Materials and Metallurgy - March 16, 2000
UNITED STATES OF AMERICA
NUCLEAR REGULATORY COMMISSION
ADVISORY COMMITTEE ON REACTOR SAFEGUARDS
***
MEETING: MATERIALS AND METALLURGY
Room 2B-3
Two White Flint North
11545 Rockville Pike
Rockville, Maryland
Thursday, March 16, 2000
The subcommittees met, pursuant to notice, at 8:35
a.m.
MEMBERS PRESENT:
WILLIAM J. SHACK, Chairman,
Materials and Metallurgy Subcommittee
GEORGE APOSTOLAKIS, Chairman,
Reliability and Probabilistic
Risk Assessment Subcommittee
THOMAS S. KRESS, ACRS Member
MARIO V. BONACA, ACRS Member
DANA A. POWERS, Chairman, ACRS
PARTICIPANTS:
SAM DURAISWAMY, ACRS Staff
NOEL F. DUDLEY, ACRS Staff
EDWIN HACKETT, NRS
SHAH MALIK, NRS
DEBORAH A. JACKSON, NRS
LEE ABRAMSON, NRS
MARK CUNNINGHAM, NRS
NATHAN SIU, NRS
MARK KIRK, NRS
DOUG KALINOUSKY, NRS
ROY WOODS, NRS
WILLIAM GALYEAN, Idaho National Engineering and
Environmental Laboratory
ROBERT HARDIES, Baltimore Gas & Electric
TERRY DIXON, Oak Ridge National Laboratory. C O N T E N T S
NUMBER DESCRIPTION PAGE
1 Introductory Statement by the
Chairman of the Materials and
Metallurgy Subcommittee 4
2 Proposed Agenda 4
3 Overview of Pressurized Thermal
Shock Technical Basis Re-evaluation
Project 5
4 PTS Re-evaluation Project 19
Developing a Generalized Flaw
5 Distribution for Reactor Pressure
Vessels 94
6 Potential Rvisions to PTS Acceptance
Criterion 135 7PRA for PTS Rule Revision160 . P R O C E E D I N G S
[8:35 a.m.]
DR. SHACK: The meeting will now come to order.
This is a joint meeting of the ACRS Subcommittee on
Materials and Metallurgy and on Reliability and
Probabilistic Risk Assessment.
I am Dr. William Shack, Chairman of the Materials
and Metallurgy Subcommittee. Dr. George Apostolakis is
Chairman of the Reliability and Probabilistic Risk
Assessment Subcommittee.
The other ACRS members in attendance are Mario
Bonaca, Thomas Kress, and Dana Powers.
The purpose of this meeting is for the
subcommittees to review the status of activities related to
the staff's pressurized thermal shock screening criterion
reevaluation project. The subcommittees will gather
information, analyze relevant issues and facts, formulate
proposed positions and actions, as appropriate, for
deliberation by the full committee.
Mr. Noel Dudley is the Cognizant ACRS Staff
Engineer for this meeting.
The rules for participation in today's meeting
have been announced as part of the notice of this meeting
previously published in the Federal Register on February 25,
2000.
A transcript of this meeting is being kept and
will be made available as stated in the Federal Register
notice. It is requested that speakers first identify
themselves and speak with sufficient clarity and volume so
that they can be readily heard.
We have received no written comments or requests
for time to make oral statements from members of the public.
I don't think I have any comments here to start
with and we will now proceed with the meeting and I will
call upon Mr. Ed Hackett, acting Chief of the Materials
Engineering Branch, Office of Nuclear Regulatory Research,
to begin.
MR. HACKETT: Thank you, Dr. Shack. I'm pleased
to be able to be back here to go over some progress. I
think it was about a year ago that it was Mike Mayfield,
Farouk Eltawila, and Mark Cunningham briefed the committee
on the project. This information is already stated.
Where we started off, and I guess this is even
more than a year ago now, was with at least the hope that
recent technical developments indicated the potential for
increasing the accuracy in these analyses, and these are
just some of the categories; improved estimates for flaw
density and distribution, embrittlement correlations, and
statistical bases for fracture toughness for the first time.
We initiated the project about April last year.
It's fully participatory with the industry. The industry is
represented here in the form of the MRP and NEI and EPRI.
We have briefed the committee, as I mentioned. I think the
first one was last February, but also last summer, and then,
of course, today. We are also planning on a briefing, I
believe it's in the fall will be the next one.
The project is organized in three key technical
areas. I think the subcommittee, the Thermal Hydraulics
Subcommittee already heard some of the results of progress
in thermal hydraulics yesterday. Today we will be focusing
on probabilistic fracture mechanics, after this
introduction, and then in the afternoon, the information
probabilistic risk assessment.
Just to put a few bullets down on the overall
approach. One of the key points is that this overall
approach is for a best estimate analysis for these
individual technical inputs, with uncertainty addressed
explicitly at each point in the evaluation, and this is a
departure from what we've done historically, as you know. A
lot of what's been done in the vessel area has been done in
a bounding sense, particularly with regard to the fracture
toughness evaluation and the fracture toughness curves.
The idea then also is to update the technical
inputs, as I mentioned, in probabilistic fracture mechanics,
thermal hydraulics and PRA, and redo the IPTS studies with
this new information.
The IPTS studies, you might recall, were conducted
on three plants. It was Calvert Cliffs, Oconee and H.B.
Robinson. I'll come to a glitch in our progress in a little
bit regarding H.B. Robinson, but the idea was to redo those.
Those were done in the 1980s. I don't remember the exact
completion dates, but largely in the 1980s. So the idea was
to redo those, which was the basis for the original rule.
In parallel, an important part of this that we
felt had to go on in parallel was a reassessment of the risk
acceptance criteria, and that's what you'll hear about this
afternoon. Of course, that was set, at that time. The
basis for that is SECY 82-465, from 1982. That was set at
the level of 5E-minus-6. Of course, the NRC has changed its
outlook on that area significantly since that time.
DR. APOSTOLAKIS: Would you explain the first
bullet? I don't understand what the best estimate analysis
with uncertainty addressed at each point means.
MR. HACKETT: Yes. This is the way, I guess you
would argue, it should have been done all along. A good
example where it's not done right now is the fracture
toughness analysis. When the analysis for PTS is done,
either to set the screening criteria or to evaluate an
individual plant against the criteria, they're using lower
bound curves from ASME, with no uncertainty. It's not a
best estimate case.
DR. APOSTOLAKIS: I guess you are using best
estimate and uncertainty in the same sentence, and that's
what confuses me.
MR. HACKETT: Okay.
DR. APOSTOLAKIS: Best estimate usually does not
go with uncertainty analysis, does it?
MR. HACKETT: In this case, the entire PTS
analysis is designed to be a best estimate analysis. But in
the past, criticism, I think valid criticism, we've gotten
from the industry is that we've taken -- it's a nested chain
of correlations and so on that get you to the screening
criteria or are assessed against the screening criteria, and
in each one of those, we typically, in the past, have made
bounding assumptions.
Now we're trying real hard to make best estimate
assumptions and --
DR. APOSTOLAKIS: So you mean to use the best
models.
MR. HACKETT: To use the best estimate model,
right.
DR. APOSTOLAKIS: Then you put an uncertainty on
it.
MR. HACKETT: Right, and then build uncertainty
in. I just flagged this up because that is very different
from what we've done in the past.
DR. APOSTOLAKIS: Okay.
DR. KRESS: Well, your re-look at the risk
acceptance criteria you think incorporate the uncertainty in
some way then.
MR. HACKETT: Yes, absolutely, and I don't know
who is going to address that this afternoon.
MR. MALIK: Mark Cunningham.
MR. HACKETT: Mark will address that. Okay. I
think that's at 1:00. So that will be the case then. I
just thought I'd summarize status real quick. We have made
some significant progress, kind of in fits and starts, I
think. There's been a lot of meetings between us and the
industry and there's been a lot of progress, there's also
been a lot of discussion, a lot of arguments, but I think
it's moving.
Sometimes it's one step forward, two steps back,
but it is moving forward.
In particular, in probabilistic fracture mechanics
area, we have an expert elicitation which is hopefully going
to give us a generic flaw distribution that's really based
on cutting up old vessel welds and looking at those
carefully and also statistically.
We're hoping to have that largely in hand by about
May of this year. That's underway right now.
We do have revised embrittlement correlations,
thanks to the work of Ernie Eason at Modeling and Computing
Services, and also Bob Odette at the University of
California-Santa Barbara.
They have a basis now, a database that supports
these correlations. It's about five times larger than the
one that went into Reg Guide 1.99 Rev. 2, which is what is
used right now.
We are looking at statistical bases for fracture
toughness. The Oak Ridge Laboratory, Mark Kirk, I think
Professor Natishan and others in the room here have been
involved in doing that for the first time on a statistical
basis.
Then another important feature is plant-specific
flux maps are being developed for the plants that we will be
evaluating. I didn't mention it earlier, but Palisades is
obviously very interested in participating in this project
and has been very cooperative, and so we are also looking at
evaluating the Palisades plant.
The wrinkle that I mentioned earlier, you can see
the dates here when these are supposed to be completed, the
Beaver Valley plant is the furthest out because about a
month or two ago, the Beaver Valley plant wasn't part of
this evaluation. We were originally going to have Robinson.
Robinson had some concerns about participating in the
project and basically opted out of the project.
We are very lucky, through the work of the MRP
particularly, that the Beaver Valley plant volunteered to
become part of the project.
Without that, we would have been without a
Westinghouse plant, which I think would have been a very
weak point for this whole project.
DR. POWERS: Can I come back to your expert
elicitation for the flaw distribution? When you described
that, you said that you had lots of information from cut-up
welds.
MR. HACKETT: Right.
DR. POWERS: How about the free sheet?
MR. HACKETT: Excuse me?
DR. POWERS: How about the free sheet? The
unwelded portion.
MR. HACKETT: The unwelded portion, yes. It does
also include that. It has not been focused on that, but
typically in the cut-ups that are done, we will take at
least several inches to a foot on the sides of the welds.
So there is information in the ultrasound exams.
DR. POWERS: That only means something if I know
what it is relative to the heat-affected zone.
MR. HACKETT: Right. Typically, we mention welds,
but a lot of these defects are focused on the heat-affected
zone, and also not just the heat-affected zone adjacent to
the structural weld itself, but the heat-affected zone that
results on the weld metal or in the base metal from the
cladding application.
So those are all captured. The plate actually is
captured obviously to a lesser degree than the HAZ or the
weld, but obviously the rationale for that is you expect a
greater defect rate in the weld or the heat-affected zone.
But we are capturing plate information, too.
DR. POWERS: Is the distribution strictly size or
is it orientation, location?
MR. HACKETT: It's everything. I guess maybe one
of the biggest drivers, of course, is the density, how many
of them are there, but then, of course, a differentiation is
being made now for the first time on whether they're
volumetric or planer. When they're volumetric, like, say,
for instance, it's spherical, turns out when you run the
fracture mechanics analyses that they don't matter, they
really don't count.
Also what we're finding is that an awful lot of
the defects are small, two millimeters, three millimeters.
When you run those through the probabilistic fracture
mechanics code, what you find is they don't participate in
any kind of failure projection, either. It's only when
they're larger. Basically, they've got to be larger, at
least four millimeters, and planer, to really contribute to
the failure frequency.
DR. POWERS: What happens when I have a cluster of
defects such that they act as -- how close do they have to
be to act as a single large defect?
MR. HACKETT: Good question. ASME has what they
call proximity rules to address that, both for surface
breaking and subsurface, and those rules are incorporated
into this assessment.
DR. POWERS: Is that a hidden conservatism that
you're putting in here?
MR. HACKETT: It would be, because in a lot of
cases, as you can imagine, that's --
DR. POWERS: I think you'd really want to flag
that. I don't know that you've got any alternative on what
to do, but I think you want to make it clear where your
conservatisms are and not say that I've universally expunged
conservatisms in here. I don't think you can.
MR. HACKETT: Right. That's a good point. You
can't. You can't ever do that one completely. We do the
best that we can with that, but that will always be there.
Thermal hydraulics, some of you may have heard
about yesterday, because I know there's overlap between the
committees, but by about the April timeframe this year,
we're looking at having a determination of the key
transients to be analyzed. I think these will follow
probably closely what was done before for 82-465, but we
have been examining more than that.
There is also this time, hopefully, going to be
some verification from testing at the Oregon State
University, the APEX facility, which my understanding is
that that will model very closely the behavior of the
Palisades plant.
With PRA, and this will be this afternoon, Mark
Cunningham's presentation, the idea is to take the criteria
that was used previously and then look at consistency with
the more recent NRC risk-informed guidance, particularly
these areas here, the reg guide, core damage frequency and
in LERF, also.
The way it's being done right now is Mark has
drafted a Commission paper that presents options for policy
decisions in this regard that will be presented for your
consideration and also for the Commission's consideration,
and Mark will be discussing that this afternoon.
DR. POWERS: It seems to me that there is a
potential difficulty in acquiring some feedback from the
fracture mechanics folks and the people doing these thermal
hydraulics and PRA. Unfortunately, I didn't attend the
thermal hydraulics meeting yesterday, so I don't know what
they said, but I do know that there is a tendency in the PRA
community to analyze accidents that are as if the operators
went away and took a break.
MR. HACKETT: Right.
DR. POWERS: Something like that. There's no
human involvement. And they're attempting to be bounding
when they do that. But that leads to some peculiarities in
the accident analysis that you get accidents that don't look
like TMI.
MR. HACKETT: Right.
DR. POWERS: Okay. And it's not clear that the
accidents that are bounding or somehow poles that are useful
in PRA for risk analysis will, in fact, be suitable for
looking at the fracture mechanics problem.
It seems to me that you guys would be very
concerned about accidents in which operators inadvertently
turned on water or something like that.
MR. HACKETT: Right. That historically hasn't
been addressed previously. My understanding is it will be
addressed this time around much more explicitly. That's a
concern.
The linkage particularly between the three areas
is critical in this project, as you noted. One of the
things that's critical, for instance, just as an aside or it
could be far more than aside, depending on how it pans out,
is our assumption of the effects of any kind of thermal
plume or thermal streaming in the thermal hydraulics sense.
We're assuming -- the assumption going in is that
there is good mixing there, that we're not going to have to
worry about more than realistically a 1D or 2D problem. If
it's a 3D problem, for instance, our fracture mechanics code
doesn't address that right now, so that kind of thing could
be a showstopper.
We have evidence that seems to indicate that's not
the case, but that's something we need to look at closely.
DR. KRESS: On that slide, before you take it off.
If you would give me a little more detail about that last
bullet. Risk-informed guidance, what is that?
MR. HACKETT: Well, basically, and Mark can
probably talk about this much more articulately than I can,
but the Regulatory Guide 1.174, as you know, has become kind
of a motherhood document for the NRC on how to evaluate risk
or evaluate issues like this on a risk-informed basis.
DR. KRESS: So the first sub-bullet really means
this kind of stuff that's in Reg Guide 1.174.
MR. HACKETT: Right. Basically, first off, the
consideration of risk, but also proper consideration of
defense-in-depth and the other elements that go into the reg
guide. A lot of those criteria, of course, are set
nominally at the 1E-minus-6 level when you're looking at
core damage frequency. The PTS criteria is set at
5E-minus-6.
Nathan may want to make a few remarks.
DR. KRESS: You're going to try to make those two
consistent some way.
MR. SIU: This is Nathan Siu, Office of Research,
PRA Branch. Again, Mark will talk about this more in the
afternoon, but I think the point is that he's raising
options that might be broader than just Reg Guide 1.174.
There's a whole variety of guidance concerning how to use
risk in decision-making.
So we're opening the question what's the
appropriate guidance here and I think the first bullet just
simply says we want to be consistent with past guidance to a
reasonable extent.
DR. KRESS: Thank you.
MR. HACKETT: The major issues I thought I'd
summarize here. Like I said, actually, things have been
going fairly well. But we did become aware about a month or
so ago that the H.B. Robinson plant was not going to be able
to participate in the project. As I said, that was a major
wrinkle, since that would have eliminated a Westinghouse
plant.
Roy, did you have a comment?
MR. WOODS: Yes. I'm Roy Woods, Office of
Research, PRA Branch. You've got the right bottom line.
H.B. Robinson will not be the plant that we're using for the
Westinghouse example. However, they didn't exactly decline
to participate. They were participating and they were
giving us information about thermal hydraulic
characteristics and we had been talking to them and they ran
into a problem with the status of updating their PRA model.
They were a few months away from putting out a
revised PRA model and they were afraid it would cause them
problems if they released the old model to us and it was
going to end up in our having to develop a model ourselves,
which would be quite inefficient.
So we went to see if we could find someone else
that would be able to participate in a more timely fashion
with their PRA model and we made the change. But they were
willing to participate to a fairly high degree and it just
wasn't quite enough to do what we needed to do.
DR. KRESS: That might even be an advantage to
have a newer plant rather than the same three.
MR. WOODS: I agree. Right.
MR. HACKETT: Another particularly interesting
aspect of this shift is, of course, Robinson is projected
right now by the NRC or themselves to have relatively no
problem on pressurized thermal shock, even for their license
renewal term, whereas Beaver Valley is projected to be right
about at the criteria at the end of their current license.
So there's a higher level of interest there on the part of
the plant.
The other thing is Dr. Powers mentioned the plate
defect distribution. Beaver Valley is a plate-limited
plant. So that will put another interesting spin on it from
the materials perspective.
I guess in summary, of course, we're here to do
these presentations as an informational briefing for the
committee. We are obviously very interested in any
feedback, particularly if you think we may be heading off in
the wrong direction somewhere. But probably it would be
good to have some kind of feedback in writing from the
committee on a periodic basis and maybe after this would be
an appropriate time after these several days of briefings.
With that, I guess I'll sit down and have Dr.
Malik come up and start to go through the probabilistic
fracture mechanics, unless there are any other questions.
Thank you.
MR. MALIK: I am Shah Malik. This will be the
presentation on progress made in probabilistic fracture
mechanics as it relates to PTS reevaluation project. I will
be helped by Mark Kirk and Doug Kalinousky in several of the
subject matters that I present here.
We will be going through the status of the PFM,
probabilistic fracture mechanics, activities, and also we'll
look where it fits into the PTS reevaluation project as a
whole, and then we'll go step by step in progress made in
major PFM technical areas and some concluding remarks after
that.
In the PFM area, we have this being a fully
participatory type of project. We are having open public
meetings involving staff, contractors, industry
representatives, as well as public, and we had several
meetings here in '99 and at least one in 2000 year, as well
as we are going to have some more recent.
In these meetings, we decide about what are the
order of issues and what could be a near-term and long-term
action plan that we need to work on, and depending upon
that, we are assigning some tasks.
In addition, we coordinate with PRA as well as the
thermal hydraulics group, so that we can have proper
interface from their output or input together.
DR. SHACK: What does this fully participatory
mean?
MR. MALIK: It means from the very beginning, the
industry and public are very much involved in the process.
We laid out all the items that we are doing, what our
thinkings are, and they come up and provide their feedback;
well, this is not the way it should be done, it should be
done this way.
So we kind of have a mutual understanding of each
other's viewpoint, rather than doing it in the end when
everything is done, and then it's not easy to interface and
bring new ideas into the picture.
DR. APOSTOLAKIS: But this confuses me a little
bit. What are the issues that have to be discussed with the
public? Is this a technical issue?
MR. MALIK: Yes, they are technical issues.
DR. APOSTOLAKIS: Like what?
MR. MALIK: Technical issues, how do we implement
fracture toughness, how do we implement multiple flaws, how
do we implement embrittlement correlation, what are the
different -- because those things are still continuing to be
developed, and at what stage we put it in, because you
always find some more time to do some more work and bring
that in.
DR. APOSTOLAKIS: So you settled on one model or
one approach for each one of these.
MR. MALIK: We are trying to settle on those, yes.
DR. APOSTOLAKIS: And there are no disagreements,
no dissenting views?
MR. MALIK: There will always be some dissenting
views, because you can always find a better mousetrap. So
we keep on working on that.
DR. APOSTOLAKIS: But the question is why didn't
you do it like NUREG-1150, handling the severe accidents? I
mean, if there is a number of approaches, then you try to
accommodate all of them and you simply assign weights to
them by eliciting expert judgment.
MR. MALIK: Approaches are still like ideas that
are being thought out and made. So they are not mature
technologies. Those are ideas that --
DR. APOSTOLAKIS: That's where you need this kind
of approach.
MR. HACKETT: Maybe I'll try. This is Ed Hackett,
again. That's a real good point. The major place that's
being done right now -- well, actually, it is -- that sort
of integration is being done all throughout the project, but
the one where it's most striking is this issue with the flaw
distribution. That's a very -- has been historically a
fairly contentious aspect of this evaluation.
Also, it happens to be a very large driver to what
comes out in failure frequencies.
So what we decided to do, I guess it was about
eight or nine months ago now, was exactly to take that type
of suggestion and we're doing an expert elicitation process
there. So not only do we have the data, but then we're
talking, we're eliciting the expert opinion of various
experts throughout the country, also internationally, to get
opinions on flaw distribution, fabrication techniques,
welding, metallurgy, distributions like that.
So that type of thing is going on continuously in
the project. That's just the case where it's most
explicitly being done.
DR. KRESS: Along the same line of George's
question, I would be interested in whether such public
meetings with all the stakeholder participation have been
useful or not. Have you changed your mind about anything
you were going to do as a result of these meetings?
MR. MALIK: Well, it brings some fresh ideas to
look into and to improve our technical basis. So it has
helped us in ways to have all the things ready before we are
ready to present all those things. Yes, it has helped us.
DR. KRESS: You think it's been worthwhile.
MR. MALIK: Yes, it has been worthwhile.
DR. APOSTOLAKIS: Six of them, all six of them
have been worthwhile.
MR. MALIK: Well, there are times we had some
heated discussions in those, as well, yes.
MR. HACKETT: I guess I could make another comment
there. This is Ed Hackett, again. Also, the industry has
also brought a significant amount of resources to bear on
this project which are well over and above what the NRC was
able to do. So I think that's been a significant help in
the project. Bob wanted to make a few remarks.
MR. HARDIES: This is Bob Hardies, from Baltimore
Gas & Electric Company, and I'm chairman of a reactor vessel
integrity group with the MRP, an EPRI group, an we're
participating in this task.
Our participation includes sort of coordinating
the efforts of all the utilities who are providing input to
this effort. So a significant portion of those six public
meetings is coordinating our contributions of our PRA, our
thermal hydraulics models and the data on the materials in
the plants.
In addition to that, we have technical input and
technical opinions, and you asked for an example of an area
of disagreement and one was warm pre-stressing. The way the
models were performed in the past, when you had an
unisolable leak, they're still treated as if that leak is
isolated, and we make the argument that if it's unisolable,
then it should be treated as if it's isolable.
The way we work that out is that the modeling gets
done with warm pre-stressing not credited, but we do, the
industry does sensitivity studies using that model to figure
out what the effect would be if it was incorporated. In
that way, our needs are accommodated and NRC needs are
accommodated.
DR. KRESS: On the last bullet there, FAVOR, is
that a new and improved version of OCA-P?
MR. MALIK: Yes. It includes OCA-P plus VISA,
which was an NRC code, also. So it combines the best effect
of both.
DR. KRESS: Are we going to sometime see the
details?
MR. MALIK: Yes. In this presentation, we're
going to have details of that, as well.
DR. KRESS: Is there plans to have it be given a
peer review of some sort?
MR. MALIK: In these meetings, we are doing some
comparative analyses and comparison.
DR. KRESS: So sort of.
MR. MALIK: So it's an ongoing process and if the
committee wants to hear more details, we can work on that
one, too.
You mentioned that the PFM, probabilistic fracture
mechanics, code developed by Oak Ridge has a release of
October to the industry for their review and application and
see what things they need to work on.
This is sort of an overall flowchart. As you can
see, it starts out from the right and it flows toward the
left side. Here we have differences in terms of
uncertainty. All the red boxes here show where there are
uncertainties in the model. For example, when starting with
the probabilistic fracture mechanics, we are performing a
stress analysis. So we have like a thermal mechanical
properties uncertainties, clad differential, thermal
coefficient of expansion enthalpy.
The Young's model is another quantity that goes
into developing thermal mechanical, thermal stress as well
as pressure stress.
