become determinist (water
injection) - Some assumptions are modified
- easy to implement (because of the modified
assumptions) - - relevance of conservative results for
practical applications ? - Specifications cannot be fulfilled and
accordingly results are very different
23Methods2- Classical event tree method (AREVA,
VEIKI INR)
Example (from VEIKI contribution)
24Methods3- Macro-events methods (UJV, IRSN)
- Based on software initially created for accident
progression event tree (EVNTRE, KANT) - Division of the simulation time in short time
intervals (60 seconds) - Use of the same event tree (macro-event) for each
interval - Quite comparable to the direct calculation method
with Monte-Carlo simulations (for IRSN) or to the
MCDET-method (for UJV) - use of classic tools, easy to integrate in a
global event tree - -
25Methods3- Macro-events methods (UJV, IRSN)
- IRSN macro-event UJV
macro-event - Each macro-event is duplicated in a global event
tree
26Methods4- MCDET-analysis (GRS)
- Mix of Discrete Dynamic Event Trees (DDET) and
Monte Carlo Simulation - Each DDET has the same structure with 22
sequences, only the random values change (Monte
Carlo)
allows a lot of sensibility analysis,
epistemic uncertainties and stochastic events are
considered singly
27Methods 5 - Stimulus-Driven Theory of
Probabilistic Dynamics (SDTPD) - (ULB, CSN)
- General methodology used as a basis for a Monte
Carlo simulation of dynamic reliability problems. - The SDTPD analysis is based on a particular
formalism - the process variables are the variables that
describe the system evolution (6 variables for
LEI and 9 variables for ULB) - the stimuli are the events that can happen
during the simulation (5 stimuli are defined by
ULB and 6 by LEI) - the dynamics are the different regimes of
evolution of the continuous process variables.
28Methods 5 - Stimulus-Driven Theory of
Probabilistic Dynamics (SDTPD) - (ULB, CSN)
- The evolution of each process variable is defined
for each dynamics. - To each stimulus are associated
- a probability of activation and a probability of
deactivation if needed, - an activation (and if needed deactivation) delay.
- This method allows a more precise modeling of
combustion if the flammability conditions
change, combustion can be canceled. This point
was not clearly specified in the specification of
the benchmark exercise, but it shows one of the
interesting capacities of the SDTPD. - A given number of histories (simulations) are
performed. For each simulation, the final result
is the containment integrity (saved or not). This
allows determining the containment failure
probability.
29RESULTS STEP 1
30RESULTS STEP 2
31COMMENTS
- The difference between results has not yet been
fully explained and should not be linked to the
method but also to benchmark assumptions
interpretation - The different contributions shows different
methodologies with advantages / disadvantages. - The analysis/comparison of results has shown some
needs in terms of guidance for the presentation
of uncertainties
32Some outlook
- These results are an encouragement to continue
the development of specific methods for dynamic
reliability problems, including specific
post-processing of results, especially for
uncertainties. - Some participants are interested for a step 3
with a direct use of a severe accident code like
ASTEC - Such application is seen (at IRSN) as an
interesting way for examination of robustness of
severe accident guide