In turn, depending on what are the thermal
hydraulic transients that are being brought in, so there
will be some uncertainty in them. And then you calculate
the stresses with another set of uncertainties going in and
along with that, we also include effect on weld, residual
stress in the weld parts of the region. So there is
uncertainty on that, as well.
So we feed all of those --
DR. POWERS: Let me ask. Before you feed all
that, let me ask a question. Especially under thermal
mechanical property uncertainties, you have a lot of thermal
mechanical property values that show up in there. How do
you treat the correlation among uncertainties; that is, if
your density is high, your Young modulus is going to be
high. So there has to be a correlation in the uncertainties
there someplace.
MR. MALIK: Yes. In the first set of analyses, we
will have something like what is called mean or best
estimate type of values and then there will be a set of
values selected for them to perform a set of calculations.
This will be like an overall loop here and in that we'll
have a set of best estimate values selected for that range
of values, and those will go into --
DR. POWERS: I find that an interesting
uncertainty analysis. I'm not sure how it works. You pick
a mean value for everything, there's no correlation -- I
mean, there's 100 percent correlation. The means correlate
with the means then. Is that factually correct?
If I have a mean value of the density, do I have a
mean value of thermal conductivity?
DR. APOSTOLAKIS: In other words, when you select,
after you do the mean value calculation, a set of values,
are these values correlated? If the alpha tends to be high,
would the other parameters also be high or are they sampled
independently?
MR. MALIK: They will be sampled independently, I
would think so.
DR. APOSTOLAKIS: So the correlation is ignored.
MR. MALIK: At the moment, yes.
MR. HACKETT: That's how you're doing this.
DR. POWERS: Well, I don't think that's an
advisable way to do things.
DR. APOSTOLAKIS: That is what?
DR. POWERS: I don't think that's the right way to
do things. I think you have to take into account
correlations exclusively.
DR. APOSTOLAKIS: If they are important, yes.
DR. POWERS: And what are the chances that the
material properties aren't going to exhibit an enormous
amount of correlation?
Similarly, what are the chances that the
uncertainty and the weld residual stress is then correlated
with the properties?
MR. KIRK: Mark Kirk, Office of Research. I have
a question. Do you mean correlated for physical reasons or
just they happen to trend with each other? Is there a
causal relation for the correlation?
DR. POWERS: Yes. I would assume that there is
some underlying causal relation. I mean, I don't know what
they are.
MR. KIRK: I'm on the materials side, so I can't
speak directly to anything thermal hydraulic, but the intent
in this process is if there are causal physical
relationships between the variables, if there are
uncertainties in any of the relationships that are shown by
the connection points, that that's all fully captured.
The degree to which it's captured really depends
upon our process, depends upon how well we elicit the -- and
I shouldn't say that, because that has a specific meaning.
It depends on how well the technical area experts in the
areas of materials and thermal hydraulics express their best
understanding of the physical bases for these relationships
that you're talking about.
To the extent that that knowledge is captured,
this model will capture it.
DR. POWERS: Did you ask them specifically about
correlations?
MR. KIRK: Certainly.
DR. APOSTOLAKIS: I don't think -- is the PRA
group going to see this? The PRA group will see the end,
right? The conditional probability of RPD failure.
MR. KIRK: This is the format, and we'll get into
this type of input a little bit more in -- a little bit
later when we talk about materials, but this is really going
to be the form of the input, the way that the understanding
of the physical relationships between all the input
parameters and the input models, this is the type of
information that the technical area experts at least in
materials and I assume in thermal hydraulics are feeding to
the PRA group. So it is going to get captured.
DR. SHACK: Well, some of these sources of
uncertainty, when I look at the flaw distribution
uncertainty, any uncertainty I have in Young's modulus is
going to be somewhere after the --
MR. KIRK: It will be swamped.
DR. SHACK: -- the 14th decimal place.
MR. KIRK: Yes.
DR. SHACK: And certainly the thermal mechanical
properties that I can think about, the yield stress will
probably have the widest distribution, the toughness is --
things like density and thermal conductivity.
DR. KRESS: And you're only looking for
correlations of the uncertainties, not the correlations
between the properties. That will automatically get taken
care of. The correlations of the uncertainties. If you
have a high uncertainty in one, do you have a high
uncertainty in the other?
DR. POWERS: Well, I think you also want to look
for correlations in the values, but that tends to be a lot
easier thing to do.
DR. KRESS: Normally you factor that into it
automatically. But I don't know how you go about getting
correlations between the uncertainties, unless you have just
a lot of data that tells you.
DR. POWERS: That's the only way you can, is to
find out that they're correlated or have a physical model
for how they're correlated.
DR. APOSTOLAKIS: He is right, though. The flaw
distribution.
DR. POWERS: Then just leave out all this stuff.
Just put nominal values in and just leave all this stuff
out. If you're going to make that judgment, what I do is
nonsense on the first part of it, don't make a big deal
about it.
DR. KRESS: Unless it's easy to do and doesn't
cost much time.
DR. POWERS: Well, the thing that Joe is worried
about is what you know intuitively sometimes turns out to be
wrong. I know not in Shack's case, ever, but in my case,
what I know intuitively often turns out to be flat wrong
when I do these integrated analyses like this. That's why
you like to do these integrated analyses.
DR. SHACK: Coming back to uncertainties, one of
the things I do notice is that we're always dealing strictly
with fabrication flaws and there's never any allowance for
growth. Have we done enough analyses to convince ourselves
that there is no significant growth of these flaws?
MR. MALIK: For the PWR environment, I don't think
there is any growth going on.
DR. SHACK: So all that work you did all those
years on cyclical flaw growth of BWRs wasn't necessary.
MR. MALIK: Well, in this particular case, flaws
are the most significant contributor for PTS type of
analyses, yes.
We combine this with crack flaw size to come up
with crack driving force in terms of the stress as far as
crack length and crack depth. Then we combine it, compare
it with fracture toughness, again. We'll have material
resistance uncertainty, such as fracture toughness, as well
as fluence, which go into defining the fracture toughness at
a given point in the reactor vessel's life.
Once we compare it with crack value and fracture
toughness, we also take into account how many flaws are
present. To perform this analysis, there are a number of
flaws that are present in the vessel. With that, we find
the conditional probability of failure for a particular
thermal hydraulic transient and when we combine this with
initiating event frequency to come up with an overall
probability of reactor vessel failure per reactor year, that
is vessel failure frequency.
And to perform this analysis, we are selecting
several plants, as you can see, four plants, Oconee-1,
Calvert Cliffs, Oconee is a B&W plant, Calvert Cliffs and
Palisades are CE plants, and Beaver Valley, three-loop,
Westinghouse plant.
In addition, we are also redoing generic SECY
82-465 analyses which were done in the early '80s and they
were a part of the PTS screening criteria, as well as PTS
rule development.
So we will be redoing those along with these
plant-specific analyses to come up with information related
to reengineering of that PTS screening criteria. There will
be some sort of curve coming out, early vessel failure
frequency as a function of RT, NDT or some other factor into
the vessel, and together with that we can decide whether the
screening criteria needs to be adjusted accordingly.
DR. APOSTOLAKIS: What is this criteria again?
MR. MALIK: Screening criteria presently for axial
weld and plate material, RT and limiting RTNDT should not be
more than 270 degrees three years before the plant is --
actually, they have to estimate and say three years
beforehand, when they are going to reach their 270 degrees
for axial welds and plate or for circumference welds of 300
degrees. So they have to know three years in advance of
that.
DR. KRESS: You can see that corresponds to a
given vessel failure frequency and so you can start with
vessel failure frequency as your acceptance criteria and
work down to the screening.
My question is now that you've got this nice band
around the uncertainty band, how will you factor uncertainty
into this criteria? I presume we're going to hear that this
afternoon.
MR. MALIK: Yes, there will be a whole set of
information provided.
DR. KRESS: I just wanted to alert them that
that's going to be a question we'd like to address.
DR. APOSTOLAKIS: Is there a distinction between
the square and the triangular? Do they mean different
things?
MR. MALIK: It's like a choice. It's K for
material resistance is greater than the crack value, yes or
no, and here is a selection, how many times you select.
DR. APOSTOLAKIS: But the triangle there, what
does it mean? The triangle with the circle in the middle.
MR. MALIK: Yes.
DR. APOSTOLAKIS: Yes. Is that different from the
square to the left? Does it mean anything different?
MR. MALIK: The only thing here is you're making a
selection to say yes or no for answer coming out, where here
you're selecting, picking up a value.
There are six different technical areas. The
first one or the most important one which we're working on
is fabrication flaw distribution in RT beltline materials.
That includes welds and plates and forgings.
The next item is regress statistical
representation of fracture toughness, crack initiation, as
well as crack arrest, K-1-c and K-1-a. Along with that will
be improved irradiation embrittlement correlation to predict
the shift in RTNDT and improve the stress distribution for
material chemistry like nickel and copper, as well as
initial RTNDT. RTNDT-0. So item four feeds into item three
and item three in turn feeds into item two. That's how it's
built up.
And coupled with that is a detailed map of
beltline neutron fluence for the four plants and application
of all those into the PFM computer code, as it's being
revised to accommodate all these developments.
DR. SHACK: Shah, just on that, three and four, is
one of the products that's going to come out of here a
revision of Reg Guide 1.99?
MR. MALIK: Yes. Item three, it will be discussed
in a few minutes, yes.
DR. SHACK: Okay. So there will be no
inconsistency between --
MR. MALIK: No. We want to work in parallel, yes.
I am providing a brief overview on fabrication flaw
distribution. Debbie Jackson will be presenting a good
presentation on our work for this, but I'm going to just
point the discussion on that.
DR. APOSTOLAKIS: How do you know it's going to be
good?
MR. MALIK: Pardon? The objective is to determine
generalized flaw sizes, density, that is number of flaws per
unit volume, the location of those flaws in welds, plates
and forging in the RPV beltline region, and we are using
non-destructive examination, as well as destructive
examination techniques, and coupling it with expert judgment
process, a form of expert judgment process, and the RES
contacted staff, Deborah Jackson, and Pacific Northwest
National Laboratory is performing the destructive and
non-destructive one, as well as helping with the expert
judgment process.
We have already performed destructive and
non-destructive examination of weld in one reactor vessel,
called pressure vessel research user facility. It was
located at Oak Ridge. And we are continuing to inspect
several of the vessels. There is parallel work going on
between industry and NRC, so they do similar kind of NDE
work and we do our NDE work and compare the results.
DR. KRESS: Whether you're looking for is the
number of laws per unit volume.
MR. MALIK: Yes.
DR. KRESS: How do you get that by inspecting the
vessel, just looking at that?
MR. MALIK: For example, if we have a piece of
weld we have cut out, we have done examination to find what
are the flaw indications. Once we have located those flaw
indications, we section the small pieces out from the weld
and cut it where they actually exist and what their size is.
So there is verification using destructive examination as
well.
DR. KRESS: How deep do you go in order to
determine this volume?
MR. MALIK: The depth will depend on how deep the
flaw indication is showing. There were flaws as big as 17
millimeters found. So they were destructively examined and
found that they were in some kind of repair weld, as well as
some kind of complex multiple flaws clustered together.
MR. HACKETT: This is Ed Hackett. I think I'll
make comment there, too. I think Dr. Kress may be referring
to how much of the volume is actually being examined and to
what level of detail.
DR. KRESS: Yes.
MR. HACKETT: The answer is the entire wall, of
course. If it's an eight-inch-thick wall, we're looking at
all eight inches. Not necessarily with the same level of
resolution on the entire way. The expectation, of course,
is that you'd probably find most of your laws, like the
previous conversation with Dr. Powers, in the heat-affected
zone or near the cladding interface, and we're focusing very
detailed examinations there.
DR. KRESS: So when you come up with the value for
this number of flaws, is it distributed?
MR. HACKETT: It is distributed. Right. Exactly.
DR. KRESS: Thank you.
MR. MALIK: There will be some non-destructive
examination of plate, as well, which is always from the weld
region. So we will have, in the middle of the plate, there
will be some flaw distribution coming out from there, as
well.
And the data is being collected for that during
the month of March and April for the plate material. We
expect that generalized flaw distribution using the expert
judgment process to be completed in the May to June
timeframe.
The next cycle is by Mark on fracture toughness
and he will also be going over embrittlement correlation.
MR. KIRK: Okay. Thank you. My name is Mark
Kirk, from the NRC Office of Research, Materials and
Engineering Branch, and I'm going to be -- today I'm going
to be going through with you two separate technical topics.
The first one is the uncertainty analysis for
fracture toughness and the second one is an update on our
progress on developing some new embrittlement trend curves.
So first, the first topic is fracture toughness.
The objective of this activity is to revise the
toughness distribution curves based on expanded data and
physical knowledge of the physics that underlies cleavage
fracture that's been gained since the models were developed
that we're currently using, and they were largely developed
over 25 years ago.
Those distributions that we're using today are
just simply based on data really from the early 1970s.
There are about 170 crack initiation data points, about 50
crack arrest data points, and that's the basis of the curves
that we use today, both in the ASME code, but more
importantly, for this discussion, in FAVOR.
In SECY 82-465 and the IPTS studies, ad hoc
statistical distributions were developed from these data in
the ASME lower bound curves, and I'll be showing you some
graphs of that in just a minute.
The RES staff involved in this activity include
myself, Shah Malik and Nathan Siu from PRA, and the
contractors involved in this activity include the Oak Ridge
National Laboratory and the University of Maryland, and,
again, I'll be filling you in on everybody's roles in just a
moment.
That gives you sort of an overall flowchart of who
is doing what and when in the fracture toughness evaluation.
Where we started out was assembling all available LEFM valid
K-1-c and K-1-a data. That was a task performed by us, by
our contractors at the Oak Ridge National Laboratory.
They collected the data and significantly expanded
our existing database. They performed a purely statistical
assessment to get us some interim curves based on the best
empirical data that's available today to use in some testing
runs of FAVOR that are going on now and also we can look at
those data to illustrate some likely overall changes in the
current FAVOR model relative to the model that was used in
IPTS and SECY 82-465.
I will fill you in on some of the details of that,
but that activity is basically concluded at this time.
Where we moved on to from there is to establish sources of
uncertainty in a way that's fully consistent with existing
PRA methodologies. Here we're involving contractors for the
University of Maryland, and, again, I'll go into more
details on this in a minute.
We're doing a root cause analysis of
uncertainties, so we don't just have to look at the end data
distribution, say, in fracture toughness or in RTNDT. We
can pick apart the uncertainties so that the uncertainties
are appropriately ascribed to different situations and not
just treated in bulk, as they've been done in the past.
Underlying this root cause analysis is we're
looking back at the physical basis, the physical causes for
these uncertainties, so that we can properly distinguish
between aliatory and epistemic uncertainty causes, and, as I
mentioned, we're doing with this and we're working with
Nathan to ensure that the methodologies that we're using is
consistent with the current PRA framework and we're also
working with Nathan and his contractors to make sure that
we're -- that the materials experts are describing their
state of knowledge to the PRA folks in a way that basically
everybody can understand.
So first, I'd like to just spend a few slides
reviewing our data collection effort and then I'll go on to
update you on where we are in the uncertainty analysis.
In data collection, Oak Ridge searched and
collected additional data. Basically, we had a 50 percent
increase in the crack initiation data and an over 100
percent increase in the crack arrest data relative to the
statistical basis that was used in SECY 82-465 and IPTS
studies.
They developed some Weibull distributions for us
to use in the FAVOR code, just strictly based on the data
fit. There is also a large K-1-c and K-1-a database that
was developed in Japan in the late '80s and early '90s.
It's been an ongoing activity here at the NRC, even
predating the PTS reevaluation, to obtain that data. We
hadn't succeeded on that. We still haven't succeeded on
that, but now the Japanese workers who put together this
database have released the data to the Pressure Vessel
Research Council and we're in the process of hopefully
crossing the T's and dotting the I's to get access to that
data.
So that's an ongoing activity in data collection.
This just sort of shows you on one slide the
culmination of the Oak Ridge effort. I'd like to focus your
attention first on the left-hand side, for you. This is a
plot of the initiation fracture toughness K-1-c versus the
normalized temperature. So that's the temperature of the
test relative to the reference no ductility temperatures
defined in ASME.
As I said, this represents about a 50 percent
increase in the statistical evidence that we had relative to
our previous work, and the thing that I'd like to point out,
and I'll point it out again over here, the black curves are
the statistically derived uncertainty bounds that Oak Ridge
fit to this particular data set.
The red curves are what was being used in the
FAVOR model up until about six months ago. Similarly, over
here, this is a plot of the crack arrest fracture toughness
relative to temperature normalized to the no ductility
reference temperature. Again, you see the red curves that
were being used in FAVOR versus the black curves. It's the
current best statistical representation of the data.
A message I'd like you to come away from this
slide with is that the old FAVOR scatter bands were just too
narrow to represent what was really going on.
We have performed some scoping studies using FAVOR
to see what effect going from the red distributions to the
black distributions has on the predicted probability of
vessel failure. Perhaps not surprisingly, whether the new
predictions are higher or lower than the old predictions is
highly dependent upon the transient. We've done some runs
where we get many more flaws initiating and going through
the wall and then we have other transients where we have
many less.
So the ballot is sort of still out as to the end
effect on this and just points out that we can't allow
ourselves to be moved around too much emotionally by changes
in where we believe we were versus where we are now. We
need to look at this in an integrated fashion.
On the next slide, slide number 12 in your packet,
which I will skip, is just a mathematical representation of
some of the curves that were on the previous slide for your
reference, and that's all been detailed in reports that are
now in NRC publication.
So just to orient you, I've gone through the data
collection and the statistical assessment and I'm now going
to move on to the root cause analysis.
There were questions raised earlier in the morning
about a fully participatory process and whether that's had
any practical benefits or not, and Bob Hardies certainly
addressed some areas where the EPRI and the MRP has brought
together input from the utilities.
I'd like to highlight here what I see as being a
very key benefit in terms of the industry bringing in expert
technical knowledge that wasn't available to the NRC and
wouldn't be available unless we were in a fully
participatory process.
This is the work on the uncertainty analysis of
the K-1-c and K-1-a curves being conducted at the University
of Maryland. It's being conducted at the University of
Maryland by contractors that are working from separate
funding sources. Professors Modarres and Mosleh have been
working with Nathan Siu in the PRA area for some time.
Through EPRI and the MRP, they brought in the expertise of
Professor Marjorie Natishan, who is sitting in the back of
the room, to help us out from the physical basis in
identifying the root causes of uncertainties on the
materials side.
These two researchers are collaborating in this
effort, but basically the handoff here is in Professor
Natishan's work and I will detail some of that, because
that's basically where we are.
She's been identifying the reasons for the
underlying uncertainties in the bulk data that you saw there
and describing that in a systematic way that's then taken by
the PRA folks and expressed mathematically to get us to our
end result, which is a recommended program structure for
FAVOR that treats the uncertainties in a way that's
consistent with the underlying physical process.
Now, Shah had used one of the root cause diagrams
and, again, this is -- the use of this type of diagramming
format has come about as a direct consequence of EPRI's
funding of Professor Natishan and I think it's brought us to
a very good place in terms of being able to look at existing
methodologies and express them a systematic fashion.
It was perhaps not a good idea to use this
diagramming process without explaining it first, so I'm a
little bit late on this, but we'll try it here.
The idea is that the diagram expresses both
parameter uncertainties in the input parameters and really
what can go into any of these yellow boxes is a distribution
of values. For example, and I will show you some real
examples in a minute, say this could be RTNDT and there
would be some distribution of RTNDT values which you could
look at.
Well, that arises due to uncertainties and a lot
of different things, a lot of process things and a lot of
parameter things. Back here you might have some of the
chemical composition elements. So distributions of chemical
composition, say, of copper and nickel could flow through
the physical model and give rise to a distribution of RTNDT
values, which then you'd ascribe some uncertainty to. So
that's the basic idea.
So in the diagram format, parameters with
distributions go in the boxes and at the nodes, those
represent different relationships between the parameters.
You can have equations that are correlations which, in fact,
have their own uncertainties associated with them based on
the data that they were drawn from and based on the
underlying physical basis of the correlation.
You can have nodes that are choices, you pick one
or the other, and you can have nodes that are comparisons,
min's, max's, things like that.
Just some things to say about the process, and,
like I said, this has been very helpful in both focusing our
attention on what the models are that we really are using
today, and also in involving a lot of different experts from
different technical areas and getting all their input into
one framework.
One very nice thing is it displays a complex
process in a very logical format and it's the only thing
I've personally seen that allows you to look at the big
picture, while still also capturing the details.
You can look at these diagrams at any level and
you don't have to hide anything if you don't want to. It's
been very useful in going through this process with experts
from within the NRC and also experts in the industry in
building consensus, because it really provides a common
language for discussion.
You will have people come in and say, well, copper
is very important as a cause of embrittlement. Yes, indeed,
it is very important, but where copper comes in is somewhere
way down here and if you try to treat copper way up here,
you're going to get stuck with gross empiricisms that are a
cause of a lot of the over-conservatisms that are endemic to
our current process.
So yes, everybody agrees that copper is very
important and you'll have people pounding the table and
saying that and you'll agree with them, but you're not going
to treat it properly and you're not going to capture it
properly in the mathematical model. It goes to the PRA
people and then eventually gets reflected in FAVOR, unless
you understand that copper is somewhere way past that wall
and not up there.
So this has really allowed people to put this
together and understand it as a group.
It also streamlines the critique, because you can
lay it down in front of someone and have them see how it
goes and I will warn you in advance, you may find some
errors in the diagrams that I'm about to show you, because
they are works in progress, and it seems like every time we
put them up, somebody finds something that's perhaps not
quite right. Hopefully we're converging on a solution.
One thing I do want to point out, and I think I
have pointed it out already, is this treats both
uncertainties in the input parameters, say copper and
nickel, measurements of temperature, as well as
uncertainties in the models which are represented by the
nodes.
So I have included more diagrams than this in the
packet and I would be happy to discuss them in detail, if
people would like, but what I would like to do is just sort
of show you one very high level diagram and then go into one
-- in a little bit of detail to focus on some of the things
that the process does. If you want to get into the details,
that's fine. That wasn't my initial intent.
So at the highest level, we're looking for a
distribution, and I shouldn't have used the word uncertainty
here. You assess the uncertainty as a result of the
distribution the model predicts, but we get a distribution
of K-1-c values. That's related to the K-1-c data that was
used in FAVOR and it's also related to the RTNDT in the
irradiated condition, because you do your K-1-c test and
then you plot it not versus temperature, but versus
temperature normalized to RTNDT, irradiated, in this case,
end of license.
RTNDT irradiated, based on our current modeling
methodology, is a direct function of the unirradiated RTNDT.
So the RTNDT measured before operation begins and the shift
in the Charpy 30-foot-pound energy, which we take to be
equal to the shift in RTNDT. So right there, even at this
high level, you see an assumption. We can talk about
whether it's a good assumption, bad, whether it's a big
error or a small error relative to other things that are in
the model, but that's going to get captured here because
what you put in is data and physical understanding.
I have included all of -- more diagrams here and
they then go on further. What I'd like to do is just show
you the T30 shift diagram, because that will enable me to
make a few points that I'd like to about really more the
process than what's in particular on the diagram.
I will step through it just very briefly. The way
we get the shift in 30-foot-pound transition temperature in
this -- and this is a -- this is basically a diagram of
what's in either staff position or 10 CFR 50.61/Reg Guide
1.99 Rev. 2.
First, you have to decide if you do or do not have
credible surveillance. If you don't have credible
surveillance or you have surveillance and it's -- or you
don't have surveillance, you use the embrittlement trend
curves. If you do have credible surveillance, you construct
a best estimate of the T30 shift based on testing of your
surveillance capsules. You then adjust that value for any
potential differences in the chemistry between that little
lump of material that you tested and your whole, say,
beltline weld, if that's what is limiting.
You also adjust that best estimate of T30 based on
your surveillance samples due to -- for any differences in
irradiation temperature that may have occurred, for example,
if your limiting material was irradiated in another vessel.
Then that goes on and flows down, as I said. Cooper is
obviously a key embrittling element, but you see it doesn't
occur early on in the diagram and, in fact, this diagram
then goes to another one where we get our T30 values.
In terms of the points that I would like to make,
and these are reflected on slide 20, so I will just say them
here and then we an skip slide 20.
A lot of times, when people look at these
diagrams, and I've already pointed out that this isn't the
end, this continues on, sometimes people get despondent
because they say this is possibly complex, we could never --
we could never reach the milestones that it laid out if we
go through this.
One thing I want to point out is that you can
enter your parameter data at any point on this diagram. You
don't have to go all the way to the far right to enter your
data and, in fact, in most cases, we don't. We might come
in here and say, okay, we have measured values of the
30-foot-pound transition temperature. We could go all the
way back to the raw Charpy data and refit it and do all
that, or we could enter here, or we might decide that we've
already done that and we could enter with Charpy shift
values.
That's going to be a decision that has to be made
by the technical experts involved in the process in terms of
what our quality of knowledge is at any particular level.
But I just wanted to point out that just because
we're trying to get this basically all the way down to
measurement error and material inhomogeneity, in most case,
we won't be entering the diagrams at that point with
parameter data.
I've pointed this out before. This appropriately
incorporates all the uncertainty sources, both uncertainties
in the parameters, any possible correlation between the
parameters, and also uncertainties in -- and it's not maybe
well reflected on this diagram, but any uncertainties in the
relationships between the parameters. Each of these
equations, it's not just simply the equation that the
materials folks are going to pass to PRA, but it's our best
understanding of is this an exact model, is this a
correlation, are there other potential correlations, and
that then will be treated in an appropriate way by the folks
in PRA.
It's probably obvious, from what I've said right
now, but the diagrams are much more than schematic. They,
in fact, represent mathematical models and will be used as
the basis for simulation studies to understand what the
uncertainties are.
And there are a few things that this process does
that our old way of doing things, which Ed pointed out was
lower bounding, can't do and doesn't do, is that we find
that when you diagram the process in this way, you find that
uncertainties split at certain levels. For instance, every
time you encounter a choice node, if, for any particular
situation, the uncertainty in a 30-foot-pound transition
shift is either going to be the uncertainty that's down here
from using the trend curve or the uncertainty that's up here
from using surveillance, it can't ever possibly be both,
because you have to pick one or the other.
Whereas if you just came in and did a statistical
assessment or a statistical analysis, I should say, of delta
T30 values at the end of this, you'd be wrapping all those
together and you wouldn't be appropriately treating it.
So by taking the process apart, we can make sure
that uncertainties that are appropriately burdened onto
appropriate situations and we also have the ability to
eliminate double-counting of uncertainties, which is a very
real potential, for instance, at this particular node, you
feed an embrittlement trend curve without -- right now, you
feed embrittlement trend curves without use of copper,
nickel, end of license fluents and product form, whereas the
equation at node four was, in fact, derived from some of
those same data.
That needs to be treated appropriately and will be
in this process.
So I'm going to -- I have included in your packet
diagrams for RTNDT and irradiated in K-1-c. If it's
acceptable to everyone, I'm just going to skip over those,
because like I said, I sort of viewed my role here as trying
to describe the process we were taking a bit more than going
into the details.
What I would like to do now is to shift gears and
move on to the irradiation embrittlement correlations. I
have borrowed a diagram from the last presentation to
indicate that everything that I'm about to talk about is
ultimately going to impact this box on the uncertainty
diagrams and everything to the right of it.
So the objective in this activity is to develop or
perhaps I should say revise, refine, improve a model to
predict the shift in -- and I want to be specific -- this is
a shift in the 30-foot-pound Charpy transition temperature
which we take to be equal to the shift in RTNDT in current
regulations, due to irradiation embrittlement.
Why are we doing this now? Well, we've got a heck
of a lot more data than we did the last time this was
revised, which is over a decade ago. In that larger data
set, we've got a much better coverage of the primary
variables, the primary embrittlement variables of copper and
nickel and so on. We've got much longer time exposures and
consequently we've got exposures to higher fluences.
The only data that is being directly considered in
this trend curve development is data from commercial reactor
surveillance. We're using data from test reactors and the
physical understanding from test reactors and theories to
help guide our models and that's where the physical
understanding comes in, but those data are not being
directly used in the correlations.
We're using rigorous statistical methods to try to
parse out the effects, which is a continuing challenge, and,
as I said, we're trying to bring in -- this is not going to
be a purely empirical model. It's a highly non-linear
model. The variables are -- a lot of the variables are
highly cross-correlated and in order to have any sensibility
to this, we need to bring in a fairly sophisticated
understanding of the underlying physical process of
irradiation embrittlement, so we know forms to try to fit
the data.
This activity provides guidance to -- actually,
the activity started and stands as a separate milestone on
all of our charts as Reg Guide 1.99 Rev. 3, for which we're
on the hook to provide the technical basis for in December
of this year and then Reg Guide 1.99 Rev. 3 will go out for
public comment sometime in June or July of '01.
But we also needed to sort of crank up the
activity to provide input to the PTS reevaluation project,
and what we're trying to do is to get to Shah and his group
a new embrittlement trend curve and a new assessment of the
uncertainties which will be rolled back into the model that
the University of Maryland is developing for us sometime in
the April timeframe.
The RES staff that is working on this is myself,
Carolyn Fairbanks, Shah Malik, and the NRC contractors that
are involved include the Oak Ridge National Lab, Modeling
and Computing Services out in Boulder, Colorado, and
Professor Bob Odette of the University of California at
Santa Barbara.
Just to give you a brief perspective on what's
changed data-wise. This just shows you the size of the
empirical data set that we're working for in terms of number
of Charpy shift values. So each of these values represents
at least two Charpy transition curves, one irradiated and
one at some level of fluence.
When we developed Reg Guide 1.99 Rev. 2, sometimes
known as the Randall-Guthrie-Odette correlation and Rev. 2
hit the books in '88, we had a bit shy of 200 shift values.
In the mid '90s, the NRC let contracts with both Modeling
and Computing Services, Ernie Eason and Joyce Wright, and
also with Professor Bob Odette at UCSB, to do an updated
assessment of the embrittlement trend curves. When they
published a NUREG for us in 1998, we were up just a bit over
600 data points.
That model was subsequently critiqued, sort of in an
informal sense within ASME and E-900 community. That led to
some of the NSSS vendors coming to us with about 200
additional data points which have now been included in our
assessment. So we're up just a little bit shy of 800 shift
values.
I'm going to put an equation here and not explain
it, which is the only safe thing for me to do. But I do
want to highlight how the model has changed, other than just
getting longer. The 1988 Reg Guide 1.99 Rev. 2 model is a
multiplicative model for Charpy shift. We have all the
chemistry factors in one term that's called the chemistry
factor and then we have fluence in a completely separate
term.
That reflected pretty much just a pure empirical
fit to the data. In the new equation, and this is just -- I
just put this up as an example. It's just one of the
candidates that are currently being considered and we're
hoping to finalize on the best end model sometime in the
next two months, but what we see, some features to
highlight, like I said, we've got physically motivated --
we've got physically motivated reasons for the forms and the
functions that we've selected. We've got separate terms for
the stable matrix defects in the A term and the copper rich
precipitates in the B term, which is in good agreement with
the underlying damage -- the underlying reasons for damage.
And it's not particularly apparent here, but there
are copper saturation limits being included, reflecting the
fact that beyond a certain point, copper is not soluble in
the matrix and will not cause damage.
Some terms that are currently under consideration
include terms accounting for phosphorous, and I should note
that there was a phosphorous term in Reg Guide 1.99 Rev. 1
that was subsequently removed. We're looking at long time
effects and irradiation temperature effects, largely as a
result of the fact that we can now see this in the data, and
I suppose it is open to some expert debate as to whether we
can see it or not.
Some of the models tell us that we should expect
to see it and as we collect more surveillance data, at long
times, we're beginning to see some effects.
Also, a big change, not so much in the equations,
but in the underlying philosophy, is it's quite likely --
DR. SHACK: These long time effects, this is what,
growth of the copper precipitates to some point where
they're no longer as effective?
MR. KIRK: Yes, or thermal -- a combined thermal
irradiation effect, any number of things. I'll get into
this just briefly. We spent a lot of time trying to develop
what I'm going to call a gating criteria. There is no
absolute truth here, of course. We've got some Heinz
variety of empirical knowledge and physical knowledge and
we've tried to come up with some criteria to help focus
ourselves on, okay, what gets in and what has to wait for
Rev. 4.
I'll get to that in just a minute. But one thing
I want to point out that's very much more procedural and
philosophical than the equation is there is definitely a
feeling among the staff that we want to move to the use of
surveillance data as a check of the correlation rather than
as an index to the correlation.
The diagram that I showed you before for delta
T30, if you remember, it had the choice branch, where you
decided if you had credible surveillance data or not, where
credibility was judged as to whether you had more than two
points or not.
So right now, if you've got more than two points
and they're reasonably close to the mean, you change the
whole embrittlement trend curve by moving it up or down to
those two data points. From discussions among the staff and
indeed discussions that have gone on within the ASTM
irradiation embrittlement community, there's, I would say,
definitely a consensus developing that that's not really a
very appropriate engineering procedure and what we should do
is move towards use of surveillance data as a check, which
is to say we still encourage the licensees to do
surveillance.
It provides more data. It keeps us from going
wrong. But we're not going to change unless the
surveillance data is just way off the mean curve, perhaps
more than three sigma out, and that still needs to be
determined.
It doesn't seem appropriate to change the whole
view of embrittlement of that particular material based on
two data points, when you've got 800 sitting back here
saying no, no, no, it's going some other way.
So like I said, that's a procedural and
philosophical change --
DR. POWERS: Before you take that equation off.
MR. KIRK: Yes.
DR. POWERS: It is remarkable for its level of
parameterization with 600 data points. It looks to me,
however, that you don't have a saturation effect built into
this equation for copper. You have a saturation effect
built in for fluents crossed with copper.
MR. KIRK: Yes. This is the danger of putting up
a particular equation. I honestly can't tell you if this is
the one with the copper saturation or not. But the one
that's being considered in the end is -- if I had to give
you my best guess right now, nine chances out of ten, you're
going to have the copper saturation term.
If you don't see it here, my error in putting up
the equation.
DR. POWERS: I just look at it.
MR. KIRK: Yes.
DR. POWERS: There's no cap on the effect of
copper.
MR. KIRK: Yes.
DR. POWERS: There's a cap on the cross with
copper and the fluents term.
MR. KIRK: Yes.
DR. POWERS: And it seems that you get a cross
also with nickel in a peculiar fashion, and it's remarkable
in light of the phase diagram.
MR. KIRK: Like I said, the problem of putting up
an equation for illustration purposes when you're not
prepared to talk about it.
But this was just one of many and is not going to
be the final one, because it's being revised as we speak.
So hopefully depending --
DR. POWERS: there must be an enormous amount of
structure to your data set.
MR. KIRK: Yes. We could go on forever.
DR. POWERS: I mean, usual metallurgical data like
this has enough scatter that straight lines and things like
that seem like appropriate.
MR. KIRK: That's, in fact, one -- those are some
of the things that, of course, we've struggled with and
that's one thing -- well, the process that we're going
through right now is we're writing the tech basis document
to support whatever equation one of us might show you at a
future date boxed in yellow.
And this is part of the process that I'm showing
here. Another part of the process that I think we need to
-- well, let me back up.
It's very easy to put up a graph of some effect
based on data and standing in a room and convince people
that you know what's going on. I find it much more
difficult to convince people that you know what's going on
if you force the investigators involved, and I include
myself in this, to put down the graph and basically write
the paper.
So convince me, let's write the technical basis,
and that's what we're -- that's the rigor that we're trying
to put ourselves through in terms of getting any particular
effect into this model. Let's convince ourselves that we
have an appropriate combined physical and statistical basis
for these effects, and this is our sort of provisional
strategy for focusing our attention on this, is that we sort
of divided a physical basis into a well accepted physical
basis, perhaps a plausible one, and one that's just not
established, that we don't know what's going on.
And then we can look at our statistical evidence
and say we either have strong evidence for an effect, say, a
correlation coefficient in excess of 95 percent, or a weak
statistical effect, perhaps a correlation or a confidence in
excess of 70 percent, but still with a coefficient you can
calibrate.
And what we did is we just sort of boxed this up
and said, okay, well, certainly if we had a well accepted
physical basis and a strong statistical basis, that effect
would be included in the model.
And if you had weak and not established, you'd
never consider putting it up. Obviously, there is a huge
gray zone in between. In our initial thoughts on this,
we've placed perhaps a bit more stock in the statistical
evidence than in the physical evidence and we felt that if
we had something that was a very strong and demonstrable
statistical effect within the power reactor database, even
if we couldn't establish a physical basis for it, we felt
that that was something that, from a regulatory perspective,
probably would be included in the model, accepting that it
would be going under a lot of scrutiny.
Conversely, if you had something with a weak
statistical basis and perhaps only a plausible physical
basis, that would be a little bit more dicey in terms of how
it gets in.
Obviously, there are no -- it's hard to draw a
line on this, but this is sort of a process we're trying to
put ourselves through. And ultimately what we'll be doing
is publishing a reg guide for public comment, publishing a
tech basis where each and every term is gone through in this
way.
The staff authors and the contractor authors will
basically have to come to the table and say here is where we
think it is and then it will open for -- the whole process
will be open for public critique.
So what at least the goal of this committee is to
get the debates squarely on a technical level and not on any
other level. It's the only way to proceed.
The status right now is that we're finalizing the
model or trying to. We've frozen the database. That's sort
of a necessary procedural step, because there's always one
more data point showing up and at some point, you've just
got to draw the line.
We've at least proposed a gating criteria for term
admission and right now, in order to try and get an
embrittlement model to Shah and to Terry and for them to
use, what we're doing is we're writing mini basis documents
so that we can try to get an embrittlement correlation to
them in the April timeframe that is hopefully no different
and, if anything, not much different than that which is
supported by the final tech basis document, which is due in
December.
And I think I just said all this. We're trying to
get this to Shah and his workers by April-May and then once
we've got the sort of the mean curve established, then all
of this knowledge feeds into the K-1-c and K-1-a uncertainty
framework and analysis that's being done by the University
of Maryland.
The deadlines for Reg Guide 1.99 Rev. 3, tech
basis document in December of 2000, draft for public comment
available middle of next year, and also just point out other
activities in the public domain is that ASTM E-10 has
ongoing technical interest in this area and they will be
evaluating the model for potential use in the E-900
standard.
With this, that's my last slide. So if there are
any questions now, I'd like to entertain them.
DR. POWERS: I guess I'd like know -- I'd like to
understand better about the rigor with which you are
approaching this problem. If we could look at your slide
29.
MR. KIRK: I'm sorry?
DR. POWERS: Is this one of your slides, 29?
MR. KIRK: No, sir.
DR. POWERS: Okay. Then I can't ask the question.
MR. KIRK: You'll have to get the next guy. Any
questions on slides lower than 26? Less than or equal to.
As my parting shot, I wanted to point out the next speaker
will be Doug Kalinousky, also from Office of Research,
Materials Engineering Branch. He is going to be talking
about statistical analysis of chemistry and RTNDT data and,
again, just to express it in the overall uncertainty
analysis framework, where this goes into the diagrams, RTNDT
unirradiated is up here, feeding into K-1-c uncertainty,
whereas the copper and nickel values are way back here in
the embrittlement correlation.
I'd just point out one other thing. This is a
diagram of the current embrittlement correlation process in
Reg Guide 1.99 Rev. 2. We put this up so we have something
to talk about. Ultimately what goes into this analysis is
going to be different than this because we're going to have
a new process and a new correlation and new data.
DR. POWERS: Let me ask you. You've mentioned
several times the word rigor in your statistical analysis.
It's been my experience that rigor is a relative thing. Can
you give me an idea, some understanding about the strictness
of your rigor?
MR. KIRK: The short answer is no. I think that
would have to be something that you would judge when you see
the product.
The problem -- and I'm not a statistician, so I'm
probably not going to provide you with an acceptable answer.
As I understand it, the problem with non-linear analysis
such as these is there is no one single right answer. If
this was Y equals MX plus B, we could talk about rigor.
It isn't and therein lies the problem. So you've
got a lot of engineering judgment going into what you then
apply fairly routine statistical tests, like student T and
analysis of variance type things, too.
So once -- really the points for discussion, at
least I think, and this might be a better question when we
can present you some more of those results and maybe that's
something you'd like to ask for next time, I think the
points of discussion are perhaps not going to be so much in
terms of the statistical tests that are applied, because the
statistical tests that are applied are, in fact, first year
statistics.
It's going to be on the engineering judgment that
we use to say, okay, this is an appropriate subset of the
data to try to apply a student T test to. That's what we
keep arguing about at least. So I think that's where it
comes in and so that's going to be an argument of
engineering judgment that's motivated by people's
understanding of embrittlement damage mechanisms, in that
case.
MR. HACKETT: This is Ed Hackett. Let me try a
slightly different take on Dr. Powers' question. One of the
areas where we have introduced, I believe, a high level of
rigor to this process is in screening and selection of the
data that went into the database, and there I can cite the
benefit of working jointly and cooperatively with the
industry on this.
For instance, some of the temperatures for the
irradiations previously involved melt wires and other forms
of selection. It was very rigorously scrubbed by the
industry and the ASTM folks this time around to just use
downcomer temperature.
So there was a lot of -- that's just one example,
but there was a lot of rigor that went into selection and
screening of the data that are in this database, and then I
agree with what Mark said subsequent to that, but there's a
fair bit more rigor in that process now than there was in
the previous version of the reg guide.
DR. SHACK: If there are no more questions, it's
probably time for a break. Come back in 15 minutes, 10:25.
[Recess.]
DR. SHACK: I'd like to come back into session,
since I suspect we're going to be running hard-pressed on
our schedule today.
MR. KALINOUSKY: I'm Doug Kalinousky. I'm with
the Office of Research, Materials Engineering Branch. Our
objective in this portion is to determine the chemistry
variability and RTNDT-0, initial RTNDT variability
distributions.
We used the NSE database for copper and nickel and
initial RTNDT values. We are trying to determine
heat-specific distributions, to determine the means of the
distributions and the variability, the standard deviation of
these.
We also are attempting to get the local
variability in a small area of the weld or plate.
We did this within a little sub-region that's used
in the FAVOR code and we are debating still whether the
through-thickness as the crack grows or not, because as the
crack grows, it might run into different coils that were
used in manufacturing the weld. So we're still debating
whether that be applied to the code or not.
I did this with myself, Tanny Santos, who is off
in Canada skiing right now, and Lee Abramson was our
statistician that we used as a consultant heavily.
DR. SHACK: What is a heat-specific distribution
of copper? Is that really the same thing as local
variability?
MR. KALINOUSKY: That would be the whole heat that
we have data for.
DR. SHACK: A heat.
MR. KALINOUSKY: A heat number. We would try to
find the mean from all the different data we have and the
distribution about that mean. The local variability instead
would be as if in the code, we have broken the welds down
and like two to three inch sections and we say the
variability in that section.
If we already assigned a mean to a point and we go
to a different point, what would be the variability between
those two points.
So we went through -- we used a couple of reports
in this thing, one from the CE owners group and one from the
B&W owner group, and we used all heats we could find with
five or more data points, so we'd have a fair representation
of the standard deviation of those mean values. We found 24
heats for copper and 39 for nickel.
We determined a mean value based on the five or
more data points for each heat-specific means. Then we used
that to also find the standard deviations.
We went ahead and plotted these out, you'll see
the next two plots would be the -- next two slides, that is,
would be a plot of the standard deviation and the mean. And
for the copper, we'll go into this next point as we show you
this slide.
DR. POWERS: Let me understand this last line.
The previous speaker mentioned statistical rigor. You have
uncertainty in the mean values and you have uncertainties in
the standard deviation values. So you are going to use a
linear regression technique that presumes precision in the
independent variable. Why?
MR. KALINOUSKY: Because we have very little data
to go by and there really is -- we did a -- this is what you
were referring to, obviously. That was the plot where we
had the large scatter. But we noticed that there is
definitely a trend as the mean value increases, the standard
deviation is increasing.
DR. POWERS: That's not an excuse for using
linearly squared statistical techniques. They presume that
there is no variability in the values for the independent
variable. Why wouldn't use something like a min/max
procedure?
MR. KALINOUSKY: We did what's called a K-ring
squared test to test the --
DR. POWERS: You can test it till the cows come
home. The fact is that you have assumed precision on the
horizontal axis here.
MR. KALINOUSKY: Yes.
DR. POWERS: And there's not. And there's another
technique for fitting the line, a min/max technique that
takes into account that there is uncertainty both in the
independent and the dependent variable. Why not?
MR. KALINOUSKY: Because we --
DR. KRESS: Does it matter, unless you're going to
--
DR. POWERS: It's going to change the slope of the
line substantially.
DR. KRESS: Yes, but that matters only if you're
going to extrapolate outside this data, you think?
DR. POWERS: Even if you're going to interpolate
it, it changes the slope of the line, it's significant.
What happens? When you have a non-standard deviation, it
tells you that you're plotting the wrong variable. It
should be something like the square or the square root or
something like that.
Okay. But you don't care. All you care about is
the linear variable anyway and you'll live with a varying
standard deviation, I assume. But now that slope of that
line becomes very critical to you and if you've got
uncertainty on both axes, you've got to use a statistical
technique that's appropriate for that, especially if you're
going to advertise it as statistically regressed.
MR. KIRK: Mark Kirk, RES. That's certainly --
that's a good comment and that's something that we can take
away and have Lee Abramson, our statistician look at. But
just to clarify, you're concerned about what, measurement
error in the mean value?
DR. POWERS: You surely have some variability or
you wouldn't be plotting standard deviations here.
MR. KIRK: Right.
DR. POWERS: I assume that you took -- you had
five determinations. You found the mean of those five
determinations and that's what I'm looking at down here.
MR. KIRK: That's correct.
DR. POWERS: And then you calculate the standard
deviation by squaring the differences and dividing by four
or something like that and then doing the square root, and
that gave you the standard deviation.
MR. KIRK: Right.
DR. POWERS: Okay. So there is some variability
in that one.
MR. KIRK: Well, one thing, if I could just
interject momentarily, that I do want to point out is that
what Doug is presenting is largely a data collection effort
to provide input or what I would call seed information to
the uncertainty analysis that's being conducted for us by
the University of Maryland contractors and is going to go
through the PRA process.
So ultimately, for example, if you had a parameter
box on the uncertainty diagram that was labeled copper,
you'd have two boxes coming out of that that's labeled -- or
you'd have one box at least coming out of that labeled
measurement uncertainty.
So it might not be the -- the goal here is just to
inform you as to sort of the status of our data collection
effort and present some overall trends. But the analysis
methodology is ultimately going to be captured in the
overall uncertainty analysis and therein every time we've
got a measured variable like temperature, copper, whatever,
there is the explicit question asked of do you need to
account for measurement uncertainty or not.
So I think that's a good point to bring up, but
that's also something that's going to be considered.
MR. KALINOUSKY: Also, as we did these, not all
the -- some of these -- a lot of these points are based on
ten values of measurements or more. And basically what we
did was I continually filtered out more and more. So I
removed the points like less than eight values and I'll
remove some of the points with less than ten and I'll remove
some more.
And the trend is still there and the slope of the
line really didn't change that much for the more certain
mean values. So that's one we did do to try to validate it,
but we can also -- I'll ask Lee Abramson about what you're
saying and see if we're going to do another rigorous way of
doing that and see if we come up with the same idea or not.
One thing also I wanted to point out about this
graph is that there's no difference between the CE heats and
the B&W heats in this one. They're basically all
intermingled. Some of these are CE, some are B&W welds.
There really is no trend as in one is high standard
deviation and one is low or anything like that, which, in
the nickel term, which is the next slide, there is a
difference.
Here we didn't attempt to put a line through this
once we plotted it out, because it's obviously grouped, two
separate areas. In this area here, we have -- the majority
of these are B&W heats. These are all CE welds. Up here
are high nickel addition welds, which are also B&W welds.
So we are still looking at this, how we should approach it
and how we should use the data the best we can for the heats
we're using, because also the same problem we have with the
copper is the heats we're using is in the PTS plant analysis
aren't all represented here.
Some of them only had one reading, so we can't
give it a mean or standard deviation, or might have had two,
so we didn't have very certain about the standard deviation.
So we can't just use the data that we have right now and say
that's the heat mean, the heat standard deviation. We have
to find some way of we have one plant that has one reading
of a mean, so the mean would be about .6 something, and we
don't know where to plot it on there, because we have to
find some way of making that determination, and we're still
looking at that.
So based on those plots, and by using the other
K-ring squared test I talked about, we've determined that
the copper could be either normally or lognormally
distributed. The readings for those means we have could be
either way. We'll be doing a -- in the final FAVOR code,
we'll do a sensitivity study and compare the two.
DR. POWERS: How did you determine that the copper
could be either normally or lognormally distributed?
MR. KALINOUSKY: We used -- we tested both of them
and both of them were acceptable to our limits that we were
measuring. There might be other ones that would fit. We
didn't test every --
DR. POWERS: What does it mean you could have a
normal distribution with a -- with something other than a
constant standard deviation?
MR. KALINOUSKY: Say that again, please.
DR. POWERS: Your standard deviation isn't
constant.
MR. KALINOUSKY: For a given mean, it is. At a
given point, it's constant.
DR. POWERS: For a given mean, it's constant
because --
MR. KALINOUSKY: This relation here would be based
on that line. Basically, it's the equation of that line is
what all this is. So you have a given mean, multiply it by
this constant value, gives you the standard deviation at
that point.
DR. POWERS: That's not a constant standard
deviation. Well, go on.
MR. KALINOUSKY: For nickel, we couldn't do this
and we are still looking at it for the same reasons.
DR. SHACK: What you're saying is that for a heat
with that mean, then you're getting a distribution of copper
in that heat. Is that what you're trying to say? Is that
what this is trying to do?
MR. KALINOUSKY: Right. That's right.
DR. POWERS: Whatever that means. I mean, it
seems to me that you have prima facie evidence that standard
deviation is not constant with copper. How can it possibly
be a lognormal distribution? It could well be lognormal
distribution on the square, just looking at it, as a guess,
the square of the copper concentration of some transform of
it, but it's not obvious that -- to me, at least.
MR. KALINOUSKY: The next step we did was go for a
weld local variability, which is what I said before would be
the variability in a small area. We used a CE report we
were able to get, had data for eight weldment blocks and we
had five measurements at a quarter-T depth. So we used
those five measurements from those eight blocks and we just
calculated simple standard deviation for both nickel and
copper and both of them came approximately about .01.
This was also independently done by -- it was
Matthew Vaughan and -- who was the other one? Yes. Steve
Byrne. It's Steve Byrne and Matthew Vaughan that did those
also and they also came up with the number approximately
.01, as well.
We can't classify what type of distribution it is
right now. We still have to look at that and analyze it and
-- but the reflexes say it's normal, but we'll have to
verify it through some statistical method.
Through wall variabilities still needs to be
determined. We have some data we can use, if they determine
that we should use that in the FAVOR code.
DR. SHACK: Again, I'm confused on this one.
MR. KALINOUSKY: Okay.
DR. SHACK: Are my eight blocks from a single
weld?
MR. KALINOUSKY: Different weldments, weld blocks.
So they made a weld -- a weld heat of -- one type of weld
heat, then another type, eight individual ones.
DR. SHACK: But the same weld wire.
MR. KALINOUSKY: No, different heats, different
weld wire heats. Does the backup slide show? This one.
MR. MALIK: Right here.
MR. KALINOUSKY: Okay. Anyway, it was a weld
block, with the weld heat, and simply they just analyzed it
for the content for the nickel and copper. Each individual
made by different weld wire heats.
DR. SHACK: So I've got eight different welds.
MR. KALINOUSKY: Right.
DR. SHACK: And I take a T-over-fourth block from
each one.
MR. KALINOUSKY: Right. They measured the T
depth.
DR. SHACK: What does that have to do with the
local variability?
MR. KALINOUSKY: Those points would be across the
welds, so they would only be half an inch apart, quarter
inch apart. So we've got the variability as you go across
the weld at a certain depth. These are all -- so basically
you're saying how does point --
DR. SHACK: I see. You're spacing them over the
T-over-four.
MR. KALINOUSKY: Yes.
MR. KIRK: In the FAVOR -- it's perhaps important
to point out that in the FAVOR code, there are sort of two
different versions of local. One is you start off in your
sample, you generate a sample from a region, which could be,
say, the beltline weld or the plate or whatever.
So say you take a sample from the beltline weld.
That beltline weld is then cut up into iso-fluents regions,
regions over which we treat the fluents to be constant based
on the fluents maps, which Shah is going to show you.
So now you have a region depending upon fluents
variability that may be something perhaps big enough to hold
in your hand. And the question was raised in some of the
public meetings that we had on this that, okay, now, in your
analysis of that vessel, you go through and say on run one,
you seed a flaw into the circ weld, sub-region B. Then on
loop 386, you wind up with a flaw in that same sub-region.
So then the question arises, in this analysis of
this vessel, you had some Monte Carlo simulation of what the
copper, what the nickel, what the composition of that region
of material was, and the question came up, on run 386,
should I now go and resample from the whole distribution,
which is sort of what Doug was showing you earlier, or is
there some smaller tighter standard deviation that you
should be sampling from.
So what we were trying to do was to look at what
data is available where you've got reasonably --
measurements of material composition reasonably closely
spaced to try to make an assessment as to whether that
resampling should be done from a smaller standard deviation
or not. That's sort of the goal here.
MR. KALINOUSKY: These are what they looked like.
We were taking these values here across the weld. That's
how come I got the local variability there.
If you had to do a through thickness, this is what
basically we would be using as a data set if we need to do
that.
So we moved on to plate chemistry and here we have
even less data. For every heat, we only had one or two
points, so we -- then we couldn't get a standard deviation
or any way to really analyze those.
So basically what we suggest doing here is just to
take the best estimate we have and let's do not sample about
it. Just say that's the best estimate and then we'll go the
plate local chemistry variability and sample that.
For the plate local, here we're able to get three
groups of data, once again, not much, with six points per
each group. This was -- these came from surveillance
specimens from St. Lucie and -- I can't remember the other
one offhand right now.
Anyhow, we analyzed these and we found standard
deviation to be, for the plates to be about .002 for the
copper and .005 for nickel. So it's what we expected, very,
very small, since plates are very homogeneous.
And we'll sample the previous mean using this
standard deviation to give us a final value to put into the
FAVOR code.
Let's move on to the initial RTNDT values. Once
again, the amount of data we have is always the hard part
here, getting enough data to use. So we pulled data out of
RPVDATA. We grouped them by the heats and we used every
heat we could find that had three or more measurements, so
we have some idea of a standard deviation.
And that gives a total of 19 heats and a total of
65 data points. What we did here is we did a transformation
of the data. We took the -- for each set of data, say, we
had five values for that heat, we'd take a mean of it for
the heat mean here, and we subtract the measured value to
give us a delta value. So basically just transforming the
data to a plus or minus around the average.
What we did do then was we graphed all those out
with a histogram and came up with here -- with the blue
would be the data we used and the red would be a fit about
that data, and it came out to be a normal -- with a standard
deviation of 16.6. So what we propose to do in the FAVOR
code would be to generate a random number and let's say we
get a .7 or so, go across here till we hit that, come down
here and say, oh, it's plus .8. So we had that to our best
estimate mean to give us some variability about that mean.
And we did the same approach with the plate, as
well. Here, once again, we had a little bit more actually,
more data for it. We had 128 total data points out of 37
heats. We did the same approach, transforming it with a
delta value, and we went ahead and plotted that out and we
ended up with this type of fitting, where it shows -- comes
out pretty normal, the values, in both cases.
Any other questions? If not, we'll move on to Dr.
Shah.
MR. MALIK: The next item in the presentation is
developing detailed fluents maps for application of the
plant-specific analysis and we have developed end-of-life
fluents maps for two plants, and we are using available
cycle-to-cycle fuel loading histories.
And also along with that, another objective is to
determine what is the uncertainty of the fluents. We are
starting out to perform in the FAVOR analysis an initial
estimate of one sigma in fluents to be roughly like 15
percent of the mean, which is much better than earlier from
the laboratory, 20 to 30 percent were used. So it's a real
improvement from that point on.
And the methodology for fluents calculation are in
the draft guide on dosimetry, 10.53. It was released in
1999 and another NUREG CR-6115 and this work is being
monitored at RES by Bill Jones, as well as work is being
performed at Brookhaven National Lab.
Two plant specific neutron fluents maps have been
completed. One was Palisades. There was Robinson, but they
opted out, so we weren't able to use that one, and the
plants next to be analyzed are Oconee, which we will be
finishing up in March, Calvert Cliffs in July, and Beaver
Valley we are expecting sometime later on that.
We are using defined actual circumferential as
well as radial grids to calculate fluents values. For
example, Palisades, we have 205 axial, 97 times eight,
there's one-eighth symmetry on the circumference.
Similarly, like between 20, around 12 to 20 radial grid
points have been used in those two.
Also, we found that the fluents decay in Reg Guide
1.99, which is like minus .24X, is a bit conservative and we
will show you a graph on that.
Here is a detailed plot, the circumferential
horizontal direction, as you go around the circumference
from zero degree to 360 degree and you have peaks here in
the beltline area. Mid-core area, there is a peak, and as
you go around circumference, this happens to be the area
core flats are located. Core flats are the region where the
reactor core is very close to the reactor vessel. So these
four areas are the core flats and the reactor vessel are
very close to those, so that's where you see those peaks.
And this fluents curve is right at the mid-core as
you go along the axial length. At the top of the core is
substantial drop-down.
Similarly, I have a plot like that for axial
variation. Again, in the mid-core level, here is the axial
variation, end of the core, top of the core, and here is a
mid-core area, where it's the peak values. And you go
around the -- this is the core flat area, and other angular
locations, these are the values.
And because of this variation in fluents, we have
to subdivide the region and perform the analysis in the
FAVOR code.
Here is the exponential decay I was telling you
about. These are actual radial distributions of flow
through thickness, peak volume in the inner radius, and as
you go to the outer radius, it drops down, where minus .24X
is used in Reg Guide 1.99 Rev. 2 and this is a straight line
on the log graph.
Actual -- most of the initiation and all the PTS
significant transients with the crack, quarter T on this
side, there is very little difference. And even in this
area of the graph, actually the crack was just initiated and
the crack arrest takes place in this deeper part of the
crack depth.
DR. KRESS: But cracks closer to zero or two
inches, where the curves are pretty close together, are the
ones you worry about.
MR. MALIK: Yes.
DR. KRESS: Okay.
MR. MALIK: Okay. The next item is the
development of the FAVOR code. FAVOR is number of fracture
analysis of vessels in Oak Ridge and it implements refined
PFM technology and up-to-date materials data and we are
trying to make it consistent with the current PRA, as well
as thermal hydraulic input data.
In research, it's myself, Nathan Siu and Lee
Abramson from PRA site and the contractor -- the main
contractor is Terry Dixon, who is present here. And
University of Maryland and PRA areas are Professors Modarres
and Mosleh, as well as input from Professor Natishan in the
fracture toughness area.
The code is being used to answer to the kind of
question, one, at the given -- at what point in the life of
a plant will the acceptance criteria, risk acceptance
criteria will be exceeded; for example, at present it's five
by 10E-6, because your failure per reactor year. So if you
are plotting effective from power year versus risk, in terms
of failure, then you want to find out at what point this
acceptance criteria is exceeded and plus what would happen
if you have improved methodology or mitigative action we are
taking, what will the effect of that and how much more plant
life can we improve with all of that.
As you can see, it involves a number of different
items. It starts out with a detailed fluents map, flaw
characterization, plates, as well as weldments,
embrittlement correlations to define shift in RTNDT, thermal
hydraulics transients, and PRA such as event frequency, what
are the credible sequence for PTS significance, the reactor
vessel integrity database to define material chemistry, as
well as industry database.
Also, along with that, and the extended fracture
toughness initiation and arrest, and the defined fracture
analysis methodology. They are all combined together to
come up with a method to use for PTS analysis.
In addition, we are doing some additional
development work here and trying to bring that, such as
effect of 3D code, plume and things like that. We are also
looking into that, as well.
Based on all of these integrated together to come
up on a plant specific or on a generic basis analysis to be
performed and then they feed in to finally revising the
screening criteria.
It combines -- the FAVOR code combines the two NRC
funded codes. OCA-P was historically developed at Oak
Ridge, as well as VISA-II. So those two into a single
combined code with all the best feature from the two
combined together.
It also incorporates the lessons learned from the
Yankee Rowe in early 1980s, as well as from IPTS analysis in
mid '80s. Now, the code is in the third generation, so we
have just in '99 released a version of the code. This plan
is to continue development of technology derived from NRC
analyzing history, available research and data.
This is, again, a list of the same thing, what are
the features of flaw characterization in plates and welds,
the map, embrittlement correlation, reactor vessel database,
fracture toughness, and here we are not using surface
breaking, as well as embedded flaw. Both types of flaws are
being looked into. This is the first time we are analyzing.
So we are taking one big step instead of assuming all flaws
to be surface breaking. We are using surface breaking, as
well as embedded flaws.
And as well as we're including the through wall
residual stresses in the reactor vessel welds.
As I said earlier, an interim version of the code
has been released in October and the next version is planned
to be available by May, and we are implementing, with a lot
of industry, a discussion on ways to improve it, make it
user-friendly and efficient. One of our end goal is to have
common understanding, what are the methods that are going
into the analysis.
Here is a little bit -- a few slides to show what
kind of independent verification we are doing. For example,
here we went from FAVOR, which is an asymmetric code. We
perform analysis using ABAQUS code and tried to compare what
were the total gradient through thickness for PTS
transients, FAVOR results shown in the red, as well as --
sorry -- in the rectangular black color, and ABAQUS results
are shown as well.
And similarly, this is the resulting hoop stress
from a thermal gradient shown here.
DR. KRESS: The hoop stress is varying because
your pressure is varying. So that's just a plot of how good
it predicts pressure.
MR. MALIK: This is a plot of temperature and this
is the hoop stress through the thickness.
DR. SHACK: There is a thermal stress contribution
to that.
MR. MALIK: Yes.
DR. SHACK: But what was this temperature -- I
mean, what was the -- you just changed the temperature on
the surface? What problem are we really looking at?
MR. MALIK: It's for exponential decay
temperature.
MR. DIXON: Terry Dixon, from Oak Ridge National
Laboratory. Actually, there's been many verification and
validation problems done. This particular one is for a
stylized exponential cool-down rate, but I could just as
easily put together slides for discontinuous functions such
as repressurizations.
It's a finite element based code. So it will
handle any thermal hydraulic boundary conditions that you
want to impose on the inner surface of the vessel. This
particular one is for an exponential decay, thermal, and I
believe it was a constant pressure, but this also includes
the through wall weld residual stress, as well as the clad
based differential thermal expansion.
DR. SHACK: But this is all -- this is a truly
axisymmetric problem.
MR. DIXON: Yes.
DR. POWERS: So you're not looking at what the
limits are on how much variation you could have asmuthally
and still get it.
MR. DIXON: No. This is a finite element
analysis.
MR. MALIK: This is the verification of the
stress, again, for this one, I think it was for the region
around the crack front and as you can see, both comparing
them with FAVOR, using ABAQUS solution as well as FAVOR, now
here is the depth point and here is the point along the
circumference of the crack, and the K solution are pretty
much matching.
The reason they are matching here is because the
equation that went into the FAVOR code originated from the
finite element itself, this should match up very closely.
And here is the comparison of -- this is for the
surface breaking flaws. Now, this is for the embedded
flaws. Here we are showing three definition solutions here
and both open and close symbol are showing here for our
calculation in FAVOR. These three are for three different
distances away from the inner surface, but this one is very
close, this is a little bit away from the inner surface,
this is farther away from the inner surface. So there three
different solutions.
Now, this shows a detailed fluents map and fluents
in the mid-core area through the circumference, very
significantly, and to match that, we need to divide the
beltline area into a number of segments, called sub-regions.
Here is the axial weld, the plate area, then the
circumferential welds, and the lower axial weld and the
lower plate area.
So we are dividing into a number of sub-regions to
as closely as possible see the distribution through the
vessel beltline area, both axially as well as
circumferentially.
This is a sample calculation. It shows that
application of this methodology in the FAVOR code can
improve the or extend the life of an operating plant. It
was done for Calvert Cliffs, NUREG report CR-4022. We have
taken the same PRA, as well as the same thermal hydraulic
results, but only the fracture mechanics part has been
varied. There are four different codes showing over here.
The effect of full power year or the RTNDT values
versus what is the probability of failure per reactor year.
The first -- the top curve is for surface flaw distribution.
That was the flaw distribution available before we have our
own distribution, we are just working on that.
So there were just surface breaking flaws and Reg
Guide 1.99 Rev. 2, the correlation. The top curve. Here is
the acceptance criteria for the risk, five times 10E-6, and
it shows up like 32 effective power year.
Whereas if we take the surface flaw distribution
and improve the correlation, it's one -- the embedded
correlation has been used, not the one that you're going to
be using.
DR. SHACK: A revised.
MR. MALIK: A revised, yes. And you see a
significant improvement, as you can see. At this point,
it's almost like it doubled the life in the plant.
The next step, what happens if you just use the
PVD -- a distribution in the process. It has only embedded
flaws. There were no surface breaking flaws in it. But
using Reg Guide 1.99 Rev. 2, earlier correlation, you see
this curve here. A significant improvement compared with
the top one.
Now, what happens if you combine the two together?
Here is the last one in which you have used embedded flaw
distribution from PDF and the revised correlation gives at
least an order of magnitude on the curve.
DR. SHACK: Now, these are presumably done with
the old K-1-c and K-1-a distributions.
MR. MALIK: Yes. There will be some more benefit
derived from that as well, yes.
DR. SHACK: I thought they got worse when they did the
statistical analysis.
MR. MALIK: You cannot say that it has to -- there
are three different transients considered here and in some
cases, it goes up. So this is all three transients
considered together.
In summary, work in the PFM area is coming along very
vigorously, actively, and some of the major technical
activities are in the correlation of fracture toughness.
We'll be completing in the April to May timeframe. The
plant release the FAVOR code with those in May to June
timeframe.
We are implementing those technical enhancements
as they become available. We don't want to wait around for
them.
And we come to new coordination and interaction
with PRA and thermal hydraulics sub-group to bring their
ideas into especially the uncertainty analysis part of the
program.
And there are some delays, as you can see, one of
the plants moving out and replacing it with another plant
means we have to go do fluents calculations and some of it
is materials related, as well as frequency and all those
things, systems, it needs to be done again.
At least for the PFM part, we see about two month
lag on that.
This is my part of the presentation, if there are
any questions.
DR. SHACK: I don't see any further questions, so
we can move on to the flaw distribution, I guess.
MR. MALIK: All right.
DR. SHACK: From rocket science to expert opinion.
MR. MALIK: Debbie Jackson will be the one who
will tell us about that.
MS. JACKSON: I am going to give you updated
information on what's going on with the development of the
flaw distribution. That's part of this PFM work. This is
just a quick list of some of the topics that I'm going to
discuss today.
I'm going to go over the background, which I think
Shah has touched on today; the approach that we're using; a
little bit of information about the reactor vessel
fabricators; the material that we're using for developing
the flaw distributions; the expert elicitation process and
some concluding remarks.
DR. POWERS: When you say flaw distribution, are
you speaking strictly of density and size or do you include
orientation and location?
MS. JACKSON: Density, size, location,
orientation. I'm going to get all that information.
DR. APOSTOLAKIS: Is it really expert opinion
elicitation process rather than expert elicitation process?
MS. JACKSON: Expert, yeah, expert judgment
process. We kind of -- I was using that interchangeably,
but actually, yeah, it's expert judgment and elicitation is
one section of it.
These are the objectives of the presentation.
I'll discuss the need for the generalized flaw distribution,
talk about the process, and then discuss the status.
This is the background, which was discussed a
little earlier today, as to why we're doing all this work
we're doing with the PTS and the flaw distribution is an
important input to the fracture mechanics calculation, so
that's why we're going through this effort.
And we believe that the fabrication process
presents a number of variables that we need to review for
the flaw distribution; specifically, the fabrication process
and the different welding processes that are used.
We're going to go over a little bit about the
expert judgment, why we're doing it. It's needed to review,
interpret and supplement available information on the
reactor vessel fabrication process. A lot of the people who
are involved in the actual fabrication processes for reactor
vessels are getting up in age and we don't have a lot of the
information here. So that's why we decided put together
this expert panel, so we could get people who are actually
involved in the fabrication process.
This is a list of some of the reference documents
that I've used. The NRC has done some expert judgment
processes in the past for other subjects and these were some
of the documents that I used just for reference in terms of
determining how you go through the expert judgment process.
In addition, Lee Abramson, who is going to do a part of this
presentation, he has been involved with the majority of
these elicitations or expert judgment processes.
This is a list of the domestic reactor vessel
fabricators, Combustion Engineers fabricated a majority of
the vessels. Babcock & Wilcox, Chicago Bridge & Iron,
Rotterdam, and New York Ship Building, and this data was
obtained from the reactor vessel integrity database, the
RVID, which NRR is responsible for putting together.
Those numbers were just the operating reactors,
the ones that are presently on line.
This slide shows the material that we're using.
Midland was done some time ago and PVRUF, Shoreham, River
Bend and Hope Creek, which were being examined by Pacific
Northwest National Lab, they are all done using an upgraded
SAFT UT system. So this is the current pieces that we're
using.
The PVRUF, Shah mentioned this briefly, this was
completed. One issue came up in one of the meetings that we
had with industry sometime late last year. They asked a
question, they said a lot of the -- the majority of the
material that we have was weld material, so what are you
going to do with the base metal, because there's so much
more base metal, and the numbers that are presently being
used for the base metal were just kind of developed through
discussions with some of the experts.
So what we have decided to do, we have started
actually inspecting some of the base metal so that we can
get a valid distribution for that.
DR. SHACK: Now, EPRI is also doing some
evaluation of the flaws in these weldments, right?
MS. JACKSON: Right. The Shoreham material
specifically is what we're working on with EPRI. PNL has
done some exams of the Shoreham vessel material and then
we've sent the material to EPRI so that they can use the
methods that are currently used in the plant, because the
SAFT UT method isn't presently used in the plants. So
that's what EPRI is doing.
DR. SHACK: So their goal is not to characterize
the flaw distribution, then. It's to benchmark the current
techniques through the SAFT.
MS. JACKSON: And also to verify some of the data
that we have, just kind of like a backup of the information
we have.
I just have a very old photograph that I have that
I found going through some paperwork that I had. This shows
one of the vessels being fabricated at Combustion
Engineering.
This is one of the methods where they -- you can
see the weldment here. There are two methods that they used
to make the rings. One of them, they actually did the
forgings, and another one, they used three plates and they
weld them together to form a shell.
As you can see by the by this picture, it's very
old. This was taken in the early '60s.
The data that PNL is gathering from the PVRUF,
this is how it was determined that they were going to
categorize the flaws just for ease of classification and
determining what we would use, because there is a different
flaw distribution -- well, a different number of flaws in
the welds versus the base metal.
And a lot of the flaws so far from the PVRUF were
found in the fusion lines or they were found in repairs,
weld repairs. The largest flaw in the PVRUF was found in a
weld repair and that was 17 millimeters.
This graph shows the comparison between the
Marshall distribution, which was the existing flaw
distribution that was used for many years, and this is the
PVRUF data that we have. There are approximately 2,500
indications that were found in PVRUF.
DR. SHACK: These are combined flaws, right?
You're not discriminating here between this is the planer
flaw, this is --
MS. JACKSON: Right. These are just all the --
DR. SHACK: All the indications.
MS. JACKSON: Yes. These are just all the flaws.
All of the different flaws. And we have some data from the
Shoreham vessel. They've just finished doing the UT exams
of the Shoreham vessel and this compares the Shoreham to the
PVRUF. They found a lot more flaws in the Shoreham vessel
than they did the PVRUF. Both of those vessels were
fabricated by Combustion Engineering, but they were
fabricated in different timeframes.
There were no surface breaking flaws located in
either of the vessels so far to date, and they just started
doing the UT exam on River Bend.
Now, I'm going to go through some of the steps
that were involved with the expert judgment process to
determine the generalized flaw distribution. First of all,
the staff and the contractors, we discussed some different
issues that we felt needed to be addressed and information
that we wanted out of this expert panel.
We determined the level of complexity and what we
had decided, we had wanted information specifically on the
weldments, the base metal. We broke the base metal up into
two groups, the forgings and the plate material, and the
cladding. We identified an expert panel. We developed the
issues and we sent them to the panel for their review, to
see if they had any comments, if there were anything that we
were overlooking.
We had a panel meeting. This was our first --
I'll go over this more in detail a little later. And we had
elicitation training. Elicitation training is important
because during the individual elicitation sessions, you want
to eliminate as much bias as you can from each individual
expert. So we spent a day and a half going through
elicitation training with each of the experts.
DR. SHACK: You were looking at Prodigal for a
while, which is another expert judgment approach to the
characterizing flaws in weldments.
MS. JACKSON: Yes. Prodigal is actually a
simulation. They don't have -- we did put the PVRUF data
into a Prodigal simulation code and it came out, the results
were pretty similar to what we actually got from the data
from PVRUF. But the -- two of the people who are actually
on the expert panel for the Prodigal are on this expert
panel that we have for the flaw distribution.
And so far, we've elicited one of the experts so
far who was on Prodigal and we -- he had some interesting
comments, so we just need to talk with him a little more to
verify some of the issues that he stated during his
elicitation session.
DR. SHACK: What is the expert judgment supposed
to -- I mean, are they supposed to come up with a
hypothesized distribution? Prodigal sort of constructs a
distribution based on judgment. Are these guys supposed to
-- a beauty contest or what, five flaw distributions?
MS. JACKSON: What we've done, initially, we gave
them a list of issues to try to get them thinking along the
lines. We presented them the PVRUF data that PNL did and we
made a presentation on the Prodigal work that was done to
date.
What we want them to do is from their own expert
-- well, from their experience, each expert has individual
experience. Some of them were actually involved in the
fabrication process. Some of them did the NDE inspections
of the individual vessels. One particular expert provided
some of the welding material to the vessel fabricators.
So we want their own individual opinion from their
area of expertise on what we've done so far to date, if they
feel that's the correct path to go through to get the
generalized flaw distribution, and also if they think a
generalized flaw distribution can be developed, one flaw
distribution.
DR. APOSTOLAKIS: What is flaw distribution,
again?
MS. JACKSON: Excuse me?
DR. APOSTOLAKIS: A flaw distribution, what is it?
MS. JACKSON: It's the --
DR. APOSTOLAKIS: Probability distribution of
what?
MS. JACKSON: It's the measurement of the number
of flaws per cubic meter in the vessel material.
DR. APOSTOLAKIS: Independent of length or just
flaws?
MS. JACKSON: Just flaws, but the flaws have been
broken down into the different sizes. Some of them in the
inner 25 millimeters of the vessel and then the outer
vessel, those flaws that are in the weldment.
DR. APOSTOLAKIS: Now, the experts are going to
give you the whole distribution? I think that's what --
MS. JACKSON: No, they're not going to give us a
distribution. That's -- they're going to -- well, Lee will
go into a little bit more detail about that, because he's
going to go through as to how we go through the statistical
process to actually develop the flaw distribution.
DR. APOSTOLAKIS: Okay.
MS. JACKSON: Through these experts.
MR. HACKETT: Let me make a quick comment on that,
too, again, because Debbie touched on it. This is Ed
Hackett. One of the things, the key things that we're
looking for from the expert elicitation process is, is there
a generalized flaw distribution or is that some kind of
fantasy construct. Just speaking as a metallurgist myself,
I could say that there would be good reason to expert a
standard or generalized distribution for CE vessels that
were fabricated with submerged arc welding over some time
period.
Whether or not you can extrapolate that kind of
thing to cover all vessels that were manufactured in the
United States over the last 20 years and is there a
generalized flaw distribution, I think we know there are
some exceptions to that already, just based on the fact that
we know B&W used electroslag as a process.
It's not a multi-pass process. It's very, very
different from the other populations.
So there is a big question just in terms of is
there a generalized flaw distribution or do we have to get
more specific about it.
MS. JACKSON: Thanks. Yes, because of the varying
processes that they used for the different vessels, it may
-- we hope that we can get one distribution.
DR. APOSTOLAKIS: Who is your technical
facilitator in the group?
MS. JACKSON: The technical, Lee Abramson. The
TFI?
DR. APOSTOLAKIS: Yes.
MS. JACKSON: Lee Abramson is heading it, but it's
going to be a group of us who are going to be doing --
actually analyzing the results. There will be three to four
of us who will be doing that.
DR. APOSTOLAKIS: What kind of expertise will be
represented there?
MS. JACKSON: What type of expertise do this --
DR. APOSTOLAKIS: I know Lee's expertise.
MS. JACKSON: Lee's -- we're going to have
metallurgists, NDE experts, fracture mechanics, and the --
Lee, being the statistics expert.
DR. APOSTOLAKIS: The three NUREGs that you cited
earlier, are they using this concept of TFI? I know
NUREG-1150 did not.
MS. JACKSON: They didn't actually use the TFI,
because they have -- what I was looking for was what the
process they used in terms of getting their experts and how
they analyzed the data. There is another document that was
put out by ASME that -- this is more of a formal process
using the technical facilitator integrator and developing
the panel. It's a document that ASME has put out. I can't
think of the exact number right now.
DR. APOSTOLAKIS: A standard? Are you referring
to the PRA standard?
MS. JACKSON: No. I don't -- I'll have to get
back with you, but I used that document to get the format
for going through this process and discussions with Lee.
DR. POWERS: I had thought you did use the
technical facilitator. They didn't use the terminology.
DR. APOSTOLAKIS: They didn't really use the TFI.
The TFI -- I think NUREG-1150 tried to be more neutral. The
TFI, according to the original definition, to, in fact, put
things together if the experts disagree, according to his
judgment.
MS. JACKSON: Different documents --
DR. APOSTOLAKIS: Is this what you intend? 1150
didn't do that. 1150 elicited rates and processed them.
DR. POWERS: They made a decision on how they were
going to run things, but in those cases where they had
difficulties, and there were a couple that did have
difficulties, the equivalent of TFI --
DR. APOSTOLAKIS: It comes close.
DR. POWERS: -- made a judgment and they went with
it.
MS. JACKSON: Right. Some documents use different
terminology, but it's basically the point of the process
where you aggregate all the results from the experts.
DR. APOSTOLAKIS: That's a technical integrator.
MR. ABRAMSON: This is Lee Abramson. Perhaps I
could clarify that. Here, the TFI we're just referring to
is the team of people. I guess the NRC and maybe some of
our contractors who are going to pull everything together
and come up with the -- I guess, in effect, the input which
can be used for this generalized flaw distribution, based on
the expert panel elicitation, on the rationales and so on.
DR. APOSTOLAKIS: I understand that. Well, there
is a NUREG on the probabilistic seismic hazard analysis
which defines this thing and makes a distinction between a
technical integrator and a technical facilitator. So that's
why I'm pressing the point, because there is a difference.
MS. JACKSON: What was the number that you said,
again, please?
DR. APOSTOLAKIS: It's in NUREG report on
probabilistic seismic hazard.
MR. ABRAMSON: That's the Shack report, right?
MS. JACKSON: Okay.
MR. ABRAMSON: The Shack report.
DR. APOSTOLAKIS: Yes. And there is a distinction
between a TFI and a TI. And from what you are saying now,
you are really going to be technical integrators, more like
1150, with maybe some --
MR. ABRAMSON: That's probably correct. We may be
a little lose in the language here.
DR. APOSTOLAKIS: If you put the word facilitator
there, it means something specific.
MS. JACKSON: Okay. We'll remember that, because
that's something we've been using. Okay. The expert panel
that we put together, there are a total of 17 people on the
expert panel. We have people from the U.S. Navy, from
academia, EPRI, independent consultants, and retirees from
different organizations.
DR. APOSTOLAKIS: How many you have total?
MS. JACKSON: Seventeen.
DR. APOSTOLAKIS: Seventeen.
MS. JACKSON: This is areas of expertise of the
various experts. The construction code failure analysis,
fracture mechanics, metallurgy, NDE, reactor vessel
fabrication, reliability of flawed welding structures, and
actually welding.
We also have people who are involved with the
steel fabrication process for the vessels.
This is the schedule. These next two slides, I'm
going to go over the schedule. The items that have checks
on them are items that have been completed to date. These
two group -- these three items actually happened when we had
the Atlanta meeting. We had the first meeting of all of the
experts and Lee performed the elicitation training and we
discussed issues and we also have the elicitation team
identified.
We're going through the elicitation of the experts
right now. We've already completed the elicitation of four.
We're doing one elicitation tomorrow, one of the experts.
This process, where we're going to take all of the
elicitation data from the experts and integrate, that's
going to happen late this month and sometime in April.
We're going to have another meeting of the expert
panel, so that all of their responses and their rationales
can be reviewed. That will be done the first part of May.
The final responses and rationales will be put together in
the end of May and then we're going to have a workshop at
the end of June where we're going to present all of the
information from this expert judgment process, and that will
be the 27th and 28th of June here at the NRC.
The next two slides are going to have a list of
the issues that were presented to the experts to develop
conversation and so that they could get a general idea as to
what type of information we wanted from them.
From the PVRUF data, we haven't found any surface
breaking flaws, so we wanted to find particularly if anyone
knew of any existence of any surface breaking flaws and we
also have the two experts, one from -- who has information
from the UK Navy and the US Navy. So we have people outside
of the nuclear industry also.
This particular issue with Hatch, there is a flaw
that was fond in a nozzle region in the Hatch vessel and
after that was found, they had changed the inspection
methods for vessels at CE. They increased the inspection
process, so that resulted in additional weld repairs and
from the PVRUF data, we found out that a lot of the flaws
were found in the weld repairs.
And this particular event happened in the early
'70s and in the mid '70s, they said that maybe they were a
little bit too reactive and they were doing too many weld
repairs, so they back and changed the inspection process,
not to what it was before the Hatch incident, but it was so
that they wouldn't have to do so many weld repairs, because
the weld repairs were just increasing at an alarming rate.
This is just a brief summary of what went on
during the first expert panel meeting. The definition that
we came up with for flaw was an unintentional discontinuity
that had the potential to compromise vessel integrity.
That's what the definition of the flaw that's going to be
used through this process when we're eliciting the
individual experts.
DR. POWERS: Can I ask a couple questions? You
chose distributions which consist of density versus --
MS. JACKSON: The through wall extant.
DR. POWERS: And extant, right. Do you have
anything that you can show us on how you're handling
orientation?
MS. JACKSON: I don't have a backup slide with
that information, but I can give that to you. That is one
of the other presentations, the location and orientation of
the various flaws.
DR. POWERS: The other question is, in the
densities, is there any likelihood that the flaws are not
uniformly distributed within the local volume, but are, in
fact, clustered? And if you do, how do you handle that?
MS. JACKSON: Some of the flaws were clustered.
They used the ASME proximity rules to separate them, because
when we initially went through the NDE exam, some of them
did appear to be clustered.
DR. POWERS: They can separate them for the
measurement purposes, but now how do they transmit into the
rest of the process to say what's the probability that you
have a cluster of flaws in this particular piece of metal?
MR. HACKETT: I think I'll comment on that, also.
Ed Hackett. Dr. Powers raised this question earlier in the
day and it's a good question. The answer does basically
relate back to the ASME proximity rules, which are going to
take a series of flaws that are grouped together, as you
say, in some kind of cluster and then look at the dimensions
and the orientation and decide if those should be counted as
a single bounding flaw, which is then what you would feed
into the fracture mechanics.
So the short answer to it is that the ASME
proximity rules would be applied to any clusters and then
are there clusters, I think the answer is absolutely yes.
You certainly see a very large cluster of discontinuities,
as Debbie put it, at the clad-base metal interface with the
heat affected zone for the cladding, basically, which is an
expectation you would have from the metallurgy in this
situation.
So that's the short answer. The good news is
that, as Debbie pointed out, we're not seeing surface
breaking flaws and these discontinuities that we do see that
are clustered are generally inconsequential when it comes to
the single dominant flaw fracture mechanics type driving
force.
The ones that the clad-base metal interface, I
believe, in PVRUF, for instance, were largely of the two
millimeter type extent. A lot of them were also volumetric.
So a lot of those are just not participating in the -- in
contributing to the failure frequency of the vessel and the
probabilistic assessment.
DR. SHACK: But is that saying, in that size
distribution we're looking at, then some of those are
actually clustered, that they've decided to build together
based on the ASME rules?
MR. HACKETT: I'd have to go back and check that,
Bill. I'm not entirely sure. It should be. The answer to
that, if that's the case, they are clustered and they're
close enough, like you have this grouping of flaws that are
nominally two millimeters, but they're only a half a
millimeter apart, well, then, I think the ASME rules would
say, no, you better add those all up and count them and make
the -- they're close enough to the surface, you're also
going to have to count that as a surface breaking flaw.
So those things should be addressed as part of the
flaw distribution.
MS. JACKSON: Right.
MR. DIXON: I've got a couple of comments to try
to address your question. The question with regard to
orientation, flaws that reside in circumferential welds are
considered to be circumferential flaws. Flaws that reside
on axial welds and plate are assumed to be axial flaws.
So in the axis of the principal stress, to answer
your question, there is no sampling.
DR. POWERS: Okay. That's really the question.
MR. DIXON: There is no sampling with regard to
orientation.
DR. POWERS: Whatever the axis of the stress is.
MR. DIXON: Right. However, with regard to the
second question, Ed addressed the fact that putting together
the flaw size distributions, proximity rules are used, but
in the sampling, there is no proximity. The way the flaw
distributions are, it's something like this. The first 15
percent are postulated to reside in maybe the first
one-eighth of the wall thickness. The next 25 percent are
between one-eighth and three-eighths.
So the wall thickness are partitioned. So when
you are in the loop, if you want to call it a loop, of
placing flaws, you're going to first decide is it a category
one, two, in other words, in with partition does it exist.
Then the other assumption is that it has equal
probability of being at any location in that partition.
Does that address your question?
DR. POWERS: Maybe. Maybe I have to see exactly
-- go through the mechanics exactly. Let me see if I've got
it.
MR. DIXON: Okay.
DR. POWERS: You end up with a flaw distribution.
That has some big flaws in it.
MR. DIXON: Yes.
DR. POWERS: Okay. There is as fair probability
that the big flaws are in -- were, in fact, stemmed from
identifying a cluster of flaws that you added all together.
MR. DIXON: Yes.
DR. POWERS: You may not have ever seen a flaw
that big, but just saw a cluster of them that was
effectively that big. So now when you apply the
distribution in your analysis, you sample, as statistics
would dictate, from the whole distribution.
Sometimes you're putting in a big flaw which
corresponds to that part of the distribution that came from
both big flaws and from clusters that were effectively big
flaws.
So you don't actually say there's -- okay, there's
flaw, flaw, flaw, cluster of flaws, then flaw, flaw, flaw,
cluster of flaws.
MR. DIXON: No.
DR. POWERS: I think I understand what you're
doing.
MR. DIXON: Every flaw is treated independently.
DR. POWERS: Okay. It would, incidentally, be
useful for the benefit of mankind and possible future people
that want to go in and further improve in your work if you
did, in the documentation, keep track of clusters and their
distributions. Maybe not be part of your work, but the next
guy that comes along might be interested in what you found
there.
MR. ABRAMSON: I would like to describe how we're
going through the elicitation sessions. First, we're doing
this individually with each experts, each of the 17 experts.
And we have a team there and normative expert, I'm serving
as that, and then we have various subject matter experts
available, and also the recorder, and Debbie has generally
been doing that.
Then we present a list of characteristics to each expert,
and I'll have a detailed list of that in a moment, and then
we ask the experts to identify and discuss the pair-wise
interaction between the characteristics, and let me explain
what I mean by that.
We generally -- we start off the session by just
giving each expert a copy of this interaction matrix. Now,
here are the -- we have identified 14 what we call
characteristics, the product form, forgings, plate,
cladding, weldment, weld processes, form mechanisms, and so
on.
And these are just the headings. We have a very
detailed discussion of each one of these. Like for the form
mechanisms, there are any number of them, for example and so
on.
We say, all right, each flaw can be characterized
by each of these characteristics. Each flaw can be
characterized like in 14 ways or 14 dimensional flaw and it
has a particular product form and has a weld process that it
was formed by and it has -- the flaw has a particular
mechanism, et cetera, et cetera, et cetera. So each flaw is
unique in this point of view.
Now, what we ask them to do is we know that these
aren't necessarily -- that -- what we're going to be asking
them, in effect, is the likelihood that each one of these
will lead to a flaw of a particular size and we know that
there can be interactions between these.
For example, the welder skill could be very
important as to whether or not you have a flaw and that
could interact with the flaw mechanism, for example and so
on. The experts are going to tell us all this.
So we ask -- we go through this one by one,
basically each one of these characteristics and we ask them
to discuss any possible interactions with all of the others.
And, of course, we're recording all of this.
DR. APOSTOLAKIS: But, Lee, just to know that the
welder skill is important gives you half the picture. Don't
you have to know how skilled the actual welders were? I
mean, you're talking about the significance of each one of
these. How do you know that?
MR. ABRAMSON: Yes. Well, this is what we ask the
experts, whether they consider welder skill. I mean, all
the welders are qualified and so on. And so we talk about
the effect of the particular skill of a welder and whether
that might make a difference or not.
DR. APOSTOLAKIS: But, I mean, let's say that they
tell you yes it makes a difference. Now what do you do?
Wouldn't you have to decide --
MR. ABRAMSON: We're going to ask them -- I'm
going to tell -- I'm going to come to that in just a moment
as to how we're going to use this.
DR. APOSTOLAKIS: Okay.
MR. ABRAMSON: In effect, we're doing this --
there are no numbers. Eventually, we're going to have to
elicit some numbers in order to be able to get a
distribution, but here, this is all qualitative and what it
does is assess the stage, as I see it, it gives the experts
a chance to discuss how they view each one of these
characteristics and, in particular, they're going to focus
generally on their own areas of expertise.
And I ask them to talk about interactions. Again,
I think very useful material as far as the rationales for
everything like that. It kind of sets the stage. We don't
ask for any numbers at this point.
So this discussion goes on for maybe a half an
hour or longer, going through this matrix. And I think it
serves that useful purpose, also to get the experts oriented
into the mode of thinking that as to how each of these
characteristics might possibly affect the likelihood of a
flaw.
DR. APOSTOLAKIS: And that's a scale from one to
14?
MR. ABRAMSON: I'll come to that in just a moment.
DR. APOSTOLAKIS: So what are the columns?
MR. ABRAMSON: Pardon me? On, the columns. When
I say interactions, you have 14 characteristics and here are
14 columns.
DR. APOSTOLAKIS: You just put X's.
MR. ABRAMSON: You just put X's, that's right.
They put X's there. So that this -- as I said, this gives
the experts an opportunity to give us a benefit of their
experience, how they see these particular characteristics,
and to ring in how they see it affecting, in a qualitative
way, the various likelihood of a flaw.
All right. And then we get to, I guess literally
it will be the bottom line that we're going to need in order
to get the distribution, although we consider -- this is an
essential part of the process of getting these rationales
out in the open and we're going to report these back, as
Debbie indicated, to the experts and, of course, in the
final report.
So after we've gone through this discussion, we go
through the characteristics one at a time. For each one, we
ask the experts to identify that alternative with the
largest likelihood of leading to a flaw. Now, for each of
the characteristics, we have a number of alternatives, and
Debbie is going to talk about those.
For example, the weld processes, there's automatic
and unautomatic, we have a number of them, versus manual.
So these are the alternatives for the characteristics. So
we have these sub-categories. We have a number of these for
each one of them and we say, all right, which is the most
important, in your opinion, that's going to be number one.
And then what we do is we don't ask them for any
absolute numbers. We ask them for only relative numbers.
And we say compare each alternative with the highest ranked
alternative, how much less likely is it to create a flaw.
We get a factor, a factor of two, a factor of three,
whatever, ten percent less, 15 percent less and so on.
And we ask them for that number and, also, in
addition, we ask them for three numbers. First of all, I
ask them for high, mid and low value. The mid value is one
that's where they say their best guess, if you like, a 50/50
chance. And we went over all of this in detail when we did
the expert elicitation, what a mid value and a high value
are.
A high value is supposed to be a subjective 90
percentile -- excuse me -- 95 percent. So we say a high
value is such that you're almost sure that it's not going to
be higher than this. You've got about a five percent chance
roughly. And a low value, you're pretty sure it's not going
to be lower, it will be less than five percent.
So you've got the high value, which is 90 percent,
mid value is about the median, all subjective, of course,
low value is five percent, so the difference between the
high value and the low value is like a 90 percent confidence
level. So we ask all of this.
DR. APOSTOLAKIS: I don't understand that bullet,
frankly.
MR. ABRAMSON: Pardon me?
DR. APOSTOLAKIS: I understand the first two. So
you're comparing each alternative with the highest ranked
alternative.
MR. ABRAMSON: That's right.
DR. APOSTOLAKIS: What is the relative change in
likelihood? I don't understand that. What do you mean by
that?
MR. ABRAMSON: Okay.
DR. APOSTOLAKIS: Let's take the example on slide
25, the processes, you have automatic --
MR. ABRAMSON: Okay.
DR. APOSTOLAKIS: -- and then manual.
MR. ABRAMSON: Right.
DR. APOSTOLAKIS: So now somebody says the highest
ranked alternative is manual.
MR. ABRAMSON: Manual, right.
DR. APOSTOLAKIS: So now I compare the three
automatic alternatives to the manual.
MR. ABRAMSON: Exactly.
DR. APOSTOLAKIS: As you say, somebody says SMAW
is a factor of two less likely and so on. We've done all
that.
MR. ABRAMSON: Right.
DR. APOSTOLAKIS: What is the relative change in
likelihood of a flaw and how that plays into this?
MR. ABRAMSON: Okay. Let me say how we're going
to use this. You have no question about we're making the
relative -- you get the relative values. The question -- I
think what you're asking is, and that's, of course,
essential for this process, is how is all this going to be
used in order to get what we call a generalized flaw
distribution.
DR. APOSTOLAKIS: Because that's where we're
headed.
MR. ABRAMSON: That's where we're headed. Okay.
Let me tell you how this is going to be done.
DR. APOSTOLAKIS: It's not clear. I really don't
understand what you mean by assess relative change in
likelihood.
MR. ABRAMSON: All right. We're going to start
with the PVRUF distribution, because that's based on data.
That's the only thing we have, and we've got some hard --
we've got some numbers out of that and Debbie has gone over
that and you've heard presentations on that.
Now, the PVRUF flaws all have their
characteristics. It was a CE vessel, some of them are
automatic, some are manual, some are repaired and so on and
so forth.
Therefore, for every kind of flaw, for every kind
of flaw there, we can characterize it -- flaw size, and we
have the distribution, for every flaw size, we can
characterize the PVRUF data according to this 14
characteristics in the matrix.
And we have -- we know what the flaw distribution
is. We know what the likelihood, what the probability of a
getting a flaw of a particular size is. That's the data --
that's what the data gave us.
Now, we have another pressure vessel, with other
characteristics. Let's, for example, say one of the PVRUF
flaws was a manual weld. All right. Another pressure
vessel had an automatic weld. Now, the experts are telling
us that, say, an automatic weld is half as likely to have a
flaw of a particular size. So what we do then is we're
going to take that distribution and we're going to divide by
two.
DR. APOSTOLAKIS: So you are, in essence, adopting
the original distribution to the new vessel with the new
characteristics using input from the experts.
MR. ABRAMSON: Precisely, that's right. We have
this benchmarked distribution, it's a PVRUF, and then we
have all the relative comparisons.
DR. APOSTOLAKIS: So the experts never give you
absolute results.
MR. ABRAMSON: No.
DR. APOSTOLAKIS: It's always relative to the
original distribution.
MR. ABRAMSON: That's right. Frankly, I think
that this is -- it's fortunate that we have the PVRUF data,
because it's much harder to give absolute numbers than it is
to give relative numbers, especially when they have no basis
for it. They have no basis.
We're fortunate -- I mean, obviously, that's what
we did in the project to get this PVRUF data and we intend
to use this as an anchor in order to be able to get the
generalized distribution, with, of course, the uncertainties
and so on and so forth.
So that's the program and that's how we intend to
use this information.
DR. APOSTOLAKIS: Again, I don't understand the
inspector skill or the welder skill. How does that enter?
I understand the materials, the procedure, the weld
processes we just discussed, because they're more or less
objective. But when you come to welder skill, what does
that mean?
MR. ABRAMSON: Well, the experts have told us, of
course, that the particular skill of the welder can matter.
The problem is, of course, I think many of these welds are
-- well, I don't know if any records exist as to which
welders did which welds and what their skill level was and
so on and so forth.
Of course, we assume they're all qualified
welders. Recognizing that there could be some variability,
one way this could enter into it is to say, well, we may
want to try to put some kind of a fudge factor or an
uncertainty factor based -- let me back up a minute.
Let's say that the experts tell us that for a
particular kind of weld characteristics, welder skill is
important. Maybe it isn't, maybe it is, but let's say it
does. The particular kind of weld, the manual welds, it's a
very complex weld for repairs, for example, repairs. It's a
repaired weld and welder skill is important, but we don't
know what the welder skill is.
So what this tells us then is since we don't know,
that maybe what we should do is we should add some factor
for increasing the uncertainty in the effect, because
they're telling us that welder skill is important. We don't
know what a welder skill is, so this, in effect, would add
to the uncertainty on to the flaw distribution.
So that would be how we could use it, and, again,
we're going to be guided, of course, to a great extent by
what the experts are telling us and our own judgment of how
to incorporate this.
DR. APOSTOLAKIS: The last question has to do with
your 14 by 14 matrix. So you've explained now what the
third bullet meant, but you had the original distribution as
the reference point.
Now, if you had these correlations, how do you
handle adjusting the values?
MR. ABRAMSON: Again, we're going to do what the
experts tell us and we're asking them for a particular
product form, for example, what are the answers. What we're
doing is where it does matter with these interactions, we
elicit different values for these relative changes.
DR. APOSTOLAKIS: So is it possible then that you
say, well, look, welders kill is important and it's strongly
correlated with inspector skill?
MR. ABRAMSON: Yes.
DR. APOSTOLAKIS: So we're not going to count
inspector skill because we have already done the other one.
MR. ABRAMSON: That's right.
DR. APOSTOLAKIS: These are the kind of judgments.
MR. ABRAMSON: Exactly.
DR. APOSTOLAKIS: That I would have to make.
MR. ABRAMSON: That's right. Exactly. Now, we
recognize that some of these, like welder skill and
inspector skill, you're really not going to be able to get
any numbers for, but, again, what we're trying to do is to
identify all -- as Debbie said, all of the issues which
could be important and listen to what the experts are
telling us and to try to incorporate as much as possible.
DR. APOSTOLAKIS: The 14 by 14 matrix then
protects you against double-counting. That's really what it
does.
MR. ABRAMSON: Yes, that's right, I mean, assuming
that things are -- inspector skill and welder skill, that's
right, we're not doing it together, of course.
DR. APOSTOLAKIS: That's a clever idea.
MR. ABRAMSON: Yes.
DR. POWERS: I guess I didn't understand how you
handle the correlation.
MR. ABRAMSON: What we do is where there is a
significant correlation, we'll elicit different values from
the experts for each of those. For example, they tell us
that the difference between weldments and plate, so we'll
do, all right, first for weldments, what are your values for
this, then for plate, what are your values for this, and so
on.
So when we do this initial discussion with the
experts on the interactions with the 14 by 14 matrix, we
make a note of what's important and, of course, we don't
forget to come back to it and ask the experts say, yeah,
this is really important, we'll come back and we'll just
re-elicit it.
In effect, we're getting it conditional on what
they say are the important values.
DR. POWERS: I mean, I understand that you might
do plates and welds differently.
MR. ABRAMSON: Right.
DR. POWERS: But suppose you come back and you
say, gee, inspection procedure and inspector skill are
highly correlated. You use the worst possible procedure
with the worst possible inspector. They combine.
MR. ABRAMSON: Right.
DR. POWERS: Whereas by the time you get down the
best possible inspector, it's pretty much independent of
procedure. He does a good job no matter what procedure is
there.
MR. ABRAMSON: Right.
DR. POWERS: How do you recognize this?
MR. ABRAMSON: Well, we ask them about it. They
tell us this. We'll say, all right, what would it be for
this particular kind of -- assume, say, you've got a good
inspector and we're dealing with -- what was the
characteristic you were dealing with, with the procedure,
say, so say you've got a good inspector and you have a
procedure.
By the way, I should emphasize one thing which we
tell the experts right away going in. What we are
interested in is the flaw distribution as -- a pressure
vessel, as installed and ready to operate. This is after
it's gone through all the pre-service inspection. So this
isn't the flaw distribution that may have existed and then
was caught by inspectors and so on and so forth. So then
the question with inspector skill has to do with, well, are
there some things which might have escaped the inspector
because there weren't the skills.
DR. POWERS: You're going to clip this
distribution somehow?
MR. ABRAMSON: You mean truncate it?
DR. POWERS: Yes, because you're going to say
certain kinds of flaws get caught.
MR. ABRAMSON: Yes, absolutely. Absolutely.
DR. POWERS: And you're going to get some
assessment of the inspector's skill and that's going to
cause you -- for poor inspectors, you will clip less than
you will for good inspectors, and some procedures are better
than others.
What I'm asking is how do you decide when you've
got correlation between them? That is, you have a bad
inspector and a bad procedure. Does that -- how does that
change where you clip this distribution, truncate the
distribution?
MR. ABRAMSON: Well, let's say, all right, well,
you see, we would have to -- in order to be able to actually
apply this information about the quality of the inspections,
we would have to know for a particular pressure vessel
whether the inspector was good or bad.
DR. POWERS: We don't know that.
MR. ABRAMSON: We don't know, so we've got a
random sample of inspectors. So I think a way we would
handle that, and I mentioned it previously, is to increase
the variability and increase the uncertainty on what the
distribution is, because we don't know whether the inspector
was good, bad or indifferent. However, we do know that
depending upon his skill, you might have a different
distribution.
Well, the way to handle that would be you'd have
to have an uncertainty bound range of some sort on the
distribution.
DR. POWERS: I can see how you'd handle the
individual. Now what I'm asking is you've got both, you've
got to account for both the inspector and the procedure that
was adopted.
MR. ABRAMSON: I think we would know the procedure
from the records.
DR. POWERS: Go back to the records.
MR. ABRAMSON: Go back to the records when you try
to do that.
DR. POWERS: And if it turned out, lo and behold,
that you used the worst possible procedure you could, the
worst one you've ever heard of, you've already corrected the
distribution for the fact that you know that the inspectors
are of a random sample, some of them were bad and some of
them were good, whatnot. Now, what do you do with the
procedure? Is it just completely independent of the
inspector or do you add another fudge factor on top of it or
do you say no, bad inspectors, I've already added enough
fudge factor, I'll add no more, but for the good one, I
haven't added enough, so I have to add some.
MR. ABRAMSON: I think it will have to be a matter
of our judgment based on what the experts are telling us how
to interpret this. That's the best I can tell you. Each
one, in effect, each distribution is going to be custom
made.
DR. KRESS: You would have to ask the experts, if
I had a high-high or a high-medium or a high-low, you would
have six different things, you would have to ask them what
factor goes in to those. I don't see any other way you do
it. Wouldn't you have to -- you would have to have them
define the correlation for you.
DR. POWERS: You're going to have to know. It
could well be that good inspectors are doing a fantastic job
and it doesn't matter what procedure you use.
DR. KRESS: Absolutely.
DR. POWERS: And then bad inspectors do a bad job,
but it's a little bit better with a good procedure, but not
a lost worse with a really bad procedure. You've got to
know that information, somebody has got to tell you that.
DR. KRESS: And then they have to extrapolate this
to suppose you have a three-way correlation. You've got a
three-dimensional matrix you have to deal with.
DR. APOSTOLAKIS: The uncertainty is in the
result. That's probably overkill.
DR. POWERS: I don't know that it's overkill,
George. The problem is if you just go through and do it
randomly, you are going to put a tail on this distribution,
that when you're talking about things at
six-times-ten-to-the-minus-fifth amounts to a bunch. But
because it's correlated, you shouldn't have that tail.
It's the classic problem of dealing with the tails
of distributions, correlations count out there.
DR. APOSTOLAKIS: Sure.
DR. POWERS: They don't affect the means very much
at all, but they sure affect those tails
MR. ABRAMSON: Recognizing that, is that -- that's
why we emphasize these interactions when we're going to try
to -- not to double count or triple count or whatever we're
going to do, we recognize that.
DR. POWERS: Good.
MR. ABRAMSON: If there are no more questions, I
Debbie has a few final remarks to make.
DR. POWERS: It gets up to about 16,000 different
ways that you have to handle things.
DR. KRESS: Yes, I think so. That's asking a
little too much of the experts.
DR. POWERS: We've got really good experts. They
all come from Oak Ridge. They're great experts. We don't
want any of the Argonne guys coming to the expert
elicitation.
MS. JACKSON: These are from the discussion,
you've gone through these. One point I want to make in
terms of the inspection procedure, the inspection procedure
is a final inspection procedure after the vessel is fully
assembled, because the welding procedures themselves have
individual inspection procedures for different points.
So the inspection procedure that's listed in the
list of characteristics is the final inspection procedure.
So I'll just go to the --
DR. POWERS: This is after the cladding?
MS. JACKSON: Yes, after the cladding. After it's
ready to be --
DR. POWERS: Then we can throw that one away.
MS. JACKSON: So these are just some concluding
remarks that we've put together so far. The expert
elicitation process is complex, as well as the expert
judgment process, and we want to identify some significant
issues in the development of flaw distribution. We want to
address the combination of the relative effects of the
characteristics in the PVRUF distribution and that the flaw
distribution may vary by vessel fabricator.
Are there any other questions?
DR. SHACK: We'll know the answer by June.
MS. JACKSON: Yes.
DR. APOSTOLAKIS: Are we writing a letter this
time?
DR. POWERS: Can they ask 16,000 questions by
June?
MS. JACKSON: I'd like to get the title of that
NUREG that you mentioned, that you mentioned before, the
title of that NUREG.
DR. APOSTOLAKIS: Abramson knows. The Shack
report, she would like to have it.
MS. JACKSON: Are you familiar with that?
MR. ABRAMSON: Yes, I've got it.
DR. SHACK: What we'd like to propose is to come
back into session at quarter to one, since we're likely to
be a little pressed for time this afternoon.
[Whereupon, at 12:02 p.m., the meeting was
recessed, to reconvene at 2:45 p.m., this same day.]. A F T E R N O O N S E S S I O N
[12:45 p.m.]
DR. SHACK: I'd like to come back into session and
I guess we're going to have Mark Cunningham who is going to
give us the big picture.
MR. CUNNINGHAM: My nickel?
DR. SHACK: Your nickel.
MR. CUNNINGHAM: Good afternoon. My name is Mark
Cunningham. I'm in the PRA Branch in the Office of Nuclear
Regulatory Research.
I'm here this afternoon to give you kind of an overview of
where we're at and where we may be going in terms of
re-looking at the acceptance criterion that's established
for the PTS rule.
Basically, just as an overview, we have a deadline
in May of this year to provide a Commission paper describing
what changes or recommending potential changes the
acceptance criteria that are used in the PTS rule or a
recommendation maybe to leave it the way it is or whatever.
We wanted to take on this issue early on, because
if the policy decision took us in a certain direction, we
wanted to know that early enough in the process so that we
could adjust the rest of the program to accommodate it.
So basically what we'll have is that what I'm
going to do today is walk you through a number of items that
will be in that Commission paper or kind of the structure of
the Commission paper, talk about the acceptance criterion
itself as it currently is, talk about two issues of things
that have arisen since 1983 or whatever when the rule was
established in terms of guidance on use of PRA, and then
information on severe accident phenomenology, and then talk
about, at least introduce some potential revisions or ways
that we could change the acceptance criterion, talk a little
bit then about how we plan to finish up the paper over the
next couple of months, including coming back to the
committee perhaps in late April or May or something like
that.
At this point, we're not looking for a letter or
anything, but we may at the -- in the May timeframe.
You probably heard a great deal about this the
last couple of days, but the rule was established in 1983 as
an adequate protection rule, on contrast to some of the
other rules that we'll talk about later, like the station
blackout rule that were cost-beneficial safety enhancements.
So it was developed under different provisions of the
backfit rule.
The rule itself established an embrittlement
screening criterion that licensees had to evaluate their
plants against to determine whether or not they had adequate
safety margins in their vessel.
The acceptance criterion is in the form of a
frequency of a through wall crack. Basically, if you could
demonstrate that the frequency of that through wall crack
was less than five-times-ten-to-the-minus-six per year, then
you could continue to operate that plant.
If you went above that, then you had to
demonstrate that, through additional analyses or changes to
the vessel design or changes to how you're operating the
plant, to reduce the frequency down to acceptable level.
There's a couple of key underlying assumptions in
that five-times-ten-to-the-minus-six. Basically, you may
have heard about this today, but it's a
five-times-ten-to-the-minus-six of basically having a
certain no ductility temperature or whatever you call it,
the RTNDT or RTPTS.
From a risk standpoint, there's a couple of key
aspects to it. One is that if you talk about a through wall
crack, we made the presumption that the through wall cracks
equivalent to a large opening in the vessel and it's
equivalent to core damage, that you're not going to have a
capability once you start one of these through wall cracks
in a PTS accident to mitigate it in erms of preventing core
damage.
When the rule was established, there was an
argument made that the containment performance was not
particularly an issue in these accidents.
DR. KRESS: Is that assumption going to be
revisited there?
MR. CUNNINGHAM: Yes. I'll come back to that, but
that's one of the things that we need to think about. The
argument at the time was that the types of accidents that
get you into a PTS are accidents where there is a great deal
of water around, that you're over-pressurizing or
over-cooling the vessel. So you've got a lot of water in
the core, in the vessel.
You also have availability and presumably
operability of containment sprays. So the effects of that
was even if you opened up the vessel and weren't able to
cool the core, that you're not threatening the containment
itself, and depending on where we go in some of the
discussions of how we might re-look at the rule, what the
acceptance criterion that may or may not be an issue, but
we'll come back to that or I'll come back to that.
There's at least four key pieces of Commission
guidance that have been established since the rule was
established in the early '80s. You're well familiar with
these. We've got the safety goal policy statement. We
established two other rules that are similar in some
respects, the station blackout rule and ATWS rule dealing
with accidents that were identified in PRAs as being very
important to risk or core damage frequency at least.
The backfit rule became a little more codified and
well established and we -- in these timeframes and the
regulatory analysis guidelines that went with the backfit
rule that introduced risk information into the backfit rule
process in a particular way was also established.
Then just in the last couple of years, we've come
up with Reg Guide 1.174. So I'm going to talk about each of
these in a little more detail. As you know, the safety goal
policy statement defined qualitative and quantitative goals
for acceptable risk. That was in the 1986 statement.
Later on, in 1990, the Commission approved having
a ten-to-the-minus-four subsidiary core damage frequency
goal. That has an impact on defining what's an acceptable
overall core damage frequency and then that starts to impact
decisions on what could be an acceptable frequency of
particular initiators, and as we'll get to in a little bit,
it kind of reflects our thinking in the station blackout
rules and the ATWS rules in terms of what was an acceptable
frequency of having core damage accidents from those
initiators.
Again, it was intended for generic decisions using
industry average information, I think. So in one respect,
it's very relevant to the PTS rule in the sense that this is
a rule that -- it's a generic rule and that sort of thing.
So let's come back to some of the options that
deal with do we have the potential for using -- how do we
use the safety goal information in re-thinking the
acceptance criterion.
In the late '80s, we had two new rules
established, as I said, with the station blackout and the
ATWS rules were established as cost-beneficial safety
enhancements. So the staff had to argue why the benefit of
achieving these rules and what core damage frequency or risk
reduction we achieved was wroth the cost of implementation.
In both cases, there was a goal established of
ten-to-the-minus-five per reactor year. So in the sense,
this starts to lay out and says that we want to have -- even
if we have an overall core damage frequency goal of
ten-to-the-minus-four, we don't want to have any particular
initiator or group of accidents contributing more than about
ten percent.
DR. KRESS: That's a real significant item.
MR. CUNNINGHAM: Yes, and it comes back and when
we come back to some of the options, it kind of precludes, I
think, some options that we might have in terms of how you
would re-established or re-think the acceptable criterion
for the PTS rule.
DR. APOSTOLAKIS: How are these groups of
accidents defined?
MR. CUNNINGHAM: Not very precisely,
unfortunately.
DR. APOSTOLAKIS: I mean, the LOCAs, how do you
treat the LOCAs? As a group or small LOCA and the medium
LOCA?
MR. CUNNINGHAM: In this case, most of the station
blackout issue was a transient-initiated. So it could be --
it was basically any transient that would get you into a
situation of loss of off-site power and on-site power.
DR. APOSTOLAKIS: So it's specific for this.
MR. CUNNINGHAM: Yes, very specific for this, with
--
DR. BONACA: Would the LOCA in design basis, you
consider core damage?
MR. CUNNINGHAM: I'm sorry.
DR. BONACA: You can see the core damage also from
a LOCA that meets design basis, which is a limited amount of
fuel oxidation.
MR. CUNNINGHAM: In the context of these, those
would not be station blackouts that would have to meet the
goal of ten-to-the-minus-five.
DR. APOSTOLAKIS: But it is apportionment of risk
to certain categories of accidents.
MR. CUNNINGHAM: Okay.
DR. APOSTOLAKIS: I'm trying to understand what
you meant by core damage.
MR. CUNNINGHAM: Really core melt, if you will.
DR. APOSTOLAKIS: Core melt. Okay.
MR. CUNNINGHAM: Core melting.
DR. APOSTOLAKIS: If you add them all together,
you get where you want to be. All right.
MR. CUNNINGHAM: Yes. And just to be clear, there
is no Commission guidance that really says we're going to
allocate ten percent, there is not that -- we had talked at
one time ten or 15 years ago about the idea of reliability
allocation or risk allocation, but it wasn't formally
established for this. It was more general guidelines.
In fact, these rules were established a little
before the Commission formally approved the
ten-to-the-minus-four as an overall goal for acceptable
frequency, but it was always in people's minds of having
roughly those numbers, if you will.
The rules themselves, these two rules, were
justified basically on an off-site risk analysis. So at
this time and using -- when they were justified, you didn't
-- there was no specific guidance on containment
performance. So it was basically you've got this initiators
and the final decision metric, if you will, was averted
off-site population dose. So it was, to some degree,
irrelevant what specific containment performance -- how
containment performed in these accidents.
It could have been good or bad or whatever. It
was kind of -- the analysis was indifferent to that.
Then came up with the backfit rule and the
regulatory analysis guidelines. It has two parts to it, one
of which -- the first part is an initial screening on
potential reductions in CDF and conditional probability of
early containment failure. So at this point, we introduced
containment performance as a particular issue into the
backfit rule process.
One of the things we'll talk about a little bit
later is the idea of using the same type of information in a
reverse sort of way to justify potential increases. This is
focusing on what is the potential benefit of a proposed
change in terms of a reduction in core damage frequency and
a reduction in -- and an analysis and evaluation of
containment performance.
So if a proposed change did not gain you much in
terms of core damage frequency, then very often they were
just excluded and said you can't pursue the backfit with
those. If they passed that test and said, yeah, it might
have this substantial benefit, then you went on to look at
the off-site risk averted associated with the accident, but
this is the place where the backfit rule and the safety
goals started to come together in terms of using the safety
goals to define that initial screening.
Last, but not least, of course, is Reg Guide
1.174. It goes off and it has a little bit different flavor
to it. One is that it introduces a set of general
principals, as you know. We discussed them for many, many
times here. But the five principals that we talk about in
Reg Guide 1.174 are not explicitly laid out in some of this
other earlier guidance, like the backfit rule.
So when we come back to it, it has some advantages
in terms of how we would use -- might use some of this
guidance to look at the PTS rule. It introduces
probabilistic guidelines in terms of CDF goals and delta CDF
and LERF, so, again, it's a little different than what was
in the reg guide analysis guidelines.
It was conditional probabilities of containment
performance. Again, I think we're basically consistent in
terms of the numerics of it to show how changes in risk, in
this case, going up, might be consistent with the backfit
rule, which is intended to look at changes in the risk going
down.
As you may recall, when we talked about 1.174, one
of the goals was that we would allow increases in core
damage frequency, fairly small increases in core damage
frequency. One of the goals was that we don't, on the one
hand, allow core damage frequency to go up to a magnitude
where if we applied the backfit rule, we'd take them back to
where they were to begin with. So we wanted to avoid that
situation.
So that's some of the more recent guidance type
information. The other part of it is more recent work
that's been going on in accident phenomenology. As I said,
the rule itself was -- at the time of the rule, the staff
opinion or judgment was that there was not a strong
correlation between having a PTS event and containment
performance, that you were likely to keep the containment in
place.
Needless to say, in the last 15 years, there's
been a lot of work going on in trying to better understand
severe accident phenomena, not the least of which is
described, if you will, in 1150, and then a lot of work
that's been done since 1150 in trying to understand the
impacts of direct containment heating. There's probably a
lot of other things.
So part of what we're going to have to address is
depending on how we go on establishing the -- re-thinking
the acceptable criterion, we may have to bring -- re-think
the issue of containment performance. The question is, is
there anything that we have not learned in the last 15 years
that would run counter to what we decided 15 years ago, that
the containment performance was not much of an issue.
We've got -- the issues I've got at the bottom of
the slide here, we're going to think about what about the
dynamic loadings on the core and in the internals and the
vessel and the piping. Can you --
DR. KRESS: Is this the rocket ship?
MR. CUNNINGHAM: The rocket is part of it, but
it's also a question of tilting and that sort of thing, just
general motions of the vessel that -- one possibility is
that that can pull penetrations, that you move the piping
enough that you pull a penetration out.
DR. KRESS: Fail containment.
MR. CUNNINGHAM: Fail containment and then you've
got to decide is that a large -- could you have a large
release under those circumstances. Combined with some of
these other things.
DR. KRESS: Is it implicit in there the thinking
that at the bottom of the vessel, that you have no way to
get a lot of ECCS through the core? So that what you have
is a passageway for natural convection for air and you may
have air combustion to the team, which changes your hydrogen
thinking and your energy thinking and what goes into
containment. Is that part of this?
MR. CUNNINGHAM: I hadn't thought about that, but
yes, that belongs.
DR. KRESS: It's part of the thinking.
MR. CUNNINGHAM: Yes, that's right. That's a good
point. So the dynamics aspects at the time of the PTS
event. You're going to have some pressure loadings at that
point from the steam escaping and that sort of thing, but
again, it's a little different in the sense that you're --
the reason you're breaking this vessel is because you've got
a lot of water inside.
So that's a little different scenario.
DR. KRESS: When we use a large break LOCA, we
have this low-down calculation to get the loads, steam going
in. If you just suddenly break off the bottom --
MR. CUNNINGHAM: Yes.
DR. KRESS: -- of the primary vessel, I don't know
how you would redo the choke flow equation. You're going to
get a definition loading.
MR. CUNNINGHAM: That's right.
DR. KRESS: Versus timing.
MR. CUNNINGHAM: That's right and it would be
different, too, if you were to take the bottom head off or
having one of the axial welds go.
DR. KRESS: Yes.
MR. CUNNINGHAM: And open up that way. That's
right. So related to that is the -- are the loadings such
that you might tend to disperse the core. One possibility
is -- especially with a core that's kind of old, you might
be breaking it apart and things like that and what impacts
does that have. You're doing this before you would melt,
before you would expose it to air or anything like that.
So you have those sorts of things, and then you
come back to the question of what's the availability of your
containment sprays and things. This is not a scenario where
--
DR. KRESS: In the risk basis, you generally have
to assume some frequency or probability that they will be
failed.
MR. CUNNINGHAM: Yes, that's correct, but it's
different in character than, say, a station blackout, where
conditional probability of containment ESF failure is
essentially one. Here you've probably got them operational
and that is going to impact the phenomenology somehow.
DR. KRESS: The failure probability.
MR. CUNNINGHAM: Yes. That's right, if it's one
percent or something like that. You've got to bring all
these things together in some sort of way to sort out what
is -- how close -- what's our real estimation of the
containment performance and is it really any different than
what we thought about 15 years ago.
So we're trying to bring those two sets of new
information together into several potential revisions, if
you will.
One potential re-thinking of the acceptance
criteria is to focus more on the core damage frequency, and
that, in a sense, what we're talking about is bringing the
PTS rule into line with the blackout and the ATWS rules.
I'll come back to that in a minute.
Others are more focused, bring in the concept of
containment performance, as well. So they're a little more
modern in terms of our thinking about how you understand
accidents. One I have kind of alluded to earlier is you
might develop some sort of a reverse backfit process.
The second is you basically work from the Reg
Guide 1.174 guidelines, which are really oriented towards
changes, burden reduction changes, if you will, associated
with license amendments. Now, in effect, you're going to
apply that same set of principals and guidelines to a rule
change. So it has that difference in flavor, but it has the
same general concepts underlying it.
I am going to talk about all of those potential
revisions a little bit. And one idea is that you could
apply the goals for the ATWS rule and the station blackout
rule.
So one possibility is that you deal with and say
that the acceptable frequency in PTS is
ten-to-the-minus-five. So it's a little bit of a relaxation
of where we are today.
You would justify, if you will, and looked at the
rule in terms of off-site consequence risk instead of
containment performance, because that was the basis for
justifying the rules to the SBO and ATWS rules to begin
with.
So in one hand, it does establish some consistency
among these three rules. It would allow some increase, but
it doesn't introduce any particular -- no explicit
consideration of containment performance into it, and so, in
a sense, it's a little dated relative to our policies of
today.
So another option is to develop a reverse backfit
process, if you will. What we mean is basically you take
the reg analysis guidelines, which are used to justify
potential reductions in core damage frequency, and turn it
around and say, well, how can I develop some sort of mirror
to that which would allow me to justify increases in core
damage frequency.
DR. KRESS: Is it one over 2000?
MR. CUNNINGHAM: Something like one over 2000 or
some such thing. So you would have to do some sort of
cost-benefit analysis to say how much can we agree to allow
this to increase. There are several issues associated with
that, problems with that. One is that this is an adequate
protection rule.
So you're exploring very --
DR. KRESS: It's apples and oranges.
MR. CUNNINGHAM: That's right, and how you would
turn that into fruit salad or whatever is a little unclear
at this point as to what you would do in those areas.
So clearly there is a policy implication and
there's a lot of work that has to be done to sort that all
out.
Another approach then is to basically take the
principals from 1.174, which, again, were designed for
license amendment, changes, and apply it to a rule change.
It has the advantage that it ensures consistency, what we
think is the right -- is the most current, anyway, and the
best way of thinking about using -- making risk-informed
decisions.
DR. KRESS: How do you go from a backfit -- 1.174
was supposed to be tied to specific individual plants. You
now go to a rule which is supposed to cover all the
population. Do you divide those things by a hundred, those
CDFs and LERF?
MR. CUNNINGHAM: That's a good question. I think
what will happen is that the rule -- the application of the
rule is going to be a plant-specific basis. There are only
going to be a few plants --
DR. KRESS: You may just treat it with --
MR. CUNNINGHAM: Yes. And that's the way --
DR. KRESS: You're right, it would be
plant-specific.
MR. CUNNINGHAM: You set up the rule in some sort
of generic way, but it has to be applied on a plant-specific
basis. In reality, that's the way it's happening today with
the present rule, is that each plant has to evaluate their
vulnerability to the PTS and you'd have to have the same
thing here.
This has implications. If you're starting now
with a goal of five-times-ten-to-the-minus-six, the Reg
Guide 1.174 process would basically say you're probably not
going to let it get any bigger, much bigger than
five-times-ten-to-the-minus-six, but you bring in the LERG
consideration and if LERF is -- if containment performance
is not an issue, then you can end up with something like
five-times-ten-to-the-minus-six.
If containment performance is an issue, then you
could -- you may have to ratchet the
five-times-ten-to-the-minus-six down a little bit to deal --
to make it more in line with our LERF criterion in 1.174.
So in this one, one of the disadvantages of going
this way is that it introduces more explicitly the
consideration of LERF and that means we've got to nail down
some of these phenomenological issues a little bit better
than where we were, than where we are today.
So that kind of gives you an idea of where we are
on this paper right now. What we're doing is developing a
Commission paper. We'll be trying to have a draft the end
of this month that's basically going to look a lot like what
you've just seen here, with -- we want to go through and say
what was the basis for the original acceptance criterion,
what have we learned since then in terms of the Commission
guidance on PRA, and on accident phenomenology, look at some
potions for potential revisions, including this issue of
containment performance, and the one thing that would -- the
paper would have is it would have a recommendation on where
-- how to go on this.
What we would like to do is get the paper to you
sometime in the next month probably, with the idea -- let me
back up. We owe it to the Commission in early May. We
think some of these issues would be worthwhile talking to
the committee about. So maybe in late April or early May,
we would get the draft paper to you, or I guess it would
have to be late -- sometime mid to late April.
DR. KRESS: Sounds like a joint PRA and Severe
Accident subcommittee meeting.
MR. CUNNINGHAM: So that would be the idea.
DR. BONACA: Would you run something like this
through the generic issue program?
MR. CUNNINGHAM: I'm sorry?
DR. BONACA: Would you run something like this
through the generic issue program? This is a situation
where you have -- I mean --
MR. CUNNINGHAM: If the issue of PTS came up today
as a new issue and not be -- have a rule already and that
sort of thing, then you would -- one way to deal with it
would be to put it through the generic issue process and say
what's the value of pursuing a rule or some other regulatory
mechanism to deal with this.
DR. BONACA: You have a burden reduction issue
here, to some degree.
MR. CUNNINGHAM: It's a burden reduction issue,
yes, that's right. So the generic issue process is,
strictly speaking, not applicable here because we've got an
existing rule and we're talking about modifying it, because
we have a different set of processes for changing rules like
that.
The flavor of this one is a little different
because the rule itself started out as being probabilistic,
basically. So we have to re-think some of those aspects of
it, as well.
DR. APOSTOLAKIS: So you're proposing to have
another subcommittee meeting to discuss this or bring it
back before the committee?
DR. SHACK: We would have to have a full committee
meeting to write a letter.
DR. APOSTOLAKIS: Sure.
MR. CUNNINGHAM: Yes.
DR. KRESS: It's the sort of thing you might be
able to put it before the full committee. That is all we're
talking about.
MR. CUNNINGHAM: This is basically all we're
talking about and the key element --
DR. KRESS: We didn't have all the other parts of
the PTS in there, we're just talking about this right here.
MR. CUNNINGHAM: Yes. I think -- and we wouldn't
-- in the March-April paper, we wouldn't be proposing to
resolve the issues on the phenomenology. We just kind of
acknowledge them and say they have to be worked. The
principal difference between what we've seen here and the
paper would be some sort of recommendation on what's the
right fit of PRA guidance, if you will, for this and you may
have gotten some sense of where I'm coming from anyway on
this.
So it may be that a full committee meeting is all
that's needed.
DR. KRESS: That's a meaty issue, allocation of
risk among sequences.
DR. APOSTOLAKIS: The problem with a full
committee meeting is if we don't like it.
DR. KRESS: It might be better to --
DR. APOSTOLAKIS: It might be better to have --
DR. KRESS: -- subcommittee and a full committee.
DR. BONACA: I think so, too.
DR. APOSTOLAKIS: Yes, because --
DR. BONACA: One of the potential revisions you
mentioned is driven by consistency with -- among the three
principal risk-informed rules. This particular case, you
really have lost a vessel. You still have an ability of
cooling it through, I guess, injecting into the vessel and
draining and then -- or through the spray system.
MR. CUNNINGHAM: Yes.
DR. BONACA: How different is this kind of
scenario from what you had for the station blackout and ATWS
rules? In those cases, we have some fraction of scenarios
where you end up with a failed vessel, but others you don't
and you're able to cool long term. I just don't see this as
a -- I mean, if this is driven by consistency, I would say I
don't care about consistency there.
I have a situation here where I have to rely on
containment. So it seems to me that that would be driving
some. I guess this is all preliminary, so you don't have
any thoughts.
MR. CUNNINGHAM: The value of the consistency is
if somebody is looking out -- if somebody is looking in from
the outside to try to understand, well, what are you really
talking about in terms of trying to have acceptable core
damage frequency from your major rules, there is an
advantage to having them all kind of line up.
There are disadvantages. The nature of this rule
is different and I think part of the reason that the present
acceptance criterion is more restrictive than that for the
ATWS rule and the station blackout rule is the recognition
of the different character of this accident. Again, right
off the bat, you've compromised one of your barriers, but
you also seem to have -- at least relative to a blackout
rule, you have perhaps more confidence in the containment
performance than you would have had.
So it is a different beast. So I guess I would be
surprised if we go the route of saying, well, just for the
purpose of consistency, we're going to set up the rule to be
like the blackout and ATWS rules.
DR. BONACA: One other question I had was it seems
the main consequence of applying these new insights to --
it's really license renewal, allows a vessel to probably be
operable for a much longer period of time. By much, I mean
some longer period of time, but the question then becomes
are there other effects that are not really within just the
rule that now come together to -- I haven't thought about
this enough, but I'm saying that as you age these plants and
you allow the vessel to continue to be operable for a long
period of time, doesn't it open up other issues, other
questions regarding --
MR. CUNNINGHAM: I'm not sure offhand whether that
comes up or not. I haven't thought much about that aspect
of it.
DR. BONACA: I haven't either, but I just --
DR. KRESS: Another thought on your consistency
question. You talk about, say, the
one-times-ten-to-the-minus-five versus the
five-times-ten-to-the-minus-six. Both of those, I presume
it is some sort of representation of a mean value.
MR. CUNNINGHAM: Yes.
DR. KRESS: The ATWS rule -- the ATWS sequence has
certain sequence-specific uncertainty associated with it.
That's a lot different in the uncertainty associated with --
and that ought to fit into the system somewhere.
MR. CUNNINGHAM: That's right.
DR. KRESS: And that either means you lower the
mean value you're dealing with or you put some sort of
confidence level on it that's different than just the mean.
MR. CUNNINGHAM: Yes.
DR. KRESS: So somehow I wanted to get across that
that thinking needs to be into this acceptance criterion.
The sequence-specific uncertainties are different and should
be accounted for when you go to this acceptance criterion
some way.
DR. BONACA: Especially, and I completely agree
with you, Tom, especially in the case where you have burden
reduction. And so that becomes a very important issue to
understand what this ten-to-the-minus-five means.
*Mr. Foley. And the
five-times-ten-to-the-minus-six, maybe this has been gone
through in the last couple of days somehow, but there is a
-- one of the things the paper needs to do is explain the --
what's the -- it's five-times-ten-to-the-minus-six of what
and that's a through wall crack frequency, but it's also
tied to a particular RTPTS or RTNDT and that value was set
based on some conservative assessments of what was really
going to happen and that sort of thing, and all of that
needs to be laid out a little more carefully in the paper
and, in a sense, re-thought of how we would do -- how we
would address the uncertainties in the acceptance criterion
as we go forward.
So it's another piece that belongs in this paper.
DR. BONACA: And also just one last comment. We
talked about rigor this morning.
MR. CUNNINGHAM: I'm sorry?
DR. BONACA: WE talked about rigor in the
calculations. I think that because of what's happening
here, I mean, rigor is not any more a desirable thing and is
an expectation. We understand how this is derived and there
is rigor.
MR. CUNNINGHAM: Okay. If there's nothing else on
that.
DR. SHACK: Comments from the committee? Perhaps
we can then start with Nathan's presentation.
MR. CUNNINGHAM: Yes. We can move into a
discussion of how we're going to do some of the PRA
calculations that assess the performance of the plants.
DR. KRESS: I did want to say I think it's crucial
that you look very carefully at this question and whether
changes to containment failure probability impacts it.
MR. CUNNINGHAM: Yes. Okay. I'm going to stay
here. We've got three other folks who are going to join me
and do most of the work. Nathan Siu and Roy Woods from PRA
staff in the Office of Research and then Bill Galyean, who
is a contractor to us from Idaho National Engineering and
Environmental Laboratory.
MR. WOODS: As mark said, I'm Roy Woods. I'm from
Mark's branch, he's my branch chief, the Probabilistic Risk
Analysis Branch in our Office of Research.
With me at the table is Nathan Siu, on the far
side there, who is senior technical advisor in the PRA and
human reliability analysis parts of this PTS effort. Nathan
is also one of the driving forces behind the uncertainty
analysis for the entire PTS effort, including the thermal
hydraulics and the probabilistic fracture mechanics and the
PRA and HRA.
DR. POWERS: I can't help but say it's better to
have him back working on the fire risk assessment.
MR. WOODS: I'm pointed out he has several hats
and I've mentioned three or four of them right there.
DR. POWERS: He's got an important hat on most of
the time.
MR. WOODS: And I think Ali Mosleh, Professor
Mosleh, from University of Maryland, Materials and Nuclear
Engineering Department is here, back there somewhere. He is
heavily involved in the uncertainty analysis, also.
Also with me here is Bill Galyean from Idaho
National Engineering and Environmental Laboratory. He is
Research's contractor for the PRA and the PRA now includes
HRA. He doesn't have those contractors, but they're working
very closely together, as I will get to in a minute here.
Anyway, that's the work that his doing for us.
The objective of the PRA part of this, of the
whole project actually, is to support development of a
technical basis for revised pressurized thermal shock rule.
In doing that, we want to ensure that the overall process is
coherent and risk-informed and that there is a good
integration of the different aspects.
As I pointed out, I'm the leader of the PRA team
which now includes HRA. That, of course, identifies the
sequences and various errors that you would be worried about
and failures that you would be worried about.
That determines the sequences that we need to do,
the thermal hydraulics analyses for which I think David
Bessette talked about. He's the leader of that team. And
then the output of the thermal hydraulics analyses tells you
the input conditions for the probabilistic fracture
mechanics, which Shah Malik is the head of that team. So
those are basically the three teams.
Throughout all of these efforts, we are doing a
unified effort to take into account the uncertainties and we
are dividing them into aliatory and epistemic uncertainties,
which George wants, and it's a very good idea. That's what
we are trying to do here.
All of this is in support of the development of a
screening criteria which will probably be very much like the
type of screening criteria we have now at least, which is
based on the reference temperature for the nil ductility
transition, which is an embrittlement parameter, really.
In developing this, we will be looking at trying
to relate whatever criteria we have to risk figures of
merit; that is, through wall crack frequency or one of the
others that Mark referred to a few minutes ago.
Right now we are aiming it mostly toward through
wall crack frequency, which we are hoping to be able to
equate to a core damage and if that comes out acceptably,
then what might go after that wouldn't make any difference
in the conclusion, then we can stop there. That's where we
are kind of hoping we will at moment.
Also, as I mentioned, we are definitely doing
treatment of uncertainty, which will be related to the
qualitative issues; in other words, where you have a great
uncertainty is where you might want to maintain your
defense-in-depth to attempt to compensate for the
uncertainty that you have.
The way we're approaching this while thing is to
update the early 1980 PRA studies that we did. Those were
for Oconee, Calvert Cliffs, and H.B. Robinson. What we are
doing in updating these studies is reflecting changes to the
operation of the plant and changes to the hardware of the
plant. For example, emergency operating procedures have
changed a great deal since the early '80s.
They are now symptom oriented instead of event
oriented. An example of the changes to the plants
themselves, we are currently working on Oconee and they've
made significant changes to their integrated control system.
So we have to take those changes into account.
Those are just examples. We're looking at the
whole plant.
We also are reflecting changes to the PRA
state-of-the-art and the example I would use there is HRA,
human reliability analysis. We're using basically the
ATHENA team in this effort and the ATHENA team is meeting
with the PRA people. They are indistinguishable now, in my
mind. We sit down and we meet together and we talk about
what sequences are going to be modeled and what's going to
be in the sequences, both hardware and people oriented
things in those sequences.
DR. POWERS: What is it that you are looking for
to get from ATHENA that you wouldn't get from something like
THERP?
MR. WOODS: One of the things is errors of
commission, plainly. What might the operator -- when might
the operator be misled and think that he should do one
thing, when actually that's not what he should do in the
particular situation. He thinks he's in one place, but he's
actually in another place and he takes the right action for
where he thinks he is, but it's the wrong action for it,
that type of thing.
That can be very important. It can be a
significant contributor to the risk and that's not in there
now and we're trying to put that in there.
MR. SIU: The other thing I think they can say is
that we're going to have a more causally beast description
of why the error occurs, whether it's an omission or a
commission error, and it's going to reflect what's happening
during the sequence.
That's something that you can include in the THERP
analysis, but it's not tied in quite as explicitly, I would
say, as in what we're going to be doing.
MR. WOODS: And on the other side of the coin,
also, they're better able to look at recovery actions.
DR. POWERS: Both those things that you mentioned
there, the causality and the recovery, aren't those going to
get terribly plant-specific?
MR. WOODS: Yes. As are some of the other issues,
some of the hardware issues. We're finding -- in fact, I'll
get to that in a minute, where we talk about wrapping their
arms around the total population of plants from basically
four analyses.
That's a difficult issue because all of these
things are -- I mean, it's not unexpected, but it's turning
out the more we look at it, the more we realize how
plant-specific they are. That is a problem.
In fact, when I get to that, if you guys have any
good ideas on how to handle that, that's one place we'd
really appreciate input.
DR. POWERS: It raises the issue of how
representative are the plants that are being run through
this thing. How big of a sample set does it take. Have you
wrestled with that issue?
MR. WOODS: That's exactly the next point at the
bottom of this slide, address other plants. Let me get to
that now.
What we need to do is make sure that within the
scope of the analyses we do, we somehow include all plants
that have a significant PTS risk at the end of their
license, and we need to do this in a defendable manner. We
want to -- I guess what I'm trying to say is we end up with
four analyses and we might find that some plant that's not
among those four has a higher safety injection pressure or
safety injection flow capability or something.
So we need to somehow take that into account.
Now, this is assuming that that high capability exists in
the plant where there will be a significant embrittlement at
the end of the license. If there isn't, then for this
purpose, it's not of concern.
If you find such a plant, then we would have to
somehow also, in all fairness, take a look and see if there
is some other feature of that plant that might tend to
counter that. Maybe they have better whatever capability
somewhere else and take all that into account, but we have
to somehow do that without doing a full-blown PRA, because
we don't have the budget or the time to do a PRA for each
and every plant.
We're struggling with that. If there are any
constructive ideas, we'd welcome them.
DR. POWERS: They're mostly desperation ideas. I
can see how you can screen out based on embrittlement, there
are data that you could go to. You might even be able to
screen that out in the hardware because you can certainly
look at the FSAR.
But if indeed errors of commission are important,
screening based on procedures is a very tough thing to do,
because you have to read the procedures.
MR. WOODS: Right.
DR. POWERS: You have to get them, and that's an
enormous task.
MR. WOODS: That's exactly what we're in the
process of doing at Oconee right now. We were down there --
these three, Mark wasn't with us, but we were there
yesterday and the day before talking in some detail, well,
great detail actually, with everybody we wanted to talk to
at Oconee. They were cooperating quite well with us.
But the more we got into it, the more we realized,
hey, they have certain procedures, they approach these
problems in a certain way, and you can't assume that someone
else will. It's different and we're struggling with how to
handle that. You've hit on a very significant problem we're
facing.
MR. SIU: If I could add, Roy. I think there are
two parts of this screening which Mark pointed out in the
previous presentation. One is this initial screening
criteria, which is based on embrittlement, and what we need
to do is to be able to pick the embrittlement screening
criteria that gives us confidence that if the plant passes
that, there's just no problem, period.
Once you get past that point, then there will be a
plant-specific analysis that will demonstrate that the
particular risk criteria are satisfied. So at that point, I
imagine that's where your procedure issues are going to come
in and that's not something that we're going to perform.
Our main task is to set the embrittlement
criterion appropriately and to set the right level for the
second step.
MR. GALYEAN: Also, if I could add. We are
engaged in an effort right now to try and categorize plant
to plant differences that we feel are relevant to the PTS
issue, things like turbine bypass capacity, high pressure
injection capacity.
DR. POWERS: Things you can read about the plant.
MR. GALYEAN: Right. And our expectation is that
-- and, in fact, it is in the program plan that towards the
end, we are going to do sensitivity studies on the PRA
models to quantify what impact these plant to plant
differences could have on at least the frequency of these
PTS sequences.
DR. POWERS: Do you want to give me the risk
achievement worth of the operator? Nobody wants to do that
for me.
MR. WOODS: That leads right into this. We've
already covered a good deal of this slide, but basically
we're trying to calculate through all correct frequencies
from four plants, including uncertainties, that we're doing
PTS and PRA models for Oconee and Beaver Valley. The NRC is
-- or Bill Galyean, at INEL, with our sponsorship, are
developing those models.
Two other plants already include PTS sequences in
their PRA models and that's Calvert Cliffs and Palisades,
and we are planning on obtaining those models. Bill is
putting them in the SAPPHIRE code, so we can manipulate it
and change it and massage it and do sensitivity studies and
that sort of thing and use all four of those.
And what we'll end up -- then there's a
significant time at the end of this last point here, a
significant time after we develop those things and use them,
we realize we'll have four different models with four
different sets of assumptions and we're going to have to
somehow come to grips with how to put it all together in a
coherent way.
But what we'll end up with is four or more points
which each point from that graph behind me represents one
plant and what you do is you evaluate that -- you evaluate
that plant for its through wall crack frequency assuming
that the material condition is at an RTNDT which is
evaluated in a certain way as required by the PTS rule at
the end of that plant's license.
And by definition, that RTNDT is RTPTS, that's
just what we mean by that. And once we come up with an
acceptable through wall crack frequency, based on safety
goals or whatever, as Mark discussed, then that determines
the through wall crack frequency star on the vertical axis
and you could read across to some representation of those
points that you have and determine what the correlated RTPTS
is.
That would then be your screening limit. The
problem is, as you pointed out, Dr. Powers, that you've got
four points at most and you need to somehow come to grips
with how to handle the other plants.
That's the point of this slide, really. I'm
pressing the end here at this part.
Open questions, in addition to the ones that we've
talked about, at the moment, we're not treating internal
fires, floods, external events in these analyses. We
realize that the resulting failures, for example, for
internal fire that causes cables to burn and causes hot
shorts and causes various equipment to fail, which might
confuse the operator, this could involve the whole process,
that could cause PTS events to initiate or it could make
ones that have initiated for some other reason worse or
both.
DR. POWERS: Maybe we should stop all this and
just get into that fire problem right away.
MR. WOODS: Put Nathan's fire hat back on and keep
it on. I understand.
DR. POWERS: Take the resources from this, devote
them all to the risk assessment and validated models, sounds
good to me.
MR. WOODS: So you have several hats. You
probably understand why it's necessary to have several hats.
Anyway, we've already mentioned the problem with
coming to grips with the relationship between through wall
crack frequency. Well, maybe we haven't, but the problem is
when you go beyond through wall crack frequency and you're
trying to say that's not equal to core damage frequency,
what you're looking at is something that's very, very
uncertain and we're not sure that we can predict how big the
hole is and whether or not the core would actually be
damaged with enough certainty to actually take credit for
it, and that's the problem with going beyond through wall
crack frequency and just assuming that's equal to CDF.
Also, if you go from CDF to LERF, it's a similar
uncertainty. So as I said, really all we're doing at the
moment is we have a task in place to identify the various
issues that would be involved if we had to or wanted to, for
whatever reason, go beyond through wall crack frequency and
we're sort of keeping track of those, but we aren't spending
a lot of our resources on that at the moment.
DR. KRESS: On your previous slide, would you put
it back up?
MR. WOODS: Certainly.
DR. KRESS: I had a question. You implied that
the three points were three different plants.
MR. WOODS: That's correct, yes.
DR. KRESS: But the PTS, RTPTS is extrapolated out
to the end of current life.
MR. WOODS: The RTPTS for each plant would be the
RTPTS for that plant at its end of license, either extended
license or license now, if it hasn't applied for an
extension, or whatever problem you --
DR. KRESS: My point is there is a time involved
in there and you have to extrapolate something about the
fluences and so forth.
MR. WOODS: Yes.
DR. KRESS: Why can't you just continue that
extrapolation and have more than one point per plant and
define what this curve looks like for each plant? And isn't
it like having more data points to fit this curve?
MR. WOODS: No, it's not.
DR. KRESS: It's not.
MR. SIU: Again, don't take the graph too
seriously. This is just an example. One of the things
we're showing, for example, is a monotype relationship
between RTPTS and through wall crack frequency, and that may
not exist just because of the system differences or the
procedure definitions.
DR. KRESS: It would be monotonic for a plant.
MR. SIU: For a plant, that's right, and you could
plot --
DR. KRESS: That's why I was suggesting it.
MR. SIU: That's right. You could do that,
certainly. I think Mark Kirk had a comment.
MR. KIRK: The only thing I wanted to point out is
that RTPTS is, by definition, fluence, it is at end of
license fluence.
DR. KRESS: But maybe you could plot it versus
effect of full power year or something.
MR. WOODS: I was going to turn this over to Bill
Galyean now to give you some more details on the PRA and
with the incorporated HRA model that we're developing.
DR. POWERS: Having convinced us that the problem
is impossible.
MR. WOODS: That was not my intent.
MR. GALYEAN: I'm going to just -- I have these
three slides that I'm going to talk about just to give you a
feel for the general philosophy of the PRA analysis.
Afterwards, I will turn it over to Nathan and he
will get into more details on the uncertainty and the
integration aspects of the process.
As has been mentioned before, our intention and
our approach is to build on the original PTS PRA analyses.
We have the benefit of their results that we can allow us to
more cleverly develop the PRA models and develop the
accident sequence, the PTS accident sequences and evaluate
the importance of the various initiating events.
Again, as has been mentioned, we intend to update
these models in the analyses based on the current plant
designs, operating procedures, operating practices, and also
update on our current understanding of reliability for the
various systems, components, and also, in particular, the
initiating event frequencies.
So basically it's just an update both on the
state-of-the-art of PRA and -- and when I say PRA, I also
mean HRA. And also to update them based on the current
designs and operations of the plants we're looking at.
DR. POWERS: I'm just curious. In setting up and
deciding how you're going to update the PRAs and what not,
you had some basis for deciding you were going to do these
things, but you were going to leave out fire.
MR. GALYEAN: As was pointed out, external events
is still an open issue, and so the decision to leave out
fire has not yet been made. It's still being talked about.
We're still trying to understand what the implications are.
There was one event that occurred at Oconee, in fact, that
did result in some over-cooling. When I say one event, I
mean a fire in a switch gear.
And so we are certainly aware of that and aware of
the potential, but as far as how significant a contributor
external events are in comparison to all the other
initiating events, that's still something we're wrestling
with and still trying to decide what the -- whether it's
worthwhile to pursue that.
Again, that decision has not yet been made.
DR. POWERS: Are we ever going to get the IPEEE
insights document, report?
MR. CUNNINGHAM: Yes. The insights report is, on
the present schedule, I believe we're supposed to have a
draft this summer. That will not happen because we've had
-- we want to develop the insights report after we got the
reviews done and the reviews won't be done this summer, for
a variety of reasons, some of which are resource
limitations, some of which are related to fire issues that
we're dealing with with a number of utilities.
So I believe that realistically it will be early
next year -- late this year or early next year.
DR. KRESS: Couldn't you ask yourself whether any
fire events will activate the ECCS and sort of estimate the
effect on the frequency, initiating frequency?
MR. GALYEAN: Well, we do, in fact, the -- an
obvious area where we can improve on the original is the
initiating event frequency. We do have quite a bit of
operating experience data that we have collected and
analyzed through another program sponsored by the NRC and in
there we do have a frequency of inadvertent SI actuation,
for example. So theoretically, any contribution --
DR. KRESS: Due to fire.
MR. GALYEAN: Due to fire would be in there.
MR. SIU: I think it's fair to say that the tools
and techniques that we have now can be applied with the same
degree of certainty that we have with other core damage
scenarios associated with fire.
The problem is in the data-gathering, because the
concern actually with Oconee, this was non-safety switch
gear that was affected and you're talking about affecting
control systems on the balance of plant side. We don't
trace those cables.
MR. GALYEAN: This slide is intended to be more
illustrative of kind of the approach we're taking. It lists
the initiating events that we're looking at. It compares
the frequency from the original Oconee IPTS analysis and the
-- to the frequency that we anticipate using in the current
analysis.
The initiating event frequencies come from NUREG
CR-5750 initiating event frequency report, came out
recently. Also, in the last column, we just have some
comments or observations that we've concluded based on our
look at these various initiating events.
An obvious point of comparison are the top -- is
the top event, the reactor trip, turbine trip event, where,
in the original analysis, they assumed six events per year
and the current industry performance is less than one a
year.
Some of the others are not so different. But we
are also looking at a number of initiating events that were
not included in the original IPTS analysis. Also note that
we are looking at both at power events and events that occur
at essentially hot zero power, which because of the thermal
hydraulics response of the plant, could be more severe than
at power events.
The other obvious area for improvement over the
original analysis is in the HRA portion, which we've already
touched on. In the original analysis, it took a very
conservative and a very crude type of approach toward
quantifying human errors and we -- the state-of-the-art, I
think, the current state-of-the-art will allow us to
significantly improve over that application of -- that was
done in the original.
In particular, and, again, as mentioned, we will
be utilizing the ATHENA folks in the development of the
human reliability analysis and they will be looking at,
again, kind of a broader range of human interactions in the
response to a PTS type of transient.
That pretty much concludes my prepared comments.
If there are no questions on the PRA portion of this
analysis, I will turn it over to Nathan and he can talk
about the uncertainty and integration issues.
MR. SIU: Thanks. The issue of uncertainty has
come up a number of times in discussion here, so we just
wanted to talk briefly about what we're planning to do, what
we are doing, and I guess I will start off by saying that a
lot of this is discussed in the white paper, which I believe
was distributed to the committee, and I know it's a lot to
read there. But if you have any comments on it, by all
means, we'd appreciate them.
I think one of the main points to raise is this
framework diagram. It's kind of hard to see on the screen
there, but, again, it's in the paper and it's in the
handout. Basically, that shows how we go from the PRA event
sequence analysis, which identifies sequences at a certain
level of detail, such as you have an initiating event and
subsequent successes and failures of your safety systems.
Obviously each PRA sequence can represent a bundle
of thermal hydraulic sequences, actual realizations, because
of, for example, different timings of events within the
definition of the PRA sequence, CF sub-scenarios that have
to be analyzed.
One of our problems, of course, is deciding which
sub-scenarios to analyze to represent the PRA sequence.
Once we have identified those sequences and they
have associated frequencies, then you pass them on to the
probabilistic fracture mechanics analysis, which is
basically all the material embedded in the FAVOR code. In
fact, the FAVOR code takes a lot of this information and
does the integration. So we're talking on a conceptual
level rather than the level of what actually is going to be
done.
And if you're interested in the mechanics, we can
talk a bit about that a little bit later.
What I did want to point out here is that the PRA
analysis does identify sequence frequencies. There will be
sub-scenario frequencies associated with the thermal
hydraulics analysis and each of these frequencies, of
course, are uncertain. There will be uncertainty
quantified.
How we do that in PRA space is it's the standard
procedure, it's well known, and we can talk about that, if
you wish, but I was going to touch briefly on what we're
doing in thermal hydraulics and TFM, because that's
something I think that's certainly a little bit unusual for
the kinds of analysis that we usually perform.
I did want to point out also that in the PFM
analysis, you see this little -- these two distributions
overlapping. That's supposed to be a representation of
stress and strength. So basically what we're saying is that
some fraction of times that the vessel is hit with a
particular thermal hydraulic sub-scenario, some pressure or
temperature characteristic curves, it will fail.
But it's some fraction of time, it's not
necessarily one, it's not necessarily a zero.
Of course, we're uncertain about a lot of the
parameters that go in here, so there's a layer of
uncertainty that's not explicitly represented in this
diagram. That's what the note at the bottom of the diagram
indicates.
Regarding the probabilistic fracture mechanics
parameter and our treatment of uncertainty, the white paper
talked about what are the sources of uncertainty in the key
model parameters, the ones that we have been told are the
ones that seem to drive the results, and based on some
guiding principals as to how we're doing this modeling,
those uncertainties were characterized as being either
aliatory or epistemic.
DR. KRESS: I gather that wasn't as
straightforward as you might think.
MR. SIU: It's not -- neither -- well, I don't
know if it's straightforward. It is something that -- there
are modeling decisions being made as you go through this.
You have to decide what's your model of the world.
Professor Apostolakis' papers talk about this.
Once you fix on that model, then you can derive
what is -- how you would categorize each of these, but I
would say at this point, the paper is still being digested
by lots of folks and I'm sure we're going to get some ideas
as to maybe whether the categorization that's in the paper
is correct or not.
I think it's a pretty good stab at it, I'd like to
think that.
The aliatory uncertainties in this -- again, I'm
talking about the probabilistic fracture mechanics part, so
that's that third box in that diagram. I'm not talking
about the whole spectrum. But certainly you have
uncertainties there arising because of the uncertainties in
the thermal hydraulics scenario. So the frequency with
which you get hit with a particular scenario trace or at
least a scenario trace that represents a bin of thermal
hydraulics sub-scenarios.
And then there is this issue of conditional
failure of the vessel given a thermal hydraulic scenario,
and that's the point I was trying to raise through that
stress-strength diagram.
We are -- so that's -- we're addressing aliatory
uncertainties through those two mechanisms, through the
scenario frequencies and through the stress-strength model.
The epistemic uncertainties, we're just using
standard estimation techniques. You heard some discussion
this morning about such things as the copper, nickel content
at, let's say, a particular position in the reactor vessel.
The point about the correlation of parameters is obviously
an important one, and I don't know that we've looked into it
as carefully as we should yet.
But once we have characterized the uncertainties
and the propagation of these uncertainties through the
model, it is done in the FAVOR code, it's a standard Monte
Carlo propagation approach and I don't know that we need to
talk about that very much.
Again, FAVOR is the tool being used to assemble
all these results.
I'd say that we're a little further behind in our treatment
of thermal hydraulic uncertainties. The white paper, as you
have seen, is focused primarily on the issue of the
probabilistic fracture mechanics issues. But certainly we
have the same objective. We need to characterize and
quantify the uncertainties, in this case, in the thermal
hydraulics analyses.
Right now, we expect that whatever we do, that
characterization will be compatible with the current version
of FAVOR, which means basically we're talking about
deterministic pressure and temperature traces over time,
also the heat transfer coefficient of the downcomer, and
that the uncertainties in the thermal hydraulics scenarios
will be represented through uncertainties in the frequencies
of those scenarios, but we won't have bands of scenarios to
propagate through the code, because of just computational
limitations. We don't think we can do that.
The University of Maryland has the lead with this
work. Professor Ali Mosleh is sitting back there. He and
Professor Modarres are our PIs, and we've initiated planning
on how to actually do this work. This is, as many of you
know, not an easy task to look at the thermal hydraulic
uncertainties.
We have a cooperative research program with the
University of Maryland, and so they will address this issue
under that task.
The first part of that task will be to look
specifically at PTS issues and later on we expect that they
will broaden out and look at non-PTS applications and maybe
broaden the approach to go beyond just the assessment of
uncertainties in the thermal hydraulics scenario
frequencies.
We do believe right now that the approach will
involve a considerable amount of screening because of,
again, the computational resources that are required. We
have to get down pretty quickly to scenarios where it
appears that a detailed analysis is needed.
We hope to use both thermal hydraulic models and
probabilistic fracture mechanics models in that screening
process.
That's all I have to say about uncertainty
analysis. Again, we have some backup slides, we'd be
willing to chat with you about that, if you have any
questions.
DR. POWERS: The thing on that last slide that's
most striking is this rapid screening and if you're going to
use Monte Carlo methods, why do you care about screening
things out?
MR. SIU: Well, as you know, you can do Monte
Carlo in the crudest fashion. You would end up simulating
things that you really don't care about. So you could use
screening in the sense of important sampling, where you
focus your Monte Carlo analysis on those parts where it
really makes a difference.
What we're talking about is trying to eliminate
scenarios where there just doesn't look like there's going
to be any PTS challenge whatsoever. That's obviously the
first screen. Then you can, from a PRA standpoint, say,
well, this is possible, but it just is highly improbable and
because of the systems failures that you require and you
throw those out, as well, and the hope is, and obviously we
don't know that this hope will be realized until we do it,
is that we really can narrow down to a smaller number of
scenarios that are reasonably tractable.
We might have to develop some sort of simplified
thermal hydraulic representation to address uncertainties,
propagation of uncertainties, but, again, that's open to
question right now.
DR. POWERS: It's just that the screening is going
to be based on intuition and judgment.
MR. SIU: Yes, that's fair, and I will also say
that I think we're way better than where we were back then
in the '80s.
DR. SHACK: Could you explain a little bit more
about the notion that the thermal hydraulic, the
uncertainties is all in the frequencies and not in the time
traces?
MR. SIU: Roy, could you go back this diagram?
MR. WOODS: Sure.
MR. SIU: As a philosophical matter, I suppose you
could say that if you define the scenarios finely enough,
let's say that you know exactly when everything occurs and
if you're comfortable that you have a very robust model for
the system behavior, that most of the uncertainties would be
in just the specification of -- I don't know what
parameters, I'm certainly not an expert here, maybe Farouk
in the back might be able to help me out here.
But if you -- there are some parameters that,
let's say, your empirical coefficients in the heat transfer
correlation, we all know, you know those within plus or
minus 20 percent, at best.
Okay. But if you've nailed everything else down
and all you have to know is that particular coefficient, you
could say, well, there could be some uncertainty there, yes,
and then I could have a bundle of scenarios rather than a
single one.
What we're saying right now is hopefully we will
carefully define the scenarios such that we can get down to
that point where if we really are talking plus or minus 20
percent, it's not really a big issue compared to some of the
other things that we've got in the other parts of the model.
And one of the concerns is that we don't do an
overkill here if we have huge uncertainties in other parts
of the analysis.
But it's clearly an approximation, but doing it
this way. We've had some discussions with the thermal
hydraulics modelers and have some sense of feeling at this
point that a lot of the uncertainties have to do with the
input to their models and if that's the case, then I think
we know how to handle that.
MR. WOODS: As we pointed out, I guess this is
just summarizing. The development of the Oconee PTS PRA
model is going very well. The integrated PRA team, HRA
team, is developing a plant model and has visited the plant.
It's 2:00. They're still visiting the plant,
aren't they?
MR. GALYEAN: That's right.
MR. WOODS: We were there Tuesday and Wednesday of
this week, the three of us, plus three HRA people and --
well, anyway, you get the idea. There was a part of the
meeting regarding integrated control systems. The guy that
we needed to talk to was only available today. So they
stayed and the three of us had another important engagement.
So we left and left them there to do it.
DR. POWERS: And then you dropped by here, right?
MR. WOODS: I'm sorry?
DR. POWERS: Then you dropped by here.
MR. WOODS: Yes, right, then we dropped by here.
I think this is an accurate statement here, screening level
results are expected shortly. It depends. You don't want
to say shortly as this afternoon, but like toward the end of
this month, middle of next month, we expect to have some
idea where the through wall crack frequency -- no, no. I'm
sorry -- where the frequencies of some of these significant
sequences are. We will not have run it through the thermal
hydraulic analysis and we will not have run it through the
PFM calculations.
But we will begin to have PRA results, PRA/HRA
results at that point.
And that's the one we're working on now. We have
made some initial contacts with Beaver Valley. I think we
pointed out they are the ones that are going to step in for
the Westinghouse three-loop plant. We wanted a three-loop
plant because H.B. Robinson is a three-loop plant. We had
the previous analyses back in the mid '80s for H.B.
Robinson, so if you choose a similar plant, you know some of
that's applicable. The thermal hydraulics models are
applicable more so than they would be for a four-loop plant
or something.
So anyway, that's been initiated. We're getting
some requests to them. I guess I have to back up and say
for the Oconee people, that the cooperation has just been
excellent. If they had it or could imagine where it might
be or could dredge it out or call somebody in, then we had
it just as quickly as they could provide it.
So that really is going very well.
And the last item on this slide, uncertainty
analysis, I guess we just talked about that. There's not
much else to add. That's the presentation. Do you have
questions?
There can be several reasons for no questions.
Some of them are not complimentary and some of them are.
DR. POWERS: I know this committee pretty well.
They're all being complimentary right now. That was a very
nice presentation.
DR. KRESS: If we had criticisms that were severe,
we wouldn't be reluctant to say them.
MR. WOODS: I've seen that over the years maybe.
DR. KRESS: Actually, I think this looks pretty
good.
DR. POWERS: You can get back to some good fire
analysis.
DR. SHACK: I guess there is sort of one comment.
I look at that embedded analysis and it sort of looks like
it makes the whole problem go away. If I live with embedded
flaws, everything else goes away. Is this overkill? Have
you seen anything that indicates that you're unconservative
somewhere else?
So that if they produce the new flaw analysis, you
could declare victory. Dana wants it done completely, but
he wants it done quickly so you can get back to the fire
analysis.
DR. POWERS: But you have to do fire analysis to
do it completely.
MR. CUNNINGHAM: There could be a couple of places
where we're under-estimating, if you will, the frequencies.
One is the human element of it, the human performance
element of it. We're adding some different wrinkles to that
that we haven't done before. So that could change our
perspective on the frequencies to some of these challenges.
The other part comes back to the acceptance
criterion that I talked about, is that I wouldn't imagine
that it gets any less conservative, if you will, or higher
value of an acceptance criterion today.
Under some scenarios that I don't think are
probable, but under some scenarios, that could become
tighter. So it offsets, to some degree, some of the
benefits we get in the materials area.
I don't think that's a likely scenario, but I
think we need to nail that down. So there are at least a
couple of places where it could come into play, where other
features of the analysis could come into play to counteract
some of the benefits we're getting out of the materials
research.
MR. SIU: There are some other places, like
treatment of support systems, where, like Bill pointed out,
some new initiators that were not in the old studies and
might raise the numbers. Again, the hope is that it doesn't
raise them tremendously, but you don't know until you do it.
MR. DIXON: Also, Terry Dixon, from Oak Ridge. I
assume you're referring to the plants that Shah put up this
morning. Those analyses were done in 1998 based on the
PVRUF data and it's my understanding that the Shoreham data
is coming in with higher flaw densities than PVRUF. So
that's one thing that could be negative relative to the
analysis results that Shah put up this morning.
Also, the statistical distribution of the K-1-c
database, it was discussed this morning that the effect is
transient dependent, so who knows. So those are two
possibilities that could go counter to what you saw this
morning.
DR. SHACK: Just to finish up, does anybody have
any -- go around the table, if anybody wants to add any
comments.
DR. POWERS: Well, the probabilistic is going to
be looked at and we'll see what we get and the plan seems to
be fine. My biggest concern is that when people start
telling screening, I think of babies and bath water and
things like that, because the intuition just doesn't work.
We wouldn't go to PRA if our intuition was so good
on these things. But these are cautious people that have a
lot of expertise in doing this and so I have a great deal of
confidence in them.
We talked this morning about rigor in the
statistical analysis and whatnot and quite frankly, I really
didn't understand the rigor there. I think what they really
mean is they're doing a pretty careful job and to an
engineering detail and they don't really mean they're going
to go through a rigorous statistical analysis on this stuff
that would leave us all confused and befuddled. They're
doing things that are pretty obvious, is what I think
actually, and it looks very promising.
This is one of the really nifty research programs,
because it brings together three disciplines and a focused
attack that probably is a lot of fun to work on, actually,
because you probably learn a heck of a lot in the project
meetings.
So I guess I'm pretty positive on this, except for
the fact that it deters a really good fire safety analyst,
so he's not available to work on one of the really important
problems.
DR. BONACA: I can only say that I am favorably
impressed by the effort, by the comprehensiveness of all the
elements coming together. This is a very good example of a
lot of deterministic and probabilistic analysis coming
together.
The area where I have still questions, in my mind,
is regarding criteria that will be used to modify the rule.
That's really a much more, I guess, sensitive issue, because
of all the things we discussed before.
I'm sure that I recognize that you recognize that
it is a sensitive issue and I'll be very alert to how it's
being modified, because there is a lot of information coming
together here, but, again, this is a quite unique scenario
we're talking about, more different than most.
So that's where I have more questions.
DR. SHACK: George?
DR. APOSTOLAKIS: The presentation this afternoon
was fairly high level. I think the implementation is really
where difficulties will be. So I guess I'll form an opinion
then.
DR. POWERS: One aspect of it that was not pursued
other than just to bring it up that will be really
interesting to see what they do is the hot standby analyses
and how you approach those problems. That will be new and
different.
DR. KRESS: Yes.
DR. SHACK: I just basically thought the
presentations were very good. It seemed to me a very
comprehensive and interesting program. We're looking
forward to sort of seeing how it all plays out.
DR. KRESS: I frankly was very impressed. I think
this is can serve as a model program on how to risk-inform
regulation. I think it's very good. I'm quite glad to see
this very nice uncertainty incorporation in the process. I
think, as a follow-on to that, I think we need to really
think about how are we going to use those uncertainties in
the decision-making process, and I didn't really see that
come through.
Now, I think it has to do with the acceptance
criteria and I think acceptance criteria, to me, is a matter
of policy and it's something that really could impact this
whole thing as much as anything, because moving it just a
little bit one way or the other can make a big difference.
The other thing is I wasn't -- I had some minor
concerns about the expert elicitation process, but that may
just be my bit. I don't like expert elicitation. But I
recognize that there are some places where that's the only
way you can get the uncertainty and so you have to use it.
But I agree with Dana that you have to watch out
for correlations and there may be better ways to correlate
the mean versus -- or the variance versus the mean and what
they have, but those are minor issues.
I really think you have a good thing going here
and I urge you to continue with it. It's a good way to --
you've wrapped up all the data, you've got all the models
wrapped up, you've done an uncertainty analysis. I think
it's a complete package and that's really what I like about
it.
People can come back ten years from now and look
at your report and say, whoa, they'll know exactly what you
did and it will all be retrievable. It's good stuff, I
think. You guys can be proud of it.
DR. BONACA: Just one thing. In addition to that,
I would really -- I really enjoyed the documentation you
provided. I think the paper on uncertainty analysis was
very clear, helpful.
MR. SIU: Thank you.
DR. SHACK: Tom, are you ready to make a decision
on whether we need a subcommittee meeting on the Commission
paper topic?
DR. KRESS: I think we ought to have a
subcommittee meeting and then bring it to the full.
DR. SHACK: Rather than just a full committee
meeting.
DR. KRESS: Yes. And just on this part that Mark
talked about.
DR. APOSTOLAKIS: Half a day?
DR. KRESS: Half a day would be plenty, I think.
MR. DUDLEY: And we would be looking at a full
committee meeting in May.
DR. KRESS: I don't know what the timing was. I
think we'll have to --
DR. SHACK: Because of the way they plan to do it,
it almost has to be.
MR. CUNNINGHAM: We owe a Commission paper in May.
DR. KRESS: That would be a good time to do it.
MR. CUNNINGHAM: It would be good. I'm not
expecting that the Commission will have to make an immediate
decision on where we go on this, so I don't know that it --
if the letter happens in June versus May, that it will make
that much difference, quite frankly.
DR. SHACK: It sort of has to be in that
timeframe.
MR. CUNNINGHAM: Yes.
DR. POWERS: Do we have it in our future
activities list?
MR. DUDLEY: No, we don't. So this meeting was to
define what future meetings we would have.
DR. BONACA: That would mean a subcommittee meeting
next month.
MR. DUDLEY: That's correct.
MR. HACKETT: Just a point of clarification. This
is Ed Hackett. The full committee then, Noel, in June,
would address the entire project or are we looking at just
addressing the acceptance criterion?
MR. DUDLEY: Well, there would be one full
committee meeting in May to discuss the risk criteria and
then the expert elicitation would be heard either in June or
July, based on your progress.
MR. HACKETT: Okay. Thanks.
DR. POWERS: Yes. We don't want to schedule an
expert elicitation process until they're ready. I don't
want it cascading, June, and then next it's July, and then
it's September and October. Give yourselves some padding on
your schedule.
I assume experts are a little bit like herding
cats and you'll not go wrong.
MR. HACKETT: It's been tough. I got to say,
Debbie probably deserves some kind of award for what she's
been able to do so far, Debbie and Lee.
MR. CUNNINGHAM: Thank you very much.
DR. SHACK: We are adjourned.
[Whereupon, at 2:16 p.m., the meeting was
concluded.]
Page Last Reviewed/Updated Tuesday, July 12, 2016